Tomás J.

Aragón

Applied Epidemiology Using R

An Open Access Book

November 25, 2012

University of California, Berkeley School of Public Health,

and the San Francisco Department of Public Health

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

 For Tomás Martı́n, Luis Miguel, and Ángela Isabel

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Preface

I wrote this book to introduce R—a language and environment for statistical
computing and graphics—to epidemiologists and health data analysts con-
ducting epidemiologic studies. From my experience in public health practice,
sometimes even formally trained epidemiologists lack the breadth of analytic
skills required at health departments where resources are very limited. Re-
cent graduates come prepared with a solid foundation in epidemiological and
statistical concepts and principles and they are ready to run a multivariable
analysis (which is not a bad thing we are grateful for highly trained staff).
However, what is sometimes lacking is the practical knowledge, skills, and
abilities to collect and process data from multiple sources (e.g., Census data;
reportable diseases, death and birth registries) and to adequately implement
new methods they did not learn in school. One approach to implementing
new methods is to look for the “commands” among their favorite statistical
packages (or to buy a new software program). If the commands do not exist,
then the method may not be implemented. In a sense, they are looking for a
custom-made solution that makes their work quick and easy.
In contrast to custom-made tools or software packages, R is a suite of basic
tools for statistical programming, analysis, and graphics. One will not find a
“command” for a large number of analytic procedures one may want to exe-
cute. Instead, R is more like a set of high quality carpentry tools (hammer,
saw, nails, and measuring tape) for tackling an infinite number of analytic
problems, including those for which custom-made tools are not readily avail-
able or affordable. I like to think of R as a set of extensible tools to implement
one’s analysis plan, regardless of simplicity or complexity. With practice, one
not only learns to apply new methods, but one also develops a depth of un-
derstanding that sharpens one’s intuition and insight. With understanding
comes clarity, focused problem-solving, and confidence.
This book is divided into two parts. First, I cover how to process, manip-
ulate, and operate on data in R. Most books cover this material briefly or
leave it for an appendix. I decided to dedicate a significant amount of space
to this topic with the assumption that the average epidemiologist is not fa-

vii
viii Preface

miliar with R and a good grounding in the basics will make the later chapters
more understandable. Second, I cover basic epidemiology topics addressed in
most books but we infuse R to demonstrate concepts and to exercise your
intuition. Readers may notice a heavier emphasis on descriptive epidemiol-
ogy which is what is more commonly used at health departments, at least
as a first step. In this section we do cover regression methods and graphical
displays. I have also included “how to” chapters on a diversity of topics that
come up in public health. My goal is not to be comprehensive in each topic
but to demonstrate how R can be used to implement a diversity of methods
relevant to public health epidemiology and evidence-based practice.
To help us spread the word, this book is available on the World Wide Web
(http://www.medepi.com). I do not want financial or geographical barriers
to limit access to this material. I am only presenting what I have learned from
the generosity of others. My hope is that more and more epidemiologists will
embrace R for epidemiological applications, or at least, include it in their
toolbox.

Berkeley, California Tomás J. Aragón

November 25, 2012

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Acknowledgements

I would like to acknowledge the professors at University of California at Berke-

ley that not only taught me the principles and methods of epidemiology, bio-
statistics, and demography, but also introduced me to the S language as a tool
for exploring and analyzing epidemiologic data. More specifically, I owe spe-
cial thanks to Professor Steve Selvin, Chair of the Division of Biostatistics at
the School of Public Health, Professor Kenneth Wachter, Chair of the Depart-
ment of Demography, and Professor Arthur Reingold, Chair of the Division of
Epidemiology. Professor Selvin convinced me that the investment in learning
S would pay off in increased productivity and understanding. From Profes-
sor Wachter I learned the fundamentals of demography and how to use S
to program functions for demographic analysis. Professor Reingold recruited
me to direct the UC Berkeley Center for Infectious Diseases & Emergency
Readiness where I have had the opportunity and challenge to think about
how to make this material more accessible to public health epidemiologists.
Using the S language is so much fun! It becomes an extension of one’s
analytic mind. Thanks for getting me started and and for giving me the
opportunity to learn and teach!

Berkeley, California Tomás J. Aragón

ix
Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Contents

Part I Working with R

1 Getting Started With R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.1 What is R? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Who should learn R? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 Why should I learn R? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.3 Where can I get R? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 How do I use R? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Using R on our computer . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Does R have epidemiology programs? . . . . . . . . . . . . . . . 6
1.2.3 How should I use these notes? . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Just do it! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 Using R as your calculator . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.2 Useful R concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.3 Useful R functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.4 How do I get help? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.5 RStudio—An integrated development environment
for R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.6 Is there anything else that I need? . . . . . . . . . . . . . . . . . . 15
1.3.7 What’s ahead? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2 Working with R data objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.1 Data objects in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.1 Atomic vs. recursive data objects . . . . . . . . . . . . . . . . . . . 25
2.1.2 Assessing the structure of data objects . . . . . . . . . . . . . . 28
2.2 A vector is a collection of like elements . . . . . . . . . . . . . . . . . . . . 29
2.2.1 Understanding vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.2 Creating vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2.3 Naming vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2.4 Indexing vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

xi
xii Contents

2.2.5 Replacing vector elements (by indexing and assignment) 41

2.2.6 Operations on vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.3 A matrix is a 2-dimensional table of like elements . . . . . . . . . . . 48
2.3.1 Understanding matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3.2 Creating matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.3.3 Naming a matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.4 Indexing a matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.3.5 Replacing matrix elements . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.3.6 Operations on a matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.4 An array is a n-dimensional table of like elements . . . . . . . . . . 62
2.4.1 Understanding arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.4.2 Creating arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.4.3 Naming arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.4.4 Indexing arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.4.5 Replacing array elements . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.4.6 Operations on arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.5 A list is a collection of like or unlike data objects . . . . . . . . . . . 77
2.5.1 Understanding lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.5.2 Creating lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.5.3 Naming lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.5.4 Indexing lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.5.5 Replacing lists components . . . . . . . . . . . . . . . . . . . . . . . . 83
2.5.6 Operations on lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.6 A data frame is a list in a 2-dimensional tabular form . . . . . . . 85
2.6.1 Understanding data frames and factors . . . . . . . . . . . . . . 85
2.6.2 Creating data frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
2.6.3 Naming data frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.6.4 Indexing data frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.6.5 Replacing data frame components . . . . . . . . . . . . . . . . . . 94
2.6.6 Operations on data frames . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.7 Managing data objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
2.8 Managing our workspace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

3 Managing epidemiologic data in R . . . . . . . . . . . . . . . . . . . . . . . . 107

3.1 Entering and importing data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.1.1 Entering data at the command prompt . . . . . . . . . . . . . . 107
3.1.2 Importing data from a file . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.1.3 Importing data using a URL . . . . . . . . . . . . . . . . . . . . . . . 118
3.2 Editing data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3.2.1 Text editor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3.2.2 The data.entry, edit, or fix functions . . . . . . . . . . . . 119
3.2.3 Vectorized approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
3.2.4 Text processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3.3 Sorting data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Contents xiii

3.4 Indexing (subsetting) data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

3.4.1 Indexing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.4.2 Subsetting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.5 Transforming data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.5.1 Numerical transformation . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.5.2 Creating categorical variables (factors) . . . . . . . . . . . . . . 130
3.5.3 “Re-coding” levels of a categorical variable . . . . . . . . . . 132
3.5.4 Use factors instead of dummy variables . . . . . . . . . . . . . . 135
3.5.5 Conditionally transforming the elements of a vector . . 136
3.6 Merging data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
3.7 Executing commands from, and directing output to, a file . . . 140
3.7.1 The source function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
3.7.2 The sink and capture.output functions . . . . . . . . . . . . 141
3.8 Working with missing and “not available” values . . . . . . . . . . . 143
3.8.1 Testing, indexing, replacing, and recoding . . . . . . . . . . . 144
3.8.2 Importing missing values with the read.table function146
3.8.3 Working with NA values in data frames and factors . . . 146
3.8.4 Viewing number of missing values in tables . . . . . . . . . . 148
3.8.5 Setting default NA behaviors in statistical models . . . . 150
3.8.6 Working with finite, infinite, and NaN numbers . . . . . . 150
3.9 Working with dates and times . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
3.9.1 Date functions in the base package . . . . . . . . . . . . . . . . . 152
3.9.2 Date functions in the chron and survival packages . . 159
3.10 Exporting data objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
3.10.1 Exporting to a generic ASCII text file . . . . . . . . . . . . . . . 159
3.10.2 Exporting to R ASCII text file . . . . . . . . . . . . . . . . . . . . . 162
3.10.3 Exporting to R binary file . . . . . . . . . . . . . . . . . . . . . . . . . 164
3.10.4 Exporting to non-R ASCII text and binary files . . . . . . 165
3.11 Working with regular expressions . . . . . . . . . . . . . . . . . . . . . . . . . 165
3.11.1 Single characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
3.11.2 Character class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
3.11.3 Concatenation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
3.11.4 Repetition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
3.11.5 Alternation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
3.11.6 Repetition > Concatenation > Alternation . . . . . . . . . . 171
3.11.7 Metacharacters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
3.11.8 Other regular expression functions . . . . . . . . . . . . . . . . . . 173
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

Part II Applied Epidemiology

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

xiv Contents

4 Analyzing simple epidemiologic data . . . . . . . . . . . . . . . . . . . . . . 181

4.1 Overview and concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.1.1 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
4.1.2 Confidence intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
4.1.3 Hypothesis testing and p-values . . . . . . . . . . . . . . . . . . . . 190
4.1.4 Power and Type II error . . . . . . . . . . . . . . . . . . . . . . . . . . 193
4.1.5 Sample size calculations for analytic studies . . . . . . . . . . 194
4.2 Evaluating a single measure of occurrence . . . . . . . . . . . . . . . . . 194
4.2.1 Poisson count (incidence) and rate data . . . . . . . . . . . . . 194
4.2.2 Binomial risk and prevalence data . . . . . . . . . . . . . . . . . . 198
4.3 Evaluating two measures of occurrence . . . . . . . . . . . . . . . . . . . . 200
4.3.1 Comparing two rate estimates: Rate ratio . . . . . . . . . . . 200
4.3.2 Comparing two risk estimates: Risk ratio and disease
odds ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
4.3.3 Comparing two odds estimates from Case-Control
data: Exposure odds ratio (EOR) . . . . . . . . . . . . . . . . . . . 205
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

5 Graphical display of epidemiologic data . . . . . . . . . . . . . . . . . . . 207

5.1 Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
5.1.1 Arithmetic-scale line graphs . . . . . . . . . . . . . . . . . . . . . . . 207
5.1.2 Semi-logarithmic-scale line graphs . . . . . . . . . . . . . . . . . . 211
5.1.3 Histograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
5.1.4 Frequency polygons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.1.5 Cumulative frequency and survival curves . . . . . . . . . . . 213
5.1.6 Scatter diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.2 Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.2.1 Bar charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.2.2 Groups bar charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.2.3 Deviation bar charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.2.4 Proportional component bar charts . . . . . . . . . . . . . . . . . 213
5.2.5 Pie charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.2.6 Dot plots and box plots . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.2.7 Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.3 Color selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.4 Miscellaneous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213

6 Control confounding with stratification methods . . . . . . . . . . 215

6.1 Pooling methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
6.1.1 Cohort studies with risk (binomial) data . . . . . . . . . . . . 215
6.1.2 Cohort studies with incidence rate data . . . . . . . . . . . . . 216
6.1.3 Case control studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
6.2 Standardization methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
6.2.1 Direct standardization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
6.2.2 Indirect standardization . . . . . . . . . . . . . . . . . . . . . . . . . . . 218

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Contents xv

7 Control confounding with regression methods . . . . . . . . . . . . . 219

7.0.3 Interpreting the regression coefficients . . . . . . . . . . . . . . . 219
7.1 Linear regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
7.2 Logistic regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
7.2.1 Unconditional logistic regression . . . . . . . . . . . . . . . . . . . . 226
7.2.2 Conditional logistic regression . . . . . . . . . . . . . . . . . . . . . . 231
7.3 Poisson regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
7.4 Indirect standardization using Poisson regression . . . . . . . . . . . 234
7.5 Cox proportional hazards regression . . . . . . . . . . . . . . . . . . . . . . 239

A Available data sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243

A.1 Latina Mothers and their Newborn . . . . . . . . . . . . . . . . . . . . . . . 243
A.2 Oswego County (outbreak) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
A.3 Western Collaborative Group Study (cohort) . . . . . . . . . . . . . . . 244
A.4 Evans County (cohort) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
A.5 Myocardial infarction case-control study . . . . . . . . . . . . . . . . . . . 247
A.6 AIDS surveillance cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
A.7 Hepatitis B surveillance cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
A.8 Measles surveillance cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
A.9 University Group Diabetes Program . . . . . . . . . . . . . . . . . . . . . . 248
A.10 Novel influenza A (H1N1) pandemic . . . . . . . . . . . . . . . . . . . . . . 248
A.10.1 United States reported cases and deaths as of July
23, 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248

B Outbreak analysis template in R . . . . . . . . . . . . . . . . . . . . . . . . . . 249

C Programming and creating R functions . . . . . . . . . . . . . . . . . . . 257

C.1 Basic programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
C.2 Intermediate programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
C.2.1 Control statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
C.2.2 Vectorized approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
C.2.3 Looping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
C.3 Writing R functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
C.3.1 Arguments default values . . . . . . . . . . . . . . . . . . . . . . . . . . 266
C.3.2 Passing optional arguments using the ... function . . . . 268
C.4 Advanced topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
C.4.1 Lexical scoping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269

Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

 Part I
Working with R

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Chapter 1
Getting Started With R

1.1 What is R?

R is a freely available “computational language and environment for data

analysis and graphics.” R is indispensable for anyone that uses and interprets
data. As medical, public health, and research epidemiologists, we use R in
the following ways:

• Full-function calculator
• Extensible statistical package
• High-quality graphics tool
• Multi-use programming language

We use R to explore, analyze, and understand epidemiological data. We

analyze data straight out of tables provided in reports or articles as well
as analyze usual data sets. The data might be a large, individual-level data
set imported from another source (e.g., cancer registry); an imported matrix
of group-level data (e.g, population estimates or projections); or some data
extracted from a journal article we are reviewing. The ability to quantitatively
express, graphically explore, and describe epidemiologic data and processes
enables one to work and strengthen one’s epidemiologic intuition.
In fact, we only use a very small fraction of the R package. For those who
develop an interest or have a need, R also has many of the statistical modeling
tools used by epidemiologists and statisticians, including logistic and Poisson
regression, and Cox proportional hazard models. However, for many of these
routine statistical models, almost any package will suffice (SAS, Stata, SPSS,
etc.). The real advantage of R is the ability to easily manipulate, explore,
and graphically display data. Repetitive analytic tasks can be automated
or streamlined with the creation of simple functions (programs that execute
specific tasks). The initial learning curve is steep, but in the long run one is
able to conduct analyses that would otherwise require a tremendous amounts
of programming and time.

3
4 1 Getting Started With R

Some may find R challenging to learn if they are not familiar with statisti-
cal programming. R was created by statistical programmers and is more often
used by analysts comfortable with matrix algebra and programming. How-
ever, even for those unfamiliar with matrix algebra, there are many analyses
one can accomplish in R without using any advanced mathematics, which
would be difficult in other programs. The ability to easily manipulate data in
R will allow one to conduct good descriptive epidemiology, life table methods,
graphical displays, and exploration of epidemiologic concepts. R allows one
to work with data in any way they come.

1.1.1 Who should learn R?

Anyone that uses a calculator or spreadsheet, or analyzes numerical data at

least weekly should seriously consider learning and using R. This includes
epidemiologists, statisticians, physician researchers, engineers, and faculty
and students of mathematics and science courses, to name just a few. We
jokingly tell our staff analysts that once they learn R they will never use a
spreadsheet program again (well almost never!).

1.1.2 Why should I learn R?

To implement numerical methods we need a computational tool. On one end

of the spectrum are calculators and spreadsheets for simple calculations, and
on the other end of the spectrum are specialized computer programs for such
things as statistical and mathematical modeling. However, many numerical
problems are not easily handled by these approaches. Calculators, and even
spreadsheets, are too inefficient and cumbersome for numerical calculations
whose scope and scale change frequently. Statistical packages are usually
tailored for the statistical analysis of data sets and often lack an intuitive,
extensible, open source programming language for tackling new problems
efficiently.1 R can do the simplest and the most complex analysis efficiently
and effectively.
When we learn and use R regularly we will save significant amounts of time
and money. It’s powerful and it’s free! It’s a complete environment for data
analysis and graphics. Its straightforward programming language facilitates
the development of functions to extend and improve the efficiency of our
analyses.

1 Read my recommendations for mostly free and open source software (FOSS) at medepi.
com.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

1.2 How do I use R? 5

1.1.3 Where can I get R?

R is available for many computer platforms, including Mac OS, Linux, Mi-
crosoft Windows, and others. R comes as source code or a binary file. Source
code needs to be compiled into an executable program for our computer.
Those not familiar with compiling source code (and that’s most of us) just
install the binary program. We assume most readers will be using R in the
Mac OS or MS Windows environment. Listed here are useful R links:

• R Project home page at http://www.r-project.org/;

• R download page at http://cran.r-project.org;
• Numerous free tutorials are at http://cran.r-project.org/other-docs.
html;
• R Wikibook at http://en.wikibooks.org/wiki/R_Programming; and
• R Journal at http://journal.r-project.org/.

To install R for Windows, do the the following:

1. Go to http://www.r-project.org/;
2. From the left menu list, click on the “CRAN” (Comprehensive R Archive
Network) link;
3. Select a nearby geographic site (e.g., http://cran.cnr.berkeley.edu/);
4. Select appropriate operating system;
5. Select on “base” link;
6. For Windows, save R-X.X.X-win32.exe to the computer; and for Mac OS,
save the R-X.X.X-mini.dmg disk image.
7. Run the installation program and accept the default installation options.
That’s it!

1.2 How do I use R?

1.2.1 Using R on our computer

Use R by typing commands at the R command line prompt (>) and pressing
Enter on our keyboard. This is how to use R interactively. Alternatively, from
the R command line, we can also execute a list of R commands that we have
saved in a text file (more on this later). Here is an example of using R as a
calculator:

> 8*4
[1] 32
> (4 + 6)^3
[1] 1000

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

6 1 Getting Started With R

Use the scan function to enter data at the command line. At the end of each
line press the Enter key to move to the next line. Pressing the Enter key
twice completes the data entry.
> quantity <- scan()
1: 34 56 22
4:
Read 3 items
> price <- scan()
1: 19.95 14.95 10.99
4:
Read 3 items
> total <- quantity*price
> cbind(quantity, price, total)
quantity price total
[1,] 34 19.95 678.30
[2,] 56 14.95 837.20
[3,] 22 10.99 241.78

1.2.2 Does R have epidemiology programs?

The default installation of R does not have packages that specifically imple-
ment epidemiologic applications; however, many of the statistical tools that
epidemiologists use are readily available, including statistical models such
as unconditional logistic regression, conditional logistic regression, Poisson
regression, Cox proportional hazards regression, and much more.
To meet the specific needs of public health epidemiologists and health
data analysts, I maintain a freely available suite of Epidemiology Tools: the
epitools R package can be directly installed from within R.
For example, to evaluate the association of consuming jello with a diarrheal
illness after attending a church supper in 1940, we can use the epitab function
from the epitools package. In the R code that follows, the # symbol precedes
comments that are not evaluated by R.

> library(epitools) #load ’epitools’ package

> data(oswego) #load Oswego dataset
> attach(oswego) #attach dataset
> round(epitab(jello, ill, method = "riskratio")$tab, 3)
Outcome
Predictor N p0 Y p1 riskratio lower upper p.value
N 22 0.423 30 0.577 1.000 NA NA NA
Y 7 0.304 16 0.696 1.206 0.844 1.723 0.442
> round(epitab(jello, ill, method = "oddsratio")$tab, 3)
Outcome

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

1.2 How do I use R? 7

Predictor N p0 Y p1 oddsratio lower upper p.value

N 22 0.759 30 0.652 1.000 NA NA NA
Y 7 0.241 16 0.348 1.676 0.59 4.765 0.442
> detach(oswego) #detach dataset
The risk of illness among those that consumed jello was 69.6% compared to
57.7% among those that did not consume jello. Both the risk ratio and the
odds ratio were elevated but we cannot exclude random error as an expla-
nation (p value = 0.44). We also notice that the odds ratio is not a good
approximation to the risk ratio. This occurs when the risks among the ex-
posed and unexposed is greater than 10%. In this case, the risks of diarheal
illness were 69.6% and 57.7% among exposed and nonexposed, respectively.

1.2.3 How should I use these notes?

The best way to learn R is to use it! Use it as our calculator! Use it as
our spreadsheet! Finally read these notes sitting at a computer and use R
interactively (this works best sitting in a cafe that brews great coffee and
plays good music). In this book, when we display R code it appears as if we
are typing the code directly at the R command line:
> x <- matrix(1:9, nrow = 3, ncol = 3)
> x
[,1] [,2] [,3]
[1,] 1 4 7
[2,] 2 5 8
[3,] 3 6 9
Sometimes the R code appears as it would in a text editor (e.g., Notepad)
before it has been evaluated by R. When a text editor version of R code is
displayed it appears without the command line prompt and without output:
x <- matrix(1:9, nrow = 3, ncol = 3)
x
When the R code displayed exceeds the page width it will continue on the
next line but indented. Here’s an example:
agegrps <- c("Age < 1", "Age 1 to 4", "Age 5 to 14", "Age 15
to 24", "Age 25 to 44", "Age 45 to 64", "Age 64+")
Although we encourage initially to use R interactively by typing expres-
sions at the command line, as a general rule, it is much better to type our
code into a text editor. We save our code with a convenient file name such as
job01.R2 . For convenience, R comes with its own text editor. For Windows,
2 The .R extension, although not necessary, is useful when searching for R command files.

Additionally, this file extension is recognized by some text editors.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

8 1 Getting Started With R

Table 1.1 Selected math operators

Operator Description Try these examples
+ addition 5+4
− subtraction 5-4
∗ multiplication 5*4
/ division 5/4
ˆ exponentiation 5^4
− unary minus (change current sign) -5
abs absolute value abs(-23)
exp exponentiation (e to a power) exp(8)
log logarithm (default is natural log) log(exp(8))
sqrt square root sqrt(64)
%/% integer divide 10%/%3
%% modulus 10%%3
%*% matrix multiplication xx <- matrix(1:4, 2, 2)
xx%*%c(1, 1)
c(1, 1)%*%xx

under File, select New script to open an empty text file. For Mac OS, under
File, select New Document. As before, save this text file with a .R extension.
Within R’s text editor, we can highlight and run selected commands.
The code in our text editor can be run in the following ways:
• Highlight and run selected command in the R editor;
• Paste the code directly into R at the command line;
• Run the file in batch mode from the R command line using the source("job01.R").

1.3 Just do it!

1.3.1 Using R as your calculator

Open R and start using it as our calculator. The most common math opera-
tors are displayed in Table 1.1. From now on make R our default calculator!
Study the examples and spend a few minutes experimenting with R as a
calculator. Use parentheses as needed to group operations. Use the keyboard
Up-arrow to recall what we previously entered at the command line prompt.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

1.3 Just do it! 9

Table 1.2 Types of evaluable expressionsa

Expression type Try these examples
literal "hello, I’m John Snow" #character
3.5 #numeric
TRUE; FALSE #logical
math operation 6*7
assignment x <- 4*4
data object x
function log(x)
a lines preceded with # are not evaluated

1.3.2 Useful R concepts

1.3.2.1 Types of evaluable expressions

Every expression that is entered at the R command line is evaluated by R

and returns a value. A literal is the simplist expression that can be evaluated
(number, character string, or logical value). Mathematical operations involve
numeric literals. For example, R evaluates the expression 4*4 and returns the
value 16). The exception to this is when an evaluable expression is assigned
an object name: x <- 4*4. To display the assigned expression, wrap the
expression in parentheses: (x <- 4*4), or type the object name.
> 4*4
[1] 16
> x <- 4*4
> x
[1] 16
> (x <- 4*4)
[1] 16
Finally, evaluable expressions must be separated by either newline breaks or
a semicolon.
> x <- 4*4; x
[1] 16
Table 1.2 summarizes evaluable expressions.

1.3.2.2 Using the assignment operator

Most calculators have a memory function: the ability to assign a number or

numerical result to a key for recalling that number or result at a later time.
The same is true in R but it is much more flexible. Any evaluable expression

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

10 1 Getting Started With R

can be assigned a name and recalled at a later time. We refer to these variables
as data objects. We use the assignment operator (<-) to name an evaluable
expression and save it as a data object.
> xx <- "hello, what’s your name"
> xx
[1] "hello, what’s your name"
Wrapping the assignment expression in parentheses makes the assignment
and displays the data object value(s).
> yy <- 5^3 #assignment; no display
> (yy <- 5^3) #assignment; displays evaluation
[1] 125
In this book, we might use (yy <- 5^3) to display the value of yy and save
space on the page. In practice, this is more common:
> yy <- 5^3
> yy
[1] 125
Multiple assignments work and are read from right to left:
> aa <- bb <- 99
> aa; bb
[1] 99
[1] 99
Finally, the equal sign (=) can be used for assignments, although I prefer and
the <- symbol:
> ages = c(34, 45, 67)
> ages
[1] 34 45 67
The reason I prefer <- for assigning object names in the workspace is be-
cause later we use = for assigning values to function arguments. For example,
> x <- 20:25 #object name assignment
> x
[1] 20 21 22 23 24 25
> sample(x = 1:10, size = 5) #argument assignments
[1] 9 6 3 2 5
> x
[1] 20 21 22 23 24 25
The first x is an object name assignment in the workspace which persist
during the R session. The second x is a function argument assignment which
is only recognized locally in the function and only for the duration of the
function execution. For clarity, it is better to keep these types of assignments
separate in our mind by using different assignment symbols.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

1.3 Just do it! 11

Study these previous examples and spend a few minutes using the assign-
ment operator to create and call data objects. Try to use descriptive names
if possible. For example, suppose we have data on age categories; we might
name the data agecat, age.cat, or age cat3 . These are all acceptable.

1.3.3 Useful R functions

When we start R you have opened a workspace. The first time we use R,
the workspace is empty. Every time we create a data object, it is in the
workspace. If a data object with the same name already exists, the old data
object will be overwritten withou warning, so be careful. To list the objects
in your workspace use the ls or objects functions:

> ls() ##display workspace objects

character(0)
> x <- 1:5
> ls()
[1] "x"
> x
[1] 1 2 3 4 5
> x <- 10:15 ##overwrites without warning
> x
[1] 10 11 12 13 14 15

Data objects can be saved between sessions. We will be prompted with

“Save workspace image?” (You can also use save.image() at the command
line.) The workspace image is saved in a file called .RData.4 Use getwd() to
display the file path to the .RData file. Table 1.3 on the following page has
more useful R functions.

1.3.3.1 What are packages?

R has many available functions. When we open R, several packages are at-
tached by default. Each package has its own suite of functions. To display
the list of attached packages use the search function.
> search() # Linux
[1] ".GlobalEnv" "package:stats" "package:graphics"

3 In older versions of R, the underscore symbol ( ) could be used for assignments, but
this is no longer permitted. The “ ” symbol can be used to make your object name more
readable and is consistent with other software.
4 In some operating systems files names that begin with a period (.) are hidden files and

are not displayed by default. You may need to change the viewing option to see the file.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

12 1 Getting Started With R

Table 1.3 Useful R functions

Function Description Try these examples
q Quit R q()
ls List objects ls()
objects objects() #equivalent
rm Remove object(s) yy <- 1:5; ls()
remove rm(yy); ls()
#remove everything: caution!
rm(list = ls()); ls()
help Open help instructions; or get help()
help on specific topic. help(plot)
?plot #equivalent
help.search Search help system given help.search("print")
character string
help.start Start help browser help.start()
apropos Displays all objects in the search apropos(plot)
list matching topic
getwd Get working directory getwd()
setwd Set working directory setwd("c:\mywork\rproject")
args Display arguments of function args(sample)
example Runs example of function example(plot)
data Information on available R data data() #displays data sets
sets; load data set data(Titanic) #load data set
save.image Saves current workspace to save.image()
.RData

[4] "package:grDevices" "package:utils" "package:datasets"

[7] "package:methods" "Autoloads" "package:base"
To display the file paths to the packages use the searchpaths function.
> searchpaths() # Linux
[1] ".GlobalEnv" "/usr/lib/R/library/stats"
[3] "/usr/lib/R/library/graphics" "/usr/lib/R/library/grDevices"
[5] "/usr/lib/R/library/utils" "/usr/lib/R/library/datasets"
[7] "/usr/lib/R/library/methods" "Autoloads"
[9] "/usr/lib/R/library/base"
To learn more about a specific package enter library(help=package name).
Alternatively, we can get more detailed information by entering help.start()
which opens the HTML help page. On this page click on the Packages
link to see the available packages. If we need to load a package enter
library(package name). For example, when we cover survival analysis we
will need to load the survival package.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

1.3 Just do it! 13

1.3.3.2 What are function arguments?

We will be using many R functions for data analysis, so we need to know

some function basics. Suppose we are interested in taking a random sample
of days from the month of June, which has 30 days. We want to use the
{sample} function but we forgot the syntax. Let’s explore:
> sample
function (x, size, replace = FALSE, prob = NULL)
{
if (length(x) == 1 && x >= 1) {
if (missing(size))
size <- x
.Internal(sample(x, size, replace, prob))
}
else {
if (missing(size))
size <- length(x)
x[.Internal(sample(length(x), size, replace, prob))]
}
}
<environment: namespace:base>
Whoa! What happened? Whenever we type the function name without any
parentheses it usually returns the whole function code. This is useful when
we start programming and we need to alter an existing function, borrow code
for our own functions, or study the code for learning how to program. If we
are already familiar with the sample function we may only need to see the
syntax of the function arguments. For this we use the args function:
> args(sample)
function (x, size, replace = FALSE, prob = NULL)
NULL
The terms x, size, replace, and prob are the function arguments. First,
notice that replace and prob have default values; that is, we do not need to
specify these arguments unless we want to override the default values. Second,
notice the order of the arguments. If you enter the argument values in the
same order as the argument list we do not need to specify the argument.
> dates <- 1:30
> sample(dates, 18)
[1] 29 12 28 8 5 2 13 24 20 18 11 14 23 1 21 4 22 9
Third, if we enter the arguments out of order then we will get either an error
message or an undesired result. Arguments entered out of their default order
need to be specified.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

14 1 Getting Started With R

> sample(18, dates) #gives undesired results

[1] 2
> #No! We wanted sample of size = 18
> sample(size = 18, x = dates) #gives desired result
[1] 6 29 1 27 13 7 24 19 21 5 22 12 14 23 25 15 18 11
Fourth, when we specify an argument we only need to type a sufficient number
of letters so that R can uniquely identify it from the other arguments.
> sample(s = 18, x = dates, r = T) #sampling with replacement
[1] 23 10 23 27 13 14 1 7 23 26 28 3 23 28 9 6 23 5
Fifth, argument values can be any valid R expression (including functions)
that yields to an appropriate value. In the following example we see two sam-
ple functions that provide random values to the sample function arguments.
> sample(s = sample(1:100, 1), x = sample(1:10, 5), r=T)
[1] 3 4 9 3 3 9 10 3 10 3 10 4 9 3 5 9 4 5
Finally, if we need more guidance on how to use the sample function enter
?sample or help(sample).

1.3.4 How do I get help?

R has extensive help capabilities. From the main menu select Help to get
you started (Figure 1.1 on the next page). The Frequently Asked Questions
(FAQ) and R manuals are available from this menu. The R functions (text). . . ,
Html help, Search help. . . , and Apropos. . . selections are also available from
the command line.
From the command line, we have several options. Suppose you are inter-
ested in learning abouting help capabilities.
> ?help #opens help page for ’help’ function
> help() #opens help page for ’help’ function
> help(help) #opens help page for ’help’ function
> help.start() #starts HTML help page
> help.search("help") #searches help system for "help"
> apropos("help") #displays ’help’ objects in search list
> apropos("help")
[1] "help" "help.request" "help.search" "help.start"
To learn about about available data sets use the data function:
> data() #displays available data sets
> try(data(package = "survival")) #lists survival pkg data sets
> help(pbc, package = "survival") #displays pbc data help page
Figure 1.1 on the facing page shows that a R Graphical User Interface (GUI)
main menu will have a Help selection.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

1.3 Just do it! 15

Fig. 1.1 Select Help from the main menu in the MS Windows R GUI.

1.3.5 RStudio—An integrated development

environment for R

RStudio5 is a free and open source integrated development environment

(IDE) for R that runs on any desktop (Windows, Mac, or Linux) (Figure 1.2
on the next page). For our purposes, I highly recommend the installation of
RStudio: it has all the tools we need to learn and apply R.

1.3.6 Is there anything else that I need?

Maybe.6 A good text editor will make your programming and data processing
easier and more efficient. A text editor is a program for, we guessed it, editing
text! The functionality we look for in a text editor are the following:
1. Toggle between wrapped and unwrapped text
2. Block cutting and pasting (also called column editing)
3. Easy macro programming
4. Search and replace using regular expressions
5. Ability to import large datasets for editing
When we are programming we want our text to wrap so we can read all
of your code. When we import a data set that is wider than the screen, we
do not want the data set to wrap: we want it to appear in its tabular format.
Column editing allows us to cut and paste columns of text at will. A macro is
5 http://www.rstudio.org/
6 If your only goal is to learn R, then RStudio is more than sufficient.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

16 1 Getting Started With R

Fig. 1.2 RStudio—An integrated development environment for R that runs in Linux, Mac
OS, or MS Windows. In this Figure, RStudio is running in MS Windows.

just a way for the text editor to learn a set of keystrokes (including search and
replace) that can be executed as needed. Searching using regular expressions
means searching for text based on relative attributes. For example, suppose
you want to find all words that begin with “b,” end with “g,” have any number
of letters in between but not “r” and “f.” Regular expression searching makes
this a trivial task. These are powerful features that once we use regularly, we
will wonder how we ever got along without them.
If we do not want to install a text editing program then we can use the
default text editor that comes with our computer operating system (gedit
in Ubuntu Linux, TextEdit in Mac OS, Notepad in Windows). However, it
is much better to install a text editor that works with R. My favorite text
editor is the free and open source GNU Emacs. GNU Emacs can be extended
with the “Emacs Speaks Statistics” (ESS) package. For more information on
Emacs and ESS pre-installed for Windows, visit http://ess.r-project.
org/. For the Mac OS, I recommend GNU Emacs for Mac OS7 (Figure 1.3
on the facing page) or Aquamacs.8
7 http://emacsformacosx.com/
8 http://aquamacs.org/

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

1.3 Just do it! 17

Fig. 1.3 Emacs and ESS in the Mac OS

1.3.7 What’s ahead?

To the novice user, R may seem complicated and difficult to learn. In fact, for
its immense power and versatility, R is easier to learn and deploy compared to
other statistical software (e.g. SAS, Stata, SPSS). This is because R was built
from the ground up to be an efficient and intuitive programming language and
environment. If one understands the logic and structure of R, then learning
proceeds quickly. Just like a spoken language, once we know its rules of
grammar, syntax, and pronunciation, and can write legible sentences, we can
figure out how to communicate almost anything. Before the next chapter,
we want to describe the “forest”: the logic and structure of working with R
objects and epidemiologic data.

1.3.7.1 Working with R objects

For our purposes, there are only five types of data objects in R9 and five types
of actions we take on these objects (Table 1.4 on the next page). That’s it! No
more, no less. You will learn to create, name, index (subset), replace compo-
nents of, and operate on these data objects using a systematic, comprehensive

9 The sixth type of R object is a function. Functions can create, manipulate, operate on, and
store data; however, we will use functions primarily to execute a series of R “commands”
and not as primary data objects.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

18 1 Getting Started With R

Table 1.4 Types of actions taken on R data objects and where to find examples
Action Vector Matrix Array List Data Frame
Creating Table 2.4 Table 2.11 Table 2.18 Table 2.24 Table 2.30
(p. 34) (p. 51) (p. 66) (p. 80) (p. 89)
Naming Table 2.5 Table 2.12 Table 2.19 Table 2.25 Table 2.31
(p. 38) (p. 55) (p. 69) (p. 81) (p. 90)
Indexing Table 2.6 Table 2.13 Table 2.20 Table 2.26 Table 2.32
(p. 39) (p. 57) (p. 70) (p. 81) (p. 91)
Replacing Table 2.7 Table 2.14 Table 2.21 Table 2.27 Table 2.33
(p. 41) (p. 58) (p. 70) (p. 84) (p. 94)
Operating on Table 2.8 Table 2.15 Table 2.22 Table 2.28 Table 2.34
(p. 42) (p. 59) (p. 71) (p. 84) (p. 95)
Table 2.9
(p. 46)

approach. As you learn about each new data object type, it will reinforce and
extend what you learned previously.
A vector is a collection of elements (usually numbers):
> x <- c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12)
> x
[1] 1 2 3 4 5 6 7 8 9 10 11 12
A matrix is a 2-dimensional representaton of a vector:
> y <- matrix(x, nrow = 2)
> y
[,1] [,2] [,3] [,4] [,5] [,6]
[1,] 1 3 5 7 9 11
[2,] 2 4 6 8 10 12
An array is a 3 or more dimensional represention of a vector:
> z <- array(x, dim = c(2, 3, 2))
> z
, , 1

[,1] [,2] [,3]

[1,] 1 3 5
[2,] 2 4 6

, , 2

[,1] [,2] [,3]

[1,] 7 9 11
[2,] 8 10 12
A list is a collection of “bins,” each containing any kind of R object:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

1.3 Just do it! 19

> mylist <- list(x, y, z)

> mylist
[[1]]
[1] 1 2 3 4 5 6 7 8 9 10 11 12

[[2]]
[,1] [,2] [,3] [,4] [,5] [,6]
[1,] 1 3 5 7 9 11
[2,] 2 4 6 8 10 12

[[3]]
, , 1

[,1] [,2] [,3]

[1,] 1 3 5
[2,] 2 4 6

, , 2

[,1] [,2] [,3]

[1,] 7 9 11
[2,] 8 10 12
A data frame is a list in tabular form where each “bin” contains a data vector
of the same length. A data frame is the usual tabular data set familiar to
epidemiologists. Each row is an record and each column (“bin”) is a field.
> kids <- c("Tomasito", "Lusito", "Angelita")
> gender <- c("boy", "boy", "girl")
> age <- c(8, 7, 4)
> mydf <- data.frame(kids, gender, age)
> mydf
kids gender age
1 Tomasito boy 8
2 Lusito boy 7
3 Angelita girl 4
In the next chapter we explore these R data objects in greater detail.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

20 1 Getting Started With R

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

1.3 Just do it! 21

Problems

1.1. If you have not done so already, install R on your personal computer.
What is the R workspace file on your operating systems? What is the file
path to your R workspace file? What is the name of this workspace file?

1.2. By default, which R packages come already loaded? What are the file
paths to the default R packages?

1.3. List all the object in the current workspace. If there are none, create
some data objects. Using one expression, remove all the objects in the current
workspace.

1.4. One inch equals 2.54 centimeters. Correct the following R code and
create a conversion table.
inches <- 1:12
centimeters <- inches/2.54
cbind(inches, centimeters)

1.5. To convert between temperatures in degrees Celsius (C) and Farenheit

(F ), we use the following conversion formulas:
5
C = (F − 32)
9
9
F = C + 32
5
At standard temperature and pressure, the freezing and boiling points of
water are 0 and 100 degrees Celsius, respectively. What are the freezing and
boiling points of water in degrees Fahrenheit?

1.6. For the Celsius temperatures 0, 5, 10, 15, 20, 25, ..., 80, 85, 90, 95, 100,
construct a conversion table that displays the corresponding Fahrenheit tem-
peratures. Hint: to create the sequence of Celsius temperatures use seq(0,
100, 5).

1.7. BMI is a reliable indicator of total body fat, which is related to the risk
of disease and death. The score is valid for both men and women but it does
have some limits. BMI does have some limitations: it may overestimate body
fat in athletes and others who have a muscular build, it may underestimate
body fat in older persons and others who have lost muscle mass.
Body Mass Index (BMI) is calculated from your weight in kilograms and
height in meters:
kg
BM I =
m2
1 kg ≈ 2.2 lb

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

22 1 Getting Started With R

Table 1.5 Body mass index classification

BMI Classification
< 18.5 Underweight
[18.5, 25) Normal weight
[25, 30) Overweight
≥ 30 Obesity

1 m ≈ 3.3 ft
Calculate your BMI (don’t report it to us).

1.8. Using Table 1.1 on page 8, explain in words, and use R to illustrate, the
difference between modulus and integer divide.

1.9. In mathematics, a logarithm (to base b) of a number x is written logb (x)

and equals the exponent y that satisfies x = by . In other words,

y = logb (x)

is equivalent to
x = by
In R, the log function is to the base e. Implement the following R code
and study the graph:
curve(log(x), 0, 6)
abline(v = c(1, exp(1)), h = c(0, 1), lty = 2)
What kind of generalizations can you make about the natural logarithm and
its base—the number e?

1.10. Risk (R) is a probability bounded between 0 and 1. Odds is the follow-
ing transformation of R:
R
Odds =
1−R
Use the following code to plot the odds:
curve(x/(1-x), 0, 1)
Now, use the following code to plot the log(odds):
curve(log(x/(1-x)), 0, 1)
What kind of generalizations can you make about the log(odds) as a trans-
formation of risk?

1.11. Use the data in Table 1.6 on the next page. Assume one is HIV-negative.
If the probability of infection per act is p, then the probability of not getting
infected per act is (1 − p). The probability of not getting infected after 2

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

1.3 Just do it! 23

Table 1.6 Estimated per-act risk (transmission probability) for acquisition of HIV, by
exposure route to an infected source. Source: CDC [1]

Exposure route Risk per 10,000 exposures

Blood transfusion (BT) 9,000

Needle-sharing injection-drug use (IDU) 67
Receptive anal intercourse (RAI) 50
Percutaneous needle stick (PNS) 30
Receptive penile-vaginal intercourse (RPVI) 10
Insertive anal intercourse (IAI) 6.5
Insertive penile-vaginal intercourse (IPVI) 5
Receptive oral intercourse on penis (ROI) 1
Insertive oral intercourse with penis (IOI) 0.5

consecutive acts is (1− p)2 , and after 3 consecutive acts is (1− p)3 . Therefore,
the probability of not getting infected infected after n consecutive acts is
(1 − p)n , and the probability of getting infected after n consecutive acts is
1 − (1 − p)n . For each non-blood transfusion transmission probability (per act
risk) in Table 1.6, calculate the cumulative risk of being infected after one
year (365 days) if one carries out the same act once daily for one year with an
HIV-infected partner. Do these cumulative risks make intuitive sense? Why
or why not?
1.12. The source function in R is used to “source” (read in) ASCII text files.
Take a group of R commands that worked from a previous problem above
and paste them into an ASCII text file and save it with the name job01.R.
Then from R command line, source the file. Here is how it looked on my
Linux computer running R:
> source("/home/tja/Documents/courses/ph251d/jobs/job01.R")
Describe what happened. Now, set echo option to TRUE.
> source("/home/tja/Documents/courses/ph251d/jobs/job01.R", echo = TRUE)
Describe what happened. To improve your understanding read the help file
on the source function.
1.13. Now run the source again (without and with echo = TRUE) but each
time create a log file using the sink function. Create two log files: job01.log1a
and job01.log1b.
> sink("/home/tja/Documents/courses/ph251d/jobs/job01.log1a")
> source("/home/tja/Documents/courses/ph251d/jobs/job01.R")
> sink() #closes connection
>
> sink("/home/tja/Documents/courses/ph251d/jobs/job01.log1b")
> source("/home/tja/Documents/courses/ph251d/jobs/job01.R", echo = TRUE)
> sink() #closes connection

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

24 1 Getting Started With R

Examine the log files and describe what happened.

1.14. Create a new job file (job02.R) with the following code:
n <- 365
per.act.risk <- c(0.5, 1, 5, 6.5, 10, 30, 50, 67)/10000
risks <- 1-(1-per.act.risk)^n
show(risks)
Source this file at the R command line and describes what happens.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Chapter 2
Working with R data objects

2.1 Data objects in R

2.1.1 Atomic vs. recursive data objects

The analysis of data in R involves creating, manipulating, and operating on

data objects using functions. Data in R are organized as objects and have
been assigned a name. We have already been introduced to several R data
objects. We will now make some additional distinctions. Every data object
has a mode and length. The mode of an object describes the type of data
it contains and is available by using the mode function. An object can be of
mode character, numeric, logical, list, or function.

> fname <- c("Juan", "Miguel"); mode(fname)

[1] "character"
> age <- c(34, 20); mode(age)
[1] "numeric"
> lt25 <- age<25
> lt25
[1] FALSE TRUE
> mode(lt25)
[1] "logical"
> mylist <- list(fname, age); mode(mylist)
[1] "list"
> mydat <- data.frame(fname, age); mode(mydat)
[1] "list"
> myfun <- function(x) {x^2}
> myfun(5)
[1] 25
> mode(myfun)
[1] "function"

25
26 2 Working with R data objects

Data objects are further categorized into atomic or recursive objects. An

atomic data object can only contain elements from one, and only one, of the
following modes: character, numeric, or logical. Vectors, matrices, and arrays
are atomic data objects. A recursive data object can contain data objects of
any mode. Lists, data frames, and functions are recursive data objects. We
start by reviewing atomic data objects.
A vector is a collection of like elements without dimensions.1 The vector el-
ements are all of the same mode (either character, numeric, or logical). When
R returns a vector the [n] indicates the position of the element displayed to
its immediate right.
> y <- c("Pedro", "Paulo", "Maria")
> y
[1] "Pedro" "Paulo" "Maria"
> x <- c(1, 2, 3, 4, 5)
> x
[1] 1 2 3 4 5
> x < 3
[1] TRUE TRUE FALSE FALSE FALSE
A matrix is a collection of like elements organized into a 2-dimensional
(tabular) data object. We can think of a matrix as a vector with a 2-
dimensional structure. When R returns a matrix the [n,] indicates the nth
row and [,m] indicates the mth column.
> x <- c("a", "b", "c", "d")
> y <- matrix(x, 2, 2)
> y
[,1] [,2]
[1,] "a" "c"
[2,] "b" "d"
An array is a collection of like elements organized into a n-dimensional
data object. We can think of an array as a vector with an n-dimensional
structure. When R returns an array the [n,,] indicates the nth row and
[,m,] indicates the mth column, and so on.
> x <- 1:8
> y <- array(x, dim=c(2, 2, 2))
> y
, , 1

[,1] [,2]
[1,] 1 3
[2,] 2 4

1 In other programming languages, vectors are either row vectors or column vectors. R

does not make this distinction until it is necessary.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.1 Data objects in R 27

, , 2

[,1] [,2]
[1,] 5 7
[2,] 6 8
If we try to include elements of different modes in an atomic data object, R
will coerce the data object into a single mode based on the following hierarchy:
character > numeric > logical. In other words, if an atomic data object
contains any character element, all the elements are coerced to character.
> c("hello", 4.56, FALSE)
[1] "hello" "4.56" "FALSE"
> c(4.56, FALSE)
[1] 4.56 0.00
A recursive data object can contain one or more data objects where each
object can be of any mode. Lists, data frames, and functions are recursive
data objects. Lists and data frames are of mode list, and functions are of
mode function (Table 2.1 on the next page).
A list is a collection of data objects without any restrictions:
> x <- c(1, 2, 3)
> y <- c("Male", "Female", "Male")
> z <- matrix(1:4, 2, 2)
> mylist <- list(x, y, z)
> mylist
[[1]]
[1] 1 2 3

[[2]]
[1] "Male" "Female" "Male"

[[3]]
[,1] [,2]
[1,] 1 3
[2,] 2 4
A data frame is a list with a 2-dimensional (tabular) structure. Epidemiol-
ogists are very experienced working with data frames where each row usually
represents data collected on individual subjects (also called records or ob-
servations) and columns represent fields for each type of data collected (also
called variables).
> subjno <- c(1, 2, 3, 4)
> age <- c(34, 56, 45, 23)
> sex <- c("Male", "Male", "Female", "Male")

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

28 2 Working with R data objects

Table 2.1 Summary of six types of data objects in R

Data object Possible mode Default class
Atomic
vector character, numeric, logical NULL
matrix character, numeric, logical NULL
array character, numeric, logical NULL
Recursive
list list NULL
data frame list data frame
function function NULL

> case <- c("Yes", "No", "No", "Yes")

> mydat <- data.frame(subjno, age, sex, case)
> mydat
subjno age sex case
1 1 34 Male Yes
2 2 56 Male No
3 3 45 Female No
4 4 23 Male Yes
> mode(mydat)
[1] "list"

2.1.2 Assessing the structure of data objects

Summarized in Table 2.1 are the key attributes of atomic and recursive data
objects. Data objects can also have class attributes. Class attributes are just
a way of letting R know that an object is “special,” allowing R to use specific
methods designed for that class of objects (e.g., print, plot, and summary
methods). The class function displays the class if it exists. For our purposes,
we do not need to know any more about classes.
Frequently, we need to assess the structure of data objects. We already
know that all data objects have a mode and length attribute. For example,
let’s explore the infert data set that comes with R. The infert data comes
from a matched case-control study evaluating the occurrence of female infer-
tility after spontaneous and induced abortion.

> data(infert) #loads data

> mode(infert)
[1] "list"
> length(infert)
[1] 8

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.2 A vector is a collection of like elements 29

At this point we know that the data object named “infert” is a list of length
8. To get more detailed information about the structure of infert use the
str function (str comes from “str”ucture).
> str(infert)
‘data.frame’: 248 obs. of 8 variables:
$ education : Factor w/ 3 levels "0-5yrs",..: 1 1 1 1 ...
$ age : num 26 42 39 34 35 36 23 32 21 28 ...
$ parity : num 6 1 6 4 3 4 1 2 1 2 ...
$ induced : num 1 1 2 2 1 2 0 0 0 0 ...
$ case : num 1 1 1 1 1 1 1 1 1 1 ...
$ spontaneous : num 2 0 0 0 1 1 0 0 1 0 ...
$ stratum : int 1 2 3 4 5 6 7 8 9 10 ...
Great! This is better. We now know that infert is a data frame with 248
observations and 8 variables. The variable names and data types are dis-
played along with their first few values. In this case, we now have sufficient
information to start manipulating and analyzing the infert data set.
Additionally, we can extract more detailed structural information that
becomes useful when we want to extract data from an object for further
manipulation or analysis (Table 2.2 on the next page). We will see extensive
use of this when we start programming in R.
To get practice calling data from the command line, enter data() to dis-
play the available data sets in R. Then enter data(data set ) to load a
dataset. Study the examples in Table 2.2 on the following page and spend
a few minutes exploring the structure of the data sets we have loaded. To
display detailed information about a specific data set use ?data set at the
command prompt (e.g., ?infert).

2.2 A vector is a collection of like elements

2.2.1 Understanding vectors

A vector is a collection of like elements (i.e., the elements all have the same
mode). There are many ways to create vectors (see Table 8). The most com-
mon way of creating a vector is using the concatenation function c:
> #numeric
> chol <- c(136, 219, 176, 214, 164)
> chol
[1] 136 219 176 214 164
> #character
> fname <- c("Mateo", "Mark", "Luke", "Juan", "Jaime")
> fname

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

30 2 Working with R data objects

Table 2.2 Useful functions to assess R data objects

Function Description Try these examples
Returns summary objects
str Displays summary of data object str(infert)
structure
attributes Returns list with data object attributes(infert)
attributes
Returns specific information
mode Returns mode of object mode(infert)
class Returns class of object, if it exists class(infert)
length Returns length of object length(infert)
dim Returns vector with object dim(infert)
dimensions, if applicable
nrow Returns number of rows, if nrow(infert)
applicable
ncol Returns number of columns, if ncol(infert)
applicable
dimnames Returns list containing vectors of dimnames(infert)
names for each dimension, if
applicable
rownames Returns vector of row names of a rownames(infert)
matrix-like object
colnames Returns vector of column names colnames(infert)
of a matrix-like object
names Returns vector of names for the names(infert)
list, if applicable (for a data frame
it returns the field names)
row.names Returns vector of row names for a row.names(infert)
data frame
head Display first 6 lines of a data head(infert)
frame infert[1:6,] #equivalent

[1] "Mateo" "Mark" "Luke" "Juan" "Jaime"

> #logical
> z <- c(T, T, F, T, F)
> z
[1] TRUE TRUE FALSE TRUE FALSE

A single digit is also a vector; that is, a vector of length = 1. Let’s confirm
this.
> 5
[1] 5
> is.vector(5)
[1] TRUE

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.2 A vector is a collection of like elements 31

2.2.1.1 Boolean operations on vectors

In R, we use relational and logical operators (Table 2.3 on page 33) to conduct
Boolean queries. Boolean operations is a methodological workhorse of data
analysis. For example, suppose we have a vector of female movie stars and a
corresponding vector of their ages (as of January 16, 2004), and we want to
select a subset of actors based on age criteria.

> movie.stars
[1] "Rebecca De Mornay" "Elisabeth Shue" "Amanda Peet"
[4] "Jennifer Lopez" "Winona Ryder" "Catherine Zeta Jones"
[7] "Reese Witherspoon"
> ms.ages
[1] 42 40 32 33 32 34 27

Let’s select the actors who are in their 30s. This is done using logical
vectors that are created by using relational operators (<, >, <=, >=, ==, !=).
Study the following example:

> #logical vector for stars with ages >=30

> ms.ages >= 30
[1] TRUE TRUE TRUE TRUE TRUE TRUE FALSE
> #logical vector for stars with ages <40
> ms.ages < 40
[1] FALSE FALSE TRUE TRUE TRUE TRUE TRUE
> #logical vector for stars with ages >=30 and <40
> (ms.ages >= 30) & (ms.ages < 40)
[1] FALSE FALSE TRUE TRUE TRUE TRUE FALSE
> thirtysomething <- (ms.ages >= 30) & (ms.ages < 40)
> #indexing vector based on logical vector
> movie.stars[thirtysomething]
[1] "Amanda Peet" "Jennifer Lopez" "Winona Ryder"
[4] "Catherine Zeta Jones"
We also saw that we can compare logical vectors using logical operators (&,
|, !). For more examples see Table 7. The expression movie.stars[thirtysomething]
is an example of indexing using a logical vector. Now, we can use the ! func-
tion to select the stars that are not “thirtysomething.” Study the following:
> thirtysomething
[1] FALSE FALSE TRUE TRUE TRUE TRUE FALSE
> !thirtysomething
[1] TRUE TRUE FALSE FALSE FALSE FALSE TRUE
> movie.stars[!thirtysomething]
[1] "Rebecca De Mornay" "Elisabeth Shue" "Reese Witherspoon"

To summarize:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

32 2 Working with R data objects

• Logical vectors are created using Boolean comparisons,

• Boolean comparisons are constructed using relational and logical operators
• Logical vectors are commonly used for indexing (subsetting) data objects
Before moving on, we need to be sure we understand the previous exam-
ples, then study the examples in Table 2.3 on the facing page. For practice,
study the examples and spend a few minutes creating simple numeric vec-
tors, then (1) generate logical vectors using relational operators, (2) use these
logical vectors to index the original numerical vector or another vector, (3)
generate logical vectors using the combination of relational and logical oper-
ators, and (4) use these logical vectors to index the original numerical vector
or another vector.
The element-wise exclusive “or” operator (xor) returns TRUE if either
comparison element is TRUE, but not if both are TRUE. In contrast, the |
returns TRUE if either or both comparison elements are TRUE.
The && and || logical operators are used for control flow in if functions.
If logical vectors are provided, only the first element of each vector is used.
Therefore, for element-wise comparisons of 2 or more vectors, use the & and
| operators but not the && and || operators.

2.2.2 Creating vectors

Vectors are created directly, or indirectly as the result of manipulating an

R object. The c function for concatenating a collection has been covered
previously. Another, possibly more convenient, method for collecting elements
into a vector is with the scan function.

> x <- scan()

1: 45 67 23 89
5:
Read 4 items
> x
[1] 45 67 23 89

This method is convenient because we do not need to type c, parentheses,

and commas to create the vector. The vector is created after executing a
newline twice.
To generate a sequence of consecutive integers use the : function.
> -9:8
[1] -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8
However, the seq function provides more flexibility in generating sequences.
Here are some examples:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.2 A vector is a collection of like elements 33

Table 2.3 Using Boolean relational and logical operators

Operator Description Try these examples
Relational operators
< Less than pos <- c("p1", "p2", "p3", "p4", "p5")
x <- c(1, 2, 3, 4, 5)
y <- c(5, 4, 3, 2, 1)
x < y
pos[x < y]
> Greater than x > y
pos[x > y]
<= Less than or equal to x <= y
pos[x <= y]
>= Greater than or x >= y
equal to pos[x >= y]
== Equal to x == y
pos[x == y]
!= Not equal to x != y
pos[x != y]
Logical operators
! NOT x > 2
!(x > 2)
pos[!(x > 2)]
& Element-wise AND (x > 1) & (x < 5)
pos[(x > 1) & (x < 5)]
| Element-wise OR (x <= 1) | (x > 4)
pos[(x <= 1) | (x > 4)]
xor Exclusive OR; xx <- x <= 1
similar to | for yy <- x > 4
comparing two xor(xx, yy)
vectors, but only xx | yy
TRUE if one or the
other is true, not
both
Logical operators for if function
&& Similar to & but if(T && T) print("Both TRUE")
used with if # nothing printed if(F && T) print("Both
function TRUE")
|| Similar to | but used if(T || F) print("Either TRUE")
with if function # nothing printed if(F || F) print("Either
TRUE")

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

34 2 Working with R data objects

Table 2.4 Common ways of creating vectors

Function Description Try these examples

c Concatenate a x <- c(1, 2, 3, 4, 5)
collection y <- c(6, 7, 8, 9, 10)
z <- c(x, y)
scan Scan a collection (after xx <- scan()
entering data press 1 2 3 4 5
Enter twice)
yy <- scan(what = "")
"Javier" "Miguel" "Martin"

xx; yy
: Generate integer 1:10
sequence 10:(-4)
seq Generate sequence of seq(1, 5, by = 0.5)
numbers seq(1, 5, length = 3)
zz <- c("a", "b", "c")
seq(along = zz)
rep Replicate argument rep("Juan Nieve", 3)
rep(1:3, 4)
rep(1:3, 3:1)
which Integer vector from age <- c(8, NA, 7, 4)
Boolean operation which(age<5 | age>=8)
paste Paste elements paste(c("A", "B", "C"), 1:3)
creating a character paste(c("A", "B", "C"), 1:3, sep="")
string
[row#, ] Indexing a matrix xx <- matrix(1:8, nrow = 2, ncol = 4)
or returns a vector xx[2,]
[ ,col#] xx[,3]
sample Sample from a vector sample(c("H","T"), 20, replace = TRUE)
runif Generate random rnorm(10, mean = 50, sd = 19)
rnorm numbers from a runif(n = 10, min = 0, max = 1)
rbinom probability rbinom(n = 10, size = 20, p = 0.5)
rpois distribution rpois(n = 10, lambda = 15)
as.vector Coerce data objects mx <- matrix(1:4, nrow = 2, ncol = 2)
into a vector mx
as.vector(mx)
vector Create vector of vector("character", 5)
specified mode and vector("numeric", 5)
length vector("logical", 5)
character Create vector of character(5)
numeric specified type numeric(5)
logical logical(5)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.2 A vector is a collection of like elements 35

> seq(1, 5, by = 0.5) ##specify interval

[1] 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
> seq(1, 5, length = 8) ##specify length
[1] 1.0000 1.5714 2.1429 2.7143 3.2857 3.8571 4.4286 5.0000
> x <- 1:8
> seq(1, 5, along = x) ##by length of other object
[1] 1.0000 1.5714 2.1429 2.7143 3.2857 3.8571 4.4286 5.0000
These types of sequences are convenient for plotting mathematical equa-
tions.2 For example, suppose we wanted to plot the standard normal curve
using the normal equation. For a standard normal curve µ = 0 (mean) and
σ = 1 (standard deviation)

−(x − µ)2
 
1
f (x) = √ exp
2πσ 2 2σ 2

Here is the R code to plot this equation (see Figure 2.1):

mu <- 0; sigma <- 1
x <- seq(-4, 4, .01)
fx <- (1/sqrt(2*pi*sigma^2))*exp(-(x-mu)^2/(2*sigma^2))
plot(x, fx, type = "l", lwd = 2)
After assigning values to mu and sigma, we assigned to x a sequence of
numbers from −4 to 4 by intervals of 0.01. Using the normal curve equation,
0.4

0.4
Density
0.2

0.2
fx

0.0
0.0

−4 0 2 4 −3 −1 1 3

x rnorm(500)

Fig. 2.1 Standard normal curve from equation and simulation

2 See also the curve function for graphing mathematical equations.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

36 2 Working with R data objects

for every value of x we calculated f (x), represented by the numeric vector fx.
We then used the plot function to plot x vs. f (x). The optional argument
type="l" produces a “line” and lwd=2 doubles the line width. For compari-
son, we also plotted a density histogram of 500 standard normal variates that
were simulated using the rnorm function (Figure 2.1 on the preceding page).3
The rep function is used to replicate its arguments. Study the examples
that follow:
> rep(5, 2) #repeat 5 2 times
[1] 5 5
> rep(1:2, 5) # repeat 1:2 5 times
[1] 1 2 1 2 1 2 1 2 1 2
> rep(1:2, c(5, 5)) # repeat 1 5 times; repeat 2 5 times
[1] 1 1 1 1 1 2 2 2 2 2
> rep(1:2, rep(5, 2)) # equivalent to previous
[1] 1 1 1 1 1 2 2 2 2 2
> rep(1:5, 5:1) # repeat 1 5 times, repeat 2 4 times, etc
[1] 1 1 1 1 1 2 2 2 2 3 3 3 4 4 5
The paste function pastes character strings:
> fname <- c("John", "Kenneth", "Sander")
> lname <- c("Snow", "Rothman", "Greenland")
> paste(fname, lname)
[1] "John Snow" "Kenneth Rothman" "Sander Greenland"
> paste("var", 1:7, sep="")
[1] "var1" "var2" "var3" "var4" "var5" "var6" "var7"
Indexing (subsetting) an object often results in a vector. To preserve the
dimensionality of the original object use the drop option.
> x <- matrix(1:8, 2, 4)
> x
[,1] [,2] [,3] [,4]
[1,] 1 3 5 7
[2,] 2 4 6 8
> x[2,] #index 2nd row
[1] 2 4 6 8
> x[2, , drop = FALSE] #index 2nd row; keep object structure
[,1] [,2] [,3] [,4]
[1,] 2 4 6 8
Up to now we have generated vectors of known numbers or character
strings. On occasion we need to generate random numbers or draw a sample
from a collection of elements. First, sampling from a vector returns a vector:

3 hist(rnorm(500), freq = FALSE, breaks = 25, main="")

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.2 A vector is a collection of like elements 37

> # toss 8 coins

> sample(c("Head","Tail"), size = 8, replace = TRUE)
[1] "Head" "Head" "Tail" "Head" "Tail" "Tail" "Head" "Head"
> # toss 2 die
> sample(1:6, size = 2, replace = TRUE)
[1] 1 4
Second, generating random numbers from a probability distribution returns
a vector:
> # toss 8 coins twice using the binomial distribution
> rbinom(n = 2, size = 8, p = 0.5)
[1] 5 2
> # generate 6 standard normal distribution values
> rnorm(6)
[1] 1.52826 -0.50631 0.56446 0.10813 -1.96716 2.01802
There are additional ways to create vectors. To practice creating vec-
tors study the examples in Table 2.4 on page 34 and spend a few minutes
creating simple vectors. If we need help with a function remember enter
?function name or help(function name ).
Finally, notice that we use vectors as arguments to functions:
# character vector used in ’sample’ function
sample(c("head", "tail"), 100, replace = TRUE)
# numeric vector used in ’rep’ function
rep(1:2, rep(5, 2))
# numeric vector used in ’matrix’ function
matrix(c(23, 45, 16, 17), nrow = 2, ncol = 2)

2.2.3 Naming vectors

The first way of naming vector elements is when the vector is created:
> x <- c(chol = 234, sbp = 148, dbp = 78, age = 54)
> x
chol sbp dbp age
234 148 78 54
The second way is to create a character vector of names and then assign that
vector to the numeric vector using the names function:
> z <- c(234, 148, 78, 54)
> z
[1] 234 148 78 54
> names(z) <- c("chol", "sbp", "dbp", "age")
> z

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

38 2 Working with R data objects

Table 2.5 Common ways of naming vectors

Function Description Try these examples
c Name vector x <- c(a = 1, b = 2, c = 3, d = 4); x
elements at time
that vector is
created
names Name vector y <- 1:4; y
elements names(y) <- c("a", "b", "c", "d"); y
#return names, if they exist
names(y)
unname Remove names y <- unname(y)
y #equivalent: names(y) <- NULL

chol sbp dbp age

234 148 78 54
The names function, without an assignment, returns the character vector of
names, if it exist. This character vector can be used to name elements of
other vectors.
> names(z)
[1] "chol" "sbp" "dbp" "age"
> z2 <- c(250, 184, 90, 45)
> z2
[1] 250 184 90 45
> names(z2) <- names(z)
> z2
chol sbp dbp age
250 184 90 45
The unname function removes the element names from a vector:
> unname(z2)
[1] 250 184 90 45
For practice study the examples in Table 2.5 and spend a few minutes
creating and naming simple vectors.

2.2.4 Indexing vectors

Indexing a vector is subsetting or extracting elements from a vector. A vector

is indexed by position(s), by name(s), and by logical vector. Positions are
specified by positive or negative integers.
> x <- c(chol = 234, sbp = 148, dbp = 78, age = 54)
> x[c(2, 4)] #extract 2nd and 4th element

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.2 A vector is a collection of like elements 39

Table 2.6 Common ways of indexing vectors

Indexing Try these examples
By position x <- c(chol = 234, sbp = 148, dbp = 78, age = 54)
x[2] #positions to include
x[c(2, 3)]
x[-c(1, 3, 4)] #positions to exclude
x[-c(1, 4)]
x[which(x<100)]
By name (if exists) x["sbp"]
x[c("sbp", "dbp")]
By logical x < 100
x[x < 100]
(x < 150) & (x > 70)
bp <- (x < 150) & (x > 70)
x[bp]
Unique values samp <- sample(1:5, 25, replace=T); samp
unique(samp)
Duplicated values duplicated(samp) #generates logical
samp[duplicated(samp)]

sbp age
148 54
> x[-c(2, 4)] #exclude 2nd and 4th element
chol dbp
234 78

Although indexing by position is concise, indexing by name (when the names

exists) is better practice in terms of documenting our code. Here is an exam-
ple:

> x[c("sbp", "age")] #extract 2nd and 4th element

sbp age
148 54

A logical vector indexes the positions that corresponds to the TRUEs. Here
is an example:
> x<=100 | x>200
chol sbp dbp age
TRUE FALSE TRUE TRUE
> x[x<=100 | x>200]
chol dbp age
234 78 54
Any expression that evaluates to a valid vector of integers, names, or
logicals can be used to index a vector.
> (samp1 <- sample(1:4, 8, replace = TRUE))

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

40 2 Working with R data objects

[1] 1 3 3 3 1 3 4 1
> x[samp1]
chol dbp dbp dbp chol dbp age chol
234 78 78 78 234 78 54 234
> (samp2 <- sample(names(x), 8, replace = TRUE))
[1] "dbp" "sbp" "sbp" "dbp" "dbp" "age" "dbp" "sbp"
> x[samp2]
dbp sbp sbp dbp dbp age dbp sbp
78 148 148 78 78 54 78 148
Notice that when we indexed by position or name we indexed the same po-
sition repeatly. This will not work with logical vectors. In the example that
follows NA means “not available.”
> (samp3 <- sample(c(TRUE, FALSE), 8, replace = TRUE))
[1] FALSE FALSE TRUE FALSE TRUE TRUE TRUE TRUE
> x[samp3]
dbp <NA> <NA> <NA> <NA>
78 NA NA NA NA
We have already seen that a vector can be indexed based on the charac-
teristics of another vector.
> kid <- c("Tomasito", "Irene", "Luisito", "Angelita", "Tomas")
> age <- c(8, NA, 7, 4, NA)
> age<=7 # produces logical vector
[1] FALSE NA TRUE TRUE NA
> kid[age<=7] # index ’kid’ using ’age’
[1] NA "Luisito" "Angelita" NA
> kid[!is.na(age)] # remove missing values
[1] "Tomasito" "Luisito" "Angelita"
> kid[age<=7 & !is.na(age)]
[1] "Luisito" "Angelita"

In this example, NA represents missing data. The is.na function returns a

logical vector with TRUEs at NA positions. To generate a logical vector to
index values that are not missing use !is.na.
For practice study the examples in Table 2.6 on the previous page and
spend a few minutes creating, naming, and indexing simple vectors.

2.2.4.1 The which function

A Boolean operation that returns a logical vector contains TRUE values where
the condition is true. To identify the position of each TRUE value we use the
which function. For example, using the same data above:
> which(age<=7) # which positions meet condition

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.2 A vector is a collection of like elements 41

Table 2.7 Common ways of replacing vector elements

Replacing Try these examples
By position x <- c(chol = 234, sbp = 148, dbp = 78, age = 54)
x[1]
x[1] <- 250
x
By name (if exists) x["sbp"]
x["sbp"] <- 150
x
By logical x[x<100]
x[x<100] <- NA
x

[1] 3 4
> kid[which(age<=7)]
[1] "Luisito" "Angelita"
Notice that is was unnecessary to remove the missing values.

2.2.5 Replacing vector elements (by indexing and

assignment)

To replace vector elements we combine indexing and assignment. Any ele-

ments of a vector that can be indexed can be replaced. Replacing vector
elements is one method of recoding a variable.
> # simulate vector with 1000 age values
> age <- sample(0:100, 1000, replace = TRUE)
> mean(age)
[1] 50.378
> sd(age)
[1] 28.25947
> agecat <- age
> agecat[age<15] <- "<15"
> agecat[age>=15 & age<25] <- "15-24"
> agecat[age>=25 & age<45] <- "25-44"
> agecat[age>=45 & age<65] <- "45-64"
> agecat[age>=65] <- "65+"
> table(agecat)
agecat
<15 15-24 25-44 45-64 65+
125 107 207 206 355

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

42 2 Working with R data objects

Table 2.8 Selected operations on single vectors

Function Description Function Description
sum summation range range
cumsum cumulative sum rev reverse order
diff x[i+1]-x[i] order order
prod product sort sort
cumprod cumulative product rank rank
mean mean sample random sample
median median quantile percentile
min minimum var variance, covariance
max maximum sd standard deviation

First, we made a copy of the numeric vector age and named it agecat. Then,
we replaced elements of agecat with character strings for each age category,
creating a character vector.
For practice study the examples in Table 2.7 on the preceding page and
spend a few minutes replacing vector elements.

2.2.6 Operations on vectors

Operations on vectors is very common in epidemiology and statistics. In

this section we cover common operations on single vectors (Table 2.8) and
multiple vectors (Table 2.9 on page 46).

2.2.6.1 Operations on single vectors

First, we focus on operating on single numeric vectors (Table 2.8). This also
gives us the opportunity to see how common mathematical notation is trans-
lated into simple R code. Pn
To sum elements of a numeric vector x of length n, ( i=1 xi ), use the sum
function:
> # generate and sum 100 random standard normal values
> x <- rnorm(100)
> sum(x)
[1] -0.34744
Pk
To calculate a cumulative sum of a numeric vector x of length n, ( i=1 xi ,
for k = 1, . . . , n), use the cumsum function which returns a vector:
# generate sequence of 2’s and calculate cumulative sum
> x <- rep(2, 10)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.2 A vector is a collection of like elements 43

> x
[1] 2 2 2 2 2 2 2 2 2 2
> cumsum(x)
[1] 2 4 6 8 10 12 14 16 18 20
Qn
To multiply elements of a numeric vector x of length n, ( i=1 xi ), use the
prod function:
> x <- c(1, 2, 3, 4, 5, 6, 7, 8)
> prod(x)
[1] 40320

QTo
k
calculate the cumulative product of a numeric vector x of length n,
( i=1 xi , for k = 1, . . . , n), use the cumprod function:

> x <- c(1, 2, 3, 4, 5, 6, 7, 8)

> cumprod(x)
[1] 1 2 6 24 120 720
5040 40320
Pn
To calculate the mean of a numeric vector x of length n, ( n1 i=1 xi ), use
the sum and length functions, or use the mean function:
> x <- rnorm(100)
> sum(x)/length(x)
[1] 0.05843341
> mean(x)
[1] 0.05843341
To calculate the sample variance of a numeric vector x of length n, use
the sum, mean, and length functions, or, more directly, use the var function.
" n #
2 1 X
2
SX = (xi − x̄)
n − 1 i=1

> x <- rnorm(100)

> sum((x-mean(x))^2)/(length(x)-1)
[1] 0.9073808
> var(x) # equivalent
[1] 0.9073808
This example illustrates how we can implement a formula in R using several
functions that operate on single vectors (sum, mean, and length). The var
function, while available for convenience, is not necessary to calculate the
sample variance.
When the var function is applied to two numeric vectors, x and y, both
of length n, the sample covariance is calculated:
" n #
1 X
SXY = (xi − x̄)(yi − ȳ)
n − 1 i=1

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

44 2 Working with R data objects

> x <- rnorm(100); y <- rnorm(100)

> sum((x-mean(x))*(y-mean(y)))/(length(x)-1)
[1] -0.09552851
> var(x, y) # equivalent
[1] -0.09552851
The sample standard deviation, of course, is just the square root of the
sample variance (or use the sd function):
v " n #
u
u 1 X
SX = t (xi − x̄) 2
n − 1 i=1

> sqrt(var(x))
[1] 0.9525654
> sd(x)
[1] 0.9525654
To sort a numeric or character vector use the sort function.
> ages <- c(8, 4, 7)
> sort(ages)
[1] 4 7 8
However, to sort one vector based on the ordering of another vector use the
order function.
> ages <- c(8, 4, 7)
> subjects <- c("Tomas", "Angela", "Luis")
> subjects[order(ages)]
[1] "Angela" "Luis" "Toms"
> # ’order’ returns positional integers for sorting
> order(ages)
[1] 2 3 1
Notice that the order function does not return the data, but rather indexing
integers in new positions for sorting the vector age or another vector. For
example, order(ages) returned the integer vector c(2, 3, 1) which means
“move the 2nd element (age = 4) to the first position, move the 3rd element
(age = 7) to the second position, and move the 1st element (age = 8) to
the third position.” Verify that sort(ages) and ages[order(ages)] are
equivalent.
To sort a vector in reverse order combine the rev and sort functions.
> x <- c(12, 3, 14, 3, 5, 1)
> sort(x)
[1] 1 3 3 5 12 14
> rev(sort(x))
[1] 14 12 5 3 3 1

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.2 A vector is a collection of like elements 45

In contrast to the sort function, the rank function gives each element of a
vector a rank score but does not sort the vector.
> x <- c(12, 3, 14, 3, 5, 1)
> rank(x)
[1] 5.0 2.5 6.0 2.5 4.0 1.0
The median of a numeric vector is that value which puts 50% of the values
below and 50% of the values above, in other words, the 50% percentile (or 0.5
quantile). For example, the median of c(4, 3, 1, 2, 5) is 3. For a vector
of even length, the middle values are averaged: the median of c(4, 3, 1,
2) is 2.5. To get the median value of a numeric vector use the median or
quantile function.
> ages <- c(23, 45, 67, 33, 20, 77)
> median(ages)
[1] 39
> quantile(ages, 0.5)
50%
39
To return the minimum value of a vector use the min function; for the max-
imum value use the max function. To get both the minimum and maximum
values use the range function.
> ages <- c(23, 45, 67, 33, 20, 77)
> min(ages)
[1] 20
> sort(ages)[1] # equivalent
[1] 20
> max(ages)
[1] 77
> sort(ages)[length(ages)] # equivalent
[1] 77
> range(ages)
[1] 20 77
> c(min(ages), max(ages)) # equivalent
[1] 20 77
To sample from a vector of length n, with each element having a default
sampling probability of 1/n, use the sample function. Sampling can be with
or without replacement (default). If the sample size is greater than the length
of the vector, then sampling must occur with replacement.
> coin <- c("H", "T")
> sample(coin, size = 10, replace = TRUE)
[1] "H" "H" "T" "T" "T" "H" "H" "H" "H" "T"
> sample(1:100, 15)
[1] 9 24 53 11 15 63 52 73 54 84 82 66 65 20 67

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

46 2 Working with R data objects

Table 2.9 Selected operations on multiple vectors

Function Description
c concatenates vectors
append Appends a vector to another vector (default is to append at
the end of the first vector)
cbind Column-bind vectors or matrices
rbind Row-bind vectors or matrices
table Creates contingency table from 2 or more vectors
xtabs Creates contingency table from 2 or more factors in a data
frame
ftable Creates flat contingency table from 2 or more vectors
outer Outer product
tapply Applies a function to strata of a vector
<, >, Relational operators, See Table 2.3 on page 33
<=, >=,
==, !=
!, Logical operators, See Table 2.3 on page 33
&, &&,
|, ||, xor

2.2.6.2 Operations on multiple vectors

Next, we review selected functions that work with one or more vectors. Some
of these functions manipulate vectors and others facilitate numerical opera-
tions.
In addition to creating vectors, the c function can be used to append
vectors.
> x <- 6:10
> y <- 20:24
> c(x, y)
[1] 6 7 8 9 10 20 21 22 23 24
The append function also appends vectors; however, one can specify at which
position.
> append(x, y)
[1] 6 7 8 9 10 20 21 22 23 24
> append(x, y, after = 2)
[1] 6 7 20 21 22 23 24 8 9 10
In contrast, the cbind and rbind functions concatenate vectors into a
matrix. During the outbreak of severe acute respiratory syndrome (SARS)
in 2003, a patient with SARS potentially exposed 111 passengers on board
an airline flight. Of the 23 passengers that sat “close” to the index case, 8
developed SARS; among the 88 passengers that did not sit “close” to the

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.2 A vector is a collection of like elements 47

index case, only 10 developed SARS [2]. Now, we can bind 2 vectors to create
a 2 × 2 table (matrix).
> case <- c("exposed" = 8, "unexposed" = 10)
> noncase <- c("exposed" = 15, "unexposed" = 78)
> cbind(case, noncase)
case noncase
exposed 8 15
unexposed 10 78
> rbind(case, noncase)
exposed unexposed
case 8 10
noncase 15 78
For the example that follows, let’s recreate the SARS data as two character
vectors.
> outcome <- c(rep("case", 8+10), rep("noncase", 15+78))
> tmp <- c("exposed", "unexposed")
> exposure <- c(rep(tmp, c(8, 10)), rep(tmp, c(15, 78)))
> cbind(exposure,outcome)[1:4,] # display 4 rows
exposure outcome
[1,] "exposed" "case"
[2,] "exposed" "case"
[3,] "exposed" "case"
[4,] "exposed" "case"
Now, use the table function to cross-tabulate one or more vectors.
> table(outcome, exposure)
exposure
outcome exposed unexposed
case 8 10
noncase 15 78
The ftable function creates a flat contingency table from one or more
vectors.
> ftable(outcome, exposure)
exposure exposed unexposed
outcome
case 8 10
noncase 15 78
This will come in handy later when we want to display a 3 or more dimen-
sional table as a “flat” 2-dimensional table.
The outer function applies a function to every combination of elements
from two vectors. For example, create a multiplication table for the numbers
1 to 5.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

48 2 Working with R data objects

> outer(1:5, 1:5, "*")

[,1] [,2] [,3] [,4] [,5]
[1,] 1 2 3 4 5
[2,] 2 4 6 8 10
[3,] 3 6 9 12 15
[4,] 4 8 12 16 20
[5,] 5 10 15 20 25
The tapply function applies a function to strata of a vector that is defined
by one or more “indexing” vectors. For example, to calculate the mean age
of females and males:
> age <- c(23, 45, 67, 88, 22, 34, 80, 55, 21, 48)
> sex <- c("M", "F", "M", "F", "M", "F", "M", "F", "M", "F")
> tapply(X = age, INDEX = sex, FUN = mean)
F M
54.0 42.6
> # equivalent
> tapply(age, sex, sum)/tapply(age, sex, length)
F M
54.0 42.6
The tapply function is an important and versatile function because it allows
us to apply any function that can be applied to a vector, to be applied to
strata of a vector. Moveover, we can use our user-created functions as well.

2.3 A matrix is a 2-dimensional table of like elements

2.3.1 Understanding matrices

A matrix is a 2-dimensional table of like elements. Matrix elements can be

either numeric, character, or logical. Contingency tables in epidemiology are
represented in R as numeric matrices or arrays. An array is the generalization
of matrices to 3 or more dimensions (commonly known as stratified tables).
We cover arrays later, for now we will focus on 2-dimensional tables.
Consider the 2 × 2 table of crude data in Table 2.10 on the facing page
[3]. In this randomized clinical trial (RCT), diabetic subjects were randomly
assigned to receive either tolbutamide, an oral hypoglycemic drug, or placebo.
Because this was a prospective study we can calculate risks, odds, a risk ratio,
and an odds ratio. We will do this using R as a calculator.
> dat <- matrix(c(30, 174, 21, 184), 2, 2)
> rownames(dat) <- c("Deaths", "Survivors")
> colnames(dat) <- c("Tolbutamide", "Placebo")
> coltot <- apply(dat, 2, sum) #column totals

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.3 A matrix is a 2-dimensional table of like elements 49

Table 2.10 Deaths among subjects who received tolbutamide and placebo in the Unver-
sity Group Diabetes Program (1970)
Tolbutamide Placebo
Deaths 30 21
Survivors 174 184

> risks <- dat["Deaths",]/coltot

> risk.ratio <- risks/risks[2] #risk ratio
> odds <- risks/(1-risks)
> odds.ratio <- odds/odds[2] #odds ratio
> # display results
> dat
Tolbutamide Placebo
Deaths 30 21
Survivors 174 184
> rbind(risks, risk.ratio, odds, odds.ratio)
Tolbutamide Placebo
risks 0.1470588 0.1024390
risk.ratio 1.4355742 1.0000000
odds 0.1724138 0.1141304
odds.ratio 1.5106732 1.0000000
Now let’s review each line briefly to understand the analysis in more detail.
dat <- matrix(c(30, 174, 21, 184), 2, 2)
We used the matrix function to take a vector and convert it into a matrix
with 2 rows and 2 columns. Notice the matrix function reads in the vector
column-wise. To read the vector in row-wise we would add the byrow=TRUE
option. Try creating a matrix reading in a vector column-wise (default) and
row-wise.
rownames(dat) <- c("Deaths", "Survivors")
colnames(dat) <- c("Tolbutamide", "Placebo")
We used the rownames and the colnames functions to assign row and column
names to the matrix dat. The row names and the column names are both
character vectors.
coltot <- apply(dat, 2, sum) #column totals
We used the apply function to sum the columns; it is a versatile function
for applying any function to matrices or arrays. The second argument is the
MARGIN option: in this case, MARGIN=2, meaning apply the sum function to
the columns. To sum the rows, set MARGIN=1.
risks <- dat["Deaths",]/coltot
risk.ratio <- risks/risks[2] #risk ratio

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

50 2 Working with R data objects

We calculated the risks of death for each treatment group. We got the nu-
merator by indexing the dat matrix using the row name "Deaths". The
numerator is a vector containing the deaths for each group and the denom-
inator is the total number of subjects in each group. We calculated the risk
ratios using the placebo group as the reference.
odds <- risks/(1-risks)
odds.ratio <- odds/odds[2] #odds ratio
Using the definition of the odds, we calculated the odds of death for each
treatment group. Then we calculated the odds ratios using the placebo group
as the reference.
dat
rbind(risks, risk.ratio, odds, odds.ratio)
Finally, we display the dat table we created. We also created a table of results
by row binding the vectors using the rbind function.
In the sections that follow we will cover the necessary concepts to make
the previous analysis routine.

2.3.2 Creating matrices

There are several ways to create matrices (Table 2.11 on the next page). In
general, we create or use matrices in the following ways:
• Contingency tables (cross tabulations)
• Spreadsheet calculations and display
• Collecting results into tabular form
• Results of 2-variable equations

2.3.2.1 Contingency tables (cross tabulations)

In the previous section we used the matrix function to create the 2 × 2 table
for the UGDP clinical trial:
> dat <- matrix(c(30, 174, 21, 184), 2, 2)
> rownames(dat) <- c("Deaths", "Survivors")
> colnames(dat) <- c("Tolbutamide", "Placebo")
> dat
Tolbutamide Placebo
Deaths 30 21
Survivors 174 184
Alternatively, we can create a 2-way contingency table using the table func-
tion with fields from a data set;

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.3 A matrix is a 2-dimensional table of like elements 51

Table 2.11 Common ways of creating a matrix

Function Description Try these examples
cbind Column-bind vectors x <- 1:3
or matrices y <- 3:1
z <- cbind(x, y); z
rbind Row-bind vectors or z2 <- rbind(x, y); z2
matrices
matrix Generates matrix mtx <- matrix(1:4, nrow=2, ncol=2); mtx
dim Assign dimensions to a mtx2 <- 1:4; mtx2
data object dim(mtx2) <- c(2, 2); mtx2
array Generates matrix when mtx <- array(1:4, dim = c(2, 2)); mtx
array is 2-dimensional
table Creates contingency table(infert$educ, infert$case)
table
xtabs Create a contingency xtabs(~education + case, data = infert)
table using a formula
interface
ftable Creates flat ftable(infert$educ, infert$spont,
contingency table infert$case)
as.matrix Coerces object into a 1:3
matrix as.matrix(1:3)
outer Outer product of two outer(1:5, 1:5, "*")
vectors
x[row, , ] Indexing an array can x <- array(1:8, c(2, 2, 2))
x[ ,col, ] return a matrix x[1, , ]
x[ , ,dep] x[ ,1, ]
x[ , ,1]

> dat2 <- read.table("http://www.medepi.net/data/ugdp.txt",

+ header = TRUE, sep = ",")
> names(dat2) #display field names
[1] "Status" "Treatment" "Agegrp"
> table(dat2$Status, dat2$Treatment)

Placebo Tolbutamide
Deaths 21 30
Survivors 184 174
Alternatively, the xtabs function cross tabulates using a formula interface.
An advantage of this function is that the field names are included.
> xtabs(~Status + Treatment, data = dat2)
Treatment
Status Placebo Tolbutamide
Deaths 21 30
Survivors 184 174

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

52 2 Working with R data objects

Finally, a multi-dimensional contingency table can be presented as a 2-

dimensional flat contingency table using the ftable function. Here we stratify
the above table by the variable Agegrp.
> xtab3way <- xtabs(~Status + Treatment + Agegrp, data=dat2)
> xtab3way
, , Agegrp = <55

Treatment
Status Placebo Tolbutamide
Deaths 5 8
Survivors 115 98

, , Agegrp = 55+

Treatment
Status Placebo Tolbutamide
Deaths 16 22
Survivors 69 76

> ftable(xtab3way) #convert to flat table

Agegrp <55 55+
Status Treatment
Deaths Placebo 5 16
Tolbutamide 8 22
Survivors Placebo 115 69
Tolbutamide 98 76
> #alternative and more consistent with xtab3way
> ftable(xtabs(~Agegrp + Status + Treatment, data=dat2))
Treatment Placebo Tolbutamide
Agegrp Status
<55 Deaths 5 8
Survivors 115 98
55+ Deaths 16 22
Survivors 69 76

2.3.2.2 Spreadsheet calculations and display

Matrices are commonly used to display spreadsheet-like calculations. In fact,

a very efficient way to learn R is to use it as our spreadsheet. For exam-
ple, assuming the rate of seasonal influenza infection is 10 infections per 100
person-years, let’s calculate the individual cumulative risk of influenza infec-
tion at the end of 1, 5, and 10 years. Assuming no competing risk, we can
use the exponential formula:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.3 A matrix is a 2-dimensional table of like elements 53

R(0, t) = 1 − e−λt

where , λ = infection rate, and t = time.

> lamb <- 10/100
> years <- c(1, 5, 10)
> risk <- 1 - exp(-lamb*tim)
> cbind(rate = lamb, years, cumulative.risk = risk)
rate years cumulative.risk
[1,] 0.1 1 0.09516258
[2,] 0.1 5 0.39346934
[3,] 0.1 10 0.63212056
Therefore, the cumulative risk of influenza infection after 1, 5, and 10 years
is 9.5%, 39%, and 63%, respectively.

2.3.2.3 Collecting results into tabular form

A 2-way contingency table from the table or xtabs functions does not have
margin totals. However, we can construct a numeric matrix that includes the
totals. Using the UGDP data again,
> dat2 <- read.table("http://www.medepi.net/data/ugdp.txt",
+ header = TRUE, sep=",")
> tab2 <- xtabs(~Status + Treatment, data = dat2)
> rowt <- tab2[,1] + tab2[,2]
> tab2a <- cbind(tab2, Total = rowt)
> colt <- tab2a[1,] + tab2a[2,]
> tab2b <- rbind(tab2a, Total = colt)
> tab2b
Placebo Tolbutamide Total
Deaths 21 30 51
Survivors 184 174 358
Total 205 204 409
This table (tab2b) is primarily for display purposes.

2.3.2.4 Results of 2-variable equations

When we have an equation with 2 variables, we can use a matrix to display

the answers for every combination of values contained in both variables. For
example, consider this equation:

z = xy

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

54 2 Working with R data objects

And suppose x = {1, 2, 3, 4, 5} and y = {6, 7, 8, 9, 10}. Here’s the long way to
create a matrix for this equation:
> x <- 1:5; y <- 6:10
> z <- matrix(NA, 5, 5) #create empty matrix of missing values
> for(i in 1:5){
+ for(j in 1:5){
+ z[i, j] <- x[i]*y[j]
+ }
+ }
> rownames(z) <- x; colnames(z) <- y
> z
6 7 8 9 10
1 6 7 8 9 10
2 12 14 16 18 20
3 18 21 24 27 30
4 24 28 32 36 40
5 30 35 40 45 50
Okay, but the outer function is much better for this task:
> x <- 1:5; y <- 6:10
> z <- outer(x, y, "*")
> rownames(z) <- x; colnames(z) <- y
> z
6 7 8 9 10
1 6 7 8 9 10
2 12 14 16 18 20
3 18 21 24 27 30
4 24 28 32 36 40
5 30 35 40 45 50
In fact, the outer function can be used to calculate the “surface” for any
2-variable equation (more on this later).

2.3.3 Naming a matrix

We have already seen several examples of naming components of a matrix.

Table 2.12 on the facing page summarizes the common ways of naming matrix
components. The components of a matrix can be named at the time the
matrix is created, or they can be named later. For a matrix, we can provide
the row names, column names, and/or field names. For example, consider the
UGDP clinical trial 2 × 2 table:
> tab
Treatment

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.3 A matrix is a 2-dimensional table of like elements 55

Table 2.12 Common ways of naming a matrix

Function Try these examples
matrix #name rows and columns only
dat <- matrix(c(178, 79, 1411, 1486), 2, 2,
dimnames = list(c("Type A", "Type B"),
c("Yes", "No")))
dat
#name rows, columns, and fields
dat <- matrix(c(178, 79, 1411, 1486), 2, 2,
dimnames = list(Behavior = c("Type A", "Type B"),
"Heart attack" = c("Yes", "No")))
dat
rownames #name rows only
dat <- matrix(c(178, 79, 1411, 1486), 2, 2)
rownames(dat) <- c("Type A", "Type B")
dat
colnames #name columns only
dat <- matrix(c(178, 79, 1411, 1486), 2, 2)
colnames(dat) <- c("Yes", "No"); dat
dimnames #name rows and columns only
dat <- matrix(c(178, 79, 1411, 1486), 2, 2)
dimnames(dat) <- list(c("Type A", "Type B"),
c("Yes", "No"))
dat
#name rows, columns, and fields
dat <- matrix(c(178, 79, 1411, 1486), 2, 2)
dimnames(dat) <- list(Behavior = c("Type A", "Type B"),
"Heart attack" = c("Yes", "No"))
dat
names #name fields when row and column names already exist
dat <- matrix(c(178, 79, 1411, 1486), 2, 2,
dimnames = list(c("Type A", "Type B"),
c("Yes", "No")))
dat #display w/o field names
names(dimnames(dat)) <- c("Behavior", "Heart attack")
dat #display w/ field names

Outcome Tolbutamide Placebo

Deaths 30 21
Survivors 174 184
In the “Treatment” field, the possible values, “Tolbutamide” and “Placebo,”
are the column names. Similarly, in the “Outcome” field, the possible values,
“Deaths” and “Survivors,” are the row names.
To review, the components of matrix can be named at the time the matrix
is created:
> tab <- matrix(c(30, 174, 21, 184), 2, 2,
+ dimnames = list(Outcome = c("Deaths", "Survivors"),
+ Treatment = c("Tolbutamide", "Placebo")))

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

56 2 Working with R data objects

> tab
Treatment
Outcome Tolbutamide Placebo
Deaths 30 21
Survivors 174 184
If a matrix does not have field names, we can add them after the fact, but
we must use the names and dimnames functions together. Having field names
is necessary if the row and column names are not self-explanatory, as this
example illustrates.
> y <- matrix(c(30, 174, 21, 184), 2, 2)
> rownames(y) <- c("Yes", "No")
> colnames(y) <- c("Yes", "No")
> y #labels not informative
Yes No
Yes 30 21
No 174 184
> #add field names
> names(dimnames(y)) <- c("Death", "Tolbutamide")
> y
Tolbutamide
Death Yes No
Yes 30 21
No 174 184
Study and test the examples in Table 2.12 on the previous page.

2.3.4 Indexing a matrix

Similar to vectors, a matrix can be indexed by position, by name, or by

logical. Study and practice the examples in Table 2.13 on the facing page.
An important skill to master is indexing rows of a matrix using logical vectors.
Consider the following matrix of data, and suppose I want to select the rows
for subjects age less than 60 and systolic blood pressure less than 140.
> dat
age chol sbp
[1,] 45 145 124
[2,] 56 168 144
[3,] 73 240 150
[4,] 44 144 134
[5,] 65 210 112
> dat[,"age"]<60
[1] TRUE TRUE FALSE TRUE FALSE

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.3 A matrix is a 2-dimensional table of like elements 57

Table 2.13 Common ways of indexing a matrix

Indexing Try these examples
By position dat <- matrix(c(178, 79, 1411, 1486), 2, 2)
dimnames(dat) <- list(Behavior = c("Type A","Type B"),
"Heart attack" = c("Yes","No"))
dat[1, ]
dat[1,2]
dat[2, , drop = FALSE]
By name (if dat["Type A", ]
exists) dat["Type A", "Type B"]
dat["Type B", , drop = FALSE]
By logical dat[, 1] > 100
dat[dat[, 1] > 100, ]
dat[dat[, 1] > 100, , drop = FALSE]

> dat[,"sbp"]<140
[1] TRUE FALSE FALSE TRUE TRUE
> tmp <- dat[,"age"]<60 & dat[,"sbp"]<140
> tmp
[1] TRUE FALSE FALSE TRUE FALSE
> dat[tmp,]
age chol sbp
[1,] 45 145 124
[2,] 44 144 134
Notice that the tmp logical vector is the intersection of the logical vectors
separated by the logical operator &.

2.3.5 Replacing matrix elements

Remember, replacing matrix elements is just indexing plus assignment: any-

thing that can be indexed can be replaced. Study and practice the examples
in Table 2.14 on the next page.

2.3.6 Operations on a matrix

In epidemiology books, authors have preferences for displaying contingency

tables. Software packages have default displays for contingency tables. In
practice, we may need to manipulate a contingency table to facilitate further
analysis. Consider the following 2-way table:
> tab

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

58 2 Working with R data objects

Table 2.14 Common ways of replacing matrix elements

Replacing Try these examples
By position dat <- matrix(c(178, 79, 1411, 1486), 2, 2)
dimnames(dat) <- list(c("Type A","Type B"), c("Yes","No"))
dat[1, ] <- 99
dat
By name (if dat["Type A", ] <- c(178, 1411)
exists) dat
By logical qq <- dat[ ,1]<100
qq
dat[qq, ] <- 99
dat
dat[dat[ ,1]<100, ] <- c(79, 1486)
dat

Treatment
Outcome Tolbutamide Placebo
Deaths 30 21
Survivors 174 184
We can transpose the matrix using the t function.
> t(tab)
Outcome
Treatment Deaths Survivors
Tolbutamide 30 174
Placebo 21 184
We can reverse the order of the rows and/or columns.
> tab[2:1,] #reverse rows
Treatment
Outcome Tolbutamide Placebo
Survivors 174 184
Deaths 30 21
> tab[,2:1] #reverse columns
Treatment
Outcome Placebo Tolbutamide
Deaths 21 30
Survivors 184 174
> tab[2:1,2:1] #reverse rows and columns
Treatment
Outcome Placebo Tolbutamide
Survivors 184 174
Deaths 21 30

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.3 A matrix is a 2-dimensional table of like elements 59

Table 2.15 Common ways of operating on a matrix

Function Description Try these examples

t Transpose matrix dat #from Table 2.14 on the facing page
t(dat)
apply Apply a function to apply(X = dat, MARGIN = 2, FUN = sum)
the margins of a apply(dat, 1, FUN=sum)
matrix apply(dat, 1, mean)
apply(dat, 2, cumprod)
sweep Return an array rsum <- apply(dat, 1, sum)
obtained from an rdist <- sweep(dat, 1, rsum, "/")
input array by rdist
sweeping out a csum <- apply(dat, 2, sum)
summary statistic cdist <- sweep(dat, 2, csum, "/")
cdist

The following short-cuts use apply and/or sweep functions.

margin.table For a contingency margin.table(dat)

rowSums table in array form, margin.table(dat, 1)
colSums compute the sum of rowSums(dat) #equivalent
table entries for a apply(dat, 1, sum) #equivalent
given index. These margin.table(dat, 2)
functions are really colSums(dat) #equivalent to previous
just the apply apply(dat, 2, sum)
function using sum.
addmargins Calculate and display addmargins(dat)
marginal totals of a
matrix
rowMeans For a contingency rowSums(dat)
colMeans table in array form, apply(dat, 1, mean) #equivalent
compute the mean of colSums(dat)
table entries for a apply(dat, 2, mean) #equivalent
given index. These
functions are really
just the apply
function using mean.
prop.table Short cut that uses prop.table(dat)
the sweep and apply dat/sum(dat)
functions to get prop.table(dat, 1)
margin and joint sweep(dat, 1, apply(y, 1, sum), "/")
distributions prop.table(dat, 2)
sweep(y, 2, apply(y, 2, sum), "/")

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

60 2 Working with R data objects

2.3.6.1 The apply function

The apply function is an important and versatile function for conducting

operations on rows or columns of a matrix, including user-created functions.
The same functions that are used to conduct operations on single vectors
(Table 2.8 on page 42) can be applied to rows or columns of a matrix.
To calculate the row or column totals use the apply with the sum function:
> tab
Treatment
Outcome Tolbutamide Placebo
Deaths 30 21
Survivors 174 184
> apply(tab, 1, sum) #row totals
Deaths Survivors
51 358
> apply(tab, 2, sum) #column totals
Tolbutamide Placebo
204 205
These operations can be used to calculate marginal totals and have them
combined with the original table into one table.
> tab
Treatment
Outcome Tolbutamide Placebo
Deaths 30 21
Survivors 174 184
> rtot <- apply(tab, 1, sum) #row totals
> tab2 <- cbind(tab, Total = rtot)
> tab2
Tolbutamide Placebo Total
Deaths 30 21 51
Survivors 174 184 358
> ctot <- apply(tab2, 2, sum) #column totals
> rbind(tab2, Total = ctot)
Tolbutamide Placebo Total
Deaths 30 21 51
Survivors 174 184 358
Total 204 205 409
For convenience, R provides some functions for calculating marginal totals,
and calculating row or column means (margin.table, rowSums, colSums,
rowMeans, and colMeans). However, these functions just use the apply func-
tion4 .
Here’s an alternative method to calculate marginal totals:
4 More specifically, rowSums, colSums, rowMeans, and colMeans are optimized for speed.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.3 A matrix is a 2-dimensional table of like elements 61

> tab
Treatment
Outcome Tolbutamide Placebo
Deaths 30 21
Survivors 174 184
> tab2 <- cbind(tab, Total=rowSums(tab))
> rbind(tab2, Total=colSums(tab2))
Tolbutamide Placebo Total
Deaths 30 21 51
Survivors 174 184 358
Total 204 205 409
For convenience, the addmargins function calculates and displays the
marginals totals with the original data in one step.
> addmargins(tab)
Treatment
Outcome Tolbutamide Placebo Sum
Deaths 30 21 51
Survivors 174 184 358
Sum 204 205 409
The power of the apply function comes from our ability to pass many
functions (including our own) to it. For practice, combine the apply function
with functions from Table 2.8 on page 42 to conduct operations on rows and
columns of a matrix.

2.3.6.2 The sweep function

The sweep function is another important and versatile function for conduct-
ing operations across rows or columns of a matrix. This function “sweeps”
(operates on) a row or column of a matrix using some function and a value
(usually derived from the row or column values). To understand this, we con-
sider an example involving a single vector. For a given integer vector x, to
convert the values of x into proportions involves two steps:
> x <- c(1, 2, 3, 4, 5)
> sumx <- sum(x) #Step 1: summation
> propx <- x/sumx #Step 2: division (the "sweep")
> propx
[1] 0.066667 0.133333 0.200000 0.266667 0.333333
To apply this equivalent operation across rows or columns of a matrix requires
the sweep function.
For example, to calculate the row and column distributions of a 2-way
table we combine the apply (step 1) and the sweep (step 2) functions:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

62 2 Working with R data objects

> tab
Treatment
Outcome Tolbutamide Placebo
Deaths 30 21
Survivors 174 184
> rtot <- apply(tab, 1, sum) #row totals
> tab.rowdist <- sweep(tab, 1, rtot, "/")
> tab.rowdist
Treatment
Outcome Tolbutamide Placebo
Deaths 0.58824 0.41176
Survivors 0.48603 0.51397
> ctot <- apply(tab, 2, sum) #column totals
> tab.coldist <- sweep(tab, 2, ctot, "/")
> tab.coldist
Treatment
Outcome Tolbutamide Placebo
Deaths 0.14706 0.10244
Survivors 0.85294 0.89756
Because R is a true programming language, these can be combined into single
steps:
> sweep(tab, 1, apply(tab, 1, sum), "/") #row distribution
Treatment
Outcome Tolbutamide Placebo
Deaths 0.58824 0.41176
Survivors 0.48603 0.51397
> sweep(tab, 2, apply(tab, 2, sum), "/") #column distribution
Treatment
Outcome Tolbutamide Placebo
Deaths 0.14706 0.10244
Survivors 0.85294 0.89756
For convenience, R provides prop.table. However, this function just uses
the apply and sweep functions.

2.4 An array is a n-dimensional table of like elements

2.4.1 Understanding arrays

While a matrix is a 2-dimensional table of like elements, an array is the

generalization of matrices to n-dimensions. Stratified contingency tables in
epidemiology are represented as array data objects in R. For example, the ran-

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.4 An array is a n-dimensional table of like elements 63

Table 2.16 Deaths among subjects who received tolbutamide and placebo in the Unver-
sity Group Diabetes Program (1970), stratifying by age
Age<55 Age≥55 Combined
Tolbutamide Placebo Tolbutamide Placebo Tolbutamide Placebo
Deaths 8 5 22 16 30 21
Survivors 98 115 76 69 174 184
Total 106 120 98 85 204 205

domized clinical trial previously shown comparing the number deaths among
diabetic subjects that received tolbutamide vs. placebo is now also stratified
by age group (Table 2.16):
This is 3-dimensional array: outcome status vs. treatment status vs. age
group. Let’s see how we can represent this data in R.
> tdat <- c(8, 98, 5, 115, 22, 76, 16, 69)
> tdat <- array(tdat, c(2, 2, 2))
> dimnames(tdat) <- list(Outcome=c("Deaths", "Survivors"),
+ Treatment=c("Tolbutamide", "Placebo"),
+ "Age group"=c("Age<55", "Age>=55"))
> tdat
, , Age group = Age<55

Treatment
Outcome Tolbutamide Placebo
Deaths 8 5
Survivors 98 115

, , Age group = Age>=55

Treatment
Outcome Tolbutamide Placebo
Deaths 22 16
Survivors 76 69
R displays the first stratum (tdat[,,1]) then the second stratum (tdat[,,2]).
Our goal now is to understand how to generate and operate on these types of
arrays. Before we can do this we need to thoroughly understand the structure
of arrays.
Let’s study a 4-dimensional array. Displayed in Table 2.17 on the following
page is the year 2000 population estimates for Alameda and San Francisco
Counties by age, ethnicity, and sex. The first dimension is age category, the
second dimension is ethnicity, the third dimension is sex, and the fourth
dimension is county. Learning how to visualize this 4-dimensional sturcture
in R will enable us to visualize arrays of any number of dimensions.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

64 2 Working with R data objects

Table 2.17 Example of 4-dimensional array: Year 2000 population estimates by age,
ethnicity, sex, and county
Ethnicity
County/Sex Age White AfrAmer AsianPI Latino Multirace AmerInd
Alameda
Female <=19 58160 31765 40653 49738 10120 839
20–44 112326 44437 72923 58553 7658 1401
45–64 82205 24948 33236 18534 2922 822
65+ 49762 12834 16004 7548 1014 246
Male <=19 61446 32277 42922 53097 10102 828
20–44 115745 36976 69053 69233 6795 1263
45–64 81332 20737 29841 17402 2506 687
65+ 33994 8087 11855 5416 711 156
San Francisco
Female <=19 14355 6986 23265 13251 2940 173
20–44 85766 10284 52479 23458 3656 526
45–64 35617 6890 31478 9184 1144 282
65+ 27215 5172 23044 5773 554 121
Male <=19 14881 6959 24541 14480 2851 165
20–44 105798 11111 48379 31605 3766 782
45–64 43694 7352 26404 8674 1220 354
65+ 20072 3329 17190 3428 450 76

Fig. 2.2 Schematic representation of a 4-dimensional array (Year 2000 population esti-
mates by age, race, sex, and county)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.4 An array is a n-dimensional table of like elements 65

Fig. 2.3 Schematic representation of a theoretical 5-dimensional array (possibly popula-

tion estimates by age (1), race (2), sex (3), party affiliation (4), and state (5)). From this
diagram, we can infer that the field “state” has 3 levels, and the field “party affiliation” has
2 levels; however, it is not apparent how many age levels, race levels, and sex levels have
been created. Although not displayed, age levels would be represented by row names (along
1st dimension), race levels would be represented by column names (along 2nd dimension),
and sex levels would be represented by depth names (along 3rd dimension).

Displayed in Figure 2.4.1 on the preceding page is a schematic represen-

tation of the 4-dimensional array of population estimates in Table 2.17 on
the facing page. The left cube represents the population estimates by age,
race, and sex (dimensions 1, 2, and 3) for Alameda County (first component
of dimension 4). The right cube represents the population estimates by age,
race, and sex (dimensions 1, 2, and 3) for San Francisco County (second
component of dimension 4). We see, then, that it is possible to visualize data
arrays in more than three dimensions.
To convince ourselves further, displayed in Figure 2.4.1 is a theorecti-
cal 5-dimensional data array. Suppose this 5-D array contained data on
age (“Young”, “Old”), ethnicity (“White”, “Nonwhite”), sex (“Male”, “Fe-
male”), party affiliation (“Democrat”, “Republican”), and state (“Califor-
nia”, “Washington State”, “Florida”). For practice, using fictitious data, try
the following R code and study the output:
tab5 <- array(1:48, dim = c(2,2,2,2,3))
dn1 <- c("Young", "Old")
dn2 <- c("White", "Nonwhite")
dn3 <- c("Male", "Female")
dn4 <- c("Democrat", "Republican")
dn5 <- c("California", "Washington State", "Florida")
dimnames(tab5) <- list(Age=dn1, Race=dn2, Sex=dn3, Party=dn4,
State=dn5)
tab5

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

66 2 Working with R data objects

Table 2.18 Common ways of creating arrays

Function Description Try these examples
array Reshapes vector into an aa <- array(1:12, dim = c(2, 3, 2))
array aa
table Creates n-dimensional data(infert) # load infert data set
contingency table from table(infert$educ, infert$spont,
n vectors infert$case)
xtabs Creates a contingency xtabs(~education + case + parity,
table from data = infert)
cross-classifying factors
contained in a data
frame using a formula
interface
as.table Creates n-dimensional ft <- ftable(infert$educ, infert$spont,
contingency table from infert$case)
n-dimensional ftable ft
as.table(ft)
dim Assign dimensions to a x <- 1:12
data object x
dim(x) <- c(2, 3, 2)
x

2.4.2 Creating arrays

In R, arrays are most often produced with the array, table, or xtabs func-
tions (Table 2.18). As in the previous example, the array function works
much like the matrix function except the array function can specify 1 or
more dimensions, and the matrix function only works with 2 dimensions.
> array(1, dim = 1)
[1] 1
> array(1, dim = c(1, 1))
[,1]
[1,] 1
> array(1, dim = c(1, 1, 1))
, , 1

[,1]
[1,] 1
The table function cross tabulates 2 or more categorical vectors: character
vectors or factors. In R, categorical data are represented as factors (more
on this later). In contrast, using a formula interface, the xtabs function
cross tabulates 2 or more factors from a data frame. Additionally, the xtabs
function includes field names (which is highly preferred). For illustration, we
will cross tabulate character vectors.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.4 An array is a n-dimensional table of like elements 67

> #read in data "as is" (no factors created)

> udat1 <- read.csv("http://www.medepi.net/data/ugdp.txt",
+ as.is = TRUE)
> str(udat1)
’data.frame’: 409 obs. of 3 variables:
$ Status : chr "Death" "Death" "Death" "Death" ...
$ Treatment: chr "Tolbutamide" "Tolbutamide" "Tolbutamide" ...
$ Agegrp : chr "<55" "<55" "<55" "<55" ...
> table(udat1$Status, udat1$Treatment, udat1$Agegrp)
, , = 55+

Placebo Tolbutamide
Death 16 22
Survivor 69 76

, , = <55

Placebo Tolbutamide
Death 5 8
Survivor 115 98
The xtabs function will not work on character vectors.
By default, R converts character fields into factors. With factors, both the
table and xtabs functions cross tabulate the fields.
> #read in data and convert character vectors to factors
> udat2 <- read.csv("http://www.medepi.net/data/ugdp.txt")
> str(udat2)
’data.frame’: 409 obs. of 3 variables:
$ Status : Factor w/ 2 levels "Death","Survivor": 1 1 1 1 ...
$ Treatment: Factor w/ 2 levels "Placebo","Tolbutamide": 2 2 ...
$ Agegrp : Factor w/ 2 levels "55+","<55": 2 2 2 2 2 2 2 ...
> table(udat2$Status, udat1$Treatment, udat1$Agegrp)
, , = 55+

Placebo Tolbutamide
Death 16 22
Survivor 69 76

, , = <55

Placebo Tolbutamide
Death 5 8
Survivor 115 98

> xtabs(~Status + Treatment + Agegrp, data = udat2)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

68 2 Working with R data objects

, , Agegrp = 55+

Treatment
Status Placebo Tolbutamide
Death 16 22
Survivor 69 76

, , Agegrp = <55

Treatment
Status Placebo Tolbutamide
Death 5 8
Survivor 115 98
Notice that the xtabs function above included the field names. Field names
can be added manually with the table functions:
> table(Outcome = udat2$Status, Therapy = udat1$Treatment,
+ Age = udat1$Agegrp)
, , Age = 55+

Therapy
Outcome Placebo Tolbutamide
Death 16 22
Survivor 69 76

, , Age = <55

Therapy
Outcome Placebo Tolbutamide
Death 5 8
Survivor 115 98
Recall that the ftable function creates a flat contingency from categorical
vectors. The as.table function converts the flat contingency table back into
a multidimensional array.
> ftab <- ftable(udat2$Agegrp, udat1$Treatment, udat1$Status)
> ftab
Death Survivor

<55 Placebo 5 115

Tolbutamide 8 98
55+ Placebo 16 69
Tolbutamide 22 76
> as.table(ftab)
, , = Death

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.4 An array is a n-dimensional table of like elements 69

Table 2.19 Common ways of naming arrays

Function Try these examples
array # name components at time array is created x <- c(140, 11, 280,
56, 207, 9, 275, 32)
# create labels for values of each dimension
rn <- c(">=1 cups per day", "0 cups per day")
cn <- c("Cases", "Controls")
dn <- c("Females", "Males")
x <- array(x, dim = c(2, 2, 2), dimnames = list(Coffee = rn,
Outcome = cn, Gender = dn))
x
dimnames x <- c(140, 11, 280, 56, 207, 9, 275, 32)
# create labels for values of each dimension
rn <- c(">=1 cups per day", "0 cups per day")
cn <- c("Cases", "Controls")
dn <- c("Females", "Males")
x <- array(x, dim = c(2, 2, 2))
dimnames(x) <- list(Coffee = rn, Outcome = cn, Gender = dn)
x
names x <- c(140, 11, 280, 56, 207, 9, 275, 32)
# create labels for values of each dimension
rn <- c(">=1 cups per day", "0 cups per day")
cn <- c("Cases", "Controls")
dn <- c("Females", "Males")
x <- array(x, dim = c(2, 2, 2))
dimnames(x) <- list(rn, cn, dn)
x # display w/o field names
names(dimnames(x)) <- c("Coffee", "Case status", "Sex")
x # display w/ field names

Placebo Tolbutamide
<55 5 8
55+ 16 22

, , = Survivor

Placebo Tolbutamide
<55 115 98
55+ 69 76

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

70 2 Working with R data objects

Table 2.20 Common ways of indexing arrays

Indexing Try these examples
By position # use x from Table 2.19
x[1, , ]
x[ ,2, ]
By name (if exists) x[ , ,"Males"]
x[ ,"Controls","Females"]
By logical vector zz <- x[ ,1,1]>50
zz
x[zz, , ]

Table 2.21 Common ways of replacing array elements

Replacing Try these examples
By position # use x from Table 2.19
x[1, 1, 1] <- NA
x
By name (if exists) x[ ,"Controls","Females"] <-
99
x
By logical x>200
x[x>200] <- 999
x

2.4.3 Naming arrays

Naming components of an array is an extension of naming components of

a matrix (Table 2.12 on page 55). Study and implement the examples in
Table 2.19 on the preceding page.

2.4.4 Indexing arrays

Indexing an array is an extension of indexing a matrix (Table 2.13 on page 57).

Study and implement the examples in Table 2.20.

2.4.5 Replacing array elements

Replacing elements of an array is an extension of replacing elements of a

matrix (Table 2.14 on page 58). Study and implement the examples in Ta-
ble 2.21.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.4 An array is a n-dimensional table of like elements 71

Table 2.22 Common ways of operating on an array

Function Description Try these examples

aperm Transpose an array by x <- array(1:24, c(2, 3, 2, 2))
permuting its x
dimensions and aperm(x, c(3, 2, 1, 4))
optionally resizing it
apply Apply a function to the apply(x, 1, sum)
margins of an array apply(x, c(2, 3), sum)
apply(x, c(1, 2, 4), sum)
sweep Return an array zz <- apply(x, c(1, 2), sum)
obtained from an input sweep(x, c(1, 2), zz, "/")
array by sweeping out a
summary statistic
The following short-cuts use apply and/or sweep functions
margin.table For a contingency table margin.table(x); sum(x)
in array form, compute margin.table(x, c(1, 2))
the sum of table entries apply(x, c(1, 2), sum) #equiv
for a given index margin.table(x, c(1, 2, 4))
apply(x, c(1, 2, 4), sum) #equiv
rowSums Sum across rows of an rowSums(x) # dims = 1
array apply(x, 1, sum) #equiv
rowSums(x, dims = 2)
apply(x, c(1, 2), sum) #equiv
colSums Sum down columns of colSums(x) # dims = 1
an array apply(x, c(2, 3, 4), sum) #equiv
colSums(x, dims = 2)
apply(x, c(3, 4), sum) #equiv
addmargins Calculate and display addmargins(x)
marginal totals of an
array
rowMeans Calculate means across rowMeans(x) # dims = 1
rows of an array apply(x, 1, mean) #equiv
rowMeans(x, dims = 2)
apply(x, c(1, 2), mean) #equiv
colMeans Calculate means down colMeans(x) # dims = 1
columns of an array apply(x, c(2, 3, 4), mean) #equiv
colMeans(x, dims = 2)
apply(x, c(3, 4), mean) #equiv
prop.table Generates distribution prop.table(x)
for dimensions that are prop.table(x, margin = 1)
summed in the prop.table(x, c(1, 2))
margin.table function prop.table(x, c(1, 2, 3))

2.4.6 Operations on arrays

With the exception of the aperm function, operating on an array (Table 2.22)
is an extension of operating on a matrix (Table 2.15 on page 59). Consider

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

72 2 Working with R data objects

Table 2.23 Example of 3-dimensional array with marginal totals: Primary and secondary
syphilis morbidity by age, race, and sex, United State, 1989
Ethnicity
Age (years) Sex White Black Other Total
≤ 14 Male 2 31 7 40
Female 14 165 11 190
Total 16 196 18 230
15-19 Male 88 1412 210 1710
Female 253 2257 158 2668
Total 341 3669 368 4378
20-24 Male 407 4059 654 5120
Female 475 4503 307 5285
Total 882 8562 961 10405
25-29 Male 550 4121 633 5304
Female 433 3590 283 4306
Total 983 7711 916 9610
30-34 Male 564 4453 520 5537
Female 316 2628 167 3111
Total 880 7081 687 8648
35-44 Male 654 3858 492 5004
Female 243 1505 149 1897
Total 897 5363 641 6901
45-54 Male 323 1619 202 2144
Female 55 392 40 487
Total 378 2011 242 2631
55+ Male 216 823 108 1147
Female 24 92 15 131
Total 240 915 123 1278

Total (all ages) Male 2804 20376 2826 26006

Female 1813 15132 1130 18075

Total 4617 35508 3956 44081

the number of primary and secondary syphilis cases in the United State,
1989, stratified by sex, ethnicity, and age (Table 2.23). This table contains
the marginal and joint distribution of cases. Let’s read in the original data
and reproduce the table results.
> sdat3 <- read.csv("http://www.medepi.net/data/syphilis89c.txt")
> str(sdat3)
‘data.frame’: 44081 obs. of 3 variables:
$ Sex : Factor w/ 2 levels "Male","Female": 1 1 1 1 1 1 1 1 ...
$ Race: Factor w/ 3 levels "White","Black",..: 1 1 1 1 1 1 ...

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.4 An array is a n-dimensional table of like elements 73

$ Age : Factor w/ 8 levels "<=14","15-19",..: 1 1 2 2 2 2 2 ...

> sdat3[1:5,] #display first 5 lines
Sex Race Age
1 Male White <=14
2 Male White <=14
3 Male White 15-19
4 Male White 15-19
5 Male White 15-19
> sdat <- xtabs(~Sex+Race+Age, data=sdat3) #create array
> sdat
, , Age = <=14

Race
Sex White Black Other
Male 2 31 7
Female 14 165 11

, , Age = 15-19

Race
Sex White Black Other
Male 88 1412 210
Female 253 2257 158

, , Age = 20-24

Race
Sex White Black Other
Male 407 4059 654
Female 475 4503 307

, , Age = 25-29

Race
Sex White Black Other
Male 550 4121 633
Female 433 3590 283

, , Age = 30-34

Race
Sex White Black Other
Male 564 4453 520
Female 316 2628 167

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

74 2 Working with R data objects

, , Age = 35-44

Race
Sex White Black Other
Male 654 3858 492
Female 243 1505 149

, , Age = 45-54

Race
Sex White Black Other
Male 323 1619 202
Female 55 392 40

, , Age = 55+

Race
Sex White Black Other
Male 216 823 108
Female 24 92 15
To get marginal totals for one dimension, use the apply function and
specify the dimension for stratifying the results.
> sum(sdat) #total
[1] 44081
> apply(X = sdat, MARGIN = 1, FUN = sum) #by sex
Male Female
26006 18075
> apply(sdat, 2, sum) #by race
White Black Other
4617 35508 3956
> apply(sdat, 3, sum) #by age
<=14 15-19 20-24 25-29 30-34 35-44 45-54 55+
230 4378 10405 9610 8648 6901 2631 1278
To get the joint marginal totals for 2 or more dimensions, use the apply
function and specify the dimensions for stratifying the results. This means
that the function that is passed to apply is applied across the other, non-
stratified dimensions.
> apply(sdat, c(1, 2), sum) #by sex and race
Race
Sex White Black Other
Male 2804 20376 2826
Female 1813 15132 1130
> apply(sdat, c(1, 3), sum) #by sex and age

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.4 An array is a n-dimensional table of like elements 75

Age
Sex <=14 15-19 20-24 25-29 30-34 35-44 45-54 55+
Male 40 1710 5120 5304 5537 5004 2144 1147
Female 190 2668 5285 4306 3111 1897 487 131
> apply(sdat, c(3, 2), sum) #by age and race
Race
Age White Black Other
<=14 16 196 18
15-19 341 3669 368
20-24 882 8562 961
25-29 983 7711 916
30-34 880 7081 687
35-44 897 5363 641
45-54 378 2011 242
55+ 240 915 123
In R, arrays are displayed by the 1st and 2nd dimensions, stratified by the
remaining dimensions. To change the order of the dimensions, and hence the
display, use the aperm function. For example, the syphilis case data is most
efficiently displayed when it is stratified by race, age, and sex:
> sdat.ras <- aperm(sdat, c(2, 3, 1))
> sdat.ras
, , Sex = Male

Age
Race <=14 15-19 20-24 25-29 30-34 35-44 45-54 55+
White 2 88 407 550 564 654 323 216
Black 31 1412 4059 4121 4453 3858 1619 823
Other 7 210 654 633 520 492 202 108

, , Sex = Female

Age
Race <=14 15-19 20-24 25-29 30-34 35-44 45-54 55+
White 14 253 475 433 316 243 55 24
Black 165 2257 4503 3590 2628 1505 392 92
Other 11 158 307 283 167 149 40 15
Another method for changing the display of an array is to convert it into
a flat contingency table using the ftable function. For example, to display
Table 2.23 on page 72 as a flat contingency table in R (but without the
marginal totals), we use the following code:
> sdat.asr <- aperm(sdat, c(3,1,2)) #rearrange to age, sex, race
> ftable(sdat.asr) #convert 2-D flat table
Race White Black Other

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

76 2 Working with R data objects

Age Sex
<=14 Male 2 31 7
Female 14 165 11
15-19 Male 88 1412 210
Female 253 2257 158
20-24 Male 407 4059 654
Female 475 4503 307
25-29 Male 550 4121 633
Female 433 3590 283
30-34 Male 564 4453 520
Female 316 2628 167
35-44 Male 654 3858 492
Female 243 1505 149
45-54 Male 323 1619 202
Female 55 392 40
55+ Male 216 823 108
Female 24 92 15
This ftable object can be treated as a matrix, but it cannot be transposed.
Notice that we can combine the ftable with addmargins:
> ftable(addmargins(sdat.asr))
Race Black Other White Sum
Age Sex
15-19 Female 2257 158 253 2668
Male 1412 210 88 1710
Sum 3669 368 341 4378
20-24 Female 4503 307 475 5285
Male 4059 654 407 5120
Sum 8562 961 882 10405
25-29 Female 3590 283 433 4306
Male 4121 633 550 5304
Sum 7711 916 983 9610
30-34 Female 2628 167 316 3111
Male 4453 520 564 5537
Sum 7081 687 880 8648
35-44 Female 1505 149 243 1897
Male 3858 492 654 5004
Sum 5363 641 897 6901
45-54 Female 392 40 55 487
Male 1619 202 323 2144
Sum 2011 242 378 2631
<=14 Female 165 11 14 190
Male 31 7 2 40
Sum 196 18 16 230
>55 Female 92 15 24 131

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.5 A list is a collection of like or unlike data objects 77

Male 823 108 216 1147

Sum 915 123 240 1278
Sum Female 15132 1130 1813 18075
Male 20376 2826 2804 26006
Sum 35508 3956 4617 44081
To share the U.S. syphilis data in a universal format, we could create a text
file with the data in a tabular form. However, the original, individual-level
data set has over 40,000 observations. Instead, it would be more convenient
to create a group-level, tabular data set using the as.data.frame function
on the data array object.
> sdat.df <- as.data.frame(sdat)
> str(sdat.df)
‘data.frame’: 48 obs. of 4 variables:
$ Sex : Factor w/ 2 levels "Male","Female": 1 2 1 2 1 2 1 2 1 2 ...
$ Race: Factor w/ 3 levels "White","Black",..: 1 1 2 2 3 3 1 1 2 2 ...
$ Age : Factor w/ 8 levels "<=14","15-19",..: 1 1 1 1 1 1 2 2 2 2 ...
$ Freq: num 2 14 31 165 7 ...
> sdat.df[1:8,]
Sex Race Age Freq
1 Male White <=14 2
2 Female White <=14 14
3 Male Black <=14 31
4 Female Black <=14 165
5 Male Other <=14 7
6 Female Other <=14 11
7 Male White 15-19 88
8 Female White 15-19 253
For additional practice, study and implement the examples in Table 2.22
on page 71.

2.5 A list is a collection of like or unlike data objects

2.5.1 Understanding lists

Up to now, we have been working with atomic data objects (vector, ma-
trix, array). In contrast, lists, data frames, and functions are recursive data
objects. Recursive data objects have more flexibility in combining diverse
data objects into one object. A list provides the most flexibility. Think of a
list object as a collection of “bins” that can contain any R object (see Fig-
ure 2.5.1 on the following page). Lists are very useful for collecting results

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

78 2 Working with R data objects

Fig. 2.4 Schematic representation of a list of length four. The first bin [1] contains a
smiling face [[1]], the second bin [2] contains a flower [[2]], the third bin [3] contains
a lightning bolt [[3]], and the fourth bin [[4]] contains a heart [[4]]. When indexing a
list object, single brackets [·] indexes the bin, and double brackets [[·]] indexes the bin
contents. If the bin has a name, then $name also indexes the contents.

of an analysis or a function into one data object where all its contents are
readily accessible by indexing.
For example, using the UGDP clinical trial data, suppose we perform
Fisher’s exact test for testing the null hypothesis of independence of rows
and columns in a contingency table with fixed marginals.
> udat <- read.csv("http://www.medepi.net/data/ugdp.txt")
> tab <- table(udat$Status, udat$Treatment)[,2:1]
> tab

Tolbutamide Placebo
Death 30 21
Survivor 174 184
> ftab <- fisher.test(tab)
> ftab

Fisher’s Exact Test for Count Data

data: tab
p-value = 0.1813
alternative hypothesis: true odds ratio is not equal to 1
95 percent confidence interval:
0.80138 2.88729
sample estimates:
odds ratio
1.5091
The default display only shows partial results. The total results are stored
in the object ftab. Let’s evaluate the structure of ftab and extract some
results:
> str(ftab)
List of 7

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.5 A list is a collection of like or unlike data objects 79

$ p.value : num 0.181

$ conf.int : atomic [1:2] 0.801 2.887
..- attr(*, "conf.level")= num 0.95
$ estimate : Named num 1.51
..- attr(*, "names")= chr "odds ratio"
$ null.value : Named num 1
..- attr(*, "names")= chr "odds ratio"
$ alternative: chr "two.sided"
$ method : chr "Fisher’s Exact Test for Count Data"
$ data.name : chr "tab"
- attr(*, "class")= chr "htest"
> ftab$estimate
odds ratio
1.5091
> ftab$conf.int
[1] 0.80138 2.88729
> ftab$conf.int[2]
[1] 2.887286
attr(,"conf.level")
[1] 0.95
> ftab$p.value
[1] 0.18126
Using the str function to evaluate the structure of an output object is a
common method employed to extract additional results for display or further
analysis. In this case, ftab was a list with 7 bins, each with a name.

2.5.2 Creating lists

To create a list directly, use the list function. A list is a convenient method
to save results in our customized functions. For example, here’s a function to
calculate an odds ratio from a 2 × 2 table:
orcalc <- function(x){
or <- (x[1,1]*x[2,2])/(x[1,2]*x[2,1])
pval <- fisher.test(x)$p.value
list(data = x, odds.ratio = or, p.value = pval)
}
The orcalc function has been loaded in R, and now we run the function on
the UGDP data.
> tab #display 2x2 table

Tolbutamide Placebo

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

80 2 Working with R data objects

Table 2.24 Common ways of creating a list

Function Description Try these examples
list Creates list x <- 1:3
object y <- matrix(c("a","c","b","d"), 2,2)
z <- c("Pedro", "Paulo", "Maria")
mm <- list(x, y, z)
mm
data.frame List in tabular x <- data.frame(id=1:3,sex=c("M","F","T"))
format where x
each “bin” has a mode(x)
vector of same class(x)
length
as.data.frame Coerces data x <- matrix(1:6, 2, 3)
object into a x
data frame y <- as.data.frame(x)
y
read.table Reads ASCII wcgs <- read.table(".../wcgs.txt",
read.csv text file into header=TRUE, sep=",")
read.delim data frame str(wcgs)
read.fmf object1
vector Creates empty vector("list", 2)
list of length n
as.list Coercion into list(1:2) # compare to as.list
list object as.list(1:2)
1. Try read.table("http://www.medepi.net/data/wcgs.txt", header=TRUE, sep=",")

Death 30 21
Survivor 174 184
> orcalc(tab) #run function
$data

Tolbutamide Placebo
Death 30 21
Survivor 174 184

$odds.ratio
[1] 1.5107

$p.value
[1] 0.18126
For additional practice, study and implement the examples in Table 2.24.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.5 A list is a collection of like or unlike data objects 81

Table 2.25 Common ways of naming lists

Function Try these examples
names z <- list(rnorm(20), "Luis", 1:3)
z
# name after creation of list
names(z) <- c("bin1", "bin2", "bin3")
z
# name at creation of list
z <- list(bin1 = rnorm(20), bin2 = "Luis", bin3 = 1:3)
z
# without assignment returns character vector
names(z)

Table 2.26 Common ways of indexing lists

Indexing Try these examples
By position z <- list(bin1 = rnorm(20), bin2 = "Luis", bin3 = 1:3)
z[1] # indexes "bin" #1
z[[1]] # indexes contents of "bin" #1
By name (if z$bin1
exists) z$bin2
Indexing by num <- sapply(z, is.numeric)
logical vector num
z[num]

2.5.3 Naming lists

Components (bins) of a list can be unnamed or named. Components of a list

can be named at the time the list is created or later using the names function.
For practice, try the examples in Table 2.25.

2.5.4 Indexing lists

If list components (bins) are unnamed, we can index the list by bin position
with single or double brackets. The single brackets [·] indexes one or more
bins, and the double brackets indexes contents of single bins only.
> mylist1 <- list(1:5, matrix(1:4,2,2), c("Juan Nieve", "Guillermo Farro"))
> mylist1[c(1, 3)] #index bins 1 and 3
[[1]]
[1] 1 2 3 4 5

[[2]]
[1] "Juan Nieve" "Guillermo Farro"

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

82 2 Working with R data objects

> mylist1[[3]] #index contents of 3rd bin

[1] "Juan Nieve" "Guillermo Farro"
When list bins are named, we can index the bin contents by name. Using
the matched case-control study infert data set, we will conduct a conditional
logistic regression analysis to determine if spontaneous and induced abortions
are independently associated with infertility. For this we’ll need to load the
survival package which contains the clogit function.
> data(infert)
> library(survival)
> mod1 <- clogit(case ~ spontaneous + induced + strata(stratum),
+ data = infert)
> mod1 #default display
Call:
clogit(case ~ spontaneous + induced + strata(stratum), data = infert)

coef exp(coef) se(coef) z p

spontaneous 1.99 7.29 0.352 5.63 1.8e-08
induced 1.41 4.09 0.361 3.91 9.4e-05

Likelihood ratio test=53.1 on 2 df, p=2.87e-12 n= 248

> str(mod1) #evaluate structure
List of 17
$ coefficients : Named num [1:2] 1.99 1.41
..- attr(*, "names")= chr [1:2] "spontaneous" "induced"
$ var : num [1:2, 1:2] 0.1242 0.0927 0.0927 0.1301
$ loglik : num [1:2] -90.8 -64.2
$ score : num 48.4
$ iter : int 5
...
> names(mod1) #names of list components
[1] "coefficients" "var" "loglik"
[4] "score" "iter" "linear.predictors"
[7] "residuals" "means" "method"
[10] "n" "terms" "assign"
[13] "wald.test" "y" "formula"
[16] "call" "userCall"
> mod1$coeff
spontaneous induced
1.9859 1.4090
The results from str(mod1) are only partially displayed. Sometimes it is
more convenience to display the names for the list rather than the complete
structure. Additionally, the summary function applied to a regression model

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.5 A list is a collection of like or unlike data objects 83

object creates a list object with more detailed results. This too has a default
display, or we can index list components by name.
> summod1 <- summary(mod1)
> summod1 #default display of more detailed results
Call:
coxph(formula = Surv(rep(1, 248), case) ~ spontaneous +
induced + strata(stratum), data = infert, method = "exact")

n= 248
coef exp(coef) se(coef) z p
spontaneous 1.99 7.29 0.352 5.63 1.8e-08
induced 1.41 4.09 0.361 3.91 9.4e-05

exp(coef) exp(-coef) lower .95 upper .95

spontaneous 7.29 0.137 3.65 14.5
induced 4.09 0.244 2.02 8.3

Rsquare= 0.193 (max possible= 0.519 )

Likelihood ratio test= 53.1 on 2 df, p=2.87e-12
Wald test = 31.8 on 2 df, p=1.22e-07
Score (logrank) test = 48.4 on 2 df, p=3.03e-11

> names(summod1) #names of list components

[1] "call" "fail" "na.action" "n" "icc"
[6] "coef" "conf.int" "logtest" "sctest" "rsq"
[11] "waldtest" "used.robust"
> summod1$coef
coef exp(coef) se(coef) z p
spontaneous 1.9859 7.2854 0.35244 5.6346 1.8e-08
induced 1.4090 4.0919 0.36071 3.9062 9.4e-05

2.5.5 Replacing lists components

Replacing list components is accomplished by combining indexing with as-

signment. And of course, we can index by position, name, or logical. Remem-
ber, if it can be indexed, it can be replaced. Study and practice the examples
in Table 2.27 on the next page.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

84 2 Working with R data objects

Table 2.27 Common ways of replacing list components

Replacing Try these examples
By position z <- list(bin1 = rnorm(20), bin2 = "Luis", bin3 = 1:3)
z[1] <- list(c(2, 3, 4)) # replaces "bin" contents
z[[1]] <- c(2, 3, 4) # replaces "bin" contents
By name (if z$bin2 <- c("Tomas", "Luis", "Angela")
exists) z
# replace name of specific "bin"
names(z)[2] <- "mykids"
z
By logical num <- sapply(z, is.numeric)
num
z[num]<- list(rnorm(10), rnorm(10))
z

Table 2.28 Common ways of operating on a list

Function Description Try these examples
lapply Applies a function to each x <- list(1:5, 6:10)
component of a list and returns a x
list lapply(x, mean)
sapply Applies a function to each sapply(x, mean)
component of a list and simplifies
do.call Calls and applies a function to do.call(rbind, x)
the list
mapply Applies a function to the first x <- list(1:4, 1:4)
elements of each argument, the x
second elements, the third y <- list(4, rep(4, 4))
elements, and so on. y
mapply(rep, x, y, SIMPLIFY=FALSE)
mapply(rep, x, y)

2.5.6 Operations on lists

Because lists can have complex structural components, there are not many
operations we will want to do on lists. When we want to apply a function
to each component (bin) of a list, we use the lapply or sapply function.
These functions are identical except that sapply “simplies” the final result,
if possible.
The do.call function applies a function to the entire list using each each
component as an argument. For example, consider a list where each bin con-
tains a vector and we want to cbind the vectors.
> mylist <- list(vec1=1:5, vec2=6:10, vec3=11:15)
> cbind(mylist) #will not work
mylist
vec1 Integer,5

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.6 A data frame is a list in a 2-dimensional tabular form 85

vec2 Integer,5
vec3 Integer,5
> do.call(cbind, mylist) #works
vec1 vec2 vec3
[1,] 1 6 11
[2,] 2 7 12
[3,] 3 8 13
[4,] 4 9 14
[5,] 5 10 15
For additional practice, study and implements the examples in Table 2.28
on the facing page.

2.6 A data frame is a list in a 2-dimensional tabular

form

A data frame is a list in 2-dimensional tabular form. Each list component

(bin) is a data field of equal length. A data frame is a list that behaves like a
matrix. Anything that can be done with lists can be done with data frames.
Many things that can be done with matrices can be done with data frames.

2.6.1 Understanding data frames and factors

Epidemiologists are familiar with tabular data sets where each row is a record
and each column is a field. A record can be data collected on individuals or
groups. We usually refer to the field name as a variable (e.g., age, gender,
ethnicity). Fields can contain numeric or character data. In R, these types of
data sets are handled by data frames. Each column of a data frame is usually
either a factor or numeric vector, although it can have complex, character,
or logical vectors. Data frames have the functionality of matrices and lists.
For example, here is the first 10 rows of the infert data set, a matched
case-control study published in 1976 that evaluated whether infertility was
associated with prior spontaneous or induced abortions.
> data(infert)
> str(infert)
‘data.frame’: 248 obs. of 8 variables:
$ education : Factor w/ 3 levels "0-5yrs",..: 1 1 ...
$ age : num NA 45 NA 23 35 36 23 32 21 28 ...
$ parity : num 6 1 6 4 3 4 1 2 1 2 ...
$ induced : num 1 1 2 2 1 2 0 0 0 0 ...
$ case : num 1 1 1 1 1 1 1 1 1 1 ...

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

86 2 Working with R data objects

$ spontaneous : num 2 0 0 0 1 1 0 0 1 0 ...

$ stratum : int 1 2 3 4 5 6 7 8 9 10 ...
$ pooled.stratum: num 3 1 4 2 32 36 6 22 5 19 ...
> infert[1:10, 1:6]
education age parity induced case spontaneous
1 0-5yrs NA 6 1 1 2
2 0-5yrs 45 1 1 1 0
3 0-5yrs NA 6 2 1 0
4 0-5yrs 23 4 2 1 0
5 6-11yrs 35 3 1 1 1
6 6-11yrs 36 4 2 1 1
7 6-11yrs 23 1 0 1 0
8 6-11yrs 32 2 0 1 0
9 6-11yrs 21 1 0 1 1
10 6-11yrs 28 2 0 1 0

The fields are obviously vectors. Let’s explore a few of these vectors to see
what we can learn about their structure in R.
> #age variable
> infert$age
[1] 26 42 39 34 35 36 23 32 21 28 29 37 31 29 31 27 30 26
...
[235] 25 32 25 31 38 26 31 31 25 31 34 35 29 23
> mode(infert$age)
[1] "numeric"
> class(infert$age)
[1] "numeric"

> #stratum variable

> infert$stratum
[1] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
...
[235] 70 71 72 73 74 75 76 77 78 79 80 81 82 83
> mode(infert$stratum)
[1] "numeric"
> class(infert$stratum)
[1] "integer"

> #education variable

> infert$education
[1] 0-5yrs 0-5yrs 0-5yrs 0-5yrs 6-11yrs 6-11yrs
...
[247] 12+ yrs 12+ yrs
Levels: 0-5yrs 6-11yrs 12+ yrs
> mode(infert$education)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.6 A data frame is a list in a 2-dimensional tabular form 87

[1] "numeric"
> class(infert$education)
[1] "factor"
What have we learned so far? In the infert data frame, age is a vector
of mode “numeric” and class “numeric,” stratum is a vector of mode “nu-
meric” and class “integer,” and education is a vector of mode “numeric”
and class “factor.” The numeric vectors are straightforward and easy to un-
derstand. However, a factor, R’s representation of categorical data, is a bit
more complicated.
Contrary to intuition, a factor is a numeric vector, not a character vector,
although it may have been created from a character vector (shown later). To
see the “true” education factor use the unclass function:
> z <- unclass(infert$education)
> z
[1] 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
...
[244] 3 3 3 3 3
attr(,"levels")
[1] "0-5yrs" "6-11yrs" "12+ yrs"
> mode(z)
[1] "numeric"
> class(z)
[1] "integer"
Now let’s create a factor from a character vector and then unclass it:
> cointoss <- sample(c("Head","Tail"), 100, replace = TRUE)
> cointoss
[1] "Tail" "Head" "Head" "Tail" "Tail" "Tail" "Head"
...
[99] "Tail" "Head"
> fct <- factor(cointoss)
> fct
[1] Tail Head Head Tail Tail Tail Head Head Head Tail Head
...
[100] Head
Levels: Head Tail
> unclass(fct)
[1] 2 1 1 2 2 2 1 1 1 2 1 1 1 2 2 2 2 2 1 1 1 1 1 1 1 2 2
[28] 1 2 2 1 1 2 1 2 2 1 1 1 1 1 2 2 1 1 2 2 2 1 1 2 2 2 1
[55] 1 1 1 1 2 1 1 2 2 2 1 1 2 2 2 2 1 2 2 1 1 1 2 1 1 2 2
[82] 1 1 2 1 1 1 2 1 1 1 1 2 1 1 1 2 1 2 1
attr(,"levels")
[1] "Head" "Tail"

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

88 2 Working with R data objects

Notice that we can still recover the original character vector using the
as.character function:
> as.character(cointoss)
[1] "Tail" "Head" "Head" "Tail" "Tail" "Tail" "Head"
...
[99] "Tail" "Head"
> as.character(fct)
[1] "Tail" "Head" "Head" "Tail" "Tail" "Tail" "Head"
...
[99] "Tail" "Head"
Okay, let’s create an ordered factor; that is, levels of a categorical vari-
able that have natural ordering. For this we set ordered=TRUE in the factor
function:
> samp <- sample(c("Low","Medium","High"), 100, replace=TRUE)
> ofac1 <- factor(samp, ordered=T)
> ofac1
[1] Low Medium High Medium Medium Medium Medium
...
[99] High High
Levels: High < Low < Medium
> table(ofac1) #levels and labels not in natural order
ofac1
High Low Medium
43 25 32
However, notice that the ordering was done in alphabetical order which is
not what we want. To change this, use the levels options in the factor
function:
> ofac2 <- factor(samp, levels=c("Low","Medium","High"), ordered=T)
> ofac2
[1] Low Medium High Medium Medium Medium Medium
...
[99] High High
Levels: Low < Medium < High
> table(ofac2)
ofac2
Low Medium High
28 35 37
Great — this is exactly what we want! For review, Table 2.29 on the next
page summarizes the variable types in epidemiology and how they are repre-
sented in R. Factors (unordered and ordered) are used to represent nominal
and ordinal categorical data. The infert data set contains nominal factors
and the esoph data set contains ordinal factors.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.6 A data frame is a list in a 2-dimensional tabular form 89

Table 2.29 Variable types in epidemiologic data and their representations in R data
frames
Representations in data Representations in R
Variable type Examples Mode Class Examples1
Numeric
Continuous 3.45, 2/3 numeric numeric infert$age
Discrete 1, 2, 3, 4, . . . numeric integer infert$stratum
Categorical
Nominal male vs. female numeric factor infert$education
Ordinal low < medium < high numeric ordered factor esoph$agegp
1. First load data: data(infert); data(esoph)

Table 2.30 Common ways of creating data frames

Function Description Try these examples
data.frame Data frames are x <- data.frame(id=1:2, sex=c("M","F"))
of mode list mode(x); x
as.data.frame Coerces data x <- matrix(1:6, 2, 3); x
object into a as.data.frame(x)
data frame
as.table Combine with x <- array(1:8, c(2, 2, 2))
ftable as.data.frame to dimnames(x) <- list(Exposure=c("Y", "N"),
convert a fully Disease = c("Y", "N"),
labeled array Confounder = c("Y", "N"))
into a data frame as.data.frame(ftable(x))
as.data.frame(as.table(x))
read.table Reads ASCII wcgs <- read.csv(".../wcgs.txt", header=T)
read.csv text file into str(wcgs)
read.delim data frame
read.fmf object1
1. Try read.csv("http://www.medepi.net/data/wcgs.txt", header=TRUE)

2.6.2 Creating data frames

In the creation of data frames, character vectors (usually representing cat-

egorical data) are converted to factors (mode numeric, class factor), and
numeric vectors are converted to numeric vectors of class numeric or class
integer.
wt <- c(59.5, 61.4, 45.2)
age <- c(11, 9, 6)
sex <- c("Male", "Male", "Female")
df <- data.frame(age, sex, wt)
df
str(df)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

90 2 Working with R data objects

Table 2.31 Common ways of naming data frames

Function Try these examples
names x <- data.frame(var1 = 1:3, var2 = c("M", "F", "F"))
x
names(x) <- c("Subjno", "Sex")
x
row.names row.names(x) <- c("Subj 1", "Subj 2", "Subj 3")
x

Factors can also be created directly from vectors as described in the previous
section.

2.6.3 Naming data frames

Everything that applies to naming list components (Table 2.25 on page 81)
also applies to naming data frame components (Table 2.31). In general, we
may be interested in renaming variables (fields) or row names of a data
frame, or renaming the levels (possible values) for a given factor (categorical
variable). For example, consider the Oswego data set.
> odat <- read.table("http://www.medepi.net/data/oswego.txt",
+ sep="", header=TRUE, na.strings=".")
> odat[1:5,1:8] #Display partial data frame
id age sex meal.time ill onset.date onset.time baked.ham
1 2 52 F 8:00 PM Y 4/19 12:30 AM Y
2 3 65 M 6:30 PM Y 4/19 12:30 AM Y
3 4 59 F 6:30 PM Y 4/19 12:30 AM Y
4 6 63 F 7:30 PM Y 4/18 10:30 PM Y
5 7 70 M 7:30 PM Y 4/18 10:30 PM Y
> names(odat)[3] <- "Gender" #Rename ’sex’ to ’Gender’
> table(odat$Gender) #Display ’Gender’ distribution

F M
44 31
> levels(odat$Gender) #Display ’Gender’ levels
[1] "F" "M"
> #Replace ’Gender’ level labels
> levels(odat$Gender) <- c("Female", "Male")
> levels(odat$Gender) #Display new ’Gender’ levels
[1] "Female" "Male"
> table(odat$Gender) #Confirm distribution is same

Female Male

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.6 A data frame is a list in a 2-dimensional tabular form 91

Table 2.32 Common ways of indexing data frames

Indexing Try these examples
By position data(infert)
infert[1:5, 1:3]
By name infert[1:5, c("education", "age", "parity")]
By logical agelt30 <- infert$age<30; agelt30
infert[agelt30, c("education","induced","parity")]
# can also use ’subset’ function
subset(infert, agelt30, c("education","induced","parity"))

44 31
> odat[1:5,1:8] #Display partial data frame
id age Gender meal.time ill onset.date onset.time baked.ham
1 2 52 Female 8:00 PM Y 4/19 12:30 AM Y
2 3 65 Male 6:30 PM Y 4/19 12:30 AM Y
3 4 59 Female 6:30 PM Y 4/19 12:30 AM Y
4 6 63 Female 7:30 PM Y 4/18 10:30 PM Y
5 7 70 Male 7:30 PM Y 4/18 10:30 PM Y
On occasion, we might be interested in renaming the row names. Currently,
the Oswego data set has default integer values from 1 to 75 as the row names.
> row.names(odat)
[1] "1" "2" "3" "4" "5" "6" "7" "8" "9" "10" "11"
[12] "12" "13" "14" "15" "16" "17" "18" "19" "20" "21" "22"
[23] "23" "24" "25" "26" "27" "28" "29" "30" "31" "32" "33"
[34] "34" "35" "36" "37" "38" "39" "40" "41" "42" "43" "44"
[45] "45" "46" "47" "48" "49" "50" "51" "52" "53" "54" "55"
[56] "56" "57" "58" "59" "60" "61" "62" "63" "64" "65" "66"
[67] "67" "68" "69" "70" "71" "72" "73" "74" "75"
We can change the row names by assigning a new character vector.
> row.names(odat) <- sample(101:199, size=nrow(odat))
> odat[1:5,1:7]
id age Gender meal.time ill onset.date onset.time
123 2 52 Female 8:00 PM Y 4/19 12:30 AM
145 3 65 Male 6:30 PM Y 4/19 12:30 AM
173 4 59 Female 6:30 PM Y 4/19 12:30 AM
138 6 63 Female 7:30 PM Y 4/18 10:30 PM
146 7 70 Male 7:30 PM Y 4/18 10:30 PM

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

92 2 Working with R data objects

2.6.4 Indexing data frames

Indexing a data frame is similar to indexing a matrix or a list: we can index

by position, by name, or by logical vector. Consider, for example, the 2004
California West Nile virus human disease surveillance data. Suppose we are
interested in summarizing the Los Angeles cases with neuroinvasive disease
(“WNND”).
> wdat <- read.csv("http://www.medepi.net/data/wnv/wnv2004fin.txt")
> str(wdat)
‘data.frame’: 779 obs. of 8 variables:
$ id : int 1 2 3 4 5 6 7 8 9 10 ...
$ county : Factor w/ 23 levels "Butte","Fresno",..: 14 ...
$ age : int 40 64 19 12 12 17 61 74 71 26 ...
$ sex : Factor w/ 2 levels "F","M": 1 1 2 2 2 2 2 1...
$ syndrome : Factor w/ 3 levels "Unknown","WNF",..: 22 3 ...
$ date.onset : Factor w/ 130 levels "2004-05-14", ..: 3 ...
$ date.tested: Factor w/ 104 levels "2004-06-02",..: 1 ...
$ death : Factor w/ 2 levels "No","Yes": 1 1 1 1 1 1 ...
> levels(wdat$county) #Review levels of ’county’ variable
[1] "Butte" "Fresno" "Glenn"
[4] "Imperial" "Kern" "Lake"
[7] "Lassen" "Los Angeles" "Merced"
[10] "Orange" "Placer" "Riverside"
[13] "Sacramento" "San Bernardino" "San Diego"
[16] "San Joaquin" "Santa Clara" "Shasta"
[19] "Sn Luis Obispo" "Tehama" "Tulare"
[22] "Ventura" "Yolo"
> levels(wdat$syndrome) #Review levels of ’syndrome’ variable
[1] "Unknown" "WNF" "WNND"
> myrows <- wdat$county=="Los Angeles" & wdat$syndrome=="WNND"
> mycols <- c("id", "county", "age", "sex", "syndrome", "death")
> wnv.la <- wdat[myrows, mycols]
> wnv.la
id county age sex syndrome death
25 25 Los Angeles 70 M WNND No
26 26 Los Angeles 59 M WNND No
27 27 Los Angeles 59 M WNND No
...
734 736 Los Angeles 71 M WNND Yes
770 772 Los Angeles 72 M WNND No
776 778 Los Angeles 50 F WNND No
In this example, the data frame rows were indexed by logical vector, and the
columns were indexed by names. We emphasize this method because it only

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.6 A data frame is a list in a 2-dimensional tabular form 93

requires application of previously learned principles that always work with R

objects.
An alternative method is to use the subset function. The first argument
specifies the data frame, the second argument is a Boolean operation that
evaluates to a logical vector, and the third argument specifies what variables
(or range of variables) to include or exclude.
> wnv.sf2 <- subset(wdat, county=="Los Angeles" & syndrome=="WNND",
+ select = c(id, county, age, sex, syndrome, death))
> wnv.sf2[1:6,]
id county age sex syndrome death
25 25 Los Angeles 70 M WNND No
26 26 Los Angeles 59 M WNND No
27 27 Los Angeles 59 M WNND No
47 47 Los Angeles 57 M WNND No
48 48 Los Angeles 60 M WNND No
49 49 Los Angeles 34 M WNND No
This example is equivalent but specifies range of variables using the : func-
tion:
> wnv.sf3 <- subset(wdat, county=="Los Angeles" & syndrome=="WNND",
+ select = c(id:syndrome, death))
> wnv.sf3[1:6,]
id county age sex syndrome death
25 25 Los Angeles 70 M WNND No
26 26 Los Angeles 59 M WNND No
27 27 Los Angeles 59 M WNND No
47 47 Los Angeles 57 M WNND No
48 48 Los Angeles 60 M WNND No
49 49 Los Angeles 34 M WNND No
This example is equivalent but specifies variables to exclude using the -
function:
> wnv.sf4 <- subset(wdat, county=="Los Angeles" & syndrome=="WNND",
+ select = -c(date.onset, date.tested))
> wnv.sf4[1:6,]
id county age sex syndrome death
25 25 Los Angeles 70 M WNND No
26 26 Los Angeles 59 M WNND No
27 27 Los Angeles 59 M WNND No
47 47 Los Angeles 57 M WNND No
48 48 Los Angeles 60 M WNND No
49 49 Los Angeles 34 M WNND No
The subset function offers some conveniences such as the ability to specify
a range of fields to include using the : function, and to specify a group of
fields to exclude using the - function.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

94 2 Working with R data objects

Table 2.33 Common ways of replacing data frame components

Replacing Try these examples
By position data(infert)
infert[1:4, 1:2]
infert[1:4, 2] <- c(NA, 45, NA, 23)
infert[1:4, 1:2]
By name names(infert)
infert[1:4, c("education", "age")]
infert[1:4, c("age")] <- c(NA, 45, NA, 23)
infert[1:4, c("education", "age")]
By logical table(infert$parity)
# change values of 5 or 6 to missing (NA)
infert$parity[infert$parity==5 | infert$parity==6] <- NA
table(infert$parity)
table(infert$parity, exclude=NULL)

2.6.5 Replacing data frame components

With data frames, as with all R data objects, anything that can be indexed
can be replaced. We already saw some examples of replacing names. For
practice, study and implement the examples in Table 2.33.

2.6.6 Operations on data frames

A data frame is of mode list, and functions that operate on components

of a list will work with data frames. For example, consider the California
population estimates and projections for the years 2000–2050.
> capop <- read.csv("http://www.dof.ca.gov/HTML/DEMOGRAP/
Data/RaceEthnic/Population-00-50/documents/California.txt")
> str(capop)
‘data.frame’: 10302 obs. of 11 variables:
$ County : int 59 59 59 59 59 59 59 59 59 59 ...
$ Year : int 2000 2000 2000 2000 2000 2000 ...
$ Sex : Factor w/ 2 levels "F","M": 1 1 1 ...
$ Age : int 0 1 2 3 4 5 6 7 8 9 ...
$ White : int 75619 76211 76701 78551 82314 ...
$ Hispanic : int 115911 113706 114177 116733 0 ...
$ Asian : int 20879 20424 21044 21920 22760 ...
$ Pacific.Islander: int 741 765 806 817 884 945 961 ...
$ Black : int 14629 15420 15783 16531 17331 ...
$ American.Indian : int 1022 1149 1169 1318 1344 1363 ...
$ Multirace : int 10731 8676 8671 8556 8621 ...

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.6 A data frame is a list in a 2-dimensional tabular form 95

Table 2.34 Common ways of operating on a data frame

Function Description Try these examples
tapply Apply a function to data(infert)
strata of a vector args(tapply)
that are defined by a tapply(infert$age, infert$education,
unique combination mean, na.rm = TRACE)
of the levels of
selected factors
lapply Apply a function to lapply(infert[,1:3], table)
each component of
the list
sapply Apply a function to sapply(infert[,c("age", "parity")],
each component of a mean, na.rm = TRUE)
list, and simplify
aggregate Splits the data into aggregate(infert[,c("age", "parity")],
subsets, computes by = list(Education = infert$education,
summary statistics Induced = infert$induced), mean)
for each, and returns
the result in a
convenient form.
mapply Apply a function to df <- data.frame(var1 = 1:4, var2 = 4:1)
the first elements of mapply("*", df$var1, df$var2)
each argument, the mapply(c, df$var1, df$var2)
second elements, the mapply(c, df$var1, df$var2, SIMPLIFY=F)
third elements, and
so on.

> capop[1:5,1:8]
County Year Sex Age White Hispanic Asian Pacific.Islander
1 59 2000 F 0 75619 115911 20879 741
2 59 2000 F 1 76211 113706 20424 765
3 59 2000 F 2 76701 114177 21044 806
4 59 2000 F 3 78551 116733 21920 817
5 59 2000 F 4 82314 119995 22760 884
Now, suppose we want to assess the range of the numeric fields. If we treat
the data frame as a list, both lapply or sapply works:
> sapply(capop[-3], range)
County Year Age White Hispanic Asian Pacific.Islander
[1,] 59 2000 0 497 110 76 1
[2,] 59 2050 100 148246 277168 46861 1890
Black American.Indian Multirace
[1,] 57 0 5
[2,] 26983 8181 17493
However, if we treat the data frame as a matrix, apply also works:
> apply(capop[,-3], 2, range)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

96 2 Working with R data objects

County Year Age White Hispanic Asian Pacific.Islander

[1,] 59 2000 0 497 110 76 1
[2,] 59 2050 100 148246 277168 46861 1890
Black American.Indian Multirace
[1,] 57 0 5
[2,] 26983 8181 17493
Some R functions, such as summary, will summarize every variable in a
data frame without having to use lapply or sapply.
> summary(capop[1:7])
County Year Sex Age
Min. :59 Min. :2000 F:5151 Min. : 0
1st Qu.:59 1st Qu.:2012 M:5151 1st Qu.: 25
Median :59 Median :2025 Median : 50
Mean :59 Mean :2025 Mean : 50
3rd Qu.:59 3rd Qu.:2038 3rd Qu.: 75
Max. :59 Max. :2050 Max. :100
White Hispanic Asian
Min. : 497 Min. : 110 Min. : 76
1st Qu.: 63134 1st Qu.: 29962 1st Qu.:19379
Median : 75944 Median :115646 Median :30971
Mean : 71591 Mean :101746 Mean :27769
3rd Qu.: 88021 3rd Qu.:154119 3rd Qu.:37763
Max. :148246 Max. :277168 Max. :46861

2.6.6.1 The aggregate function

The aggregate function is almost identical to the tapply function. Recall

that tapply allows us to apply a function to a vector that is stratified by
one or more fields; for example, calculating mean age (1 field) stratified by
sex and ethnicity (2 fields). In contrast, aggregate allows us to apply a
function to a group of fields that are stratified by one or more fields; for
example, calculating the mean weight and height (2 fields) stratified by sex
and ethnicity (2 fields):
> sex <- c("M", "M", "M", "M", "F", "F", "F", "F")
> eth <- c("W", "W", "B", "B", "W", "W", "B", "B")
> wgt <- c(140, 150, 150, 160, 120, 130, 130, 140)
> hgt <- c( 60, 70, 70, 80, 40, 50, 50, 60)
> df <- data.frame(sex, eth, wgt, hgt)
> aggregate(df[, 3:4], by = list(Gender = df$sex,
+ Ethnicity = df$eth), FUN = mean)
Gender Ethnicity wgt hgt
1 F B 135 55
2 M B 155 75

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.7 Managing data objects 97

3 F W 125 45
4 M W 145 65
For another example, in the capop data frame, we notice that the variable
age goes from 0 to 100 by 1-year intervals. It will be useful to aggregate ethnic-
specific population estimates into larger age categories. More specifically, we
want to calculate the sum of ethnic-specific population estimates (6 fields)
stratified by age category, sex, and year (3 fields). We will create a new 7-
level age category field commonly used by the National Center for Health
Statistics. Naturally, we use the aggregate function:
> capop <- read.csv("http://www.dof.ca.gov/HTML/DEMOGRAP/Data/
RaceEthnic/Population-00-50/documents/California.txt")
> to.keep <- c("White", "Hispanic", "Asian", "Pacific.Islander",
+ "Black", "American.Indian", "Multirace")
> age.nchs7 <- c(0, 1, 5, 15, 25, 45, 65, 101)
> capop$agecat7 <- cut(capop$Age, breaks = age.nchs7, right=FALSE)
> capop7 <- aggregate(capop[,to.keep], by = list(Age=capop$agecat7,
+ Sex=capop$Sex, Year=capop$Year), FUN = sum)
> levels(capop7$Age)[7] <- "65+"
> capop7[1:14, 1:6]
Age Sex Year White Hispanic Asian
1 [0,1) F 2000 75619 115911 20879
2 [1,5) F 2000 313777 464611 86148
3 [5,15) F 2000 924930 1124573 241047
4 [15,25) F 2000 868767 946948 272846
5 [25,45) F 2000 2360250 1742366 667956
6 [45,65) F 2000 2102090 735062 445039
7 65+ F 2000 1471842 279865 208566
8 [0,1) M 2000 79680 121585 21965
9 [1,5) M 2000 331193 484068 91373
10 [5,15) M 2000 979233 1175384 257574
11 [15,25) M 2000 925355 1080868 279314
12 [25,45) M 2000 2465194 1921896 614608
13 [45,65) M 2000 2074833 687549 384011
14 65+ M 2000 1075226 202299 154966

2.7 Managing data objects

When we work in R we have a workspace. Think of the workspace as our

desktop that contains the “objects” (data and tools) we use to conduct our
work. To view the objects in our workspace use the ls or objects functions
(Table 2.35 on the next page):
> ls()

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

98 2 Working with R data objects

Table 2.35 Common ways of managing data objects

Function Description Try these examples
ls List objects ls()
objects objects() #equivalent
rm Remove object(s) yy <- 1:5; ls()
remove rm(yy); ls()
# remove in objects in working environment
# Don’t do this unless we are really sure
rm(list = ls())
save.image Saves current save.image()
workspace
save Writes R objects to x <- runif(20)
load the specified y <- list(a = 1, b = TRUE, c = "oops")
external file. The save(x, y, file = "c:/temp/xy.Rdata")
objects can be read rm(x,y); x; y
back from the file at load(file = "c:/temp/xy.Rdata")
a later date using x
’load’ y

[1] "add.to.x" "add.to.y" "age" "age.nchs7"

[5] "agecat" "alt" "ar.num" "capop"
[9] "capop7" "dat" "dat.ordered" "dat.random"
[13] "dat2" "dat3" "dat4" "dd"
...
We use the pattern option to search for object names that contain the pat-
tern we specify.
> ls(pattern = "dat")
[1] "dat" "dat.ordered" "dat.random" "dat2"
[5] "dat3" "dat4" "mydat" "sdat"
[9] "sdat.asr" "sdat3" "st.dates" "udat1"
[13] "udat2" "wdat"
The rm or remove functions will remove workspace objects.
> rm(dat, dat2, dat3, dat4)
> ls(patt="dat")
[1] "dat.ordered" "dat.random" "mydat" "sdat"
[5] "sdat.asr" "sdat3" "st.dates" "udat1"
[9] "udat2" "wdat"
To remove all data objects use the following code with extreme caution:
rm(list = ls())
However, the object names may not be sufficiently descriptive to know
what these objects contain. To assess R objects in our workspace we use the
functions summarized in Table 2.2 on page 30. In general, we never go wrong
using the str, mode, and class functions.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.7 Managing data objects 99

> mode(capop7)
[1] "list"
> class(capop7)
[1] "data.frame"
> str(capop7)
‘data.frame’: 714 obs. of 10 variables:
$ Age : Factor w/ 7 levels "[0,1)",..: 1 2 6 ...
$ Sex : Factor w/ 2 levels "F","M": 1 1 1 1 ...
$ Year : Factor w/ 51 levels "2000",..: 1 1 1 ...
$ White : int 75619 313777 924930 868767 ...
$ Hispanic : int 115911 464611 1124573 946948 ...
$ Asian : int 20879 86148 241047 272846 ...
$ Pacific.Islander: int 741 3272 9741 9629 19085 9898 ...
$ Black : int 14629 65065 195533 158923 ...
$ American.Indian : int 1022 4980 15271 14301 30960 ...
$ Multirace : int 10731 34524 78716 56735 82449 ...
>
> mode(orcalc)
[1] "function"
> class(orcalc)
[1] "function"
> str(orcalc)
function (x)
- attr(*, "source")= chr [1:5] "function(x){" ...
> orcalc
function(x){
or <- (x[1,1]*x[2,2])/(x[1,2]*x[2,1])
pval <- fisher.test(x)$p.value
list(data = x, odds.ratio = or, p.value = pval)
Objects created in the workspace are available during the R session. Upon
closing the R session, R asks whether to save the workspace. To save the
objects without exiting an R session, use the save.image function:
> save.image()
The save.image function is actually a special case of the save function:
save(list = ls(all = TRUE), file = ".RData")
The save function saves an R object as an external file. This file can be
loaded using the load function.
> x <- 1:5
> x
[1] 1 2 3 4 5
> save(x, file="/home/tja/temp/x")
> rm(x)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

100 2 Working with R data objects

Table 2.36 Assessing and coercing data objects

Query data type Coerce to data type
is.vector as.vector
is.matrix as.matrix
is.array as.array
is.list as.list
is.data.frame as.data.frame
is.factor as.factor
is.ordered as.ordered
is.table as.table

is.numeric as.numeric
is.integer as.integer
is.character as.character
is.logical as.logical

is.function as.function

is.null as.null

is.na n/a
is.nan n/a
is.finite n/a
is.infinite n/a

> x
Error: object "x" not found
> load(file="/home/tja/temp/x")
> x
[1] 1 2 3 4 5
Table 2.36 provides more functions for conducting specific object queries
and for coercing one object type into another. For example, a vector is not a
matrix.
> is.matrix(1:3)
[1] FALSE
However, a vector can be coerced into a matrix.
> as.matrix(1:3)
[,1]
[1,] 1
[2,] 2
[3,] 3
> is.matrix(as.matrix(1:3))

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.7 Managing data objects 101

[1] TRUE
A common use would be to coerce a factor into a character vector.
> sex <- factor(c("M", "M", "M", "M", "F", "F", "F", "F"))
> sex
[1] M M M M F F F F
Levels: F M
> unclass(sex) #does not coerce into character vector
[1] 2 2 2 2 1 1 1 1
attr(,"levels")
[1] "F" "M"
> as.character(sex) #yes, works
[1] "M" "M" "M" "M" "F" "F" "F" "F"
In R, missing values are represented by the value NA (“not available”). The
is.na function evaluates an object and returns a logical vector indicating
which positions contain NA. The !is.na version returns positions that do not
contain NA.
> x <- c(12, 34, NA, 56, 89)
> is.na(x)
[1] FALSE FALSE TRUE FALSE FALSE
> !is.na(x)
[1] TRUE TRUE FALSE TRUE TRUE
We can use is.na to replace missing values.
> x[is.na(x)] <- 999
> x
[1] 12 34 999 56 89
In R, NaN represents “not a number” and Inf represent an infinite value.
Therefore, we can use is.nan and is.infinite to assess which positions
contain NaN and Inf, respectively.
> x <- c(0, 3, 0, -6)
> y <- c(4, 0, 0, 0)
> z <- x/y
> z
[1] 0 Inf NaN -Inf
> is.nan(z)
[1] FALSE FALSE TRUE FALSE
> is.infinite(z)
[1] FALSE TRUE FALSE TRUE

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

102 2 Working with R data objects

2.8 Managing our workspace

Our workspace is like a desktop that contains the “objects” (data and tools)
we use to conduct our work. Use the getwd function to list the file path to
the workspace file .RData.
> getwd()
[1] "/home/tja/Data/R/home"
Use the setwd function to set up a new workspace location. A new .RData
file will automatically be created there
setwd("/home/tja/Data/R/newproject")
This is one method to manage multiple workspaces for one’s projects.
Use the search function to list the packages, environments, or data frames
attached and available.
> search() # Linux
[1] ".GlobalEnv" "package:stats" "package:graphics"
[4] "package:grDevices" "package:utils" "package:datasets"
[7] "package:methods" "Autoloads" "package:base"
The global environment .GlobalEnv is our workspace. The searchpaths
function list the full paths:
> searchpaths()
[1] ".GlobalEnv" "/usr/lib/R/library/stats"
[3] "/usr/lib/R/library/graphics" "/usr/lib/R/library/grDevices"
[5] "/usr/lib/R/library/utils" "/usr/lib/R/library/datasets"
[7] "/usr/lib/R/library/methods" "Autoloads"
[9] "/usr/lib/R/library/base"

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.8 Managing our workspace 103

Table 2.37 Risk of Death in a 20-year Period Among Women in Whickham, England,
According to Smoking Status at the Beginning of the Period
Smoking
Vital Status Yes No
Dead 139 230
Alive 443 502

Table 2.38 Risk of Death in a 20-year Period Among Women in Whickham, England,
According to Smoking Status at the Beginning of the Period
Smoking
Vital Status Yes No Total
Dead 139 230 369
Alive 443 502 945
Total 582 732 1314

Table 2.39 Risk Ratio and Odds Ratio of Death in a 20-year Period Among Women in
Whickham, England, According to Smoking Status at the Beginning of the Period
Smoking
Yes No
Risk 0.24 0.31
Risk Ratio 0.76 1.00
Odds 0.31 0.46
Odds Ratio 0.68 1.00

Problems

2.1. Download and install RStudio from http://www.rstudio.org. Start a

new project. For example, I started a new project in the following directory:
/home/tja/Documents/courses/ph251d/Rproj. Inside the Rproj directory I re-
named the XXX.Rproj file to ph251d.Rproj. From the main menu select File →
New → R Script. Save the R script as ph251d.R in the Rproj directory. Write
and run R code from this script. Test R code from the tables in this chapter.
2.2. Names and describe the six type of R objects.
2.3. How many ways can we index an object?
2.4. Finish the sentence: Any R object component(s) that can be indexed,
can be . . . .
2.5. Recreate Table 2.37 using any combination of the matrix, cbind, rbind,
dimnames, or names functions.

2.6. Starting with the 2 × 2 matrix object we created previously, using only
the apply, cbind, rbind, names, and dimnames functions, recreate Table 2.38.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

104 2 Working with R data objects

2.7. Using the 2 × 2 data from Table 2.37 on the preceding page, use the
sweep and apply functions to calculate marginal and joint distributions.

2.8. Using the data from the previous problems, recreate Table 2.39 on the
previous page and interpret the results.

2.9. Read in the Whickham, England data using the R code below.
wdat = read.table("http://www.medepi.net/data/whickham-engl.txt",
sep = ",", header = TRUE)
str(wdat)
xtabs(~Vital.Status + Age + Smoking, data = wdat)
Stratified by age category, calculate the risk of death comparing smokers to
nonsmokers. Show your results. What is your interpretation.

2.10. Use the read.table function to read in the syphilis data available
at http://www.medepi.net/data/syphilis89c.txt. Evaluate structure of
data frame. Do not attach std data frame (yet). Create a 3-dimensional
array using both the table or xtabs function. Now attach the std data
frame using the attach function. Create the same 3-dimensional array using
both the table or xtabs function.

2.11. Use the apply function to get marginal totals for the syphilis 3-
dimensional array.

2.12. Use the sweep and apply functions to get marginal and joint distribu-
tions for a 3-D array.

2.13. Review and read in the group-level, tabular data set of primary and
secondary syphilis cases in the United States in 1989 available at http:
//www.medepi.net/data/syphilis89b.txt. Use the rep function on the
data frame fields to recreate the individual-level data frame with over 40,000
observations.

2.14. Working with population estimates can be challenging because of the

amount of data manipulation. Study the 2000 population estimates for Cali-
fornia Counties: http://www.medepi.net/data/calpop/CalCounties2000.
txt. Now, study and implement this R code. For each expression or group
of expressions, explain in words what the R code is doing. Be sure to display
intermediate objects to understand each step.
#1 Read county names
cty <- scan("http://www.medepi.net/data/calpop/calcounty.txt",
what="")

#2 Read county population estimates

calpop =
read.csv("http://www.medepi.net/data/calpop/CalCounties2000.txt",

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

2.8 Managing our workspace 105

header = T)

#2 Replace county number with county name

for(i in 1:length(cty)){
calpop$County[calpop$County==i] <- cty[i]
}

#3 Discretize age into categories

calpop$Agecat <- cut(calpop$Age, c(0,20,45,65,100),
include.lowest = TRUE, right = FALSE)

#4 Create combined API category

calpop$AsianPI <- calpop$Asian + calpop$Pacific.Islander

#5 Shorten selected ethnic labels

names(calpop)[c(6, 9, 10)] = c("Latino", "AfrAmer", "AmerInd")

#6 Index Bay Area Counties

baindex <- calpop$County=="Alameda" | calpop$County=="San Francisco"
bapop <- calpop[baindex,]
bapop

#7 Labels for later use

agelabs <- names(table(bapop$Agecat))
sexlabs <- c("Female", "Male")
racen <- c("White", "AfrAmer", "AsianPI", "Latino", "Multirace",
"AmerInd")
ctylabs <- names(table(bapop$County))

#8 Aggregate
bapop2 <- aggregate(bapop[,racen],
list(Agecat = bapop$Agecat, Sex = bapop$Sex,
County = bapop$County), sum)
bapop2

#9 Temp matrix of counts

tmp <- as.matrix(cbind(bapop2[1:4,racen], bapop2[5:8,racen],
bapop2[9:12,racen], bapop2[13:16,racen]))

#10 Convert into final array

bapop3 <- array(tmp, c(4, 6, 2, 2))
dimnames(bapop3) <- list(agelabs, racen, sexlabs, ctylabs)
bapop3

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Chapter 3
Managing epidemiologic data in R

3.1 Entering and importing data

There are many ways of getting our data into R for analysis. In the section
that follows we review how to enter the Unversity Group Diabetes Program
data (Table 3.1) as well as the original data from a comma-delimited text
file. We will use the following approaches:
• Entering data at the command prompt
• Importing data from a file
• Importing data using an URL

3.1.1 Entering data at the command prompt

We review four methods. For Methods 1 and 2, data are entered directly at
the command prompt. Method 3 uses the same R expressions and data as
Methods 1 and 2, but they are entered into a text editor, saved as an text
file with a .R extension (e.g., job02.R), and then executed from the command
prompt using the source function. Alternatively, the R expressions and data

Table 3.1 Deaths among subjects who received tolbutamide and placebo in the Unversity
Group Diabetes Program (1970), stratifying by age
Age<55 Age≥55 Combined
Tolbutamide Placebo Tolbutamide Placebo Tolbutamide Placebo
Deaths 8 5 22 16 30 21
Survivors 98 115 76 69 174 184

107
108 3 Managing epidemiologic data in R

can be copied and pasted into R.1 And, for Method 4 we use R’s spreadsheet
editor (least preferred).

3.1.1.1 Method 1

For review, a convenient way to enter data at the command prompt is to use
the c function:

> #enter data for a vector

> vec1 <- c(8, 98, 5, 115); vec1
[1] 8 98 5 115
> vec2 <- c(22, 76, 16, 69); vec2
[1] 22 76 16 69
>
> #enter data for a matrix
> mtx1 <- matrix(vec1, 2, 2); mtx1
[,1] [,2]
[1,] 8 5
[2,] 98 115
> mtx2 <- matrix(vec2, 2, 2); mtx2
[,1] [,2]
[1,] 22 16
[2,] 76 69
>
> #enter data for an array and sum across strata
> udat <- array(c(vec1, vec2), c(2, 2, 2)); udat
, , 1

[,1] [,2]
[1,] 8 5
[2,] 98 115

, , 2

[,1] [,2]
[1,] 22 16
[2,] 76 69

> udat.tot <- apply(udat, c(1, 2), sum); udat.tot

[,1] [,2]
[1,] 30 21
[2,] 174 184
>
1 In RStudio, select from main menu Code → Run Region, etc.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.1 Entering and importing data 109

> #enter a list

> x <- list(crude.data = udat.tot, stratified.data = udat)
> x$crude.data
[,1] [,2]
[1,] 30 21
[2,] 174 184
> x$stratified
, , 1

[,1] [,2]
[1,] 8 5
[2,] 98 115

, , 2

[,1] [,2]
[1,] 22 16
[2,] 76 69

>
> #enter simple data frame
> subjname <- c("Pedro", "Paulo", "Maria")
> subjno <- 1:length(subjname)
> age <- c(34, 56, 56)
> sex <- c("Male", "Male", "Female")
> dat <- data.frame(subjno, subjname, age, sex); dat
subjno subjname age sex
1 1 Pedro 34 Male
2 2 Paulo 56 Male
3 3 Maria 56 Female
>
> #enter a simple function
> odds.ratio <- function(aa, bb, cc, dd){ aa*dd / (bb*cc)}
> odds.ratio(30, 174, 21, 184)
[1] 1.510673

3.1.1.2 Method 2

Method 2 is identical to Method 1 except one uses the scan function. It does
not matter if we enter the numbers on different lines, it will still be a vector.
Remember that we must press the Enter key twice after we have entered the
last number.
> udat.tot <- scan()
1: 30 174

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

110 3 Managing epidemiologic data in R

3: 21 184
5:
Read 4 items
> udat.tot
[1] 30 174 21 184
To read in a matrix at the command prompt combine the matrix and
scan functions. Again, it does not matter on what lines we enter the data, as
long as they are in the correct order because the matrix function reads data
in column-wise.
> udat.tot <- matrix(scan(), 2, 2)
1: 30 174 21 184
5:
Read 4 items
> udat.tot
[,1] [,2]
[1,] 30 21
[2,] 174 184

> udat.tot <- matrix(scan(), 2, 2, byrow = T) #read row-wise

1: 30 21 174 184
5:
Read 4 items
> udat.tot
[,1] [,2]
[1,] 30 21
[2,] 174 184
To read in an array at the command prompt combine the array and scan
functions. Again, it does not matter on what lines we enter the data, as long
as they are in the correct order because the array function reads the numbers
column-wise. In this example we include the dimnames argument.

> udat <- array(scan(), dim = c(2, 2, 2),

+ dimnames = list(Vital.Status = c("Dead","Survived"),
+ Treatment = c("Tolbutamide", "Placebo"),
+ Age.Group = c("<55", "55+")))
1: 8 98 5 115 22 76 16 69
9:
Read 8 items
> udat
, , Age.Group = <55

Treatment
Vital.Status Tolbutamide Placebo
Dead 8 5

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.1 Entering and importing data 111

Survived 98 115

, , Age.Group = 55+

Treatment
Vital.Status Tolbutamide Placebo
Dead 22 16
Survived 76 69

To read in a list of vectors of the same length (“fields”) at the command

prompt combine the list and scan function. We will need to specify the
type of data that will go into each “bin” or “field.” This is done by specifying
the what argument as a list. This list must be values that are either logical,
integer, numeric, or character. For example, for a character vector we can use
any expression, say x, that would evaluate to TRUE for is.character(x ). For
brevity, use "" for character, 0 for numeric, 1:2 for integer, and T or F for
logical. Look at this example:
> dat <- scan("", what = list(1:2, "", 0, "", T))
1: 3 "John Paul" 84.5 Male F
2: 4 "Jane Doe" 34.5 Female T
3:
Read 2 records
> dat
[[1]]
[1] 3 4

[[2]]
[1] "John Paul" "Jane Doe"

[[3]]
[1] 84.5 34.5

[[4]]
[1] "Male" "Female"

[[5]]
[1] FALSE TRUE

> str(dat)
List of 5
$ : int [1:2] 3 4
$ : chr [1:2] "John Paul" "Jane Doe"
$ : num [1:2] 84.5 34.5
$ : chr [1:2] "Male" "Female"
$ : logi [1:2] FALSE TRUE

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

112 3 Managing epidemiologic data in R

> #same example with field names

> dat <- scan("", what = list(id = 1:2, name = "", age = 0,
+ sex = "", dead = TRUE))
1: 3 "John Paul" 84.5 Male F
2: 4 "Jane Doe" 34.5 Female T
3:
Read 2 records
> dat
$id
[1] 3 4

$name
[1] "John Paul" "Jane Doe"

$age
[1] 84.5 34.5

$sex
[1] "Male" "Female"

$dead
[1] FALSE TRUE

> str(dat)
List of 5
$ id : int [1:2] 3 4
$ name: chr [1:2] "John Paul" "Jane Doe"
$ age : num [1:2] 84.5 34.5
$ sex : chr [1:2] "Male" "Female"
$ dead: logi [1:2] FALSE TRUE

To read in a data frame at the command prompt combine the data.frame,

scan, and list functions.

> dat <- data.frame(scan("", what = list(id=1:2, name="",

+ age=0, sex="", dead=T)) )
1: 3 "John Paul" 84.5 Male F
2: 4 "Jane Doe" 34.5 Female T
3:
Read 2 records
> dat
id name age sex dead
1 3 John Paul 84.5 Male FALSE
2 4 Jane Doe 34.5 Female TRUE

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.1 Entering and importing data 113

3.1.1.3 Method 3

Method 3 uses the same R expressions and data as Methods 1 and 2, but
they are entered into a text editor, saved as an ASCII text file with a .R
extension (e.g., job01.R), and then executed from the command prompt using
the source function. Alternatively, the R expressions and data can be copied
and pasted into R.2
For example, the following expressions are in a text editor and saved to a
file named job01.R.
x <- 1:10
x
One can copy and paste this code into R at the commmand prompt.
> x <- 1:10
> x
[1] 1 2 3 4 5 6 7 8 9 10
However, if we execute the code using the source function, it will only display
to the screen those objects that are printed using the show or print function.
Here is the text editor code again, but including show.
x <- 1:10
show(x)
Now, source job01.R using source at the command prompt.
> source("/home/tja/Documents/Rproj/job01.R")
[1] 1 2 3 4 5 6 7 8 9 10
In general, we highly recommend using a text editor for all our work. The
program file (e.g., job01.R) created with the text editor facilitates document-
ing our code, reviewing our code, debugging our code, replicating our analytic
steps, and auditing by external reviewers.

3.1.1.4 Method 4 (optional read)

Method 4 uses R’s spreadsheet editor.3 This is not a preferred method because
we like the original data to be in a text editor or read in from a data file. We
will be using the data.entry and edit functions. The data.entry function
allows editing of an existing object and automatically saving the changes to
the original object name. In contrast, the edit function allows editing of an
existing object but it will not save the changes to the original object name; we
must explicitly assign it to an object name (event if it is the original name).

2 In RStudio, select from main menu Code → Run Region, etc.

3 Editing of matrix and data frame objects is not currently supported in RStudio.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

114 3 Managing epidemiologic data in R

Fig. 3.1 Select Help from the main menu.

To enter a vector we need to initialize a vector and then use the data.entry
function (Figure 3.1).
> x <- numeric(10) #Initialize vector with zeros
> x
[1] 0 0 0 0 0 0 0 0 0 0
> data.entry(x) #Enter numbers, then close window
> x
[1] 1 2 3 4 5 6 7 8 9 10
However, the edit function applied to a vector does not open a spread-
sheet. Try the edit function and see what happens.
xnew <- edit(numeric(10)) #Edit number, then close window
To enter data into a spreadsheet matrix, first initialize a matrix and then
use the data.entry or edit function. Notice that the editor added default
column names. However, to add our own column names just click on the
column heading with our mouse pointer (unfortunately we cannot give row
names).
> xnew <- matrix(numeric(4),2,2)
> data.entry(xnew)
> xnew <- edit(xnew) #equivalent
>
> #open spreadsheet editor in one step
> xnew <- edit(matrix(numeric(4),2,2))
> xnew
col1 col2

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.1 Entering and importing data 115

[1,] 11 33
[2,] 22 44
Arrays and nontabular lists cannot be entered using a spreadsheet editor.
Hence, we begin to see the limitations of spreadsheet-type approach to data
entry. One type of list, the data frame, can be entered using the edit function.
To enter a data frame use the edit function. However, we do not need
to initialize a data frame (unlike with a matrix). Again, click on the column
headings to enter column names.
> df <- edit(data.frame()) #Spreadsheet screen not shown
> df
mykids age
1 Tomasito 7
2 Luisito 6
3 Angelita 3
When using the edit function to create a new data frame we must assign
it an object name to save the data frame. Later we will see that when we
edit an existing data object we can use the edit or fix function. The fix
function differs in that fix(data object ) saves our edits directly back to
data object without the need to make a new assignment.
mypower <- function(x, n){x^n}
fix(mypower) # Edits saved to ’mypower’ object
mypower <- edit(mypower) #equivalent

3.1.2 Importing data from a file

3.1.2.1 Reading an ASCII text data file

In this section we review how to read the following types of text data files:
• Comma-separated variable (csv) data file (± headers and ± row names)
• Fixed width formatted data file (± headers and ± row names)
Here is the University Group Diabetes Program randomized clinical trial
text data file that is comma-delimited, and includes row names and a header
(ugdp.txt).4 The header is the first line that contains the column (field)
names. The row names is the first column that starts on the second line
and uniquely identifies each row. Notice that the row names do not have a
column name associated with it. A data file can come with either row names
or header, neither, or both. Our preference is to work with data files that
have a header and data values that are self-explanatory. Even without a data
dictionary one can still make sense out of this data set.
4 Available at http://www.medepi.net/data/ugdp.txt

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

116 3 Managing epidemiologic data in R

Status,Treatment,Agegrp
1,Dead,Tolbutamide,<55
2,Dead,Tolbutamide,<55
...
408,Survived,Placebo,55+
409,Survived,Placebo,55+
Notice that the header row has 3 items, and the second row has 4 items. This
is because the row names start in the second row and have no column name.
This data file can be read in using the read.table function, and R figures
out that the first column are row names.5
> ud <- read.table("http://www.medepi.net/data/ugdp.txt",
+ header = TRUE, sep = ",")
> head(ud) #displays 1st 6 lines
Status Treatment Agegrp
1 Dead Tolbutamide <55
2 Dead Tolbutamide <55
3 Dead Tolbutamide <55
4 Dead Tolbutamide <55
5 Dead Tolbutamide <55
6 Dead Tolbutamide <55
Here is the same data file as it would appear without row names and
without a header (ugdp2.txt).
Dead,Tolbutamide,<55
Dead,Tolbutamide,<55
...
Survived,Placebo,55+
Survived,Placebo,55+
This data file can be read in using the read.table function. By default, it
adds row names (1, 2, 3, . . . ).
> cnames <- c("Status", "Treatment", "Agegrp")
> udat2 <- read.table("http://www.medepi.net/data/ugdp2.txt",
+ header = FALSE, sep = ",", col.names = cnames)
> head(udat2)
Status Treatment Agegrp
1 Dead Tolbutamide <55
2 Dead Tolbutamide <55
3 Dead Tolbutamide <55
4 Dead Tolbutamide <55
5 Dead Tolbutamide <55
6 Dead Tolbutamide <55

5 If row names are supplied, they must be unique.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.1 Entering and importing data 117

Here is the same data file as it might appear as a fix formatted file. In this
file, columns 1 to 8 are for field #1, columns 9 to 19 are for field #2, and
columns 20 to 22 are for field #3. This type of data file is more compact.
One needs a data dictionary to know which columns contain which fields.
Dead Tolbutamide<55
Dead Tolbutamide<55
...
SurvivedPlacebo 55+
SurvivedPlacebo 55+
This data file would be read in using the read.fwf function. Because the
field widths are fixed, we must strip the white space using the strip.white
option.
> cnames <- c("Status", "Treatment", "Agegrp")
> udat3 <- read.fwf("http://www.medepi.net/data/ugdp3.txt",
+ width = c(8, 11, 3), col.names = cnames, strip.white = TRUE)
> head(udat3)
Status Treatment Agegrp
1 Dead Tolbutamide <55
2 Dead Tolbutamide <55
3 Dead Tolbutamide <55
4 Dead Tolbutamide <55
5 Dead Tolbutamide <55
6 Dead Tolbutamide <55
Finally, here is the same data file as it might appear as a fixed width
formatted file but with numeric codes (ugdp4.txt). In this file, column 1 is
for field #1, column 2 is for field #2, and column 3 is for field #3. This type
of text data file is the most compact, however, one needs a data dictionary
to make sense of all the 1s and 2s.
121
121
...
212
212
Here is how this data file would be read in using the read.fwf function.
> cnames <- c("Status", "Treatment", "Agegrp")
> udat4 <- read.fwf("http://www.medepi.net/data/ugdp4.txt",
+ width = c(1, 1, 1), col.names = cnames)
> head(udat4)
Status Treatment Agegrp
1 1 2 1
2 1 2 1
3 1 2 1

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

118 3 Managing epidemiologic data in R

4 1 2 1
5 1 2 1
6 1 2 1
R has other functions for reading text data files (read.csv, read.csv2,
read.delim, read.delim2). In general, read.table is the function used most
commonly for reading in data files.

3.1.2.2 Reading data from a binary format (e.g., Stata, Epi Info)

To read data that comes in a binary or proprietary format load the foreign
package using the library function. To review available functions in the the
foreign package try help(package = foreign). For example, here we read
in the ‘infert’ data set which is also available as a Stata data file.6
> idat <- read.dta("c:/.../data/infert.dta")
> head(idat)[,1:8]
id education age parity induced case spontaneous stratum
1 1 0 26 6 1 1 2 1
2 2 0 42 1 1 1 0 2
3 3 0 39 6 2 1 0 3
4 4 0 34 4 2 1 0 4
5 5 1 35 3 1 1 1 5
6 6 1 36 4 2 1 1 6

3.1.3 Importing data using a URL

As we have already seen, text data files can be read directly off a web server
into R using the read.table function. Here we load the Western Collabora-
tive Group Study data directly off a web server.
> wdat <- read.table("http://www.medepi.net/data/wcgs.txt",
+ header = TRUE, sep = ",")
> str(wdat)
‘data.frame’: 3154 obs. of 14 variables:
$ id : int 2001 2002 2003 2004 2005 2006 2007 2010 ...
$ age0 : int 49 42 42 41 59 44 44 40 43 42 ...
$ height0: int 73 70 69 68 70 72 72 71 72 70 ...
$ weight0: int 150 160 160 152 150 204 164 150 190 175 ...
$ sbp0 : int 110 154 110 124 144 150 130 138 146 132 ...
$ dbp0 : int 76 84 78 78 86 90 84 60 76 90 ...
$ chol0 : int 225 177 181 132 255 182 155 140 149 325 ...

6 Available at http://www.medepi.net/data/infert.dta

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.2 Editing data 119

$ behpat0: int 2 2 3 4 3 4 4 2 3 2 ...

$ ncigs0 : int 25 20 0 20 20 0 0 0 25 0 ...
$ dibpat0: int 1 1 0 0 0 0 0 1 0 1 ...
$ chd69 : int 0 0 0 0 1 0 0 0 0 0 ...
$ typechd: int 0 0 0 0 1 0 0 0 0 0 ...
$ time169: int 1664 3071 3071 3064 1885 3102 3074 1032 ...
$ arcus0 : int 0 1 0 0 1 0 0 0 0 1 ...

3.2 Editing data

In the ideal setting, our data has already been checked, errors corrected, and
ready to be analyzed. Post-collection data editing can be minimized by good
design and data collection. However, we may still need to make corrections
or changes in data values.

3.2.1 Text editor

For small data sets, it may be convenience to edit the data in our favorite
text editor. Key-recording macros, and search and replace tools can be very
useful and efficient. Figure 3.2 on the following page displays West Nile virus
(WNV) infection surveillance data.7 This file is a comma-delimited data file
with a header.

3.2.2 The data.entry, edit, or fix functions

For vector and matrices we can use the data.entry function to edit these data
object elements. For data frames and functions use the edit or fix functions.
Remember that changes made with the edit function are not saved unless we
assign it to the original or new object name. In contrast, changes made with
the fix function are saved back to the original data object name. Therefore,
be careful when we use the fix function because we may unintentionally
overwrite data.
Now let’s read in the WNV surveillance raw data as a data frame. Then,
using the fix function, we will edit the first three records where the value for
the syndome variable is “Unk” and change it to NA for missing (Figure 3.3
on page 121). We will also change “.” to NA.

7 Raw data set available at http://www.medepi.net/data/wnv2004raw.txt, and clean data

set available at http://www.medepi.net/data/wnv2004fin.txt

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

120 3 Managing epidemiologic data in R

Fig. 3.2 Editing West Nile virus human surveillance data in text editor. Source: California
Department of Health Services, 2004

> wd <- read.table("http://www.medepi.net/data/wnv/wnv2004raw.txt",

+ header = TRUE, sep = ",", as.is = TRUE)
> wd[wd$syndrome=="Unknown",][1:3,] #before edits (3 records)
id county age sex syndrome date.onset date.tested death
128 128 Los Angeles 81 M Unknown 07/28/2004 08/11/2004 .
129 129 Riverside 44 F Unknown 07/25/2004 08/11/2004 .
133 133 Los Angeles 36 M Unknown 08/04/2004 08/11/2004 No
> fix(wd) #open R spreadsheet and make edits (see figure)
> wd[c(128, 129, 133),] #after edits (3 records)
id county age sex syndrome date.onset date.tested death
128 128 Los Angeles 81 M NA 07/28/2004 08/11/2004 NA
129 129 Riverside 44 F NA 07/25/2004 08/11/2004 NA
133 133 Los Angeles 36 M NA 08/04/2004 08/11/2004 No
First, notice that in the read.table function as.is=TRUE. This means the
data is read in without R making any changes to it. In other words, character
vectors are not automatically converted to factors. We set the option because
we knew we were going to edit and make corrections to the data set, and cre-
ate factors later. In this example, I manually started changing the missing val-
ues “Unknown” to NA (R’s representation of missing values). However, this
manual approach would be very inefficient. A better approach is to specify
which values in the data frame should be converted to NA. In the read.table
function we should have set the option na.string=c("Unknown", "."), con-
verting the character strings “Unknown” and “.” into NA. Let’s replace the
missing values with NAs upon reading the data file.
> wd <- read.table("http://www.medepi.net/data/wnv/wnv2004raw.txt",
+ header = TRUE, sep = ",", as.is = TRUE,

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.2 Editing data 121

Fig. 3.3 Using the fix function to edit the WNV surveillance data frame. Unfortunately,
this approach does not facilitate documenting our edits. Source: California Department of
Health Services, 2004

+ na.string=c("Unknown", "."))
> wd[c(128, 129, 133),] #verify change
id county age sex syndrome date.onset date.tested death
128 128 Los Angeles 81 M <NA> 07/28/2004 08/11/2004 <NA>
129 129 Riverside 44 F <NA> 07/25/2004 08/11/2004 <NA>
133 133 Los Angeles 36 M <NA> 08/04/2004 08/11/2004 No

3.2.3 Vectorized approach

How do we make these and other changes after the data set has been read
into R? Although using R’s spreadsheet function is convenient, we do not
recommend it because manual editing is inefficient, our work cannot be repli-
cated and audited, and documentation is poor. Instead use R’s vectorized
approach. Let’s look at the distribution of responses for each variable to as-
sess what needs to be “cleaned up,” in addition to converting missing values
to NA.
> wd <- read.table("http://www.medepi.net/data/wnv/wnv2004raw.txt",
+ header = TRUE, sep = ",", as.is = TRUE)
> str(wd)
‘data.frame’: 779 obs. of 8 variables:
$ id : int 1 2 3 4 5 6 7 8 9 10 ...
$ county : chr "San Bernardino" "San Bernardino" ...
$ age : chr "40" "64" "19" "12" ...

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

122 3 Managing epidemiologic data in R

$ sex : chr "F" "F" "M" "M" ...

$ syndrome : chr "WNF" "WNF" "WNF" "WNF" ...
$ date.onset : chr "05/19/2004" "05/22/2004" ...
$ date.tested: chr "06/02/2004" "06/16/2004" ...
$ death : chr "No" "No" "No" "No" ...
> lapply(wd, table) #apply ’table’ function to fields
$id

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
...
768 769 770 771 772 773 774 775 776 777 778 779 780 781
1 1 1 1 1 1 1 1 1 1 1 1 1 1

$county

Butte Fresno Glenn Imperial

7 11 3 1
Kern Lake Lassen Los Angeles
59 1 1 306
Merced Orange Placer Riverside
1 62 1 109
Sacramento San Bernardino San Diego San Joaquin
3 187 2 2
Santa Clara Shasta Sn Luis Obispo Tehama
1 5 1 10
Tulare Ventura Yolo
3 2 1

$age

. 1 10 11 12 13 14 15 16 17 18 19 2 20 21 22 23 24 25 26
6 1 1 1 3 2 3 3 1 4 6 5 1 4 2 3 6 8 3 9
...
82 83 84 85 86 87 88 89 9 91 93 94
10 5 6 4 2 2 1 6 1 4 1 1

$sex

. F M
2 294 483

$syndrome

Unknown WNF WNND

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.2 Editing data 123

105 391 283

$date.onset

02/02/2005 05/14/2004 05/16/2004 05/19/2004 05/22/2004

1 1 1 1 2
...
10/28/2004 10/29/2004 10/30/2004 11/08/2004 11/12/2004
2 1 1 4 2

$date.tested

01/21/2005 02/04/2005 02/23/2005 06/02/2004 06/16/2004

1 1 1 1 4
...
11/29/2004 12/02/2004 12/03/2004 12/07/2004
8 1 2 1

$death

. No Yes
66 686 27
What did we learn? First, there are 779 observations and 781 id’s; there-
fore, 3 observations were removed from the original data set. Second, we see
that the variables age, sex, syndrome, and death have missing values that
need to be converted to NAs. This can be done one field at a time, or for the
whole data frame in one step. Here is the R code.
#individually
wd$age[wd$age=="."] <- NA
wd$sex[wd$sex=="."] <- NA
wd$syndrome[wd$syndrome=="Unknown"] <- NA
wd$death[wd$death=="."] <- NA

#or globally
wd[wd=="." | wd=="Unknown"] <- NA
After running the above code, let’s evaluate one variable to verify the
missing values were converted to NAs.
> table(wd$death)

No Yes
686 27
> table(wd$death, exclude=NULL)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

124 3 Managing epidemiologic data in R

No Yes <NA>
686 27 66
We also notice that the entry for one of the counties, San Luis Obispo,
was misspelled (Sn Luis Obispo). We can use replacement to make the cor-
rections:
> wd$County[wd$County=="Sn Luis Obispo"] <- "San Luis Obispo"

3.2.4 Text processing

On occasion, we will need to process and manipulate character vectors using

a vectorized approach. For example, suppose we need to convert a character
vector of dates from “mm/dd/yy” to “yyyy-mm-dd”.8 We’ll start by using the
substr function. This function extracts characters from a character vector
based on position.
> bd <- c("07/17/96","12/09/00","11/07/97")
> mon <- substr(bd, start=1, stop=2); mon
[1] "07" "12" "11"
> day <- substr(bd, 4, 5); day
[1] "17" "09" "07"
> yr <- as.numeric(substr(bd, 7, 8)); yr
[1] 96 0 97
> yr2 <- ifelse(yr<=19, yr+2000, yr+1900); yr2
[1] 1996 2000 1997
> bdfin <- paste(yr2, "-", mon, "-", day, sep=""); bdfin
[1] "1996-07-17" "2000-12-09" "1997-11-07"
In this example, we needed to convert “00” to “2000”, and “96” and “97 to
“1996” and “1997”, respectively. The trick here was to coerce the character
vector into a numeric vector so that 1900 or 2000 could be added to it. Using
the ifelse function, for values ≤ 19 (arbitrarily chosen), 2000 was added,
otherwise 1900 was added. The paste function was used to paste back the
components into a new vector with the standard date format.
The substr function can also be used to replace characters in a character
vector. Remember, if it can be indexed, it can be replaced.
> bd
[1] "07/17/96" "12/09/00" "11/07/97"
> substr(bd, 3, 3) <- "-"
> substr(bd, 6, 6) <- "-"
> bd
[1] "07-17-96" "12-09-00" "11-07-97"
8 ISO 8601 is an international standard for date and time representations issued by the

International Organization for Standardization (ISO). See http://www.iso.org

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.3 Sorting data 125

Table 3.2 R functions for processing text in character vectors

Function Description Try these examples
nchar Returns the number x <- c("a", "ab", "abc", "abcd")
of characters in nchar(x)
each element of a
character vector
substr Extract or replace #extraction
substrings in a mon <- substr(some.dates, 1, 2); mon
character vector day <- substr(some.dates, 4, 5); day
yr <- substr(some.dates, 7, 8); yr
#replacement
mdy <- paste(mon, day, yr); mdy
substr(mdy, 3, 3) <- ’/’
substr(mdy, 6, 6) <- ’/’
mdy
paste Concatenate vectors rd <- paste(mon, "/", day, "/", yr, sep="")
after converting to rd
character
strsplit Split the elements some.dates <- c("10/02/70", "02/04/67")
of a character some.dates
vector into strsplit(some.dates, "/")
substrings

3.3 Sorting data

The sort function sorts a vector as expected:

> x <- sample(1:10, 10); x
[1] 4 3 6 1 7 9 5 8 2 10
> sort(x)
[1] 1 2 3 4 5 6 7 8 9 10
> sort(x, decreasing = TRUE) #reverse sort
[1] 10 9 8 7 6 5 4 3 2 1
> rev(sort(x)) #reverse sort
[1] 10 9 8 7 6 5 4 3 2 1
However, if we want to sort one vector based on the ordering of elements
from another vector, use the order function. The order function generates
an indexing/repositioning vector. Study the following example:
> x <- sample(1:20, 5); x
[1] 18 10 6 13 11
> sort(x) #sorts as expected
[1] 6 10 11 13 18
> y <- sample(1:20, 5); y
[1] 11 10 13 3 9
> order(y) #4th element to 1st position, 5th to 2nd, etc.
[1] 4 5 2 1 3

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

126 3 Managing epidemiologic data in R

> x[order(y)] #use order(y) to sort elements of x

[1] 13 11 10 18 6
Based on this we can see that sort(x) is just x[order(x)].
Now let us see how to use the order function for data frames. First, we
create a small data set.
> sex <- rep(c("Male", "Female"), c(4, 4))
> ethnicity <- rep(c("White", "African American", "Latino",
+ "Asian"), 2)
> age <- sample(1:100, 8)
> dat <- data.frame(age, sex, ethnicity)
> dat <- dat[sample(1:8, 8),] #randomly order rows
> dat
age sex ethnicity
5 57 Female White
8 93 Female Asian
1 7 Male White
4 65 Male Asian
6 38 Female African American
3 27 Male Latino
2 66 Male African American
7 72 Female Latino
Okay, now we will sort the data frame based on the ordering of one field, and
then the ordering of two fields:
> dat[order(dat$age),] #sort based on 1 variable
age sex ethnicity
1 7 Male White
3 27 Male Latino
6 38 Female African American
5 57 Female White
4 65 Male Asian
2 66 Male African American
7 72 Female Latino
8 93 Female Asian
> dat[order(dat$sex, dat$age),] #sort based on 2 variables
age sex ethnicity
6 38 Female African American
5 57 Female White
7 72 Female Latino
8 93 Female Asian
1 7 Male White
3 27 Male Latino
4 65 Male Asian
2 66 Male African American

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.4 Indexing (subsetting) data 127

3.4 Indexing (subsetting) data

For this section, please load the well known Oswego foodborne illness dataset:
> odat <- read.table("http://www.medepi.net/data/oswego.txt",
+ header = TRUE, as.is = TRUE, sep = "")
> str(odat)
‘data.frame’: 75 obs. of 21 variables:
$ id : int 2 3 4 6 7 8 9 10 14 16 ...
$ age : int 52 65 59 63 70 40 15 33 10 32 ...
$ sex : chr "F" "M" "F" "F" ...
$ meal.time : chr "8:00 PM" "6:30 PM" "6:30 PM" ...
$ ill : chr "Y" "Y" "Y" "Y" ...
$ onset.date : chr "4/19" "4/19" "4/19" "4/18" ...
$ onset.time : chr "12:30 AM" "12:30 AM" ...
$ baked.ham : chr "Y" "Y" "Y" "Y" ...
...
$ vanilla.ice.cream : chr "Y" "Y" "Y" "Y" ...
$ chocolate.ice.cream: chr "N" "Y" "Y" "N" ...
$ fruit.salad : chr "N" "N" "N" "N" ...

3.4.1 Indexing

Now, we will practice indexing rows from this data frame. First, we create a
new data set that contains only cases. To index the rows with cases we need
to generate a logical vector that is TRUE for every value of odat$ill that “is
equivalent to” "Y". For “is equivalent to” we use the == relational operator.
> cases <- odat$ill=="Y"
> cases
[1] TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE
...
[73] FALSE FALSE FALSE
> odat.ca <- odat[cases, ]
> odat.ca[, 1:8]
id age sex meal.time ill onset.date onset.time baked.ham
1 2 52 F 8:00 PM Y 4/19 12:30 AM Y
2 3 65 M 6:30 PM Y 4/19 12:30 AM Y
3 4 59 F 6:30 PM Y 4/19 12:30 AM Y
4 6 63 F 7:30 PM Y 4/18 10:30 PM Y
...
43 71 60 M 7:30 PM Y 4/19 1:00 AM N
44 72 18 F 7:30 PM Y 4/19 12:00 AM Y
45 74 52 M <NA> Y 4/19 2:15 AM Y

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

128 3 Managing epidemiologic data in R

46 75 45 F <NA> Y 4/18 11:00 PM Y

It is very important to understand what we just did: we extracted the rows
with cases by indexing the data frame with a logical vector.
Now, we combine relational operators with logical operators to extract
rows based on multiple criteria. Let’s create a data set with female cases, age
less than the median age, and consumed vanilla ice cream.
> fem.cases.vic <- odat$ill=="Y" & odat$sex=="F" &
+ odat$vanilla.ice.cream=="Y" & odat$age < median(odat$age)
> odat.fcv <- odat[fem.cases.vic, ]
> odat.fcv[ , c(1:6, 19)]
id age sex meal.time ill onset.date vanilla.ice.cream
8 10 33 F 7:00 PM Y 4/18 Y
10 16 32 F <NA> Y 4/19 Y
13 20 33 F <NA> Y 4/18 Y
14 21 13 F 10:00 PM Y 4/19 Y
18 27 15 F 10:00 PM Y 4/19 Y
23 36 35 F <NA> Y 4/18 Y
31 48 20 F 7:00 PM Y 4/19 Y
37 58 12 F 10:00 PM Y 4/19 Y
40 65 17 F 10:00 PM Y 4/19 Y
41 66 8 F <NA> Y 4/19 Y
42 70 21 F <NA> Y 4/19 Y
44 72 18 F 7:30 PM Y 4/19 Y
In summary, we see that indexing rows of a data frame consists of using re-
lational operators (<, >, <=, >=, ==, !=) and logical operators (&, |, !)
to generate a logical vector for indexing the appropriate rows.

3.4.2 Subsetting

Subsetting a data frame using the subset function is equivalent to using log-
ical vectors to index the data frame. In general, we prefer indexing because it
is generalizable to indexing any R data object. However, the subset function
is a convenient alternative for data frames. Again, let’s create data set with
female cases, age < median, and ate vanilla ice cream.
> odat.fcv <- subset(odat, subset = {ill=="Y" & sex=="F" &
+ vanilla.ice.cream=="Y" & age < median(odat$age)},
+ select = c(id:onset.date, vanilla.ice.cream))
> odat.fcv
id age sex meal.time ill onset.date vanilla.ice.cream
8 10 33 F 7:00 PM Y 4/18 Y
10 16 32 F . Y 4/19 Y

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.5 Transforming data 129

13 20 33 F . Y 4/18 Y
14 21 13 F 10:00 PM Y 4/19 Y
18 27 15 F 10:00 PM Y 4/19 Y
23 36 35 F . Y 4/18 Y
31 48 20 F 7:00 PM Y 4/19 Y
37 58 12 F 10:00 PM Y 4/19 Y
40 65 17 F 10:00 PM Y 4/19 Y
41 66 8 F . Y 4/19 Y
42 70 21 F . Y 4/19 Y
44 72 18 F 7:30 PM Y 4/19 Y
In the subset function, the first argument is the data frame object name,
the second argument (also called subset) evaluates to a logical vector, and
third argument (called select) specifies the fields to keep. In the second
argument,
subset = {...}
the curly brackets are included for convenience to group the logical and re-
lational operations. In the select argument, using the : operator, we can
specify a range of fields to keep.

3.5 Transforming data

Transforming fields in a data frame is very common. The most common

transformations include the following:
• Numerical transformation of a numeric vector
• Discretizing a numeric vector into categories or levels (“categorical vari-
able”)
• Re-coding integers that represent levels of a categorical variable
For each of these, we must decide whether the newly created vector should be
a new field in the data frame, overwrite the original field in the data frame, or
not be a field in the data frame (but rather a vector object in the workspace).
For the examples that follow load the well known Oswego foodborne illness
dataset:
> odat <- read.table("http://www.medepi.net/data/oswego.txt",
+ header = TRUE, as.is = TRUE, sep = "")

3.5.1 Numerical transformation

> # transform age variable centering it

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

130 3 Managing epidemiologic data in R

> # create new field in same data frame

> odat$age
[1] 52 65 59 63 70 40 15 33 10 32 62 36 33 13 7 3 59 15
...
[73] 17 36 14
> odat$age.centered <- odat$age - mean(odat$age)
> odat$age.centered
[1] 15.1866667 28.1866667 22.1866667 26.1866667
...
[73] -19.8133333 -0.8133333 -22.8133333
>
> # overwrite original field in same data frame (not recommended!!!)
> # odat$age <- odat$age - mean(odat$age)
>
> # create new vector in workspace; data frame remains unchanged
> age.centered <- odat$age - mean(odat$age)
> age.centered
[1] 15.1866667 28.1866667 22.1866667 26.1866667
...
[73] -19.8133333 -0.8133333 -22.8133333

For convenience, the transform function facilitates the transformation of

numeric vectors in a data frame. The transform function comes in handy
when we plan on transforming many fields: we do not need to specify the
data frame each time we refer to a field name. For example, the following
lines are are equivalent. Both add a new transformed field to the data frame.
odat$age.centered <- odat$age - mean(odat$age)
odat <- transform(odat, age.centered = age - mean(age))

3.5.2 Creating categorical variables (factors)

Now, reload the Oswego data set to recover the original odat$age field. We
are going to create a new field with the following seven age categories (in
years): < 1, 1 to 4, 5 to 14, 15 to 24, 25 to 44, 45 to 64, and 65+. We will
demonstrate this using several methods:

3.5.2.1 Using cut function (preferred method)

> agecat <- cut(odat$age, breaks = c(0, 1, 5, 15, 25, 45,

+ 65, 100))
> agecat
[1] (45,65] (45,65] (45,65] (45,65] (65,100] (25,45]

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.5 Transforming data 131

...
[73] (15,25] (25,45] (5,15]
Levels: (0,1] (1,5] (5,15] (15,25] (25,45] ... (65,100]

Note that the cut function generated a factor with 7 levels for each interval.
The notation (15, 25] means that the interval is open on the left boundary
(> 15) and closed on the right boundary (≤ 25). However, for age categories,
it makes more sense to have age boundaries closed on the left and open on
the right: [a, b). To change this we set the option right = FALSE
> agecat <- cut(odat$age, breaks = c(0, 1, 5, 15, 25, 45,
+ 65, 100), right = FALSE)
> agecat
[1] [45,65) [65,100) [45,65) [45,65) [65,100) [25,45)
...
[73] [15,25) [25,45) [5,15)
Levels: [0,1) [1,5) [5,15) [15,25) [25,45) ... [65,100)
> table(agecat)
agecat
[0,1) [1,5) [5,15) [15,25) [25,45) [45,65) [65,100)
0 1 14 13 18 20 9
Okay, this looks good, but we can add labels since our readers may not be
familiar with open and closed interval notation [a, b).
> agelabs <- c("<1", "1-4", "5-14", "15-24", "25-44", "45-64",
+ "65+")
> agecat <- cut(odat$age, breaks = c(0, 1, 5, 15, 25, 45,
+ 65, 100), right = FALSE, labels = agelabs)
> agecat
[1] 45-64 65+ 45-64 45-64 65+ 25-44 15-24
...
[71] 5-14 5-14 15-24 25-44 5-14
Levels: <1 1-4 5-14 15-24 25-44 45-64 65+
> table(agecat, case = odat$ill)
case
agecat N Y
<1 0 0
1-4 0 1
5-14 8 6
15-24 5 8
25-44 8 10
45-64 5 15
65+ 3 6

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

132 3 Managing epidemiologic data in R

3.5.2.2 Using indexing and assignment (replacement)

The cut function is the preferred method to create a categorical variable.

However, suppose one does not know about the cut function. Applying basic
R concepts always works!
> agegroup <- odat$age
> agegroup[odat$age<1] <- 1
> agegroup[odat$age>=1 & odat$age<5] <- 2
> agegroup[odat$age>=5 & odat$age<15] <- 3
> agegroup[odat$age>=15 & odat$age<25] <- 4
> agegroup[odat$age>=25 & odat$age<45] <- 5
> agegroup[odat$age>=45 & odat$age<65] <- 6
> agegroup[odat$age>=65] <- 7
> #create factor
> agelabs <- c("<1", "1 to 4", "5 to 14", "15 to 24",
+ "25 to 44", "45 to 64", "65+")
> agegroup <- factor(agegroup, levels = 1:7, labels = agelabs)
> agegroup
[1] 45 to 64 65+ 45 to 64 45 to 64 65+ 25 to 44
...
[73] 15 to 24 25 to 44 5 to 14
7 Levels: <1 1 to 4 5 to 14 15 to 24 25 to 44 ... 65+
> table(case = odat$ill, agegroup)
agegroup
case <1 1 to 4 5 to 14 15 to 24 25 to 44 45 to 64 65+
N 0 0 8 5 8 5 3
Y 0 1 6 8 10 15 6
In these previous examples, notice that agegroup is a factor object that
is not a field in the odat data frame.

3.5.3 “Re-coding” levels of a categorical variable

In the previous example the categorical variable was a numeric vector (1,
2, 3, 4, 5, 6, 7) that was converted to a factor and provided labels (“<1”,
“1 to 4”, “5 to 14”, . . . ). In fact, categorical variables are often represented
by integers (for example, 0 = no, 1 = yes; or 0 = non-case, 1 = case) and
provided labels. Often, ASCII text data files are integer codes that require
a data dictionary to convert these integers into categorical variables in a
statistical package. In R, keeping track of integer codes for categorical vari-
ables is unnecessary. Therefore, re-coding the underlying integer codes is also
unnecessary; however, if we feel the need to do so, here’s how.
> # Create categorical variable

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.5 Transforming data 133

> ethlabs <- c("White", "Black", "Latino", "Asian")

> ethnicity <- sample(ethlabs, 100, replace = T)
> ethnicity <- factor(ethnicity, levels = ethlabs)
> ethnicity
[1] Black Asian Latino White Black Asian White Black
...
[97] Black Black Asian Latino
Levels: White Black Latino Asian
The levels option allowed us to determine the display order, and the
first level becomes the reference level in statistical models. To display the
underlying numeric code use unclass function which preserves the levels
attribute.9
> x <- unclass(ethnicity)
> x
[1] 2 4 3 1 2 4 1 2 1 3 4 3 2 1 3 2 1 1 1 3 1 2 2 1 3 4 2 3
...
[85] 2 2 3 3 3 3 1 3 4 4 1 1 2 2 4 3
attr(,"levels")
[1] "White" "Black" "Latino" "Asian"
To recover the original factor,
> factor(x,labels=levels(x))
[1] Black Asian Latino White Black Asian White
...
[92] Latino Asian Asian White White Black Black
[99] Asian Latino
Levels: White Black Latino Asian
Although one can extract the integer code, why would one need to do so?
One is tempted to use the integer codes as a way to share data sets. However,
we recommend not using the integer codes, but rather just provide the data
in its native format.10 This way, the raw data is more interpretable and
eliminates the intermediate step of needing to label the integer code. Also, if
the data dictionary is lost or not provided, the raw data is still interpretable.
In R, we can re-label the levels using the levels function and assigning
to it a character vector of new labels. Make sure the order of the new labels
corresponds to order of the factor levels.
> levels(ethnicity2) <- c("Caucasion", "African American",
+ "Hispanic", "Asian")
> table(ethnicity2)
ethnicity2
Caucasion African American Hispanic Asian
9 The as.integer function also works but does not preserve the levels attribute.
10 For example, http://www.medepi.net/data/oswego.txt

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

134 3 Managing epidemiologic data in R

23 31 28 18
In R, we can re-order and re-label at the same time using the levels
function and assigning to it a list.
> table(ethnicity)
ethnicity
White Black Latino Asian
23 31 28 18
> ethnicity3 <- ethnicity
> levels(ethnicity3) <- list(Hispanic = "Latino", Asian = "Asian",
+ Caucasion = "White", "African American" = "Black")
> table(ethnicity3)
ethnicity3
Hispanic Asian Caucasion African American
28 18 23 31
The list function is necessary to assure the re-ordering. To re-order without
re-labeling just do the following:
> table(ethnicity)
ethnicity
White Black Latino Asian
23 31 28 18
> ethnicity4 <- ethnicity
> levels(ethnicity4) <- list(Latino = "Latino", Asian = "Asian",
+ White = "White", Black = "Black")
> table(ethnicity4)
ethnicity4
Latino Asian White Black
28 18 23 31
In R, we can sort the factor levels by using the factor function in one of
two ways:
> table(ethnicity)
ethnicity
White Black Latino Asian
23 31 28 18
> ethnicity5a <- factor(ethnicity, sort(levels(ethnicity)))
> table(ethnicity5a)
ethnicity5a
Asian Black Latino White
18 31 28 23
> ethnicity5b <- factor(as.character(ethnicity))
> table(ethnicity5b)
ethnicity5b
Asian Black Latino White
18 31 28 23

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.5 Transforming data 135

Table 3.3 Categorical variable represented as a factor or a set of dummy variables

Factor Dummy variables

Ethnicity Asian Black Latino

White 0 0 0
Asian 1 0 0
Black 0 1 0
Latino 0 0 1

In the first example, we assigned to the levels argument the sorted level
names. In the second example, we started from scratch by coercing the orig-
inal factor into a character vector which is then ordered alphabetically by
default.

3.5.3.1 Setting factor reference level

The first level of a factor is the reference level for some statistical models
(e.g., logistic regression). To set a different reference level use the relevel
function.
> levels(ethnicity)
[1] "White" "Black" "Latino" "Asian"
> ethnicity6 <- relevel(ethnicity, ref = "Asian")
> levels(ethnicity6)
[1] "Asian" "White" "Black" "Latino"
As we can see, there is tremendous flexibility in dealing with factors with-
out the need to “re-code” categorical variables. This approach facilitates re-
viewing our work and minimizes errors.

3.5.4 Use factors instead of dummy variables

A nonordered factor (nominal categorical variable) with k levels can also

be represented with k − 1 dummy variables. For example, the ethnicity
factor has four levels: white, Asian, black, and Latino. Ethnicity can also be
represented using 3 dichotomous variables, each coded 0 or 1. For example,
using white as the reference group, the dummy variables would be asian,
black, and latino (see Table 3.3). The values of those three dummy variables
(0 or 1) are sufficient to represents one of four possible ethnic categories.
Dummy variables can be used in statistical models. However, in R, it is
unnecessary to create dummy variables, just create a factor with the desired
number of levels and set the reference level.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

136 3 Managing epidemiologic data in R

3.5.5 Conditionally transforming the elements of a

vector

We can conditionally transform the elements of a vector using the ifelse

function. This function works as follows: ifelse(test, if test = TRUE do
this, else do this ).
> x <- sample(c("M", "F"), 10, replace = T); x
[1] "M" "F" "M" "F" "M" "F" "M" "F" "M" "F"
> y <- ifelse(x=="M", "Male", "Female")
> y
[1] "Male" "Female" "Male" "Female" "Male" "Female"
[7] "Male" "Female" "Male" "Female"

3.6 Merging data

In general, R’s strength is not data management but rather data analysis.
Because R can access and operate on multiple objects in the workspace it is
generally not necessary to merge data objects into one data object in order
to conduct analyses. On occasion, it may be necessary to merge two data
frames into one data frames
Data frames that contain data on individual subjects are generally of two
types: (1) each row contains data collected on one and only one individual,
or (2) multiple rows contain repeated measurements on individuals. The lat-
ter approach is more efficient at storing data. For example, here are two
approaches to collecting multiple telephone numbers for two individuals.
> tab1
name wphone fphone mphone
1 Tomas Aragon 643-4935 643-2926 847-9139
2 Wayne Enanoria 643-4934 <NA> <NA>
>
> tab2
name telphone teletype
1 Tomas Aragon 643-4935 Work
2 Tomas Aragon 643-2926 Fax
3 Tomas Aragon 847-9139 Mobile
4 Wayne Enanoria 643-4934 Work
The first approach is represented by tab1, and the second approach by
tab2.11 Data is more efficiently stored in tab2, and adding new types of

11 This approach is the basis for designing and implementing relational databases. A
relational database consists of multiple tables linked by an indexing field.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.6 Merging data 137

Table 3.4 R functions for transforming variables in data frames

Function Description Try these examples
<- Transforming a dat <- data.frame(id=1:3, x=c(0.5,1,2)); dat
vector and dat$logx <- log(x) #creates new field
assigning it to a dat
new data frame
variable name
transform Transform one or dat <- data.frame(id=1:3, x=c(0.5,1,2)); dat
more variables dat <- transform(dat, logx = log(x))
from a data frame dat
cut Creates a factor by age <- sample(1:100, 500, replace = TRUE)
dividing the range # cut into 2 intervals agecut <- cut(age, 2,
of a numeric vector right = FALSE)
into intervals table(agecut)
#cut using specified intervals agecut2 <-
cut(age, c(0, 50 100),
right = FALSE, include.lowest = TRUE)
table(agecut2)
levels Gives access to the sex <- sample(c("M","F","T"),500, replace=T)
levels attribute of a sex <- factor(sex)
factor table(sex)
# relabel each level; use same order
levels(sex) <- c("Female", "Male",
"Transgender")
table(sex)
# relabel/recombine
levels(sex) <- c("Female", "Male", "Male")
table(sex)
# reorder and/or relabel
levels(sex) <- list ("Men" = "Male",
"Women" = "Female")
table(sex)
relevel Set the reference sex2 <- relevel(sex, ref = "Women")
level for a factor table(sex2)
ifelse Conditionally age <- sample(1:100, 1000, replace = TRUE)
operate on agecat <- ifelse(age<=50, "<=50", ">50")
elements of a table(agecat)
vector based on a
test

telephone numbers only requires assigning a new value (e.g., Pager) to the
teletype field.
> tab2
name telphone teletype
1 Tomas Aragon 643-4935 Work
2 Tomas Aragon 643-2926 Fax
3 Tomas Aragon 847-9139 Mobile
4 Wayne Enanoria 643-4934 Work

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

138 3 Managing epidemiologic data in R

5 Tomas Aragon 719-1234 Pager

In both these data frames, an indexing field identifies an unique individual
that is associated with each row. In this case, the name column is the indexing
field for both data frames.
Now, let’s look at an example of two related data frames that are linked
by an indexing field. The first data frame contains telephone numbers for 5
employees and fname is the indexing field. The second data frame contains
email addresses for 3 employees and name is the indexing field.
> phone
fname phonenum phonetype
1 Tomas 643-4935 work
2 Tomas 847-9139 mobile
3 Tomas 643-4926 fax
4 Chris 643-3932 work
5 Chris 643-4926 fax
6 Wayne 643-4934 work
7 Wayne 643-4926 fax
8 Ray 643-4933 work
9 Ray 643-4926 fax
10 Diana 643-3931 work
> email
name mail mailtype
1 Tomas [email protected] Work
2 Tomas [email protected] Personal
3 Wayne [email protected] Work
4 Wayne [email protected] Work
5 Chris [email protected] Work
6 Chris [email protected] Personal
To merge these two data frames use the merge function.
> dat <- merge(email, phone, by.x="name", by.y="fname")
> dat
name mail mailtype phonenum phonetype
1 Chris [email protected] Work 643-3932 work
2 Chris [email protected] Personal 643-3932 work
3 Chris [email protected] Work 643-4926 fax
4 Chris [email protected] Personal 643-4926 fax
5 Tomas [email protected] Work 643-4935 work
6 Tomas [email protected] Personal 643-4935 work
7 Tomas [email protected] Work 847-9139 mobile
8 Tomas [email protected] Personal 847-9139 mobile
9 Tomas [email protected] Work 643-4926 fax
10 Tomas [email protected] Personal 643-4926 fax
11 Wayne [email protected] Work 643-4934 work

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.6 Merging data 139

12 Wayne [email protected] Work 643-4934 work

13 Wayne [email protected] Work 643-4926 fax
14 Wayne [email protected] Work 643-4926 fax
> dat <- merge(phone, email, by.x="fname", by.y="name")
> dat
fname phonenum phonetype mail mailtype
1 Chris 643-3932 work [email protected] Work
2 Chris 643-4926 fax [email protected] Work
3 Chris 643-3932 work [email protected] Personal
4 Chris 643-4926 fax [email protected] Personal
5 Tomas 643-4935 work [email protected] Work
6 Tomas 847-9139 mobile [email protected] Work
7 Tomas 643-4926 fax [email protected] Work
8 Tomas 643-4935 work [email protected] Personal
9 Tomas 847-9139 mobile [email protected] Personal
10 Tomas 643-4926 fax [email protected] Personal
11 Wayne 643-4934 work [email protected] Work
12 Wayne 643-4926 fax [email protected] Work
13 Wayne 643-4934 work [email protected] Work
14 Wayne 643-4926 fax [email protected] Work
The by.x and by.y options identify the indexing fields. By default, R selects
the rows from the two data frames that is based on the intersection of the
indexing fields (by.x, by.y). To merge the union of the indexing fields, set
all=TRUE:
> dat <- merge(phone, email, by.x="fname", by.y="name",
+ all=TRUE)
> dat
fname phonenum phonetype mail mailtype
1 Chris 643-3932 work [email protected] Work
2 Chris 643-4926 fax [email protected] Work
3 Chris 643-3932 work [email protected] Personal
4 Chris 643-4926 fax [email protected] Personal
5 Diana 643-3931 work <NA> <NA>
6 Ray 643-4933 work <NA> <NA>
7 Ray 643-4926 fax <NA> <NA>
8 Tomas 643-4935 work [email protected] Work
9 Tomas 847-9139 mobile [email protected] Work
10 Tomas 643-4926 fax [email protected] Work
11 Tomas 643-4935 work [email protected] Personal
12 Tomas 847-9139 mobile [email protected] Personal
13 Tomas 643-4926 fax [email protected] Personal
14 Wayne 643-4934 work [email protected] Work
15 Wayne 643-4926 fax [email protected] Work
16 Wayne 643-4934 work [email protected] Work

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

140 3 Managing epidemiologic data in R

17 Wayne 643-4926 fax [email protected] Work

To “reshape” tabular data look up and study the reshape and stack
functions.

3.7 Executing commands from, and directing output to,

a file

3.7.1 The source function

We use the source function to execute R commands are contained in an

ASCII text file. For example, consider the contents this source file (chap03.R):
i <- 1:5
x <- outer(i, i, "*")
show(x)
Here we run source from the R command prompt:
> source("/home/tja/Documents/wip/epir/r/chap03.R")
[,1] [,2] [,3] [,4] [,5]
[1,] 1 2 3 4 5
[2,] 2 4 6 8 10
[3,] 3 6 9 12 15
[4,] 4 8 12 16 20
[5,] 5 10 15 20 25
Nothing is printed to the console unless we explicitly use the the show (or
print) function. This enables us to view only the results we want to review.
An alternative approach is to print everything to the console as if the R
commands were being enter directly at the command prompt. For this we do
not need to use the show function in the source file; however, we must set the
echo option to TRUE in the source function. Here is the edited source file
(chap03.R)
i <- 1:5
x <- outer(i, i, "*")
x
Here we run source from the R command prompt:
> source("/home/tja/Documents/wip/epir/r/chap03.R", echo=TRUE)
> i <- 1:5
> x <- outer(i, i, "*")
> x
[,1] [,2] [,3] [,4] [,5]
[1,] 1 2 3 4 5

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.7 Executing commands from, and directing output to, a file 141

[2,] 2 4 6 8 10
[3,] 3 6 9 12 15
[4,] 4 8 12 16 20
[5,] 5 10 15 20 25

3.7.2 The sink and capture.output functions

We can add code to the source file to “sink” selected results to an output
file using the sink or capture.output functions. Consider our edited source
file:
i <- 1:5
x <- outer(i, i, "*")
sink("/home/tja/Documents/wip/epir/r/chap03.log")
cat("Here are the results of the outer function", fill=TRUE)
show(x)
sink()
Here we run source from the R command prompt:
> source("/home/tja/Documents/wip/epir/r/chap03.R")
>
Nothing was printed to the console because sink sent it to the output file
(chap03.log). Here are the contents of chap03.log:
Here are the results of the outer function
[,1] [,2] [,3] [,4] [,5]
[1,] 1 2 3 4 5
[2,] 2 4 6 8 10
[3,] 3 6 9 12 15
[4,] 4 8 12 16 20
[5,] 5 10 15 20 25
The first sink opened a connection and created the output file (chap03.log).
The cat and show functions printed results to output file. The second sink
close the connection.
Alternatively, as before, if we use the echo = TRUE option in the source
function, everything is either printed to the console or output file. The sink
connection determines what is printed to the output file. Here is the edited
source file (chap03.R):
i <- 1:5
x <- outer(i, i, "*")
sink("/home/tja/Documents/wip/epir/r/chap03.log")
# Here are the results of the outer function
x

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

142 3 Managing epidemiologic data in R

sink()
Here we run source from the R command prompt:
> source("/home/tja/Documents/wip/epir/r/chap03.R", echo=T)
> i <- 1:5
> x <- outer(i, i, "*")
> sink("/home/tja/Documents/wip/epir/r/chap03.log")
>
Nothing was printed to the console after the first sink because it was printed
to output file (chap03.Rout). Here are the contents of chap03.Rout:
> # Here are the results of the outer function
> x
[,1] [,2] [,3] [,4] [,5]
[1,] 1 2 3 4 5
[2,] 2 4 6 8 10
[3,] 3 6 9 12 15
[4,] 4 8 12 16 20
[5,] 5 10 15 20 25
> sink()
The sink and capture.output functions accomplish the same task: send-
ing results to an output file. The sink function works in pairs: open and
closing the connection to the output file. In contrast, capture.output func-
tion appears once and only prints the last object to the output file. Here is
the edited source file (chap03.R) using capture.output instead of sink:
i <- 1:5
x <- outer(i, i, "*")
capture.output(
{
# Here are the results of the outer function
x
}, file = "/home/tja/Documents/wip/epir/r/chap03.log")
Here we run source from the R command prompt:
> source("/home/tja/Documents/wip/epir/r/chap03.R", echo=TRUE)
> i <- 1:5
> x <- outer(i, i, "*")
> capture.output(
+ {
+ # Here are the results of the outer function
+ x
+ }, file = "/home/tja/Documents/wip/epir/r/chap03.Rout")
And, Here are the contents of chap03.log:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.8 Working with missing and “not available” values 143

[,1] [,2] [,3] [,4] [,5]

[1,] 1 2 3 4 5
[2,] 2 4 6 8 10
[3,] 3 6 9 12 15
[4,] 4 8 12 16 20
[5,] 5 10 15 20 25
Even though the capture.output function can run several R commands
(between curly brackets), only the final object (x) was printed to the output
file. This would be true even if the echo option was not set to TRUE in the
source function.
In summary, the source function runs R commands contained in an ex-
ternal source file. Setting the echo option to TRUE prints commands and
results to the console as if the commands were directly entered at the com-
mand prompt. The sink and capture.output functions print results to an
output file.

3.8 Working with missing and “not available” values

In R, missing values are represented by NA, but not all NAs represent missing
values — some are just “not available.” NAs can appear in any data object.
The NA can represent a true missing value, or it can result from an operation
to which a value is “not available.” Here are three vectors that contain true
missing values.
x <- c(2, 4, NA, 5); x
y <- c("M", NA, "M", "F"); y
y <- c(F, NA, F, T); z
However, elementary numerical operations on objects that contain NA return
a single NA (“not available”). In this instance, R is saying “An answer is ‘not
available’ until you tell R what to do with the NAs in the data object.” To
remove NAs for a calculation specify the na.rm (“NA remove”) option.
> sum(x) # answer not available
[1] NA
> mean(x) # answer not available
[1] NA
> sum(x, na.rm = TRUE) # better
[1] 11
> mean(x, na.rm = TRUE) # better
[1] 3.666667
Here are more examples where NA means an answer is not available:
> # Inappropriate coercion

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

144 3 Managing epidemiologic data in R

> as.numeric(c("4", "six"))

[1] 4 NA
Warning message:
NAs introduced by coercion
> # Indexing out of range
> c(1:5)[7]
[1] NA
> df <- data.frame(v1 = 1:3, v2 = 4:6)
> df[4, ] # There is no 4th row
v1 v2
NA NA NA
> # Indexing with non-existing name
> df["4th",]
v1 v2
NA NA NA
> # Operations with NAs
> NA + 8
[1] NA
> # Variance of a single number
> var(55)
[1] NA
In general, these NAs indicate a problem that we must or should be addressed.

3.8.1 Testing, indexing, replacing, and recoding

Regardless of the source of the NAs — missing or not available values —

using the is.na function, we can generate a logical vector to identify which
positions contain or do not contain NAs. This logical vector can be used index
the original or another vector. Values that can be indexed can be replaced
> x <- c(10, NA, 33, NA, 57)
> y <- c(NA, 24, NA, 47, NA)
> is.na(x) #generate logical vector
[1] FALSE TRUE FALSE TRUE FALSE
> which(is.na(x)) #which positions are NA
[1] 2 4
> x[!is.na(x)] #index original vector
[1] 10 33 57
> y[is.na(x)] #index other vector
[1] 24 47
> x[is.na(x)] <- 999 #replacement
> x
[1] 10 999 33 999 57

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.8 Working with missing and “not available” values 145

For a vector, recoding missing values to NA is accomplished using replace-

ment.
> x <- c(1, -99, 3, -88, 5)
> x[x==-99 | x==-88] <- NA
> x
[1] 1 NA 3 NA 5
For a matrix, we can recode missing values to NA by using replacement one
column at a time, or globablly like a vector.
> m <- m2 <- matrix (c(1, -99, 3, 4, -88, 5), 2, 3)
> m
[,1] [,2] [,3]
[1,] 1 3 -88
[2,] -99 4 5
> # Replacement one column at a time
> m[m[,1]==-99, 1] <- NA
> m
[,1] [,2] [,3]
[1,] 1 3 -88
[2,] NA 4 5
> m[m[,3]==-88, 3] <- NA
> m
[,1] [,2] [,3]
[1,] 1 3 NA
[2,] NA 4 5
> # Global replacement
> m2[m2==-99 | m2==-88] <- NA
> m2
[,1] [,2] [,3]
[1,] 1 3 NA
[2,] NA 4 5
Likewise, for a data frame, we can recode missing values to NA by using
replacement one field at a time, or globablly like a vector.
> fname <- c("Tom", "Unknown", "Jerry")
> age <- c(56, 34, -999)
> z1 <- z2 <- data.frame(fname, age)
> z1
fname age
1 Tom 56
2 Unknown 34
3 Jerry -999
> # Replacement one column at a time
> z1$fname[z1$fname=="Unknown"] <- NA
> z1$age[z1$age==-999] <- NA

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

146 3 Managing epidemiologic data in R

> z1
fname age
1 Tom 56
2 <NA> 34
3 Jerry NA
> # Global replacement
> z2[z2=="Unknown" | z2==-999] <- NA
> z2
fname age
1 Tom 56
2 <NA> 34
3 Jerry NA

3.8.2 Importing missing values with the read.table

function

When importing ASCII data files using the read.table function, use the
na.strings option to specify what characters are to be converted to NA.
The default setting is na.strings="NA". Blank fields are also considered to
be missing values in logical, integer, numeric, and complex fields. For example,
suppose the data set contains 999, 888, and . to represent missing values,
then import the data like this:
mydat <- read.table("dataset.txt", na.strings = c(999, 888, "."))
If a number, say 999, represents a missing value in one field but a valid value
in another field, then import the data using the as.is=TRUE option. Then
replace the missing values in the data frame one field at a time, and convert
categorical field to factors.

3.8.3 Working with NA values in data frames and

factors

There are several function for working with NA values in data frames. First,
the na.fail function tests whether a data frame contains any NA values,
returning an error message if it contains NAs.
> name <- c("Jose", "Ana", "Roberto", "Isabel", "Jen")
> gender <- c("M", "F", "M", NA, "F")
> age <- c(34, NA, 22, 18, 34)
> df <- data.frame(name, gender, age)
> df

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.8 Working with missing and “not available” values 147

name gender age

1 Jose M 34
2 Ana F NA
3 Roberto M 22
4 Isabel <NA> 18
5 Jen F 34
> na.fail(df) # NAs in data frame
Error in na.fail.default(df) : missing values in object
> na.fail(df[c(1, 3, 5),]) # no NAs in data frame
name gender age
1 Jose M 34
3 Roberto M 22
5 Jen F 34
Both na.omit and na.exclude remove observations for any field that contain
NAs. na.exclude differs from na.omit only in the class of the “na.action”
attribute of the result, which is “exclude” (see help for details).
> na.omit(df)
name gender age
1 Jose M 34
3 Roberto M 22
5 Jen F 34
> na.exclude(df)
name gender age
1 Jose M 34
3 Roberto M 22
5 Jen F 34
The complete.cases function returns a logical vector for observations that
are “complete” (i.e., do not contain NAs).
> complete.cases(df)
[1] TRUE FALSE TRUE FALSE TRUE
> df[complete.cases(df),] # equivalent to na.omit
name gender age
1 Jose M 34
3 Roberto M 22
5 Jen F 34

3.8.3.1 NA values in factors

By default, factor levels do not include NA. To include NA as a factor level,

use the factor function, setting the exclude option to NULL. Including NA
as a factor level enables counting and displaying the number of NAs in tables,
and analyzing NA values in statistical models.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

148 3 Managing epidemiologic data in R

> df$gender
[1] M F M <NA> F
Levels: F M
> xtabs(~gender, data = df)
gender
F M
2 2
> df$gender.na <- factor(df$gender, exclude = NULL)
> xtabs(~gender.na, data = df)
gender.na
F M <NA>
2 2 1

3.8.3.2 Indexing data frames that contain NAs

Using the original data frame (that can contain NAs), we can index sujects
with ages less than 25.
> df$age # age field
[1] 34 NA 22 18 34
> df[df$age<25, ] # index ages < 25
name gender age
NA <NA> <NA> NA
3 Roberto M 22
4 Isabel <NA> 18
The row that corresponds to the age that is missing (NA) has been converted
to NAs (“not available”) by R. To remove this uninformative row we use the
is.na function.
> df[df$age<25 & !is.na(df$age), ]
name gender age
3 Roberto M 22
4 Isabel <NA> 18
This differs from the na.omit, na.exclude, and complete.cases functions
that remove all missing values from the data frame first.

3.8.4 Viewing number of missing values in tables

By default, NAs are not tabulated in tables produced by the table and xtabs
functions. The table function can tabulate character vectors and factors.
The xtabs function only works with fields in a data frame. To tabulate NAs
in character vectors using the table function, set the exclude function to
NULL in the table function.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.8 Working with missing and “not available” values 149

> df$gender.chr <- as.character(df$gender)

> df$gender.chr
[1] "M" "F" "M" NA "F"
> table(df$gender.chr)

F M
2 2
> table(df$gender.chr, exclude = NULL)

F M <NA>
2 2 1
However, this will not work with factors: we must change the factor levels
first.
> table(df$gender) #does not tabulate NAs

F M
2 2
> table(df$gender, exclude = NULL) #does not work

F M
2 2
> df$gender.na <- factor(df$gender, exclude = NULL) #works
> table(df$gender.na)

F M <NA>
2 2 1
Finally, whereas the exclude option works on character vectors tabulated
with table function, it does not work on character vectors or factors tabu-
lated with the xtabs function. In a data frame, we must convert the charac-
ter vector to a factor (setting the exclude option to NULL), then the xtabs
functions tabulates the NA values.
> xtabs(~gender, data=df, exclude=NULL) # does not work
gender
F M
2 2
> xtabs(~gender.chr, data=df, exclude=NULL) # still does not work
gender.chr
F M
2 2
> df$gender.na <- factor(df$gender, exclude = NULL) #works
> xtabs(~gender.na, data = df)
gender.na
F M <NA>
2 2 1

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

150 3 Managing epidemiologic data in R

3.8.5 Setting default NA behaviors in statistical models

Statistical models, for example the glm function for generalized linear models,
have default NA behaviors that can be reset locally using the na.action
option in the glm function, or reset globally using the na.action option
setting in the options function.
> options("na.action") # display global setting
$na.action
[1] "na.omit"

> options(na.action="na.fail") # reset global setting

> options("na.action")
$na.action
[1] "na.fail"
By default, na.action is set to “na.omit” in the options function. Glob-
ally (inside the options function) or locally (inside a statistical function),
na.action can be set to the following:
• “na.fail”
• “na.omit”
• “na.exclude”
• “na.pass”
With “na.fail”, a function that calls na.action will return an error if the
data object contains NAs. Both “na.omit” and “na.exclude” will remove row
observations from a data frame that contain NAs.12 With na.pass, the data
object is returned unchanged.

3.8.6 Working with finite, infinite, and NaN numbers

In R, some numerical operations result in negative infinity, positive infinity,

or an indeterminate value (NAN for “not a number”). To assess whether
values are finite or infinite, use the is.finite or is.infinite functions,
respectively. To assess where a value is NAN, use the is.nan function. While
is.na can identify NANs, is.nan cannot identify NAs.
> x <- c(-2:2)/c(2, 0, 0, 0, 2)
> x
[1] -1 -Inf NaN Inf 1
> is.infinite(x)

12 If na.omit removes cases, the row numbers of the cases form the “na.action” attribute

of the result, of class “omit”. na.exclude differs from na.omit only in the class of the
“na.action” attribute of the result, which is “exclude”. See help for more details.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.9 Working with dates and times 151

[1] FALSE TRUE FALSE TRUE FALSE

> x[is.infinite(x)]
[1] -Inf Inf
> is.finite(x)
[1] TRUE FALSE FALSE FALSE TRUE
> x[is.finite(x)]
[1] -1 1
> is.nan(x)
[1] FALSE FALSE TRUE FALSE FALSE
> x[is.nan(x)]
[1] NaN
> is.na(x) # does index NAN
[1] FALSE FALSE TRUE FALSE FALSE
> x[is.na(x)]
[1] NaN
> x[is.nan(x)] <- NA
> x
[1] -1 -Inf NA Inf 1
> is.nan(x) # does not index NA
[1] FALSE FALSE FALSE FALSE FALSE

3.9 Working with dates and times

There are 60 seconds in 1 minute, 60 minutes in 1 hour, 24 hours in 1 day, 7

days in 1 week, and 365 days in 1 year (except every 4th year we have a leap
year with 366 days). Although this seems straightforward, doing numerical
calculations with these time measures is not. Fortunately, computers make
this much easier. Functions to deal with dates are available in the base,
chron, and survival packages.
Summarized in Figure 3.4 on the next page is the relationship between
recorded data (calendar dates and times) and their representation in R as
date-time class objects (Date, POSIXlt, POSIXct). The as.Date function
converts a calendar date into a Date class object. The strptime function
converts a calendar date and time into a date-time class object (POSIXlt,
POSIXct). The as.POSIXlt and as.POSIXct functions convert date-time
class objects into POSIXlt and POSIXct, respectively.
The format function converts date-time objects into human legible char-
acter data such as dates, days, weeks, months, times, etc. These functions
are discussed in more detail in the paragraphs that follow.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

152 3 Managing epidemiologic data in R

Fig. 3.4 Displayed are functions to convert calendar date and time data into R date-time
classes (as.Date, strptime, as.POSIXlt, as.POSIXct), and the format function converts
date-time objects into character dates, days, weeks, months, times, etc.

3.9.1 Date functions in the base package

3.9.1.1 The as.Date function

Let’s start with simple date calculations. The as.Date function in R converts
calendar dates (e.g., 11/2/1949) into a Date objects—a numeric vector of
class Date. The numeric information is the number of days since January
1, 1970—also called Julian dates. However, because calendar date data can
come in a variety of formats, we need to specify the format so that as.Date
does the correct conversion. Study the following analysis carefully.
> bdays <- c("11/2/1959", "1/1/1970")
> bdays
[1] "11/2/1959" "1/1/1970"
> #convert to Julian dates
> bdays.julian <- as.Date(bdays, format = "%m/%d/%Y")
> bdays.julian
[1] "1959-11-02" "1970-01-01"
Although this looks like a character vectors, it is not: it is class “Date” and
mode “numeric”.
> #display Julian dates

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.9 Working with dates and times 153

> as.numeric(bdays.julian)
[1] -3713 0
> #calculate age as of today’s date
> date.today <- Sys.Date()
> date.today
[1] "2005-09-25"
> age <- (date.today - bdays.julian)/365.25
> age
Time differences of 45.89733, 35.73169 days
> #the display of ’days’ is not correct
> #truncate number to get "age"
> age2 <- trunc(as.numeric(age))
> age2
[1] 45 35
> #create date frame
> bd <- data.frame(Birthday = bdays, Standard = bdays.julian,
+ Julian = as.numeric(bdays.julian), Age = age2)
> bd
Birthday Standard Julian Age
1 11/2/1959 1959-11-02 -3713 45
2 1/1/1970 1970-01-01 0 35
To summarize, as.Date converted the character vector of calendar dates
into Julian dates (days since 1970-01-01) are displayed in a standard format
(yyyy-mm-dd). The Julian dates can be used in numerical calculations. To
see the Julian dates use as.numeric or julian function. Because the calen-
dar dates to be converted can come in a diversity of formats (e.g., November
2, 1959; 11-02-59; 11-02-1959; 02Nov59), one must specify the format op-
tion in as.Date. Below are selected format options; for a complete list see
help(strptime).
"%a" Abbreviated weekday name.
"%A" Full weekday name.
"%b" Abbreviated month name.
"%B" Full month name.
"%d" Day of the month as decimal number (01-31)
"%j" Day of year as decimal number (001-366).
"%m" Month as decimal number (01-12).
"%U" Week of the year as decimal number (00-53) using the
first Sunday as day 1 of week 1.
"%w" Weekday as decimal number (0-6, Sunday is 0).
"%W" Week of the year as decimal number (00-53) using the
first Monday as day 1 of week 1.
"%y" Year without century (00-99). If you use this on input,
which century you get is system-specific. So don’t!
Often values up to 69 (or 68) are prefixed by 20 and

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

154 3 Managing epidemiologic data in R

70-99 by 19.
"%Y" Year with century.
Here are some examples of converting dates with different formats:
> as.Date("November 2, 1959", format = "%B %d, %Y")
[1] "1959-11-02"
> as.Date("11/2/1959", format = "%m/%d/%Y")
[1] "1959-11-02"
> #caution using 2-digit year
> as.Date("11/2/59", format = "%m/%d/%y")
[1] "2059-11-02"
> as.Date("02Nov1959", format = "%d%b%Y")
[1] "1959-11-02"
> #caution using 2-digit year
> as.Date("02Nov59", format = "%d%b%y")
[1] "2059-11-02"
> #standard format does not require format option
> as.Date("1959-11-02")
[1] "1959-11-02"
Notice how Julian dates can be used like any integer:
> as.Date("2004-01-15"):as.Date("2004-01-23")
[1] 12432 12433 12434 12435 12436 12437 12438 12439 12440
> seq(as.Date("2004-01-15"), as.Date("2004-01-18"), by = 1)
[1] "2004-01-15" "2004-01-16" "2004-01-17" "2004-01-18"

3.9.1.2 The weekdays, months, quarters, julian functions

Use the weekdays, months, quarters, or julian functions to extract infor-

mation from Date and other date-time objects in R.
> mydates <- c("2004-01-15","2004-04-15","2004-10-15")
> mydates <- as.Date(mydates)
> weekdays(mydates)
[1] "Thursday" "Thursday" "Friday"
> months(mydates)
[1] "January" "April" "October"
> quarters(mydates)
[1] "Q1" "Q2" "Q4"
> julian(mydates)
[1] 12432 12523 12706
attr(,"origin")
[1] "1970-01-01"

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.9 Working with dates and times 155

3.9.1.3 The strptime function

So far we have worked with calendar dates; however, we also need to be able to
work with times of the day. Whereas as.Date only works with calendar dates,
the strptime function will accept data in the form of calendar dates and
times of the day (HH:MM:SS, where H = hour, M = minutes, S = seconds).
For example, let’s look at the Oswego foodborne ill outbreak that occurred
in 1940. The source of the outbreak was attributed to the church supper that
was served on April 18, 1940. The food was available for consumption from 6
pm to 11 pm. The onset of symptoms occurred on April 18th and 19th. The
meal consumption times and the illness onset times were recorded.
> odat <- read.table("http://www.medepi.net/data/oswego.txt",
+ sep = "", header = TRUE, as.is = TRUE)
> str(odat)
‘data.frame’: 75 obs. of 21 variables:
$ id : int 2 3 4 6 7 8 9 10 14 16 ...
$ age : int 52 65 59 63 70 40 15 33 10 32 ...
$ sex : chr "F" "M" "F" "F" ...
$ meal.time : chr "8:00 PM" "6:30 PM" "7:30 PM" ...
$ ill : chr "Y" "Y" "Y" "Y" ...
$ onset.date : chr "4/19" "4/19" "4/19" "4/18" ...
$ onset.time : chr "12:30 AM" "10:30 PM" ...
...
$ vanilla.ice.cream : chr "Y" "Y" "Y" "Y" ...
$ chocolate.ice.cream: chr "N" "Y" "Y" "N" ...
$ fruit.salad : chr "N" "N" "N" "N" ...
To calculate the incubation period, for ill individuals, we need to subtract
the meal consumption times (occurring on 4/18) from the illness onset times
(occurring on 4/18 and 4/19). Therefore, we need two date-time objects to
do this arithmetic. First, let’s create a date-time object for the meal times:
> # look at existing data for meals
> odat$meal.time[1:5]
[1] "8:00 PM" "6:30 PM" "6:30 PM" "7:30 PM" "7:30 PM"
> # create character vector with meal date and time
> mdt <- paste("4/18/1940", odat$meal.time)
> mdt[1:4]
[1] "4/18/1940 8:00 PM" "4/18/1940 6:30 PM"
[3] "4/18/1940 6:30 PM" "4/18/1940 7:30 PM"
> # convert into standard date and time
> meal.dt <- strptime(mdt, format = "%m/%d/%Y %I:%M %p")
> meal.dt[1:4]
[1] "1940-04-18 20:00:00" "1940-04-18 18:30:00"
[3] "1940-04-18 18:30:00" "1940-04-18 19:30:00"
> # look at existing data for illness onset

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

156 3 Managing epidemiologic data in R

> odat$onset.date[1:4]
[1] "4/19" "4/19" "4/19" "4/18"
> odat$onset.time[1:4]
[1] "12:30 AM" "12:30 AM" "12:30 AM" "10:30 PM"
> # create vector with onset date and time
> odt <- paste(paste(odat$onset.date, "/1940", sep=""),
+ odat$onset.time)
> odt[1:4]
[1] "4/19/1940 12:30 AM" "4/19/1940 12:30 AM"
[3] "4/19/1940 12:30 AM" "4/18/1940 10:30 PM"
> # convert into standard date and time
> onset.dt <- strptime(odt, "%m/%d/%Y %I:%M %p")
> onset.dt[1:4]
[1] "1940-04-19 00:30:00" "1940-04-19 00:30:00"
[3] "1940-04-19 00:30:00" "1940-04-18 22:30:00"
> # calculate incubation period
> incub.period <- onset.dt - meal.dt
> incub.period
Time differences of 4.5, 6.0, 6.0, 3.0, 3.0, 6.5, 3.0, 4.0,
6.5, NA, NA, NA, NA, 3.0, NA, NA, NA, 3.0, NA, NA,
...
NA, NA, NA, NA, NA, NA, NA hours
> mean(incub.period, na.rm = T)
Time difference of 4.295455 hours
> median(incub.period, na.rm = T)
Error in Summary.difftime(..., na.rm = na.rm) :
sum not defined for "difftime" objects
> # try ’as.numeric’ on ’incub.period’
> median(as.numeric(incub.period), na.rm = T)
[1] 4
To summarize, we used strptime to convert the meal consumption date
and times and illness onset dates and times into date-time objects (meal.dt
and onset.dt) that can be used to calculate the incubation periods by simple
subtraction (and assigned name incub.period).
Notice that incub.period is an atomic object of class difftime:
> str(incub.period)
Class ’difftime’ atomic [1:75] 4.5 6 6 3 3 6.5 3 4 NA ...
..- attr(*, "tzone")= chr ""
..- attr(*, "units")= chr "hours"
This is why we had trouble calculating the median (which should not be the
case). We got around this problem by coercion using as.numeric:
> as.numeric(incub.period)
[1] 4.5 6.0 6.0 3.0 3.0 6.5 3.0 4.0 6.5 NA NA NA NA 3.0
...

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.9 Working with dates and times 157

Now, what kind of objects were created by the strptime function?

> str(meal.dt)
’POSIXlt’, format: chr [1:75] "1940-04-18 18:30:00" ...
> str(onset.dt)
’POSIXlt’, format: chr [1:75] "1940-04-19 00:30:00" ...
The strptime function produces a named list of class POSIXlt. POSIX
stands for “Portable Operating System Interface,” and “lt” stands for “legible
time”.13

3.9.1.4 The POSIXlt and POSIXct functions

The POSIXlt list contains the date-time data in human readable forms. The
named list contains the following vectors:
’sec’ 0-61: seconds
’min’ 0-59: minutes
’hour’ 0-23: hours
’mday’ 1-31: day of the month
’mon’ 0-11: months after the first of the year.
’year’ Years since 1900.
’wday’ 0-6 day of the week, starting on Sunday.
’yday’ 0-365: day of the year.
’isdst’ Daylight savings time flag. Positive if in force,
zero if not, negative if unknown.
Let’s examine the onset.dt object we created from the Oswego data.
> is.list(onset.dt)
[1] TRUE
> names(onset.dt)
[1] "sec" "min" "hour" "mday" "mon" "year" "wday"
[8] "yday" "isdst"
> onset.dt$min
[1] 30 30 30 30 30 0 0 0 0 30 30 15 0 0 0 45 45 0
...
> onset.dt$hour
[1] 0 0 0 22 22 2 1 23 2 10 0 22 22 1 23 21 21 1
...
> onset.dt$mday
[1] 19 19 19 18 18 19 19 18 19 19 19 18 18 19 18 18 18 19
...
> onset.dt$mon

13 For more information visit the Portable Application Standards Committee site at
http://www.pasc.org/

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

158 3 Managing epidemiologic data in R

[1] 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
...
> onset.dt$year
[1] 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40
...
> onset.dt$wday
[1] 5 5 5 4 4 5 5 4 5 5 5 4 4 5 4 4 4 5
...
> onset.dt$yday
[1] 109 109 109 108 108 109 109 108 109 109 109 108 108 109
...
The POSIXlt list contains useful date-time information; however, it is not
in a convenient form for storing in a data frame. Using as.POSIXct we can
convert it to a “continuous time” object that contains the number of seconds
since 1970-01-01 00:00:00. as.POSIXlt coerces a date-time object to POSIXlt.
> onset.dt.ct <- as.POSIXct(onset.dt)
> onset.dt.ct[1:5]
[1] "1940-04-19 00:30:00 Pacific Daylight Time"
[2] "1940-04-19 00:30:00 Pacific Daylight Time"
[3] "1940-04-19 00:30:00 Pacific Daylight Time"
[4] "1940-04-18 22:30:00 Pacific Daylight Time"
[5] "1940-04-18 22:30:00 Pacific Daylight Time"
> as.numeric(onset.dt.ct[1:5])
[1] -937326600 -937326600 -937326600 -937333800 -937333800

3.9.1.5 The format function

Whereas the strptime function converts a character vector of date-time in-

formation into a date-time object, the format function converts a date-time
object into a character vector. The format function gives us great flexibility
in converting date-time objects into numerous outputs (e.g., day of the week,
week of the year, day of the year, month of the year, year). Selected date-time
format options are listed on page 153, for a complete list see help(strptime).
For example, in public health, reportable communicable diseases are often
reported by “disease week” (this could be week of reporting or week of symp-
tom onset). This information is easily extracted from R date-time objects.
For weeks starting on Sunday use the “%U” option in the format function,
and for weeks starting on Monday use the “%W” option.
> decjan <- seq(as.Date("2003-12-15"), as.Date("2004-01-15"),
+ by =1)
> decjan
[1] "2003-12-15" "2003-12-16" "2003-12-17" "2003-12-18"
...

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.10 Exporting data objects 159

[29] "2004-01-12" "2004-01-13" "2004-01-14" "2004-01-15"

> disease.week <- format(decjan, "%U")
> disease.week
[1] "50" "50" "50" "50" "50" "50" "51" "51" "51" "51" "51"
[12] "51" "51" "52" "52" "52" "52" "00" "00" "00" "01" "01"
[23] "01" "01" "01" "01" "01" "02" "02" "02" "02" "02"

3.9.2 Date functions in the chron and survival

packages

The chron and survival packages have customized functions for dealing
with dates. Both packages come with the default R installation. To learn
more about date and time classes read R News, Volume 4/1, June 2004.14

3.10 Exporting data objects

On occassion, we need to export R data objects. This can be done in several

ways, depending on our needs:
• Generic ASCII text file
• R ASCII text file
• R binary file
• Non-R ASCII text files
• Non-R binary file

3.10.1 Exporting to a generic ASCII text file

3.10.1.1 The write.table function

We use the write.table function to exports a data frame as a tabular ASCII

text file which can be read by most statistical packages. If the object is not
a data frame, it converts to a data frame before exporting it. Therefore, this
function only make sense with tabular data objects. Here are the default
arguments:
> args(write.table)
function (x, file = "", append = FALSE, quote = TRUE,
sep = " ", eol = "\n", na = "NA", dec = ".",

14 http://cran.r-project.org/doc/Rnews

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

160 3 Managing epidemiologic data in R

Table 3.5 R functions for exporting data objects

Function Description Try these examples
Export to Generic ASCII text file
write.table Write tabular data as a data write.table(infert,"infert.dat")
write.csv frame to an ASCII text file; write.csv(infert,"infert.csv")
read file back in using
read.table function
write Write matrix elements to an x <- matrix(1:4, 2, 2)
ASCII text file write(t(x), "x.txt")
Export to R ASCII text file
dump “Dumps” list of R objects dump("Titanic", "titanic.R")
as R code to an ASCII text
file; read back in using
source function
dput Writes an R object as R dput(Titanic, "titanic.R")
code (but without the object
name) to the console, or an
ASCII text file; read file
back in using dget function
Export to R binary file
save “Saves” list of R objects as save(Titanic, "titanic.Rdata")
binary filename.Rdata file;
read back in using load
function
Export to non-R ASCII text file
write.foreign From foreign package: write.foreign(infert,
writes text files (SPSS, datafile="infert.dat",
Stata, SAS) and code to codefile="infert.txt", package =
read them "SPSS")
Export to non-R binary file
write.dbf From foreign package: write.dbf(infert, "infert.dbf")
writes DBF files
write.dta From foreign package: write.dta(infert, "infert.dta")
writes files in Stata binary
format

row.names = TRUE, col.names = TRUE,

qmethod = c("escape", "double"))
The 1st argument will be the data frame name (e.g., infert), the 2nd will be
the name for the output file (e.g., infert.dat), the sep argument is set to be
space-delimited, and the row.names argument is set to TRUE.
The following code:
write.table(infert,"infert.dat")
produces this ASCII text file:
"education" "age" "parity" "induced" "case" ...

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.10 Exporting data objects 161

"1" "0-5yrs" 26 6 1 1 2 1 3
"2" "0-5yrs" 42 1 1 1 0 2 1
"3" "0-5yrs" 39 6 2 1 0 3 4
"4" "0-5yrs" 34 4 2 1 0 4 2
"5" "6-11yrs" 35 3 1 1 1 5 32
...
Because row.names=TRUE, the number field names in the header (row 1) will
one less that the number of columns (starting with row 2). The default row
names is a character vector of integers. The following code:
write.table(infert,"infert.dat", sep=",", row.names=FALSE)
produces a commna-delimited ASCII text file without row names:
"education","age","parity","induced","case", ...
"0-5yrs",26,6,1,1,2,1,3
"0-5yrs",42,1,1,1,0,2,1
"0-5yrs",39,6,2,1,0,3,4
"0-5yrs",34,4,2,1,0,4,2
"6-11yrs",35,3,1,1,1,5,32
...
Note that the write.csv function produces a comma-delimited data file by
default.

3.10.1.2 The write function

The write function writes the contents of a matrix in a columnwise order

to an ASCII text file. To get the same appearance as the matrix, we must
transpose the matrix and specify the number of columns (the default is 5).
If we set file="", then the output is written to the screen:
> infert.tab1 <- xtabs(~case+parity,data=infert)
> infert.tab1
parity
case 1 2 3 4 5 6
0 66 54 24 12 4 5
1 33 27 12 6 2 3
> write(infert.tab1, file="") #not what we want
66 33 54 27 24
12 12 6 4 2
5 3
> #much better
> write(t(infert.tab1), file="", ncol=ncol(infert.tab1))
66 54 24 12 4 5
33 27 12 6 2 3

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

162 3 Managing epidemiologic data in R

To read the raw data back into R, we would use the scan function. For
example, if the data had been written to data.txt, then the following code
reads the data back into R:
> matrix(scan("data.txt"), ncol=6, byrow=TRUE)
Read 12 items
[,1] [,2] [,3] [,4] [,5] [,6]
[1,] 66 54 24 12 4 5
[2,] 33 27 12 6 2 3
Of course, all the labeling was lost.

3.10.2 Exporting to R ASCII text file

Data objects can also be exported as R code in an ASCII text file using the
dump and dput functions. This has advantages for complex R objects (e.g.,
arrays, lists) that do not have simple tabular structures, and the R code
makes the raw data human legible.

3.10.2.1 The dump function

The dump function exports multiple objects as R code as the next example
illustrates:
> infert.tab1 <- xtabs(~case+parity,data=infert)
> infert.tab2 <- xtabs(~education+parity+case,data=infert)
> infert.tab1 #display matrix
parity
case 1 2 3 4 5 6
0 66 54 24 12 4 5
1 33 27 12 6 2 3
> infert.tab2 #display array
, , case = 0

parity
education 1 2 3 4 5 6
0-5yrs 2 0 0 2 0 4
6-11yrs 28 28 14 8 2 0
12+ yrs 36 26 10 2 2 1

, , case = 1

parity
education 1 2 3 4 5 6

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.10 Exporting data objects 163

0-5yrs 1 0 0 1 0 2
6-11yrs 14 14 7 4 1 0
12+ yrs 18 13 5 1 1 1

> dump(c("infert.tab1", "infert.tab2"),"infert_tab.R") #export

The dump function produced the following R code in the infert tab.R text
file.
‘infert.tab1‘ <-
structure(c(66, 33, 54, 27, 24, 12, 12, 6, 4, 2, 5, 3),
.Dim = c(2L, 6L), .Dimnames = structure(list(case = c("0", "1"),
parity = c("1", "2", "3", "4", "5", "6")), .Names = c("case",
"parity")), class = c("xtabs", "table"), call = quote(xtabs(
formula = ~case + parity, data = infert)))
‘infert.tab2‘ <-
structure(c(2, 28, 36, 0, 28, 26, 0, 14, 10, 2, 8, 2, 0, 2,
2, 4, 0, 1, 1, 14, 18, 0, 14, 13, 0, 7, 5, 1, 4, 1, 0, 1, 1,
2, 0, 1), .Dim = c(3L, 6L, 2L), .Dimnames = structure(list(
education = c("0-5yrs", "6-11yrs", "12+ yrs"), parity = c("1",
"2", "3", "4", "5", "6"), case = c("0", "1")), .Names =
c("education", "parity", "case")), class = c("xtabs", "table"),
call = quote(xtabs(formula = ~education + parity + case,
data = infert)))
Notice that the 1st argument was a character vector of object names. The
infert tab.R file can be run in R using the source function to recreate all
the objects in the workspace.

3.10.2.2 The dput function

The dput function is similar to the dump function except that the object name
is not written. By default, the dput function prints to the screen:
> dput(infert.tab1)
structure(c(66, 33, 54, 27, 24, 12, 12, 6, 4, 2, 5, 3),
.Dim = c(2L, 6L), .Dimnames = structure(list(case = c("0", "1"),
parity = c("1", "2", "3", "4", "5", "6")), .Names = c("case",
"parity")), class = c("xtabs", "table"), call = quote(xtabs(
formula = ~case + parity, data = infert)))
To export to an ASCII text file, give a new file name as the second argument,
similar to dump. To get back the R code use the dget function:
> dput(infert.tab1, "infert_tab1.R")
> dget("infert_tab1.R")
parity

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

164 3 Managing epidemiologic data in R

case 1 2 3 4 5 6
0 66 54 24 12 4 5
1 33 27 12 6 2 3

3.10.3 Exporting to R binary file

3.10.3.1 The save function

The save function exports R data objects to binary file (filename.RData)

which is the most effient, compact method to export objects. The first argu-
ment(s) can be the names of the objects to save followed by the output file
name, or list with a character vector of object names followed by the output
file name. Here is an example of the first option:
> x <- 1:5; y <- x^3
> save(x, y, file="xy.RData")
> rm(x, y)
> ls()
character(0)
> load(file="xy.RData")
> ls()
[1] "x" "y"
Notice that we used the load function to load the binary file back into the
workspace.
Now here is an example of the second option using a list:
> x <- 1:5; y <- x^3
> save(list=c("x", "y"), file="xy.RData")
> rm(x, y)
> ls()
character(0)
> load(file="xy.RData")
> ls()
[1] "x" "y"
In fact, the save.image function we use to save the entire workspace is just
the following:
save(list = ls(all=TRUE), file = ".RData")

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.11 Working with regular expressions 165

3.10.4 Exporting to non-R ASCII text and binary files

The foreign package contains functions for exporting R data frames to non-R
ASCII text and binary files. The write.foreign function write two ASCII
text files: the first file is the data file, and the second file is the code file
for reading the data file. The code file contains either SPSS, Stata, or SAS
programming code. The write.dbf function writes a data frame to a binary
DBF file, which can be read back into R using the read.dbf function. Finally,
the write.dta function writes a data frame to a binary Stata file, which can
be read back into R using the read.dta function.

3.11 Working with regular expressions

A regular expression is a special text string for describing a search pattern

which can be used for searching text strings, indexing data objects, and
replacing object elements. For example, we applied Global Burden of Disease
methods to evaluate causes of premature deaths in San Francisco [4]. Using
regular expressions we were able to efficiently code over 14,000 death records,
with over 900 ICD-10 cause of death codes, into 117 mutually exclusive cause
of death categories. Without regular expressions, this local study would have
been prohibitively tedious.
A regular expression is built up from specifying one character at a time.
Using this approach, we cover the following:
• Single character: matching a single character;
• Character class: matching a single character from among a list of charac-
ters;
• Concatenation: combining single characters into a new match pattern;
• Repetition: specifying how many times a single character or match pattern
might be repeated;
• Alternation: a regular expression may be matched from among two or more
regular expressions; and
• Metacharacters: special characters that require special treatment.

3.11.1 Single characters

The search pattern is built up from specifying one character at a time. For
example, the pattern "x" looks for the letter x in a text string. Next, consider
a character vector of text strings. We can use the grep function to search for
a pattern in this data vector.
> vec1 <- c("x", "xa bc", "abc", "ax bc", "ab xc", "ab cx")

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

166 3 Managing epidemiologic data in R

> grep("x", vec1)

[1] 1 2 4 5 6
> vec1[grep("x", vec1)] #index by position
[1] "x" "xa bc" "ax bc" "ab xc" "ab cx"
The grep function returned an integer vector indicating the positions in the
data vector that contain a match. We used this integer vector to index by
position.
The caret ^ matches the empty string at the beginning of a line. Therefore,
to match this pattern at the beginning of a line we add the ^ character to
the regular expression:
> grep("^x", vec1)
[1] 1 2
> vec1[grep("^x", vec1)] #index by position
[1] "x" "xa bc"
The $ character matches the empty string at the end of a line. Therefore,
to match this pattern at the end of a line we add the $ character to the
regular expression:
> vec1[grep("x$", vec1)] #index by position
[1] "x" "ab cx"
The ^ and $ characters are examples of metacharacters (more on these later).
To match this pattern at the beginning of a word, but not the beginning
of a line, we add a space to the regular expression:
> vec1[grep(" x", vec1)] #index by position
[1] "ab xc"
To match this pattern at the end of a word, but not the end of a line, we add
a space to the regular expression:
> vec1[grep("x ", vec1)] #index by position
[1] "ax bc"
The period “.” matches any single character, including a space.
> vec1[grep(".bc", vec1)]
[1] "xa bc" "abc" "ax bc"

3.11.2 Character class

A character class is a list of characters enclosed by square brackets [ and ]

which matches any single character in that list. For example, "[fhr]" will
match the single character f, h, or r. This can be combined with metachar-
acters for more specificity; for example, "^[fhr]" will match the single char-
acter f, h, or r at the beginning of a line.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.11 Working with regular expressions 167

Table 3.6 Predefined character classes for regular expressions

Predefined Description Alternative
[[:lower:]] Lower-case letters in the current [a-z]
locale
[[:upper:]] Upper-case letters in the current [A-Z]
locale
[[:alpha:]] Alphabetic characters [A-Za-z]
[[:digit:]] Digits [0-9]
[[:alnum:]] Alphanumeric characters [A-Za-z0-9]
[[:punct:]] Punctuation characters: ! " # $ "[]...^...-]" (], ^, and - require
% & ’ ( ) * + , - . / : ; < = special placement in character
> ? @ [ \ ] ^ ‘ { | } ~ classes. See p. 172)
[[:space:]] Space characters: tab, newline,
vertical tab, form feed, carriage
return, and space
[[:graph:]] Graphical characters [[:alnum:][:punct:]]
[[:print:]] Printable characters [[:alnum:][:punct:][:space:]]
[[:xdigit:]] Hexadecimal digits: [0-9A-Fa-f]

> vec2 <- c("fat", "bar", "rat", "elf", "mach", "hat")

> grep("^[fhr]", vec2)
[1] 1 3 6
> vec2[grep("^[fhr]", vec2)] #index by position
[1] "fat" "rat" "hat"
As already shown, ^ character matches the empty string at the beginning
of a line. However, when ^ is the first character in a character class list, it
matches any character not in the list. For example, "^[^fhr]" will match
any single character at the beginning of a line except f, h, or r.
> vec2 <- c("fat", "bar", "rat", "elf", "mach", "hat")
> vec2[grep("^[^fhr]", vec2)] #index by position
[1] "bar" "elf" "mach"
Character classes can be specified as a range of possible characters. For
example, [0-9] matches a single digit with possible values from 0 to 9, [A-Z]
matches a single letter with possible values from A to Z, and [a-z] matches
a single letter from a to z. The pattern [0-9A-Za-z] matches any single
alphanumeric character.
For convenience, certain character classes are predefined and their inter-
pretation depend on locale.15 For example, to match a single lower case letter
we place [:lower:] inside square brackets like this: "[[:lower:]]", which
is equivalent to "[a-z]". Table 3.6 lists predefined character classes. This is
15 The locale describes aspects of the internationalization of a program. Initially most
aspects of the locale of R are set to “C” (which is the default for the C language and
reflects North-American usage).

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

168 3 Managing epidemiologic data in R

very convenient for matching punctuation characters and multiple types of

spaces (e.g., tab, newline, carriage return).
In this final example, "^.[^a].” will match any first character, followed
by any character except a, and followed by any character one or more times:
> vec2[grep("^.[^a].+", vec2)] #index by position
[1] "elf"
Combining single character matches is call concatenation.

3.11.3 Concatenation

Single characters (including character classes) can be concatenated ; for ex-

ample, the pattern "^[fhr]at$" will match the single, isolated words fat,
hat, or rat.
> vec3 <- c("fat", "bar", "rat", "fat boy", "elf", "mach",
+ "hat")
> vec3[grep("^[fhr]at$", vec3)] #index by position
[1] "fat" "rat" "hat"
The concatenation "[ct]a[br]" will match the pattern that starts with
c or t, followed by a, and followed by b or r.
> vec4 <- c("cab", "carat", "tar", "bar", "tab", "batboy",
+ "care")
> vec4[grep("[ct]a[br]", vec4)] #index by position
[1] "cab" "carat" "tar" "tab" "care"
To match single, 3-letter words use "^[ct]a[br]$".
> vec4[grep("^[ct]a[br]$", vec4)] #index by position
[1] "cab" "tar" "tab"
The period (.) is another metacharacter: it matches any single character.
For example, "f.t" matches the pattern f + any character + t.
> vec5 <- c("fate", "rat", "fit", "bat", "futbol")
> vec5[grep("f.t", vec5)] #index by position
[1] "fate" "fit" "futbol"

3.11.4 Repetition

Regular expressions (so far: single characters, character classes, and concate-
nations) can be qualified by whether a pattern can repeat (Table 3.7 on the
next page). For example, the pattern "^f.+t$" matches single, isolated words

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.11 Working with regular expressions 169

Table 3.7 Regular expressions may be followed by a repetition quantifier

Repetition quantifier Description
? Preceding pattern is optional and will be matched at most once
* Preceding pattern will be matched zero or more times
+ Preceding pattern will be matched one or more times
{n} Preceding pattern is matched exactly n times
{n,} Preceding pattern is matched n or more times
{n, m} Preceding pattern is matched at least n times, but not more than
m times

that start with f or F, followed by 1 or more of any character, and ending

with t.
> vec6 <- c("fat", "fate", "feat", "bat", "Fahrenheit", "bat",
+ "foot")
> vec6[grep("^[fF].+t$", vec6)] #index by position
[1] "fat" "feat" "Fahrenheit" "foot"
Repetition quantifiers gives us great flexibility to specify how often preceding
patterns can repeat.

3.11.5 Alternation

Two or more regular expressions (so far: single characters, character classes,
concatenations, and repetitions) may be joined by the infix operator |. The
resulting regular expression can match the pattern of any subexpression. For
example, the World Health Organization (WHO) Global Burden of Disease
(GBD) Study used International Classification of Diseases, 10th Revision
(ICD-10) codes (ref). The GBD Study ICD-10 codes for hepatitis B are the
following:
B16, B16.0, B16.1, B16.2, B16.3, B16.4, B16.5, B16.7, B16.8, B16.9, B17, B17.0,
B17.2, B17.8, B18, B18.0, B18.1, B18.8, B18.9

Notice that B16 and B16.0 are not the same ICD-10 code! The GBD Study
methods were used to study causes of death in San Francisco, California
(ref). Underlying causes of death were obtained from the State of California,
Center for Health Statistics. The ICD-10 code field did not have periods so
that the hepatitis B codes were the following.
B16, B160, B161, B162, B163, B164, B165, B167, B168, B169, B17, B170, B172,
B178, B18, B180, B181, B188, B189

To match the pattern of ICD-10 codes representing hepatitis B, the fol-

lowing regular expression was used (without spaces):

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

170 3 Managing epidemiologic data in R

"^B16[0-9]?$|^B17[0,2,8]?$|^B18[0,1,8,9]?$"
This regular expression matches ^B16[0-9]?$ or ^B17[0,2,8]?$ or ^B18[0,1,8,9]?$.
Similar to the first and third pattern, the second regular expression, ^B17[0,2,8]?$,
matches B17, B170, B172, or B178 as isolated text strings.
To see how this works, we can match each subexpression individually and
then as an alternation:
> hepb <- c("B16", "B160", "B161", "B162", "B163", "B164",
+ "B165", "B167", "B168", "B169", "B17", "B170",
+ "B172", "B178", "B18", "B180", "B181", "B188",
+ "B189")
> grep("^B16[0-9]?$", hepb) #match 1st subexpression
[1] 1 2 3 4 5 6 7 8 9 10
> grep("^B17[0,2,8]?$", hepb) #match 2nd subexpression
[1] 11 12 13 14
> grep("^B18[0,1,8,9]?$", hepb) #match 3rd subexpression
[1] 15 16 17 18 19
> #match any subexpression
> grep("^B16[0-9]?$|^B17[0,2,8]?$|^B18[0,1,8,9]?$", hepb)
[1] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
A natural use for these pattern matches is for indexing and replacement.
We illustrate this using the 2nd subexpression.
> #indexing
> hepb[grep("^B17[0,2,8]?$", hepb)]
[1] "B17" "B170" "B172" "B178"
> #replacement
> hepb[grep("^B17[0,2,8]?$", hepb)] <- "HBV"
> hepb
[1] "B16" "B160" "B161" "B162" "B163" "B164" "B165" "B167"
[9] "B168" "B169" "HBV" "HBV" "HBV" "HBV" "B18" "B180"
[17] "B181" "B188" "B189"
Using regular expression alternations allowed us to efficiently code over
14,000 death records, with over 900 ICD-10 cause of death codes, into 117
mutually exclusive cause of death categories for our San Francisco study.
Suppose sfdat was the data frame with San Francisco deaths for 2003–2004.
Then the following code would tabulate the deaths caused by hepatatis B:
> sfdat$hepb <- rep("No", nrow(sfdat)) #new field
> get.hepb <- grep("^B16[0-9]?$|^B17[0,2,8]?$|^B18[0,1,8,9]?$",
+ sfdat$icd10)
> sfdat$hepb[get.hepb] <- "Yes"
> table(sfdat$hepb)

No Yes
14125 23

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.11 Working with regular expressions 171

Therefore, in San Francisco, during the period 2003-2004, there were 23

deaths caused by hepatitis B. Without regular expressions, this mortality
analysis would have been prohibitively tedious.
In this next example we use regular expressions to correct misspelled data.
Suppose we have a data vector containing my first name (“Tomas”), but
sometimes misspelled. We want to locate the most common misspellings and
correct them:
tdat <- c("Tom", "Thomas", "Tomas", "Tommy", "tomas")
> misspelled <- grep("^[Tt]omm?y?$|^[Tt]homas$|^tomas$", tdat)
> misspelled
[1] 1 2 4 5
> tdat[misspelled] <- "Tomas"
> tdat
[1] "Tomas" "Tomas" "Tomas" "Tomas" "Tomas"

3.11.6 Repetition > Concatenation > Alternation

Repetition takes precedence over concatenation, which in turn takes prece-

dence over alternation. A whole subexpression may be enclosed in parentheses
to override these precedence rules. For example, consider how the following
regular expression changes when parentheses are used give concatenation
precedence over repetition:
> vec7 <- c("Tommy", "Tomas", "Tomtom")
> #repetition takes precedence
> vec7[grep("[Tt]om{2,}", vec7)]
[1] "Tommy"
> #concatenation takes precedence
> vec7[grep("([Tt]om){2,}", vec7)]
[1] "Tomtom"
Recall that {2,} means repeat the previous 2 or more times.

3.11.7 Metacharacters

Any character, or combination of characters, can be use to specify a pattern

except for these metacharacters:
.\|()[{^$*+?

Metacharacters have special meaning in regular expressions, and these have

already been presented and are summarized in Table 3.8. However, inside a

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

172 3 Managing epidemiologic data in R

Table 3.8 Metacharacters used by regular expressions

Char. Description Example Literal search
^ Matches empty string at beginning of "^my car" See p. 172
line
When 1st character in character class, "[^abc]" See p. 172
matches any single character not in
the list
$ Matches empty string at end of line "my car$" "[$]"
[ Character class "[[]"
. Matches any single character "p.t" "[.]"
? Repetition quantifier (Table 3.7) ".?" "[?]"
* Repetition quantifier (Table 3.7) ".*" "[*]"
+ Repetition quantifier (Table 3.7) ".+" "[+]"
( ) Grouping subexpressions "([Tt]om){2,}" "[(]" or "[)]"
| Join subexpresions, any of which can "Tomas|Luis" "[|]"
be matched
{ Not used in R regular expressions n/a "[{]"

character class, metacharacters have their literal interpretation. For example,

to search for data vector elements that contain one or more periods use:
> vec8 <- c("oswego.dat", "oswego", "infert.dta", "infert")
> vec8[grep("[.]", vec8)]
[1] "oswego.dat" "infert.dta"
If we want to include the following characters inside a character class, they
require special placement:
]-^

To include a literal ], place it first in the list. Similarly, to include a literal

-, place it first or last in the list.16 Finally, to include a literal ^, place it
anywhere but first.
> ages <- c("<1^1", "[1,15)", "15-34", "[35,65)", "[65,110]")
> ages[grep("[]^-]", ages)]
[1] "<1^1" "15-34" "[65,110]"
To search for a literal ^ as a single character is tricky because it must
be placed inside a character class but preceded by another character ("^"
will not work, and "[^]" returns an error). Because we are only interested in
finding ^, then the first character in the list should be any character we expect
not to find in the data vector ("[/^]" should work). Study the example that
follows:
> vec9 <- c("8^2", "89", "y^x", "time")
> grep("/", vec9) #test that / is not in data

16 Although the - sign is not a metacharacter, it does have special meaning inside a
character class because it is used to specify a range of characters; e.g., [A-Za-z0-9].

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.11 Working with regular expressions 173

Table 3.9 Commonly used functions that use regular expressions

Function Description
grep Searches for pattern matches within a character vector; returns integer
vector indicating vector positions containing pattern
regexpr Similar to grep but returns integer vectors with detailed information for
the first occurrence of a pattern match within text string elements of a
character vector
gregexpr Similar to regexpr but returns a list with detailed information for the
multiple occurrences of a pattern match within text string elements of a
character vector
sub Searches and replaces the first occurrence of a pattern match within text
string elements of a character vector
gsub Searches and replaces multiple occurrences of a pattern match within text
string elements of a character vector

integer(0)
> vec9[grep("[/^]", vec9)]
[1] "8^2" "y^x"
The first character in the list (/) was selected because there was no match in
the data vector.

3.11.8 Other regular expression functions

For most epidemiologic applications, the grep function will meet our regu-
lar expression needs. Table 3.9 summarizes other functions that use regular
expressions. Whereas the grep function enables indexing and replacing ele-
ments of a character vector, the sub and gsub functions searches and replaces
single or multiple pattern matches within text string elements of a character
vector. Review the following example:
> vec10 <- c("California", "MiSSISSIppi")
> grep("SSI", vec10) #can be used for replacement
[1] 2
> sub("SSI", replacement="ssi", vec10) #replace 1st occurrence
[1] "California" "MissiSSIppi"
> gsub("SSI", replacement="ssi", vec10) #replace all occurrences
[1] "California" "Mississippi"
The regexpr function provides detailed information on the first pattern
match within text string elements of a character vector. It returns two inte-
ger vectors. In the first vector, -1 indicates no match, and nonzero positive
integers indicate the character position where the first match begins within
a text string. In the second vector, the nonzero positive integers indicate the

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

174 3 Managing epidemiologic data in R

match length. In contrast, the gregexpr function provides detailed informa-

tion on multiple pattern matches within text string elements of a character
vector. It returns a list where each bin contains detailed information (similar
to regexpr) for each text string element of a character vector. Study the
following examples:
> regexpr("SSI", vec10)
[1] -1 3
attr(,"match.length")
[1] -1 3
> gregexpr("SSI", vec10)
[[1]]
[1] -1
attr(,"match.length")
[1] -1

[[2]]
[1] 3 6
attr(,"match.length")
[1] 3 3

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.11 Working with regular expressions 175

Problems

3.1. Using RStudio and the data from Table 3.1 on page 107 Create the
following data frame:
> dat
Status Treatment Agegrp Freq
1 Dead Tolbutamide <55 8
2 Survived Tolbutamide <55 98
3 Dead Placebo <55 5
4 Survived Placebo <55 115
5 Dead Tolbutamide 55+ 22
6 Survived Tolbutamide 55+ 76
7 Dead Placebo 55+ 16
8 Survived Placebo 55+ 69
3.2. Select 3 to 5 classmates and collect data on first name, last name, affil-
iation, two email addresses, and today’s date. Using a text editor, create a
data frame with this data.
3.3. Review the United States data on AIDS cases by year available at http:
//www.medepi.net/data/aids.txt. Read this data into a data frame. Graph
a calendar time series of AIDS cases.
# Hint
plot(x, y, type = "l", xlab = "x axis label", lwd = 2,
ylab = "y axis label", main = "main title")
3.4. Review the United States data on measles cases by year available at
http://www.medepi.net/data/measles.txt. Read this data into a data
frame. Graph a calendar time series of measle cases using an arithmetic and
semi-logarithmic scale.
# Hint
plot(x, y, type = "l", lwd = 2, xlab = "x axis label",
ylab="y axis label", main = "main title")
plot(x, y, type = "l", lwd = 2, xlab = "x axis label", log = "y",
ylab="y axis label", main = "main title")
3.5. Review the United States data on hepatitis B cases by year available at
http://www.medepi.net/data/hepb.txt. Read this data into a data frame.
Using the R code below, plot a times series of AIDS and hepatitis B cases.
matplot(hepb$year, cbind(hepb$cases,aids$cases),
type = "l", lwd = 2, xlab = "Year", ylab = "Cases",
main = "Reported cases of Hepatitis B and AIDS,
United States, 1980-2003")
legend(1980, 100000, legend = c("Hepatitis B", "AIDS"),
lwd = 2, lty = 1:2, col = 1:2)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

176 3 Managing epidemiologic data in R

Table 3.10 Data dictionary for Evans data set

Variable Variable name Variable type Possible values
id Subject identifier Integer
chd Coronary heart disease Categorical-nominal 0 = no
1 = yes
cat Catecholamine level Categorical-nominal 0 = normal
1 = high
age Age Continuous years
chl Cholesterol Continuous >0
smk Smoking status Categorical-nominal 0 = never smoked
1 = ever smoked
ecg Electrocardiogram Categorical-nominal 0 = no abnormality
1 = abnormality
dbp Diastolic blood pressure Continuous mm Hg
sbp Systolic blood pressure Continuous mm Hg
hpt High blood pressure Categorical-nominal 0 = no
1 = yes
(dbp ≥ 95 or
sbp ≥ 160)
ch cat × hpt Categorical product term
cc cat × chl Continuous product term

3.6. Review data from the Evans cohort study in which 609 white males
were followed for 7 years, with coronary heart disease as the outcome of
interest (http://www.medepi.net/data/evans.txt). The data dictionary
is provided in Table 3.10.
a Recode the binary variables (0, 1) into factors with 2 levels.
b Discretized age into a factor with more than 2 levels.
c Create a new hyptertension categorical variable based on the current clas-
sification scheme17 :
Normal: SBP< 120 and DBP< 80;
Prehypertension: SBP=[120, 140) or DBP=[80, 90);
Hypertension-Stage 1: SBP=[140, 160) or DBP=[90, 100); and
Hypertension-Stage 2: SBP≥ 160 or DBP≥ 100.
d Using R, construct a contigency table comparing the old and new hyper-
tension variables.

3.7. Review the California 2004 surveillance data on human West Nile
virus cases available at http://www.medepi.net/data/wnv/wnv2004raw.
txt. Read in the data, taking into account missing values. Convert the cal-
endar dates into the international standard format. Using the write.table
function export the data as an ASCII text file.

17 http://www.nhlbi.nih.gov/guidelines/hypertension/phycard.pdf

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

3.11 Working with regular expressions 177

3.8. On April 19, 1940, the local health officer in the village of Lycoming,
Oswego County, New York, reported the occurrence of an outbreak of acute
gastrointestinal illness to the District Health Officer in Syracuse. Dr. A. M.
Rubin, epidemiologist-in-training, was assigned to conduct an investigation.
(See Appendix A.2 on page 244 for data dictionary.)
When Dr. Rubin arrived in the field, he learned from the health officer
that all persons known to be ill had attended a church supper held on the
previous evening, April 18. Family members who did not attend the church
supper did not become ill. Accordingly, Dr. Rubin focused the investigation
on the supper. He completed Interviews with 75 of the 80 persons known to
have attended, collecting information about the occurrence and time of onset
of symptoms, and foods consumed. Of the 75 persons interviewed, 46 persons
reported gastrointestinal illness.
The onset of illness in all cases was acute, characterized chiefly by nau-
sea, vomiting, diarrhea, and abdominal pain. None of the ill persons reported
having an elevated temperature; all recovered within 24 to 30 hours. Approx-
imately 20% of the ill persons visited physicians. No fecal specimens were
obtained for bacteriologic examination. The investigators suspected that this
was a vehicle-borne outbreak, with food as the vehicle. Dr. Rubin put his
data into a line listing.18
The supper was held in the basement of the village church. Foods were
contributed by numerous members of the congregation. The supper began at
6:00 p.m. and continued until 11:00 p.m. Food was spread out on a table and
consumed over a period of several hours. Data regarding onset of illness and
food eaten or water drunk by each of the 75 persons interviewed are provided
in the line listing. The approximate time of eating supper was collected for
only about half the persons who had gastrointestinal illness.

a. Using RStudio plot the cases by time of onset of illness (include appropriate
labels and title). What does this graph tell you? (Hint: Process the text
data and then use the hist function.)
b. Are there any cases for which the times of onset are inconsistent with the
general experience? How might they be explained?
c. How could the data be sorted by illness status and illness onset times?
d. Where possible, calculate incubation periods and illustrate their distribu-
tion with an appropriate graph. Use the truehist function in the MASS
package. Determine the mean, median, and range of the incubation period.

18 See data set at http://www.medepi.net/data/oswego/oswego.txt.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

 Part II
Applied Epidemiology

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Chapter 4
Analyzing simple epidemiologic data

4.1 Overview and concepts

In this chapter we focus on analyzing one-sample measurements (measures of

occurrence: rate, risk, and prevalence) and two-sample measurements (mea-
sures of association: rate ratio, risk ratio, and odds ratio). At the point of
analyzing data that has already been collected we are assuming that the
study subjects are representative of the target population (no selection bias),
and that the data was measured accurately (no measurement bias). Now, for
a given measure of occurrence or measure of association we are interested in
the following:
1. What is the point estimate? (estimation)
• Measures of occurrence are one-sample measures
• Measures of association are two-sample measures
2. What is the variability and precision of this estimate? (confidence interval)
3. Is this estimate consistent with a reference value? (p value and Type I
error)
4. What is the chance of detecting a meaningful difference from the reference,
if one exists? (effect size, power, and Type II error)
5. How many subjects (sample size) are required for the following?
• To achieve a desired confidence interval width for a one-sample measure
(used for descriptive studies);
• To detect if a one-sample measure differs from a reference value; and
• To detect if a two-sample measure differs from a reference value (used
for analytic studies)
First, the point estimate enables epidemiologic inference. For example,
what is the risk of disease occurrence or the prevalence of a condition? What
is the risk ratio for developing disease among exposed persons compared to
unexposed persons? Second, the confidence interval provides information on

181
182 4 Analyzing simple epidemiologic data

the varianace of the estimate. The variance depends on the natural variability
of the measure (e.g., height in human populations), measurement random
error (e.g., instrument, technician), and sample size. Increasing the sample
size improves estimation precision, narrowing the confidence interval.
Third, we might be interested in knowing whether this point estimate is
consistent with a reference value. Under the assumption that the estimate
comes from a distribution with the mean equal to the reference value (null
hypothesis), we can calculate the probability (two-sided p-value) of getting
the test statistic value or more extreme values. If the null hypothesis is true
and we incorrectly reject the null hypothesis because the p value is lower than
α (arbitrarily chosen—usually 0.05), we call this a Type I error.
Fourth, if it was “consistent” (p > α), was there a sufficient sample size to
detect a meaningful difference if one existed? We don’t want to mistakenly
infer no difference when there was insufficent data to draw this conclusion.
This requires defining what we mean by “meaningful difference” (we will
call it “effect size”1 ), and then calculating the probability of detecting the
effect size (or larger), if it exists. This probability is called statistical power.
An effect size implies an alternative hypothesis. The probability of failing to
detect the effect size under the alternative hypothesis is β or Type II error.
Power is equal to 1 − β
Fifth, when we decide to sample a target population to estimate epidemio-
logic measures, our required sample size calculations will depend on whether
the statistic is a one-sample measure (measures of occurrence) or a two-
sample measure (measures of association). For a one-sample measure, we
may be interested in the following:
• Sufficient sample size to achieve a desired confidence interval width (used
for descriptive studies);
• Sufficient sample size for hypothesis testing (meaningful difference from
some reference value). This requires setting type I (α) and type II errors
(β).

4.1.1 Estimation

The estimation of epidemiologic measures is usually divided into point esti-

mation and confidence interval estimation. We cover CI estimation in the next
section. Displayed in Table 4.1 are the measures of occurrence and measures
of association commonly used in epidemiology. Among these, we will focus
on incidence, prevalence, rate, risk, odds, and the ratio of rates (rate ratio),
risks (risk ratio), and odds (odds ratio). Both incidence and rate are driven by
the numbers of new cases occurring during a period of time. These processes

1 From intervention trials

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.1 Overview and concepts 183

are commonly represented using a Poisson probability model. For prevalence

and risk we often use proportions for estimation, and we commonly represent
these using a binomial probability model.
Which point estimation method we might use will depend on whether we
estimate a one- or two-sample measure, and whether the sample is large or
small. The estimation methods are summarized in Table 4.2
The odds ratio from a 2 × 2 table provides an example of UMLE, CMLE,
MUE, and SSAE. We will use the notation for a crude 2 × 2 table (see
Table 4.3). The UMLE odds ratio is just the ratio of the disease odds for a
cohort study, or the ratio of the exposure odds for a case-control study.

A1 /B1 A1 B0
DOR = =
A0 /B0 A0 B1
A1 /A0 A1 B0
EOR = =
B1 /B0 A0 B1

Presented in Table 4.4 is data from a cohort study of diarrhea in breast-

fed infants colonized with Vibrio cholerae 01 [5]. The occurrence of diarrhea

Table 4.1 Epidemiologic measures of occurrence and association

Type Measuresa Description
Measures of Time Average time in a state (“survival”)
occurrence Incidenceb Number of new cases
Prevalencec Number of existing cases
Rate New cases per person-time at risk
Riskd Probability of becoming a case
Odds Odds of becoming a case
Measures of Rate ratio Comparison of two rates
association Risk ratio Comparison of two risks
Odds ratio Comparison of two odds
a
All measures have a time element that must be specified
b
Sometimes expressed as a proportion by dividing by “total population”
c
Commonly expressed as a proportion by dividing by “total population”
d
Sometimes estimated using a proportion (binomial model)

Table 4.2 Commonly used estimation methods for epidemiologic measures

Measure Large sample Small sample
One-sample UMLE MUE
Two-sample UMLE MUE, CMLE, SSA
UMLE = unconditional maximum likelihood estimation
CMLE = conditional maximum likelihood estimation
UMLE = median unbiased estimation
SSA = small sample adjustment estimation

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

184 4 Analyzing simple epidemiologic data

was assessed comparing infants with low vs. high antibody titers. The UMLE
disease odds ratio can be calculated:
12/2 (12)(9)
ORUMLE = = = 7.71
7/9 (7)(2)

The odds of developing diarrhea was 7.7 times higher in infants with low
antibodies titers compared to infants with high antibody titers.
The CMLE odds ratio calculation is based on using the hypergeometric
distribution for tables with small numbers such as Table 4.4. In this case, we
treat the margins as fixed and model the distribution of A1 . The hypergeo-
metric equation (Equation 4.1) is the basis of Fisher’s exact test.
n1

n0
a1
a1 m1 −a1 OR
Pr(A1 = a1 | m1 , m0 , n1 , n0 ) = P n1
n0
k
(4.1)
k k m1 −k OR

where k ranges over all possible values of A1 . The solution to OR in Equa-

tion 4.1 is the CMLE odds ratio. In R, we use the fisher.test function to
calculate the CMLE odds ratio:
> dat <- matrix(c(12, 2, 7, 9), 2, 2)
> fisher.test(dat)$estimate
odds ratio
7.166131
Therefore, ORCMLE = 7.17.
The MUE odds ratio is based on calculating exact lower and upper p-
values. Using this hypergeometric example, the MUE odds ratio satisfies
Equation 4.1 when Plower = Pupper = 0.5, or

0 = Plower − Pupper . (4.2)

Table 4.3 Notation for crude 2 × 2 table

Exposed Unexposed Total
Cases A1 A0 M1
Noncases B1 B0 M0
Total N1 N0 N

Table 4.4 Comparison of diarrhea in 30 breast-fed infants colonized with V. Cholerae 01,
by antibody titers in mother’s breast milk
Antibody level
Low High Total
Diarrhea 12 7 19
No diarrhea 2 9 11
Total 14 16 30

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.1 Overview and concepts 185

Therefore, with Equation 4.2 we can use the uniroot and fisher.test func-
tions to solve for the MUE odds ratio. This involves two steps:
• Create function for Equation 4.2; and
• Pass equation function and data to uniroot to solve unknown OR.
The equation function must have one unknown, in this case it will be the
odds ratio output from fisher.test.
> dat <- matrix(c(12, 2, 7, 9), 2, 2) #create data object
> or.mue.eqn <- function(x, or) {
+ fisher.test(x, or = or, alt = "less")$p.value -
+ fisher.test(x, or = or, alt = "greater")$p.value
+ }
> uniroot(function(or) {or.mue.eqn(x = dat, or)},
+ interval = c(0, 100))$root
[1] 6.88068
Therefore, ORMUE = 6.88
Here is the small-sample adjusted odds ratio formula from Jewell: [6]:

A1 B0
ORSS = (4.3)
(A0 + 1)(B1 − 1)

And here is the calculation in R:

> (12*9)/((7+1)*(2+1))
[1] 4.5
We used R to demonstrate four different methods for calculating an odds
ratio for a crude 2 × 2 table (Table 4.5). When the observed or expected
number in any cell is less than 5, then use one of the small sample methods
for estimation. If there would be a zero in the denominator of the UMLE,
then use Jewell’s small sample adjustment.

4.1.1.1 p-value function

Before getting to confidence intervals, we must review the p-value function.

We can test if our odds ratio estimate is consistent with a value of 1 (i.e.,
test of the null hypothesis).

Table 4.5 Summary of odds ratio estimation using data from Table 4.4
Method Odds ratio Comment
Unconditional MLE 7.7 Large sample
Conditional MLE 7.2 Small sample
Median Unbiased Estimate 6.9 Small sample
Small Sample Adjusted 4.5 Small sample; zero in denominator

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

186 4 Analyzing simple epidemiologic data

H0 : OR = 1
H1 : OR 6= 1

For our hypothesis test we will calculate a two-sided p-value using the
fisher.test function.
> dat <- matrix(c(12, 2, 7, 9), 2, 2) #create data object
> fisher.test(dat, or = 1, alt = "two.sided", conf.int = F)

Fisher’s Exact Test for Count Data

data: dat
p-value = 0.02589
alternative hypothesis: true odds ratio is not equal to 1
sample estimates:
odds ratio
7.166131
Therefore, under the null hypothesis, the probability of seeing the CMLE
odds ratios of 7.16 or more extreme is 0.026. Similarly, we can calculate and
plot two-sided p-values for a range of test hypotheses: this is called the p-
value function, and it enables us to view which test hypotheses are most
consistent with the observed data.
To create and plot a p-value function for our data (Table 4.4), we use the
fisher.test function to calculate the two-side exact p-values for a range of
test hypotheses. The following R code generates the p-value function plot in
Figure 4.1.
dat <- matrix(c(12, 2, 7, 9), 2, 2) #create data object
cases <- seq(0, 65, .1) #range of hypotheses
nn <- length(Cases) #length of vector
p.vals <- rep(NA, nn) #empty vector to fill
for(i in 1:nn){
p.vals[i] <- fisher.test(dat, or = Cases[i])$p.value
}
plot(cases, p.vals, cex = 0.5, log = "x",
xlab = "Cases (log scale)", ylab = "P value")
abline(v = 6.88068, h = 0.05, lty = 2)
abline(v = 1) #null hypothesis (OR = 1)
In summary, the p-value function gives the two-sided p-values for a range
of test hypotheses, including the null hypothesis. The point at which the
curve peaks is the point estimate. The concentration of the curve around
the point estimate represents the precision of the estimate. Where the p =
0.05 horizontal line intersects the curve delineates the lower and upper 95%
confidence limits. In fact, a confidence interval represents only one horizontal
slice through the p-value function. The (1 − α) confidence interval are the
test hypothesis values for which the two-sided p > α.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.1 Overview and concepts 187

1.0
0.8
0.6
P value

0.4
0.2
0.0

0.1 0.5 2.0 10.0 50.0

Cases (log scale)

Fig. 4.1 The p-value function for the CMLE odds ratio from Table 4.4. The plot is
discontinuous because the hypergeometric distribution is a discrete probability model. The
vertical solid line is the null hypothesis (OR = 1). The vertical dotted line is the MUE
odds ratio (OR = 6.88). The horizontal line is at p = 0.05, and where it intersects the
curve delineates the 95% confidence interval. Notice that the 95% CI does not contain the
null hypothesis, which is consistent with p = 0.026 for testing OR = 1.

4.1.2 Confidence intervals

The width of confidence intervals provide information on natural variability

of the measure and the precision of the measurement process. Increasing
sample size will improve precision. For starters, we are generally are interested
in calculating a 95% confidence interval (CI). But what is a 95% CI? A
95% confidence interval represents a statistical procedure, that if it could
be repeated many times, we would construct intervals that would cover the
“true” population mean value 95% of the time. The procedure we use depends
on what we know or what we assume about the distribution of a specific
measure.
First, does this measure follow a known distribution? Second, if we are
not sure, can we safely assume it follows a known distribution? Third, can we
directly construct a confidence interval using this known distribution? Using
today’s computers, this is usually straightforward. Fourth, can we construct
a reasonable confidence interval using a normal distribution approximation?
Fifth, if we have no idea of the underlying distribution of a specific measure,
can we use simulation methods to construct the confidence interval?

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

188 4 Analyzing simple epidemiologic data

In general, we use the following methods to construct confidence intervals

for epidemiologic measures:
• From p-value function (see Section 4.1.1.1);
• Normal distribution approximation methods;
• Exact methods using a known distribution;
• Exact method approximations using derived formulas; and
• Resampling (simulation) methods for unknown distributions.

4.1.2.1 Confidence interval using normal approximation

Epidemiologists are familiar with the normal distribution approximation to

construct a confidence interval for large sample data. For some estimate, say
Θ, the confidence interval is constructed using this formula:

ΘL , ΘU = Θ ± Z × SE(Θ), (4.4)
where ±Z are the quantile values of the standard normal distribution density
curve such that the area between −Z and +Z is equal to the confidence
level. SE(Θ) is the standard error of the measure Θ. For a 95% confidence
interval, P (−Z ≤ z ≤ +Z) = 0.95, where Z = 1.96. This means that P (Z ≤
−1.96) = P (z ≥ 1.96) = 0.025. In truth, 1.96 is only an approximation.
To get the precise Z quantile values that correspond to a specific confidence
level, we use the qnorm function. For a given probability p (area under the
normal distribution density curve), where p = P (Z ≤ q), the qnorm function
calculates the corresponding quantile value q. The following code does the
trick:
> conf.level <- 0.90
> qnorm((1 + conf.level)/2)
[1] 1.644854
> conf.level <- 0.95
> qnorm((1 + conf.level)/2)
[1] 1.959964
> conf.level <- 0.99
> qnorm((1 + conf.level)/2)
[1] 2.575829
Now, the SE(Θ) is often calculated from some formula that has been de-
rived. For example, for the disease odds ratio from a cohort study (Table 4.3),
here is the standard error formula:
r
1 1 1 1
SE[log(OR)] = + + +
A1 B1 A0 B0

And here is the (1 − α)% confidence interval:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.1 Overview and concepts 189

ORL , ORU = exp{log(OR) ± Zα/2 SE[log(OR)]}

Therefore, using the data of diarrheal illness in breast-fed infants, we can

calculate the 90% confidence interval in R:
> conf.level <- 0.90
> Z <- qnorm((1 + conf.level)/2)
> logOR <- log((12*9)/(7*2))
> SE.logOR <- sqrt(1/12 + 1/2 + 1/7 + 1/9)
> OR <- exp(logOR + c(-1, 1)*Z*SE.logOR)
> OR
[1] 1.712517 34.750147
HERE:

4.1.2.2 Confidence interval using exact methods

If we know (or assume to know) the underlying distribution, then it is pos-

sible to calculate an exact confidence interval. Calculating exact confidence
intervals is necessary for small sample. We will briefly review the following
approaches:
• R functions;
• Tail method; and
• Resampling (e.g., bootstrapping)

Confidence interval for crude 2 × 2 table: odds ratio and rate ratio

R provides two functions for calculating exact confidence intervals for crude
2 × 2 tables:
• fisher.test for CMLE odds ratios; and
• poisson.test for rate ratios.

Tail method

Resampling

4.1.2.3 Confidence interval using exact approximations

Before the wide availability of computing power, formulas were derived to

approximate exact confidence intervals for selected epidemiologic measures.
We’ll see a few examples in the sections that follow.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

190 4 Analyzing simple epidemiologic data

4.1.3 Hypothesis testing and p-values

A natural question that arises from evaluating an epidemiologic measure is:

how compatible is the estimate compared to some reference value? For ex-
ample, suppose that R̂ proportion of hospitalized patients suffered a severe
complication last year, and our goal was to have not more than R0 propor-
tion experiencing a severe complication. How compatible is our experience
(R̂) with a reference value of R0 . To answer this, we specify the null hy-
pothesis that R = R0 . For a given sample population, we can calculate the
probability p of observing a value of R̂ or greater under the null hypothesis:
p = P (R ≥ R̂ | R = R0 ). Because we are only interested in an increase
in complications, so a one-sided p value is appropriate. If this p value is
high, then our estimate (R̂) is more compatible with the null hypothesis. In
other words, values of R̂ or more extreme are likely to occur by chance alone
(random error). In contrast, if this p value is low, then our estimate is less
compatible with the null hypothesis. That is, values of R̂ or more extreme
are unlikely to occur by chance alone (random error). High p values (more
compatible with the null hypothesis) can occur because the null hypothesis
is true, or because the sample size is too small to detect an alternative hy-
pothesis (i.e., insufficient statistical power—more on this later). Similar to
confidence intervals, to calculate the p value we must either know or make
assumptions about the underlying distribution.
For example, suppose 5% of our hospital admissions resulted in severe
complications and our goal was to stay at or below a reference value of 3%. We
naturally want to know if our observed 5% is compatible with the reference
goal of 3% or lower. The one-sided p value depends on the magnitude of
the difference (5% vs 3%) and number of hospitalizations. If 8 out of 160
hospital admissions (5%) resulted in a severe complication, is 8 complications
compatible with the reference goal of 3% or less?

x=8 n = 160
R̂ = 8/160 = 0.05 R0 = 0.03
p = P (R ≥ R̂ | R0 ) = P (R ≥ 0.05 | R0 = 0.03)

Assuming a binomial distribution, we can use the binom.test function to

calculate the one-sided p value.
> binom.test(x = 8, n = 160, p = 0.03, alternative = "greater")

Exact binomial test

data: 8 and 160

num. of successes = 8, num. of trials = 160, p-value = 0.1101
alternative hypothesis: true probability of success is > 0.03

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.1 Overview and concepts 191

Therefore, there is an 11% probability of observing 8 or more severe compli-

cations with 160 admissions. Is this compatible with 3%? There is no right
answer, but if we select a signifance level α = 0.05, then we would conclude
“yes, it is compatible” because p > 0.05 and we do not reject the null hypoth-
esis. The problem with this approach is that by merely increasing the number
of hospitalizations (i.e., sample size), the p value becomes “significant” (less
than 0.05), even if the magnitude of the difference remains the same (5% vs.
3%). Table 4.6 summarizes this example. We will learn how to handle this
later by expliciting specifying what differences matters.

4.1.3.1 Significance level (α) and Type I error

How low should the p value be for us to conclude that R̂ is not compatible
with R0 ? There is no hard fast rule. However, sometimes a decision rule is
used to reject the null hypothesis if p ≤ α, where α is arbitrarily set to be
small (also called the significance level). Therefore, under the null hypothesis,
we are willing to incorrectly reject the null hypothesis α(×100)% of the time.
This is called the Type I error and is often set at α = 0.05.
This has been an example of a one-sample measure (R̂). These concepts
equally apply to two-sample measures of association such as rate ratios, risk
ratios, and odds ratios. The reference value for these measures of association
will be the null value 1 for no association.

4.1.3.2 The p value function

In the previous example, we calculated a one-sided p value under the null

hypothesis. More generally, for a given sample, we can calculate a two-sided
p value not only for the null hypothesis, but also for a sequence of alternative

Table 4.6 One-sided p value example: Is a hospital severe complication risk of 5% com-
patible with a reference value (“goal”) of 3% or less? Answer: It depends.
Reference Sample Complications Proportion (R̂) one-sided p value
(R0 ) (n) (x) (R̂ = x/n) P (R ≥ R̂)
0.03 20 1 0.05 0.4562
0.03 40 2 0.05 0.3385
0.03 80 4 0.05 0.2193
0.03 160 8 0.05 0.1101
0.03 640 32 0.05 0.0041

Table 4.7 Hypothetical cohort data corresponding the p value function in Figure 4.2
Exposure Cases Person-years
Yes 9 186
No 2 128

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

192 4 Analyzing simple epidemiologic data

1 0

0.9 10

0.8 20

95% Lower Confidence Limit = 0.74

95% Upper Confidence Limit = 21.0

Confidence level (%)

0.7 30

Median unbiased estimate

Null hypothesis
0.6
P−value

40

0.5 50

0.4 60

0.3 70

0.2 80

0.1 90
95% Confidence Interval
0.05 95

0 100

0.2 0.5 1.0 2.0 2.9 5.0 10.0 20.0

Rate Ratio
Fig. 4.2 The p-value function for the hypothetical cohort study in Table 4.7. The rate
ratio median unbiased estimate (rr
ˆ = 2.9) is tested against the null hypothesis (rr = 1) and
alternative hypotheses (rr 6= 1). The alternative hypotheses that correspond to p = 0.05
are the lower and upper 95% Confidence Limits, respectively.

hypotheses. The point estimate is tested against these alternative hypotheses:

the higher the p value, the more compatible the observed value is with this
alternative hypothesis. The plot of these p values is known as the p value
function or confidence interval funtion [7].
We have seen the p-value function for a Poisson count (Figure ??). Here
is a another example. Consider the hypothetical cohort data in Table 4.7
from Rothman [7]. From these two rates we can calculates the rate ratio and
two-sided p-values to construct the p-value function (Figure 4.2).2 We note
from the p-value function that the MUE is at the peak, and that we can also
determine confidence intervals.

2 Using methods described in Rothman [7], we calculated median unbiased estimates and
two-sided mid-P exact p-values.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.1 Overview and concepts 193

Null distribution Alternative distribution

H0 H1

Power
( 1 − β)

β α 2

− Z1−α 2 µ0 Z1−α 2 µ1

Fig. 4.3 Sample distribution curves to understand the relations of components that are
used to calculate sample size

4.1.4 Power and Type II error

Up to now we have asked if an estimate is consistent with a reference value.

This reference value is usually a “null” hypotheses for which we can calculate
one- or two-sided p-values. With a range of alternative hypotheses we can
generate a p-value function. However, now we may interested in detecting a
difference from some reference value. This requires an understanding of Type
II error (β) and power (1 − β). (see Figure 4.3). β, or Type II error, is the
probability of failing to reject the null hypothesis (no difference) when the
alternative hypothesis is true. Power, or 1 − β, is the probability of detecting
a difference when the alternative hypothsis is true. To be meaningful, this
difference must have clinical or public health relevance.
When a p-value is “high” there are two possible explanations: the estimate
is consistent with the reference value, or the estimate is different from the
reference value but we did not have enough power to detect this difference.
Therefore, to draw a valid conclusion that an estimate is consistent with a
reference value we must have enough power to detect a meaningful difference

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

 194 4 Analyzing simple epidemiologic data

if one exists. Power is influenced by the difference of interest, the variability,

and the sample size. We see that statistical power is very important: we need
sufficient power for (1) inferring an estimate is consistent with a reference
value, and (2) calculating a sample size to detect a difference.

4.1.5 Sample size calculations for analytic studies

Sample size to be
covered in differ-
ent section
4.2 Evaluating a single measure of occurrence

In terms of implementing analytic methods using R, we use the following

steps:
1. Assign input values
2. Do calculations
3. Collect results
Using this approach prepares one for writing functions later.

4.2.1 Poisson count (incidence) and rate data

In this section, we address the count of new cases (incidence). For convenient,
we assume the occurrence of new cases follows a Poisson distribution. The
Poisson distribution is a discrete probability distribution with the following
density function:
x−λ λx
P (X = x) = , (4.5)
x!
where X is the random variable, x is the observed count, and λ is the expected
count. And, here is the distribution distribution function:
x
X k −λ λk
P (X ≤ x) = . (4.6)
k!
k=0

In R, we use the dpois and ppois functions for the density and distri-
bution functions, respectively. For example, in the United States, the rate of
meningococcal disease is about 1 case per 100,000 population per year [8]. In
San Francisco, with a population of about 800,000, we expected about 8 cases
per year (λ). In a given year, what is the probability of observing exactly 6
cases? Using the dpois function,

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.2 Evaluating a single measure of occurrence 195

> dpois(x = 6, lambda = 8)

[1] 0.1221382
there is about a 12% chance of observing 6 cases exactly. In a given year, what
is the probability of observing more than 10 cases? The chance of observing
more than 10 cases is the 1 − P (X ≤ 10):
> 1 - ppois(q = 10, lambda = 8)
[1] 0.1841142
Therefore, there is about an 18% chance of observing more than 10 cases in
a given year.

4.2.1.1 Estimation

In epidemiology, we calculate the average or per-capita rate (r) as the count

of new cases (x) divided by the person-time at risk (P T ).
x
r= (4.7)
PT
Again, for convenience, we assume the count x has a Poisson distribution,
and we consider P T a fixed quantity.

4.2.1.2 Confidence intervals

Normal approximation

Here is the standard error of an incidence rate, and the formula to construct
a confidence interval using a normal approximation:
p
SE(r) = x/P T 2
rL ; rU = r ± Z × SE(r)

Consider 8 cases of cancer from 85,000 person-years at risk for a rate of

9.4 cases per 100,000 person-years [3]. Here we calculate a 90% confidence
interval using a normal approximation method,
> #assign input value
> conf.level <- 0.90
> Z <- qnorm(0.5*(1 + conf.level))
> x <- 8
> PT <- 85000
> mult <- 100000
> #do calculations
> r <- x/PT

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

196 4 Analyzing simple epidemiologic data

> SE.r <- sqrt(x/PT^2)

> LL <- r - Z*SE.r
> UL <- r + Z*SE.r
> #collect results
> cbind(x, PT, rate=mult*r, lower=mult*LL, upper=mult*UL)
x PT rate lower upper
[1,] 8 85000 9.4118 3.9384 14.885
If we plan on using this method often, it is convenient to convert your
steps into a function:
cipois.norm <- function(x, PT = 1, conf.level = 0.95, mult = 1){
Z <- qnorm(0.5*(1 + conf.level))
r <- x/PT
SE.r <- sqrt(x/PT^2)
LL <- r - Z*SE.r
UL <- r + Z*SE.r
cbind(x, PT, rate=mult*r, lower=mult*LL, upper=mult*UL,
conf.level=conf.level, multiplier=mult)
}
Notice, we followed the same steps: assign input values (default values were
assigned to function arguments), do calculations, and collect results (into a
matrix object). We set the default PT=1, which calculate a confidence interval
for a Poisson count. We also added a multiplier in order to change the rate
to a more interpretable number (e.g., rate per 100,00 person-years). Here we
test our function:
> cipois.norm(8, 85000, 0.90, 100000)
x PT rate lower upper conf.level multiplier
[1,] 8 85000 9.4118 3.9384 14.885 0.9 1e+05

Exact approximation

For low counts, an exact confidence interval is more accurate. A normal dis-
tribution is symmetric; however, a low count will have an asymmetric distri-
bution. We will use the Poisson distribution to calculate an exact confidence,
but first we start with an exact approximation to the Poisson distribution
using Byar’s method [3].

Byar’s confidence limits:

s !3
1 Z 1
rL , rU = (x + 0.5) 1 − ± /P T (4.8)
9(x + 0.5) 3 (x + 0.5)
Here is Byar’s method converted into a function:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.2 Evaluating a single measure of occurrence 197

cipois.byar <- function(x, PT = 1, conf.level = 0.95, mult = 1) {

Z <- qnorm(0.5*(1+conf.level))
Zinsert <- (Z/3)*sqrt(1/(x+0.5))
r <- x/PT
LL <- ((x+0.5)*(1-1/(9*(x+0.5))-Zinsert)^3)/PT
UL <- ((x+0.5)*(1-1/(9*(x+0.5))+Zinsert)^3)/PT
cbind(x, PT, rate=mult*r, lower=mult*LL, upper=mult*UL,
conf.level=conf.level, multiplier=mult)
}
Suppose we observe a rate of 3 cases per 2500 person-years (12 cases
per 10,000 person-years). What is the 90% confidence interval using Byar’s
method:
> cipois.byar(3, 2500, 0.90, 10000)
x PT rate lower upper conf.level multiplier
[1,] 3 2500 12 3.3446 25.437 0.9 10000

Exact methods

The tail method:

cipois.exact <- function (x, PT = 1, conf.level = 0.95, mult = 1) {
f1 <- function(x, ans, alpha = alp) {
ppois(x, ans) - alpha/2
}
f2 <- function(x, ans, alpha = alp) {
1 - ppois(x, ans) + dpois(x, ans) - alpha/2
}
alp <- 1 - conf.level
interval <- c(0, x * 9)
r <- x/PT
UL <- uniroot(f1, interval = interval, x = x)$root/PT
if(x == 0) {
LL <- 0
} else {
LL <- uniroot(f2, interval = interval, x = x)$root/PT
}
cbind(x, PT, rate=mult*r, lower=mult*LL, upper=mult*UL,
conf.level=conf.level, multiplier=mult)
}
Suppose we observe a rate of 3 cases per 2500 person-years (12 cases per
10,000 person-years). What is the 90% confidence interval using exact tail
method for the Poisson distribution?
> cipois.exact(3, 2500, 0.90, 10000)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

198 4 Analyzing simple epidemiologic data

x PT rate lower upper conf.level multiplier

[1,] 3 2500 12 3.2708 31.015 0.9 10000

The Daly method:

In 1992, Dr. Leslie Daly published a SAS macro that uses the Gamma distri-
bution to calculate exact confidence intervals for a Poisson count. The SAS
macro was translated into the R language.
cipois.daly <- function(x, PT = 1, conf.level = 0.95, mult = 1) {
r <- x/PT
if(x != 0) {
LL <- qgamma((1 - conf.level)/2, x)/PT
UL <- qgamma((1 + conf.level)/2, x + 1)/PT
} else {
if(x == 0) {
LL <- 0
UL <- -log(1 - conf.level)/PT
}
}
cbind(x, PT, rate=mult*r, lower=mult*LL, upper=mult*UL,
conf.level=conf.level, multiplier=mult)
}
Suppose we observe a rate of 3 cases per 2500 person-years (12 cases per
10,000 person-years). What is the 90% confidence interval using exact Daly
method for the Poisson distribution?
> cipois.daly(3, 2500, 0.90, 10000)
x PT rate lower upper conf.level multiplier
[1,] 3 2500 0.94118 3.2708 31.015 0.9 10000

4.2.1.3 Comparison

Comparison to a “fixed” reference value.

4.2.1.4 Power and sample size

4.2.2 Binomial risk and prevalence data

4.2.2.1 Estimation
x
R= (4.9)
N

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.2 Evaluating a single measure of occurrence 199

4.2.2.2 Confidence intervals

Normal approximation
r
x(N − x)
SE(R) = (4.10)
N3
R does have a function for testing a one sample proportions. However,
it was more informative to learn how to create your own function using
known statistical formulas. Why? Because R or another software package
may not have the specific function you need, and by learning how to create
you own functions you will be able to solve many more problems effectively
and efficiently. Here is the same analysis using R’s prop.test function (which
additionally provides a confidence interval).

> prop.test(x=39, n=215, p=.15)

1-sample proportions test with continuity correction

data: 39 out of 215, null probability 0.15

X-squared = 1.425, df = 1, p-value = 0.2326
alternative hypothesis: true p is not equal to 0.15
95 percent confidence interval:
0.1335937 0.2408799
sample estimates:
p
0.1813953

Exact approximation

Wilson’s confidence limits:

Baby Rothman, p. 132.
" r #
N x Z2 x(N − x) Z2
RL , RU = + ±Z + (4.11)
N + Z2 N 2N N 3 4N 2
Show R code for
Wilsons

Exact methods

To assess whether an observed one sample proportion (R = x/N ) differs from

an alternative value one can calculate a p value based on the binomial distri-
bution (binom.test) or a normal distribution approximation to the binomial
distribution (covered in previous section). The approximation using the nor-

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

200 4 Analyzing simple epidemiologic data

mal distribution is satisfactory when the expected number of “successes” (x)

and the “failures” (N-x) are both larger than 5. Now, here is the same one
sample analysis but using an exact method with R’s binom.test function:

> binom.test(x=39, n=215, p=.15)

Exact binomial test

data: 39 and 215

number of successes = 39, number of trials = 215, p-value = 0.2135
alternative hypothesis: true probability of success is not equal to 0.15
95 percent confidence interval:
0.1322842 0.2395223
sample estimates:
probability of success
0.1813953

4.2.2.3 Comparison

4.2.2.4 Power and sample size

4.3 Evaluating two measures of occurrence

4.3.1 Comparing two rate estimates: Rate ratio

4.3.1.1 Estimation

r1 a/P T1
rr = = (4.12)
r0 b/P T0

4.3.1.2 Confidence intervals

Normal approximation
r
1 1
SE[log(rr)] = + (4.13)
a b

Table 4.8 Table notation for cohort study, person-time data

Exposed Unexposed Total
Number of new cases a b M
Person-time at risk P T1 P T0 T

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.3 Evaluating two measures of occurrence 201

Exact approximation

None known to
TJA

Exact methods

Show example
using poisson.test
function
4.3.1.3 Comparison to reference value (hypothesis testing &
p-values) Compare to pois-
son.exact from
exactci package

Show example
using poisson.test
4.3.1.4 Power and sample size function

4.3.2 Comparing two risk estimates: Risk ratio and Compare to pois-
disease odds ratio son.exact from
exactci package

##Table set up
## Disease
##Exposure Yes No Total
## Yes x1 . n1
## No x0 . n0

4.3.2.1 Estimation

R1 a/N1
RR = = (4.14)
R0 b/N0
R1 /(1 − R1 ) a(N0 − b)
DOR = = (4.15)
R0 /(1 − R0 ) b(N1 − a)

Table 4.9 Cohort study, binomial data

Exposed Unexposed
Number of new cases a b
Persons at risk N1 N0

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

202 4 Analyzing simple epidemiologic data

4.3.2.2 Confidence intervals

r
1 1 1 1
SE[log(RR)] = − + − (4.16)
a N1 b N0
r
1 1 1 1
SE[log(DOR)] = + + + (4.17)
a N 1 − a b N0 − b

Normal approximation

###Risk Ratio CI from baby Rothman, p. 135

rr.wald <- function(x, conf.level = 0.95){
##prepare input
aa <- x[1,1]; n1 <- sum(x[1,])
bb <- x[2,1]; n0 <- sum(x[2,])
##calculate
p1 <- aa/n1 ##risk among exposed
p0 <- bb/n0 ##risk among unexposed
RR <- p1/p0
logRR <- log(RR)
SElogRR <- sqrt(1/aa - 1/n1 + 1/aa - 1/n0)
Z <- qnorm(0.5*(1 + conf.level))
LCL <- exp(logRR - Z*SElogRR)
UCL <- exp(logRR + Z*SElogRR)
##collect
list(x = x,
risks = c(p1 = p1, p0 = p0),
risk.ratio = RR,
conf.int = c(LCL, UCL),
conf.level = conf.level
)
}

Here we test the code:

> ##From Jewell, p. 83
> tab7.4 <- matrix(c(178, 79, 1411, 1486), 2, 2)
> dimnames(tab7.4) <- list("Behavior type" = c("Type A", "Type B"),
+ "CHD Event" = c("Yes", "No"))
> tab7.4
CHD Event
Behavior type Yes No
Type A 178 1411
Type B 79 1486
> rr.wald(tab7.4)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.3 Evaluating two measures of occurrence 203

$x
CHD Event
Behavior type Yes No
Type A 178 1411
Type B 79 1486

$risks
p1 p0
0.112020 0.050479

$risk.ratio
[1] 2.2191

$conf.int
[1] 1.7186 2.8654

$conf.level
[1] 0.95

Exact approximation

None known to
TJA

Exact methods

rr.boot <- function(x, conf.level = 0.95, replicates = 5000){

##prepare input
aa <- x[1,1]; n1 <- sum(x[1,])
bb <- x[2,1]; n0 <- sum(x[2,])
##calculate
p1 <- aa/n1 ##risk among exposed
p0 <- bb/n0 ##risk among unexposed
RR <- p1/p0
r1 <- rbinom(replicates, n1, p1)/n1
aa.boot <- aa.boot2 <- rbinom(replicates, n0, p0)
bb.boot[bb.boot2==0] <- bb.boot2 + 1
n0.denom <- rep(n0, replicates)
n0.denom[bb.boot2==0] <- n0.denom + 1
r0 <- bb.boot/n0.denom
rrboot <- r1/r0
rrbar <- mean(rrboot)
alpha <- 1 - conf.level
ci <- quantile(rrboot, c(alpha/2, 1-alpha/2))

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

 204 4 Analyzing simple epidemiologic data

##collect
list(x = x,
risks = c(p1 = p1, p0 = p0),
risk.ratio = RR,
rrboot.mean = rrbar,
conf.int = unname(ci),
conf.level = conf.level,
replicates = replicates)
}

Here we test the code:

> rr.boot(tab7.4)
$x
CHD Event
Behavior type Yes No
Type A 178 1411
Type B 79 1486

$risks
p1 p0
0.112020 0.050479

$risk.ratio
[1] 2.2191

$rrboot.mean
[1] 2.2443

$conf.int
[1] 1.7235 2.9085

$conf.level
[1] 0.95

$replicates
[1] 5000

4.3.2.3 Comparison to reference value (hypothesis testing &

p-values)

See fisher.exact in
exact2x2 package

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

4.3 Evaluating two measures of occurrence 205

4.3.2.4 Power and sample size

4.3.3 Comparing two odds estimates from

Case-Control data: Exposure odds ratio (EOR)

4.3.3.1 Estimation

4.3.3.2 Confidence intervals

Normal approximation
r
1 1 1 1
SE[log(EOR)] = + + + (4.18)
a b c d

Exact approximation

Exact methods

Compare
fisher.test vs.
fisher.exact (ex-
act2x2)
4.3.3.3 Comparison to reference value (hypothesis testing &
p-values) consider mid-
p method from
4.3.3.4 Power and sample size Rothman

Table 4.10 Case-control study, binomial data

Exposed Unexposed Exposure Odds
Cases a b a/b
Controls c d c/d

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

206 4 Analyzing simple epidemiologic data

Problems

4.1. Using the 2 × 2 table format displayed in Table 4.11, create and test a
function that takes four integers and calculate the risk ratio (RR) and the
odds ratio (OR).

a/(a + b)
RR =
c/(c + d)
ad
OR =
bc
4.2. Using the 2 × 2 table format displayed in Table 4.11, create a function
that takes a 2 × 2 matrix and calculates the risk ratio (RR) and odds ratio
(OR).

4.3. Create test a function for a measure of association that includes a 90%
confidence interval and p-value. Use input data from an actual data frame.

Table 4.11 Deaths among subjects who received tolbutamide and placebo in the Unver-
sity Group Diabetes Program (1970), stratifying by age
Disease No disease Total
Exposed a b a+b
Nonexposed c d c+d
Total a+c b+d a+b+c+d

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Chapter 5
Graphical display of epidemiologic
data

The graphical display of epidemiologic data plays a central role in descrip-

tive epidemiology, yet no epidemiology textbook dedicates adequate coverage
to the methods and techniques of graphing epidemiologic data with the sole
exception of Principles of Epidemiology: An Introduction to Applied Epi-
demiology and Biostatistics. This freely available book from the Centers for
Disease Control and Prevention has an excellent chapter on the tabular and
graphical display of epidemiologic data and we highly recommend it. Our
chapter parallels their chapter and demonstrates how to contruct their graph-
ical displays using R. Additionally, we cover how to use R for optimal colors
selection.
In this chapter we start by presenting the final graph we want to generate.
Then we provide selected R code for you to study, implement, and view in R.
Step-by-step you will implement R code until you have generated the final
graph. We repeat this process with subsequents graphs.
What you will learn is that the even the simplest appearing graphs can
be complicated to construct. That is why many graphs in epidemiology text-
books are not accompanied with the code that created them. We hope to
demystify the construction of high-quality graphical displays of epidemio-
logic data.

5.1 Graphs

5.1.1 Arithmetic-scale line graphs

In epidemiology, line graphs are commonly used to display the number of

disease cases or disease rates over calendar time. Displayed in Figure 5.1 on
the following page is the number of measles cases reported in the United
States from 1950 to 2001. Calendar time is on the x-axis and the number

207
208 5 Graphical display of epidemiologic data

Measles cases reported in United States, 1950−2001

1000
900
800
Reported Cases (x 1,000)

Vaccine
700 Licensed
600
500
400
300
200
100
0

1950 1960 1970 1980 1990 2000

Year

Fig. 5.1 Example of arithmetic line graph

of reported cases each year are on the y-axis. This type of line graph is also
called a time series plot. The number of reported cases and the y-axis scale
have not been transformed and are presented in their native scales; hence,
the term “arithmetic scale.”
The plot function was used to create Figure 5.1. Think of an arithmetic
line graph as a plotted series of {xi , yi } coordinates with a line drawn through
each point. In effect, we provide as arguments to the plot function two
numeric vectors: a vector of xi coordinates and a vector of yi coordinates.
For example, read in the measles data and draw a basic plot (Figure 5.2 on
the facing page).
md <- read.table("http://www.medepi.net/data/measles.txt",
header = TRUE, sep = "")
plot(md$year, md$cases)
What can we learn from studying this Figure 5.2 on the next page: (1)
the x and y axis labels come from the vector names provided as arguments
to the plot function, (2) the x and y axis scales are set by R defaults, (3) the
x-y coordinates are plotted as points using R’s default symbol, and (4) the

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

5.1 Graphs 209

md$cases

4 e+05
0 e+00

1950 1970 1990

md$year
Fig. 5.2 Example of basic default plot

shape of the figure is approximately square. Now try the following on your
own:
## change plot type
plot(md$year, md$cases, type = "l")
plot(md$year, md$cases, type = "b")

##add axis labels and main title

plot(md$year, md$cases, type = "l",
xlab = "Year", ylab = "Reported Cases (x 1000)",
main = "Measles cases reported in the United States, 1950-2001")

##suppress default axes, create new axes, add box frame

plot(md$year, md$cases, type = "l",
xlab = "Year", ylab = "Reported Cases (x 1000)",
main = "Measles cases reported in the United States, 1950-2001",
axes = FALSE)
axis(side = 1, at = seq(1950, 2000, 5), labels = seq(1950, 2000, 5))

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

210 5 Graphical display of epidemiologic data

axis(side = 2)
box()
In the plot function, the option type sets the type of line plot that should
be drawn (“p” = points, “l” = lines, “b” = both, etc.). The plot function has
many options which you can learn about by typing ?plot and ?plot.default
at the R prompt. Setting axes=FALSE removes all axes. The axis function
redraws an axis. The side option specifies which side to redraw (1 = bottom-
side, 2 = left-side, 3 = top-side, 4 = right-side). Setting axis(n ) alone re-
draws the original default axis for side n.
The following R code will reproduce Figure 5.1 on page 208 but with a
square plot area. Changing the plot area is left as an exercise.
plot(md$year, md$cases/1000, type = "l",
xlab = "Year", ylab = "Reported Cases (x 1000)",
main = "Measles cases reported in the United States, 1950-2001",
axes = FALSE, xlim = c(1950, 2001), ylim = c(0,1000),
xaxs = "i", yaxs = "i")
axis(1, at = md$year, labels = FALSE, tick = TRUE)
axis(1, at = seq(1950, 2000, 5), labels = seq(1950, 2000, 5),
tick = TRUE, tcl = -1)
axis(2, at = seq(0, 1000, 100), labels = seq(0, 1000, 100),
las = 2, tick = TRUE)
arrows(1963, 660, 1963, 400, length = 0.15, angle = 10)
text(1963, 700, "Vaccine\nLicensed")
box(bty="l")
The plot options xlim and ylim set the display range for the x-axis and
y-axis, respectively. By default, extra length is added to the range of the
x-axis and y-axis. The options xaxs="i" and yaxs="i" remove this extra
length (compare to previous plot to see difference). The first axis function
is used to set the x-axis tick marks at 1-year intervals. The second axis
function is used to add x-axis labels and to set longer x-axis tick marks at
5-year intervals. The option tcl changes the length. The third axis function
is used to add y-axis labels and tick marks. The las=2 option sets the labels
at 90 degrees to the y-axis. The arrows function adds an arrow starting at
{x0 = 1963, y0 = 660} and ending at {x1 = 1963, y1 = 400}. The text
function added “Vaccine Licensed” centered at {x = 1963, y = 700}. The
\n in Vaccine\nLicensed inserted a line break between the word “Vaccine”
and “Licensed”. Finally, the box option bty="l" sets the frame in the shape
of the letter “l”.
To get help for these function options, type ?box, ?axis, ?arrows, or
?text. If you cannot find a specific option, type ?par. The par function
controls many of the graphic parameter. In fact, it’s worth printing out the
par help pages and keeping it handy.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

5.1 Graphs 211

5.1.2 Semi-logarithmic-scale line graphs

What are the limitations of Figure 5.1 on page 208? From years 1959 to
2001, the number of measle cases ranged from a low of 86 cases to a high
of 763,000 cases. With a large number of cases, the the y-axis must contain
a larger number range. Before 1970, we can appreciate the trend in measles
cases; however, after 1970, we cannot visually appreciate the trend in measles
cases. One graphical solution is to take the natural logarithm of the number of
cases (y-axix values). This compresses the larger numbers and decompresses
the smaller numbers, making the measles trend after 1970 easier to visualize.
For example, try this:
plot(md$year, log(md$cases), type = "l")
A problem with this approach is that the y-axis values are difficult to in-
terpret. Although you can re-label the y-axis, it’s easier to just use the log
option. This code creates a semi-logarithmic plot.
plot(md$year, md$cases, type = "l", log = "y")
The following R code will reproduce the semi-logarithmic graph in Fig-
ure 5.3 on the following page but with a square plot area. Changing the plot
area is left as an exercise. This graph is similar to Figure 5.1 on page 208
except that we spend extra time building the y-axis labels.
plot(md$year, md$cases, type = "l", log = "y",
xlim = c(1950, 2001), ylim = c(1, 1100000),
xlab = "Year", ylab = "Reported cases",
main = "Measles cases reported in the United States, 1950-2001",
xaxs = "i", yaxs = "i", axes = FALSE)
axis(1, at = md$year, labels = FALSE, tick = TRUE)
axis(1, at = seq(1950, 2000, 5), labels = seq(1950, 2000, 5),
tick = TRUE, tcl = -1, cex.axis = cex)
axis(2, at = c(seq(1, 10, 1), seq(10, 100, 10),
seq(100, 1000, 100), seq(1000, 10000, 1000),
seq(10000, 100000, 10000), seq(100000, 1000000, 100000)),
labels = FALSE, tick = TRUE, cex.axis = cex)
axis(2, at = c(1, 5, 10, 50, 100, 500, 1000, 5000, 10000,
50000, 100000, 500000, 1000000), labels = c("1", "5",
"10", "50", "100", "500", "1,000", "5,000", "10,000",
"50,000", "100,000", "500,000", "1,000,000"),
las = 2, tick = TRUE, tcl = -0.7, cex.axis = cex)
arrows(1963, 10000, 1963, 300000, length = 0.15, angle = 10)
text(1963, 5000, "Vaccine Licensed (1963)", cex = cex)
box(bty = "l")
The trick to constructing these graphs is to build the code incrementally
and testing it often to debug errors along the way.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

212 5 Graphical display of epidemiologic data

Measles cases reported in the United States, 1950−2001

1,000,000
500,000

100,000
50,000
Reported cases

10,000
5,000 Vaccine Licensed (1963)

1,000
500

100
50

10
5

1950 1960 1970 1980 1990 2000

Year

Fig. 5.3 Example of semi-logarithmic-scale line graph

5.1.3 Histograms

A histogram displays the frequency distribution of continuous data.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

5.4 Miscellaneous 213

5.1.4 Frequency polygons

5.1.5 Cumulative frequency and survival curves

5.1.6 Scatter diagrams

5.2 Charts

5.2.1 Bar charts

5.2.2 Groups bar charts

5.2.3 Deviation bar charts

5.2.4 Proportional component bar charts

5.2.5 Pie charts

5.2.6 Dot plots and box plots

5.2.7 Maps

5.3 Color selection

5.4 Miscellaneous

locator identify

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Chapter 6
Control confounding with stratification
methods

In this chapter, we review two stratification methods to control for confound-

ing:
• Pooling methods
• Standardization methods

6.1 Pooling methods

6.1.1 Cohort studies with risk (binomial) data

6.1.1.1 Risk ratio

P ai N0i
i Ti
RRM H = P bi N1i
(6.1)
i Ti
P  M1i N1i N0i 
i − aTi bi i Ti2
Var[log(RRM H )] = P  P  (6.2)
ai N0i bi N1i
i Ti i Ti

rr.mh <- function(x, conf.level=0.95){

ai <- x[1,1,]; bi <- x[1,2,]

Table 6.1 Stratified tables for a cohort study with risk data
Exposure
Outcome Exposed Nonexposed Total
Cases ai bi M1i
Noncases ci di M0i
Total N1i N0i Ti

215
216 6 Control confounding with stratification methods

ci <- x[2,1,]; di <- x[2,2,]

N0i <- bi+di; N1i <- ai+ci
M0i <- ci+di; M1i <- ai+bi
Ti <- ai+bi+ci+di
RRmh <- sum(ai*N0i/Ti)/sum(bi*N1i/Ti)
num <- sum(M1i*N1i*N0i/Ti^2-ai*bi/Ti)
dem <- sum(ai*N0i/Ti)*sum(bi*N1i/Ti)
Z <- qnorm((1+conf.level)/2)
SElogRRmh <- sqrt(num/dem)
LL <- exp(log(RRmh)-Z*SElogRRmh)
UL <- exp(log(RRmh)+Z*SElogRRmh)
list(data=x, risk.ratio=RRmh, conf.int=c(LL,UL))
}

##test function
##read UGDP data
ud <- read.table("http://www.medepi.net/data/ugdp.txt",
header=T, sep=",")
str(ud)
ud[1:6,]
tab <- table(ud$Status,ud$Treatment,ud$Agegrp)[,2:1,2:1]
tab

rr.mh(tab)

6.1.1.2 Risk difference

P ai N0i −bi N1i
i Ti
RDM H = P N1i N0i
(6.3)
i Ti

P  N1i N0i 2 h ai d i bi c i
i
i Ti 2 (N −1)
N1i 1i
+ 2 (N −1)
N0i 0i
Var(RDM H ) = P 2 (6.4)
N1i N0i
i Ti

6.1.2 Cohort studies with incidence rate data

6.1.2.1 Rate ratio

P ai P T0i
i Ti
IRM H = P bi P T1i
(6.5)
i Ti

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

6.1 Pooling methods 217

Table 6.2 Stratified tables for a cohort study with incidence rate data
Exposure
Outcome Exposed Nonexposed Total
Cases ai bi Mi
Person-time at risk P T1i P T0i Ti

Table 6.3 Stratified tables for a case-control data

Exposure
Outcome Exposed Nonexposed Total
Cases ai bi M1i
Controls ci di M0i
Total N1i N0i Ti

P  Mi P T1i P T0i 2
i Ti
Var[log(IRM H )] = P  P  (6.6)
ai P T0i bi P T1i
i Ti i Ti

6.1.2.2 Rate difference

P ai P T0i −bi P T1i
i Ti
IDM H = P P T1i P T0i
(6.7)
i Ti

P  P T1i P T0i 2  ai bi

i Ti 2
P T1i
+ 2
P T0i
Var(IDM H ) = P 2 (6.8)
P T1i P T0i
i Ti

6.1.3 Case control studies

6.1.3.1 Odds ratio ratio

P ai d i
i Ti
ORM H = P bi ci
(6.9)
i Ti

P P P
Gi Pi
i i (Gi Q i + Hi P i ) i Hi Q i
Var[log(ORM H )] = 2 + P P + 2 (6.10)
2 ( i Gi i Hi )
P P
2 ( i Gi ) 2 ( i Hi )

where

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

218 6 Control confounding with stratification methods

a i di bi c i (ai + di ) (bi + ci )
Gi = , Hi = , Pi = , Gi =
Ti Ti Ti Ti

6.2 Standardization methods

6.2.1 Direct standardization

6.2.2 Indirect standardization

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Chapter 7
Control confounding with regression
methods

In this chapter we cover how to use regression models to control for confound-
ing for a continuous and binary outcome. We cover the following models:
• Continuous outcome
– Linear regression
• Binary outcome
– Unconditional logistic regression
– Conditional logistic regression
– Poisson regression
– Cox proportional hazards regression model
Our goal is not to teach statistical methods, but rather to briefly review se-
lected regression models and to implement them in R for conducting epidemi-
ologic analyses. Several excellent biostatistic books cover regression methods
for epidemiologists [9, 10, 11, 6].

7.0.3 Interpreting the regression coefficients

To interpret the regression coefficient, we must consider the following:

• What is the outcome measure y? (count, rate, odds, hazard)
• What is the presumed distribution of y? (binomial, Poission, etc.)
• What is the transformation of y, if any? (natural logarithm)
First, consider a continuous outcome measure y that is not transformed,
where y is normally distributed conditioned on a dichotomous predictor vari-
able x.
yi = a0 + a1 x i

219
220 7 Control confounding with regression methods

10 20 30 40 50 60 70 80 90
y

x2
50
40
30
20
10
0
0 10 20 30 40 50

x1

Fig. 7.1 A general linear model (y = a0 + a1 x1 + a2 x2 ) is fitted to a cloud of data points

(x1 , x2 , y). The model coefficients (a0 , a1 , a2 ) are the intercept and slopes of a fitted plane
that minimize the distance to the observed data. a0 is the y intercept when x1 = 0 and
x2 = 0, a1 is the line slope in the x1 -y plane, and a2 is the line slope in the x2 -y plane

For an nonexposed subject xi = 0 and for an exposed subject xi = 1. To

interpret the coefficient a1 take the difference of the equation when xi = 1
(exposed) and the equation when xi = 0 (nonexposed).

ye = a0 + a1 (1)
−[yu = a0 + a1 (0)]

(ye − yu ) = (a0 − a0 ) + (a1 − 0)

ye − yu = a1

Therefore, for this linear regression equation, the coefficient a1 is equal to the
change in y (or ye − yu ) for a unit change in x (or 1 − 0). A regression model
can incorporate multiple variables where each coefficient is now adjusted for
the presence of other variables. For example, in Figure 7.1, the coefficient
a1 represents the change y for a unit change in x1 , adjusted for x2 ; and the

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

7 Control confounding with regression methods 221

Fig. 7.2 In epidemiology, we often use two types of regression models. The dependent
variable, y, can undergo no transformation before fitting a regression model (Approach
A), or can undergo a transformation (e.g., natural logarithm) before fitting a regression
model. For example, the measure y (e.g., counts or rates), the assumed distribution of y
(e.g., Poisson), and the transformation of y (e.g., natural logirithm) can come together as
a Poisson regression model.

coefficient a2 represents the change in y for a unit change in x2 , adjusted for

x1 .
Second, consider instead y to represent an outcome of an unspecified form.
We can take the natural logarithm of y first, and repeat the same analysis.
For an unexposed subject x = 0 and for an exposed subject x = 1.

log(ye ) = a0 + a1 (1)

Again, to interpret the coefficient a1 take the difference of the equation when
x = 1 and the equation when x = 0.

−[log(yu ) = a0 + a1 (0)]

(log(ye ) − log(yu )) = (a0 − a0 ) + (a1 − 0)

ye
log( ) = a1
yu
ye
= ea 1
yu

Therefore, now the coefficient a1 is equal to the change in the log(y) (or
log(ye ) − log(yu )) for a unit change in x (or 1 − 0). Equivalently, a1 is equal
to the log(y1 /y0 ), or ea1 = y1 /y0 .
In multivariable regression, the general formula becomes

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

222 7 Control confounding with regression methods

yi = a0 + a1 xi1 + a2 xi2 + . . . + ak xjk

where yi is fitted using an additive model. In contrast, when we transform yi

using the natural logarithm we get:

log(yi ) = a0 + a1 xi1 + a2 xi2 + . . . + ak xjk (7.1)

yi = exp(a0 + a1 xi1 + a2 xi2 + . . . + ak xjk )
yi = ea0 ea1 xi1 ea2 xi2 . . . eak xjk

where yi is fitted using a multiplicative model.

Third, in addition to being a continuous measure, y can be a count, rate,
or odds. When y is an odds, then the log(y) is the link function for logistic
regression (also called logit). When y is a count or a rate, then the log(y) is
the link function for Poisson regression.
The generalized linear model (GLM) is a flexible generalization of ordinary
linear regression and, by specifying a link function, we are able to model
linear, logistic, and Poisson regression. In R, we use the glm function which
has the following general syntax:
glm(outcome_variable ~ covariable1 + covariable2 ,
family = link_function , data = data_frame )
where outcome variable, covariable1, and covariable2 are fields in data frame,
and link function can be one of the following:
gaussian(link = "identity")
binomial(link = "logit")
poisson(link = "log")
for linear, logistic, and Poisson regression, respectfully. We see applications
in the sections that follow.

7.1 Linear regression

For a continuous outcome measure, we use multiple linear regression to con-

trol for confounding. In R, we can use the lm or glm function. The lm func-
tion fits a line to the data using the linear least squares method, and the glm
function uses the iteratively reweighted least squares (IWLS) method. For
our purposes, we will use the glm function.
Consider the data collected on Latina mothers and their newborn infants
(See Appendix A.1 on page 243). Table 7.1 on the facing page summarizes
the data collected.
In the published paper [12, 13], the authors used the natural logarithm
of the birthweight as the dependent variable, making the influence of the
independent variables multiplicative on the birthweight. When specifying the

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

7.1 Linear regression 223

family option, it is not necessary to specify the link; i.e., we accept the
default link function. Here is the multiple linear regression analysis:
> ## Latina mothers birthweight of newborns
> bdat <- read.table("http://www.medepi.net/data/birthwt9.txt",
+ header =TRUE, sep = "")
> mod1 = glm(log(bwt) ~ age + parity + gest + sex + cigs + ht + wt,
+ family = gaussian, data = bdat)
> mod1

Call: glm(formula = log(bwt) ~ age + parity + gest + sex + cigs + ht +

wt, family = gaussian, data = bdat)

Coefficients:
(Intercept) age parity gest sex
6.1270855 0.0023539 0.0089355 0.0067584 -0.0326892
cigs ht wt
-0.0030201 0.0000147 0.0020536

Degrees of Freedom: 426 Total (i.e. Null); 419 Residual

Null Deviance: 9.156
Residual Deviance: 5.974 AIC: -593.2
Notice that the default display provides limited information. Therefore,
we use the summary function:
> summary(mod1)

Call:
glm(formula = log(bwt) ~ age + parity + gest + sex + cigs + ht +

Table 7.1 Data dictionary for Latina mothers and their newborn infants
Variable Description Possible values
age Maternal age In years (self-reported)
parity Parity Count of previous live births
gest Gestation Reported in days
sex Gender Male = 1, Female = 2
bwt Birth weight Grams
cigs Smoking Number of cigarettes per day
(self-reported)
ht Maternal height Measured in centimeters
wt Maternal weight Pre-pregnancy weight
(self-reported)
r1 Rate of weight gain (1st trimester) Kilograms per day (estimated)
r2 Rate of weight gain (2nd trimester) Kilograms per day (estimated)
r2 Rate of weight gain (3rd trimester) Kilograms per day (estimated)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

224 7 Control confounding with regression methods

wt, family = gaussian, data = bdat)

Deviance Residuals:
Min 1Q Median 3Q Max
-0.39748 -0.07429 0.00522 0.08219 0.39097

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 6.1270855 0.2157182 28.403 < 2e-16 ***
age 0.0023539 0.0011364 2.071 0.03893 *
parity 0.0089355 0.0062630 1.427 0.15441
gest 0.0067584 0.0005187 13.030 < 2e-16 ***
sex -0.0326892 0.0116243 -2.812 0.00515 **
cigs -0.0030201 0.0017457 -1.730 0.08437 .
ht 0.0000147 0.0009819 0.015 0.98806
wt 0.0020536 0.0006236 3.293 0.00107 **
---
Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1

(Dispersion parameter for gaussian family taken to be 0.01425768)

Null deviance: 9.1562 on 426 degrees of freedom

Residual deviance: 5.9740 on 419 degrees of freedom
AIC: -593.25

Number of Fisher Scoring iterations: 2

7.2 Logistic regression

In epidemiology we use unconditional logistic regression to model binary out-

comes with a binomial distribution. The most common study designs are the
following:
• Cross-sectional study with binomial data (1 = event, 0 = no event)
• Cohort study with binomial data (1 = event, 0 = no event)
• Cohort study with no censoring1 (at the end of the study period, except for
subjects that had the event, the remaining subjects had equal observation
times)
• Case-control study with frequency matching
For case-control studies with pairwise (1-to-n) matching we use conditional
logistic regression (see Section 7.2.2 on page 231). For cohort studies with
1 Censoring is when the event of interest is not observed due to the end of the study, loss
to follow up, study withdrawal, or a competing risk (e.g., death from another cause)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

7.2 Logistic regression 225

risk odds = R/(1−R) log(risk odds) = logit

log(R/(1−R))
4

4
R/(1−R)

0
−4

−4
0.0 0.4 0.8 0.0 0.4 0.8

R R

Fig. 7.3 The logit transformation is a double transformation. First, the odds tranfor-
mation (R/(1 − R)) unbounds the probabilities near 1; second, the logit transformation
(log(odds)) additionally unbounds the probabilities near 0.

with individual obervation times and censoring we use survival analysis (see
Cox proportional hazards regression in Section 7.5 on page 239).
We recall that the logit, or log-odds, is used so that we can map proba-
bilities values, that are between 0 and 1, to values between negative infinity
and positive infinity (see Figure 7.3). This is useful for very low or very high
probabilities that will have asymmetric distributions.
Consider a binary outcome measure Ri with a binomial distribution, where
the log-odds (or logit) is conditioned on a dichotomous predictor variable xi ,
and is represented with this formula:
Ri
log( ) = a0 + a1 x i (7.2)
1 − Ri
logit = a0 + a1 xi (7.3)

For a nonexposed subject xi = 0 and for an exposed subject xi = 1. To

interpret the coefficient a1 take the difference of the equation when xi = 1
(exposed) and the equation when xi = 0 (nonexposed).

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

226 7 Control confounding with regression methods

log(R1 /(1 − R1 )) = a0 + a1 (1)

−[log(R0 /(1 − R0 )) = a0 + a1 (0)]

(log[R1 /(1 − R1 )] − log[R0 /(1 − R0 )]) = (a0 − a0 ) + (a1 − 0)

 
R1 /(1 − R1 )
log = a1
R0 /(1 − R0 )
R1 /(1 − R1 )
= ea 1
R0 /(1 − R0 )
odds ratio = ea1

Therefore, for this logistic regression equation, the coefficient a1 is equal to

the change in the log-odds (or logit) for a unit change in x (or 1 − 0). Also
notice that exp(a1 ) is the odds ratios.
A multiple logistic regression model can incorporate multiple variables
where each coefficient is now adjusted for the remaining covariables in the
model. In multivariable regression, the general formula becomes

log(odds) = a0 + a1 xi1 + a2 xi2 + . . . + ak xjk (7.4)

odds = exp(a0 + a1 xi1 + a2 xi2 + . . . + ak xjk ) (7.5)
odds = ea0 ea1 xi1 ea2 xi2 . . . eak xjk (7.6)

where the log-odds is fitted modeling an additive relationship (or the odds
is fitted modeling a multiplicative relationship). The exp(ai ) is an adjusted
odds ratio for the ith covariable, adjusted for the remaining covariables.

7.2.1 Unconditional logistic regression

Unconditional logistic regression is used for binomial data. By binomial

we mean that each observation is considered an independent and identical
Bernoulli trial with a binary outcome, usually 1 for event, and 0 for nonevent.
This is usually a cohort, case-control, or cross-sectional study.

7.2.1.1 Individual-level data

Consider the Altman hypertension individual-level data set which consists of

433 observations and three dichotomous covariables: smoking, obesity, and
snoring [14].
> adat = read.table("http://www.medepi.net/data/altman.txt",
+ header = TRUE, sep = "")
> str(adat)
’data.frame’: 433 obs. of 4 variables:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

7.2 Logistic regression 227

$ smoking: Factor w/ 2 levels "No","Yes": 1 1 1 1 1 2 2 1 1 1 ...

$ obesity: Factor w/ 2 levels "No","Yes": 1 1 1 1 1 1 1 2 1 1 ...
$ snoring: Factor w/ 2 levels "No","Yes": 1 1 1 1 1 1 1 1 2 2 ...
$ htn : Factor w/ 2 levels "No","Yes": 2 2 2 2 2 2 2 2 2 2 ...
> ftable(adat)
htn No Yes
smoking obesity snoring
No No No 55 5
Yes 152 35
Yes No 7 1
Yes 36 15
Yes No No 15 2
Yes 72 13
Yes No 2 0
Yes 15 8
Here is the multiple logistic regression model:
> mod1 = glm(htn ~ smoking + obesity + snoring, family = binomial, data = adat)
> mod1

Call: glm(formula = htn ~ smoking + obesity + snoring, family = binomial,

data = adat)

Coefficients:
(Intercept) smokingYes obesityYes snoringYes
-2.37766 -0.06777 0.69531 0.87194

Degrees of Freedom: 432 Total (i.e. Null); 429 Residual

Null Deviance: 411.4
Residual Deviance: 398.9 AIC: 406.9

> summary(mod1)

Call:
glm(formula = htn ~ smoking + obesity + snoring, family = binomial,
data = adat)

Deviance Residuals:
Min 1Q Median 3Q Max
-0.8578 -0.6330 -0.6138 -0.4212 2.2488

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -2.37766 0.38010 -6.255 3.97e-10 ***
smokingYes -0.06777 0.27812 -0.244 0.8075

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

228 7 Control confounding with regression methods

obesityYes 0.69531 0.28508 2.439 0.0147 *

snoringYes 0.87194 0.39749 2.194 0.0283 *
---
Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1

(Dispersion parameter for binomial family taken to be 1)

Null deviance: 411.42 on 432 degrees of freedom

Residual deviance: 398.92 on 429 degrees of freedom
AIC: 406.92

Number of Fisher Scoring iterations: 4

Notice that the summary function provides more detailed results; however,
odds ratios and confidence intervals are not provided. We can fix this by
providing our own function.
The following or.glm function takes as an argument the logistic regression
model object.
or.glm <- function(x, conf.level = 0.95){
## x = glm logistic regression model object
sumx <- summary(x)$coefficients
Z <- qnorm((1 + conf.level)/2)
expor <- exp(sumx[,"Estimate"])
lcl <- exp(sumx[,"Estimate"] - Z*sumx[,"Std. Error"])
ucl <- exp(sumx[,"Estimate"] + Z*sumx[,"Std. Error"])
coef <- sumx
or <- cbind(coef=sumx[,"Estimate"], "exp(coef)"=expor, lcl, ucl)
list(coef = coef, odds.ratio = or)
}
And here is the or.glm function in action:
> or.glm(mod1)
$coef
Estimate Std. Error z value Pr(>|z|)
(Intercept) -2.37766133 0.3800998 -6.2553609 3.965976e-10
smokingYes -0.06777489 0.2781181 -0.2436911 8.074701e-01
obesityYes 0.69530959 0.2850801 2.4389972 1.472808e-02
snoringYes 0.87193919 0.3974935 2.1935938 2.826462e-02

$odds.ratio
coef exp(coef) lcl ucl
(Intercept) -2.37766133 0.09276728 0.04404061 0.1954053
smokingYes -0.06777489 0.93447081 0.54179029 1.6117596
obesityYes 0.69530959 2.00432951 1.14632691 3.5045298
snoringYes 0.87193919 2.39154401 1.09731504 5.2122522

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

7.2 Logistic regression 229

7.2.1.2 Group-level data

Group-level data can come in two forms. Here is the first form:
> adat2
smoking obesity snoring htn Freq
1 No No No No 55
2 Yes No No No 15
3 No Yes No No 7
4 Yes Yes No No 2
5 No No Yes No 152
6 Yes No Yes No 72
7 No Yes Yes No 36
8 Yes Yes Yes No 15
9 No No No Yes 5
10 Yes No No Yes 2
11 No Yes No Yes 1
12 Yes Yes No Yes 0
13 No No Yes Yes 35
14 Yes No Yes Yes 13
15 No Yes Yes Yes 15
16 Yes Yes Yes Yes 8
This data frame provides every covariable pattern and frequency of the pat-
tern.2 To model this data we use the glm function weight option:
> mod2 = glm(htn ~ smoking + obesity + snoring, weight = Freq,
+ family = binomial, data = adat2)
> or.glm(mod2)
$coef
Estimate Std. Error z value Pr(>|z|)
(Intercept) -2.37766146 0.3801835 -6.2539830 4.001145e-10
smokingYes -0.06777489 0.2781237 -0.2436861 8.074739e-01
obesityYes 0.69530960 0.2850849 2.4389559 1.472976e-02
snoringYes 0.87193932 0.3975727 2.1931569 2.829607e-02

$odds.ratio
coef exp(coef) lcl ucl
(Intercept) -2.37766146 0.09276726 0.04403337 0.1954374
smokingYes -0.06777489 0.93447081 0.54178428 1.6117775
obesityYes 0.69530960 2.00432951 1.14631607 3.5045629
snoringYes 0.87193932 2.39154432 1.09714478 5.2130624
Here is an alternative form for group-level data:
> adat3

2 Rcode: adat2 = data.frame(as.table(ftable(adat)))

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

 230 7 Control confounding with regression methods

smoking obesity snoring htn.yes htn.no

1 No No No 5 55
2 Yes No No 2 15
3 No Yes No 1 7
4 Yes Yes No 0 2
5 No No Yes 35 152
6 Yes No Yes 13 72
7 No Yes Yes 15 36
8 Yes Yes Yes 8 15
To model this we must create an outcome matrix with the outcome counts.
> htn.yes.no = as.matrix(adat3[, 4:5])
> htn.yes.no
htn.yes htn.no
[1,] 5 55
[2,] 2 15
[3,] 1 7
[4,] 0 2
[5,] 35 152
[6,] 13 72
[7,] 15 36
[8,] 8 15
We can use this outcome matrix with the glm function.
> mod3 = glm(htn.yes.no ~ smoking + obesity + snoring,
+ family = binomial, data = adat3)
> or.glm(mod3)
$coef
Estimate Std. Error z value Pr(>|z|)
(Intercept) -2.37766146 0.3801845 -6.2539671 4.001553e-10
smokingYes -0.06777489 0.2781242 -0.2436857 8.074742e-01
obesityYes 0.69530960 0.2850851 2.4389544 1.472983e-02
snoringYes 0.87193932 0.3975736 2.1931517 2.829645e-02

$odds.ratio
coef exp(coef) lcl ucl
(Intercept) -2.37766146 0.09276726 0.04403329 0.1954377
smokingYes -0.06777489 0.93447081 0.54178381 1.6117789
obesityYes 0.69530960 2.00432951 1.14631567 3.5045641
snoringYes 0.87193932 2.39154432 1.09714274 5.2130721
To summarize, we showed how to use the glm function to conduct uncon-
ditional logistic regression for individual-level data and two forms of group-
level data. We also created the or.glm function to calculate odds ratios and
confidence intervals.
HERE 2012-11-25

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

7.2 Logistic regression 231

7.2.2 Conditional logistic regression

> mdat <- read.table("http://www.medepi.net/data/mi.txt",sep="")

> names(mdat) <- c("match","person","mi","smk","sbp","ecg")
> mdat$mi <- factor(mdat$mi, levels=0:1, labels=c("No","Yes"))
> mdat$smk <- factor(mdat$smk, levels=0:1, labels=c("Not current","Current"))
> mdat$ecg <- factor(mdat$ecg, levels=0:1, labels=c("Normal","Abnormal"))
> str(mdat)
’data.frame’: 117 obs. of 6 variables:
$ match : int 1 1 1 2 2 2 3 3 3 4 ...
$ person: int 1 2 3 4 5 6 7 8 9 10 ...
$ mi : Factor w/ 2 levels "No","Yes": 2 1 1 2 1 1 2 1 1 2 ...
$ smk : Factor w/ 2 levels "Not current",..: 1 1 1 1 1 1 1 1 1 1 ...
$ sbp : int 160 140 120 160 140 120 160 140 120 160 ...
$ ecg : Factor w/ 2 levels "Normal","Abnormal": 2 1 1 2 1 1 1 1 1 1 ...
>
> library(survival)
> clog1 <- clogit(I(mi=="Yes")~smk+strata(match),data=mdat)
> summary(clog1)
Call:
coxph(formula = Surv(rep(1, 117L), I(mi == "Yes")) ~ smk + strata(match),
data = mdat, method = "exact")

n= 117, number of events= 39

coef exp(coef) se(coef) z Pr(>|z|)

smkCurrent 0.8434 2.3242 0.4661 1.809 0.0704 .
---
Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1

exp(coef) exp(-coef) lower .95 upper .95

smkCurrent 2.324 0.4303 0.9322 5.794

Rsquare= 0.028 (max possible= 0.519 )

Likelihood ratio test= 3.37 on 1 df, p=0.06655
Wald test = 3.27 on 1 df, p=0.07038
Score (logrank) test = 3.43 on 1 df, p=0.06408

> clog2 <- clogit(I(mi=="Yes")~cut(sbp,2)+strata(match),data=mdat)

> summary(clog2)
Call:
coxph(formula = Surv(rep(1, 117L), I(mi == "Yes")) ~ cut(sbp,
2) + strata(match), data = mdat, method = "exact")

n= 117, number of events= 39

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

232 7 Control confounding with regression methods

coef exp(coef) se(coef) z Pr(>|z|)

cut(sbp, 2)(140,160] 2.0369 7.6667 0.4584 4.443 8.86e-06

cut(sbp, 2)(140,160] ***

---
Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1

exp(coef) exp(-coef) lower .95 upper .95

cut(sbp, 2)(140,160] 7.667 0.1304 3.122 18.83

Rsquare= 0.198 (max possible= 0.519 )

Likelihood ratio test= 25.83 on 1 df, p=3.724e-07
Wald test = 19.74 on 1 df, p=8.859e-06
Score (logrank) test = 27.59 on 1 df, p=1.502e-07

> clog3 <- clogit(I(mi=="Yes")~ecg+strata(match),data=mdat)

> summary(clog3)
Call:
coxph(formula = Surv(rep(1, 117L), I(mi == "Yes")) ~ ecg + strata(match),
data = mdat, method = "exact")

n= 117, number of events= 39

coef exp(coef) se(coef) z Pr(>|z|)

ecgAbnormal 2.0649 7.8842 0.7855 2.629 0.00857 **
---
Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1

exp(coef) exp(-coef) lower .95 upper .95

ecgAbnormal 7.884 0.1268 1.691 36.76

Rsquare= 0.078 (max possible= 0.519 )

Likelihood ratio test= 9.51 on 1 df, p=0.002043
Wald test = 6.91 on 1 df, p=0.008568
Score (logrank) test = 9.38 on 1 df, p=0.0022

> clog4 <- clogit(I(mi=="Yes")~smk+sbp+ecg+strata(match),data=mdat)

> summary(clog4)
Call:
coxph(formula = Surv(rep(1, 117L), I(mi == "Yes")) ~ smk + sbp +
ecg + strata(match), data = mdat, method = "exact")

n= 117, number of events= 39

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

7.2 Logistic regression 233

coef exp(coef) se(coef) z Pr(>|z|)

smkCurrent 0.72906 2.07313 0.56126 1.299 0.19395
sbp 0.04564 1.04670 0.01525 2.994 0.00276 **
ecgAbnormal 1.59926 4.94938 0.85341 1.874 0.06094 .
---
Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1

exp(coef) exp(-coef) lower .95 upper .95

smkCurrent 2.073 0.4824 0.6901 6.228
sbp 1.047 0.9554 1.0159 1.078
ecgAbnormal 4.949 0.2020 0.9292 26.362

Rsquare= 0.173 (max possible= 0.519 )

Likelihood ratio test= 22.2 on 3 df, p=5.925e-05
Wald test = 13.68 on 3 df, p=0.003382
Score (logrank) test = 19.68 on 3 df, p=0.0001979

> clog5 <- clogit(I(mi=="Yes")~smk+cut(sbp,2)+ecg+strata(match),data=mdat)

> summary(clog5)
Call:
coxph(formula = Surv(rep(1, 117L), I(mi == "Yes")) ~ smk + cut(sbp,
2) + ecg + strata(match), data = mdat, method = "exact")

n= 117, number of events= 39

coef exp(coef) se(coef) z Pr(>|z|)

smkCurrent 0.6450 1.9061 0.6091 1.059 0.2896
cut(sbp, 2)(140,160] 1.9789 7.2345 0.5018 3.944 8.03e-05
ecgAbnormal 1.9467 7.0055 0.9918 1.963 0.0497

smkCurrent
cut(sbp, 2)(140,160] ***
ecgAbnormal *
---
Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1

exp(coef) exp(-coef) lower .95 upper .95

smkCurrent 1.906 0.5246 0.5776 6.29
cut(sbp, 2)(140,160] 7.234 0.1382 2.7057 19.34
ecgAbnormal 7.005 0.1427 1.0028 48.94

Rsquare= 0.237 (max possible= 0.519 )

Likelihood ratio test= 31.72 on 3 df, p=5.986e-07
Wald test = 17.42 on 3 df, p=0.00058
Score (logrank) test = 30.97 on 3 df, p=8.613e-07

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

234 7 Control confounding with regression methods

7.3 Poisson regression

log(rate) = a0 + a1 x

log(r1 ) = a0 + a1 (1)
−[log(r0 ) = a0 + a1 (0)]

log(r1 ) − log(r0 ) = (a0 − a0 ) + (a1 − 0)

 
r1
log = a1
r0
r1
= ea 1
r0
For fitting the Poisson regression model, the following relationship is im-
portant:
count
log( ) = a0 + a1 x
person-time
log(count) − log(person-time) = a0 + a1 x
log(count) = a0 + a1 x + log(person-time)

where log(person-years) is called the offset. The offset is considered a fixed

quantity and is not fit in the statistical model; however, we will need to
specify this in the Poisson model in R. We’ll see this in the examples that
follow.

7.4 Indirect standardization using Poisson regression

The direct standardization of rates involves using weights from a standard

population. In contrast, the indirect standardization of rates involves using
the crude and age-specific rates from a standard population. The calculation
involves two components:
1. Calculate the standardized incidence ratio (SIR) (also called standardized
mortality ratio [SMR]).
2. Multiply the SIR and the standard crude rate

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

7.4 Indirect standardization using Poisson regression 235

S
ISR = (SIR)(Rcrude )
 
Observed S
= Rcrude
Expected
 P 
Ai S
= P i S
Rcrude
i P T R
i i

In this example, we use the dataset of U.S. white male population estimates
and male cancer deaths in 1940 compared to 1960.
> #enter data
> dth60 <- c(141, 926, 1253, 1080, 1869, 4891, 14956, 30888,
+ 41725, 26501, 5928)
> pop60 <- c(1784033, 7065148, 15658730, 10482916, 9939972,
+ 10563872, 9114202, 6850263, 4702482, 1874619,
+ 330915)
> dth40 <- c(45, 201, 320, 670, 1126, 3160, 9723, 17935,
+ 22179, 13461, 2238)
> pop40 <- c(906897, 3794573, 10003544, 10629526, 9465330,
+ 8249558, 7294330, 5022499, 2920220, 1019504,
+ 142532)
> #housekeeping
> tab <- cbind(pop40, dth40, pop60, dth60)
> agelabs <- c("<1", "1-4", "5-14", "15-24", "25-34",
+ "35-44", "45-54", "55-64", "65-74", "75-84",
+ "85+")
> rownames(tab) <- agelabs
> #calculate crude rates
> CDR.1940 <- sum(dth40)/sum(pop40)
> CDR.1960 <- sum(dth60)/sum(pop60)
> #display data and crude rates
> tab
pop40 dth40 pop60 dth60
<1 906897 45 1784033 141
1-4 3794573 201 7065148 926
5-14 10003544 320 15658730 1253
15-24 10629526 670 10482916 1080
25-34 9465330 1126 9939972 1869
35-44 8249558 3160 10563872 4891
45-54 7294330 9723 9114202 14956
55-64 5022499 17935 6850263 30888
65-74 2920220 22179 4702482 41725
75-84 1019504 13461 1874619 26501
85+ 142532 2238 330915 5928
> round(100000*c(CDR.1940=CDR.1940, CDR.1960=CDR.1960), 1)
CDR.1940 CDR.1960

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

236 7 Control confounding with regression methods

119.5 166.1
Now, we calculate the indirect standardized rate for 1940 using the 1960
data as the standard. Using the data we prepared above, we do the calcula-
tions in two steps:
> # sir calculation
> Ri.1960 <- dth60/pop60
> CDR.1960 <- sum(dth60)/sum(pop60)
> SIR.1940 <- sum(dth40)/sum(pop40*Ri.1960)
> ISR.1940 <- SIR.1940*CDR.1960
> c(SIR.1940=SIR.1940, ISR.1940=100000*ISR.1940,
+ CDR.1960=100000*CDR.1960)
SIR.1940 ISR.1940 CDR.1960
0.8305351 137.9414537 166.0874444
Now, to use Poisson regression we need to prepare our male cancer death
data as a data frame, and set 1960 as the standard (reference) population.
> #create data frame for poisson regression
> year <- factor(c(rep(1960, 11), rep(1940, 11)),
+ levels = c(1960, 1940)) #1960 as reference
> pop <- c(pop60, pop40)
> deaths <- c(dth60, dth40)
> age <- factor(rep(agelabs, 2), levels=agelabs)
> cad <- data.frame(year, pop, deaths, age)
> cad
year pop deaths age
1 1960 1784033 141 <1
2 1960 7065148 926 1-4
3 1960 15658730 1253 5-14
...
20 1940 2920220 22179 65-74
21 1940 1019504 13461 75-84
22 1940 142532 2238 85+
We are now in the position to use the Poisson regression model to estimate
the crude death rates for the 1940 and 1960 data.
> pmod <- glm(deaths~year, family = poisson(link="log"), data = cad,
+ offset = log(pop))
> summary(pmod)

Call:
glm(formula = deaths ~ year, family = poisson(link = "log"),
data = cad, offset = log(pop))

Deviance Residuals:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

7.4 Indirect standardization using Poisson regression 237

Min 1Q Median 3Q Max

-204.71 -128.34 -55.75 130.06 268.34

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -6.400411 0.002772 -2309.11 <2e-16 ***
year1940 -0.328958 0.004664 -70.53 <2e-16 ***
---
Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1

(Dispersion parameter for poisson family taken to be 1)

Null deviance: 487587 on 21 degrees of freedom

Residual deviance: 482471 on 20 degrees of freedom
AIC: 482690

Number of Fisher Scoring iterations: 6

> pmod$coef
(Intercept) year1940
-6.4004110 -0.3289584
> cdr.1940 <- exp(pmod$coeff[1] + pmod$coeff[2])
> cdr.1940*100000 #from model
(Intercept)
119.5286
> CDR.1940*100000 #from arithmetic
[1] 119.5286
> cdr.1960 <- exp(pmod$coeff[1])
> cdr.1960*100000 #from model
(Intercept)
166.0874
> CDR.1960*100000 #from arithmetic
[1] 166.0874
We now use the Poisson model to calculate an adjusted rate ratio compar-
ing the 1940 population to the 1960 population (standard). This rate ratio
closely approximates the SIR that, when multiplied by the 1960 crude rate,
gives us the indirect standardized rate.
> pmod2 <- glm(deaths ~ year + age, family= poisson(link="log"),
+ data = cad, offset = log(pop))
> summary(pmod2)

Call:
glm(formula = deaths ~ year + age, family = poisson(link = "log"),
data = cad, offset = log(pop))

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

238 7 Control confounding with regression methods

Deviance Residuals:
Min 1Q Median 3Q Max
-10.51100 -3.69907 -0.02425 3.15316 8.81875

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -9.521870 0.073337 -129.838 < 2e-16 ***
year1940 -0.182227 0.004679 -38.945 < 2e-16 ***
age1-4 0.408585 0.079143 5.163 2.44e-07 ***
age5-14 -0.110780 0.077538 -1.429 0.15308
age15-24 0.211467 0.077125 2.742 0.00611 **
age25-34 0.830261 0.075569 10.987 < 2e-16 ***
age35-44 1.841193 0.074167 24.825 < 2e-16 ***
age45-54 3.099207 0.073601 42.108 < 2e-16 ***
age55-64 4.101147 0.073464 55.825 < 2e-16 ***
age65-74 4.806312 0.073430 65.454 < 2e-16 ***
age75-84 5.299837 0.073494 72.112 < 2e-16 ***
age85+ 5.513262 0.074154 74.349 < 2e-16 ***
---
Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1

(Dispersion parameter for poisson family taken to be 1)

Null deviance: 487587.25 on 21 degrees of freedom

Residual deviance: 517.29 on 10 degrees of freedom
AIC: 755.8

Number of Fisher Scoring iterations: 4

> pmod2$coef
(Intercept) year1940 age1-4 age5-14 age15-24
-9.5218697 -0.1822271 0.4085846 -0.1107800 0.2114672
age25-34 age35-44 age45-54 age55-64 age65-74
0.8302607 1.8411928 3.0992069 4.1011465 4.8063119
age75-84 age85+
5.2998370 5.5132617
> exp(pmod2$coeff[2]) #SIR from Poisson model
year1940
0.833412
> SIR.1940 #SIR from arithmetic
[1] 0.8305351
Poisson regression seems like a lot of work. However, an advantage is that
we have standard errors estimates for constructing confidence intervals.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

7.5 Cox proportional hazards regression 239

> cdr.1960 #crude death rate for 1960 from Poisson model
(Intercept)
0.001660874
> conf.level <- 0.95
> Z <- qnorm((1+conf.level)/2)
> sy <- summary(pmod2)$coef #this includes standard errors
> log.sir.1940 <- sy["year1940","Estimate"]
> sir.1940 <- exp(log.sir.1940)
> SE.log.sir.1940 <- sy["year1940","Std. Error"]
> LCL.sir.1940 <- exp(log.sir.1940 - Z*SE.log.sir.1940)
> UCL.sir.1940 <- exp(log.sir.1940 + Z*SE.log.sir.1940)
> isr.1940 <- cdr.1960*c(sir.1940, LCL.sir.1940, UCL.sir.1940)
> names(isr.1940) <- c("ISR","LCL","UCL")
> round(isr.1940*100000, 1)
ISR LCL UCL
138.4 137.2 139.7
Recall that the previously calculated ISR1940 = 137.9 per 100,000 person-
years.

7.5 Cox proportional hazards regression

library(survival)
pbc2 <- pbc
pbc2[pbc==-9] <- NA
str(pbc2)

cmod1 <- coxph(Surv(time, status)~ascites, data=pbc2)

summary(cmod1)
plot(survfit(cmod1))

cmod2 <- coxph(Surv(time, status)~ascites+sex, data=pbc2)

summary(cmod2)
plot(survfit(cmod2))

sfit1 <- survfit(Surv(time, status)~ascites, data=pbc2)

plot(sfit1)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

 Appendixes

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Appendix A
Available data sets

A.1 Latina Mothers and their Newborn

From 1980 to 1990 data was collected on 427 Latino mothers that gave birth
at the University of California, San Francisco [12, 13]. Data was collected
on the characteristics of the mothers and their newborn infants (Table A.1).
Mothers were weighed at each prenatal visit. Rate of weight gain during each
trimester was based on a linear regression interpolation. The data set can
be viewed and downloaded from http://www.medepi.net/data/birthwt9.
txt.

Table A.1 Data dictionary for Latina mothers and their newborn infants
Variable Description Possible values
age Maternal age In years (self-reported)
parity Parity Count of previous live births
gest Gestation Reported in days
sex Gender Male = 1, Female = 2
bwt Birth weight Grams
cigs Smoking Number of cigarettes per day
(self-reported)
ht Maternal height Measured in centimeters
wt Maternal weight Pre-pregnancy weight
(self-reported)
r1 Rate of weight gain (1st trimester) Kilograms per day (estimated)
r2 Rate of weight gain (2nd trimester) Kilograms per day (estimated)
r2 Rate of weight gain (3rd trimester) Kilograms per day (estimated)

243
244 A Available data sets

A.2 Oswego County (outbreak)

On April 19, 1940, the local health officer in the village of Lycoming, Oswego
County, New York, reported the occurrence of an outbreak of acute gastroin-
testinal illness to the District Health Officer in Syracuse. Dr. A. M. Rubin,
epidemiologist-in-training, was assigned to conduct an investigation.
When Dr. Rubin arrived in the field, he learned from the health officer
that all persons known to be ill had attended a church supper held on the
previous evening, April 18. Family members who did not attend the church
supper did not become ill. Accordingly, Dr. Rubin focused the investigation
on the supper. He completed Interviews with 75 of the 80 persons known to
have attended, collecting information about the occurrence and time of onset
of symptoms, and foods consumed. Of the 75 persons interviewed, 46 persons
reported gastrointestinal illness.
The onset of illness in all cases was acute, characterized chiefly by nau-
sea, vomiting, diarrhea, and abdominal pain. None of the ill persons reported
having an elevated temperature; all recovered within 24 to 30 hours. Approx-
imately 20 physicians. No fecal specimens were obtained for bacteriologic
examination.
The supper was held in the basement of the village church. Foods were
contributed by numerous members of the congregation. The supper began at
6:00 p.m. and continued until 11:00 p.m. Food was spread out on table and
consumed over a period of several hours. Data regarding onset of illness and
food eaten or water drunk by each of the 75 persons interviewed are provided
in the attached line listing (Oswego dataset). The approximate time of eating
supper was collected for only about half the persons who had gastrointestinal
illness.
The data set can be viewed and downloaded from http://www.medepi.
net/data/oswego.txt. The data dictionary is provided in Table A.2 on the
facing page.

A.3 Western Collaborative Group Study (cohort)

The Western Collaborative Group Study (WCGS), a prospective cohort

studye, recruited middle-aged men (ages 39 to 59) who were employees of 10
California companies and collected data on 3154 individuals during the years
1960–1961. These subjects were primarily selected to study the relationship
between behavior pattern and the risk of coronary hearth disease (CHD). A
number of other risk factors were also measured to provide the best possible
assessment of the CHD risk associated with behavior type. Additional vari-
ables collected include age, height, weight, systolic blood pressure, diastolic
blood pressure, cholesterol, smoking, and corneal arcus. The median follow
up time was 8.05 years.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

A.4 Evans County (cohort) 245

Table A.2 Data dictionary for Oswego County data set

Variable Possible values
id Subject identificaton number
age Age in years
sex Sex: F = Female, M = Male
meal.time Meal time on April 18th
ill Developed illness: Y = Yes N = No
onset.date Onset date: ”4/18” = April 18th, ”4/19” = April 19th
onset.time Onset time: HH:MM AM/PM
baked.ham Consumed item: Y = Yes; N = No
spinach Consumed item: Y = Yes; N = No
mashed.potato Consumed item: Y = Yes; N = No
cabbage.salad Consumed item: Y = Yes; N = No
jello rolls Consumed item: Y = Yes; N = No
brown.bread Consumed item: Y = Yes; N = No
milk Consumed item: Y = Yes; N = No
coffee Consumed item: Y = Yes; N = No
water Consumed item: Y = Yes; N = No
cakes Consumed item: Y = Yes; N = No
vanilla.ice.cream Consumed item: Y = Yes; N = No
chocolate.ice.cream Consumed item: Y = Yes; N = No
fruit.salad Consumed item: Y = Yes; N = No

The data set can be viewed and downloaded from http://www.medepi.

net/data/wcgs.txt. The data dictionary is provided in Table A.3 on the
next page.

A.4 Evans County (cohort)

The Evans County data set is used to demonstrate a standard logistic re-
gression (unconditional) [15]. The data are from a cohort study in which 609
white males were followed for 7 years, with coronary heart disease as the
outcome of interest.
The data set can be viewed and downloaded from http://www.medepi.
net/data/evans.txt. The data dictionary is provided in Table A.4 on the
following page.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

246 A Available data sets

Table A.3 Data dictionary for Western Collaborative Group Study data set
Variable Variable name Variable type Possible values
id Subject ID Integer 2001–22101
age0 Age Continuous 39–59 years
height0 Height Continuous 60–78 in
weight0 Weight Continuous 78–320 lb
sbp0 Systolic blood pressure Continuous 98–230 mm Hg
dbp0 Diastolic blood pressure Continuous 58–150 mm Hg
chol0 Cholesterol Continuous 103–645 mg/100 ml
behpat0 Behavior pattern Categorical 1 = Type A1
2 = Type A2
3 = Type B1
4 = Type B2
ncigs0 Smoking Integer Cigarettes/day
dibpat0 Behavior pattern Categorical 0 = Type B
1 = Type A
chd69 Coronary heart disease event Categorical 0 = None
1 = Yes
typechd Coronary heart disease event Categorical 0 = CHD event
1 = Symptomatic MI
2 = Silent MI
3 = Classical angina
time169 Observation (follow up) time Continuous 18–3430 days
arcus0 Corneal arcus Categorical 0 = None
1 = Yes

Table A.4 Data dictionary for Evans data set

Variable Variable name Variable type Possible values
id Subject identifier Integer
chd Coronary heart disease Categorical-nominal 0 = no
1 = yes
cat Catecholamine level Categorical-nominal 0 = normal
1 = high
age Age Continuous years
chl Cholesterol Continuous >0
smk Smoking status Categorical-nominal 0 = never smoked
1 = ever smoked
ecg Electrocardiogram Categorical-nominal 0 = no abnormality
1 = abnormality
dbp Diastolic blood pressure Continuous mm Hg
sbp Systolic blood pressure Continuous mm Hg
hpt High blood pressure Categorical-nominal 0 = no
1 = yes
(dbp≥ 95 or sbp≥ 160)
ch cat × hpt Categorical product term
cc cat × chl Continuous product term

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

A.8 Measles surveillance cases 247

Table A.5 Data dictionary for myocardial infarction (MI) case-control data set
Variable Variable name Variable type Possible values
match Matching strata Integer 1–39
person Subject identifier Integer 1–117
mi Myocardial infarction Categorical- 0 = No
nominal 1 = Yes
smk Smoking status Categorical- 0 = Not current smoker
nominal 1 = Current smoker
sbp Systolic blood pressure Categorical- 120, 140, or 160
ordinal
ecg Electrocardiogram Categorical- 0 = No abnormality
nominal 1 = abnormality

A.5 Myocardial infarction case-control study

The myocardial infarction (MI) data set [15] is used to demonstrate condi-
tional logistic regression. The study is a case-control study that involves 117
subjects in 39 matched strata (matched by age, race, and sex). Each stratum
contains three subjects, one of whom is a case diagnosed with myocardial
infarction and the other two are matched controls.
The data set can be viewed and downloaded from http://www.medepi.
net/data/mi.txt. The data dictionary is provided in Table A.5.

A.6 AIDS surveillance cases

http://www.medepi.net/data/aids.txt

A.7 Hepatitis B surveillance cases

http://www.medepi.net/data/hepb.txt

A.8 Measles surveillance cases

http://www.medepi.net/data/measles.txt

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

248 A Available data sets

A.9 University Group Diabetes Program

http://www.medepi.net/data/ugdp.txt

A.10 Novel influenza A (H1N1) pandemic

A.10.1 United States reported cases and deaths as of

July 23, 2009

http://www.medepi.net/data/h1n1panflu23jul09usa.txt

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Appendix B
Outbreak analysis template in R

We provide an analysis template using R (and the ’epitools’ package). Ex-

amples involve human West Nile virus surveillance, other data sets (AIDS,
measles, hepatitis B, etc.).

Read data

Human West Nile virus disease surveillance, California, 2004.

wnv <-read.table("http://www.medepi.net/data/wnv/wnv2004raw.txt",
sep = ",", header = TRUE, na.strings = ".")
str(wnv) #display data set structure
head(wnv) #display first 6 lines
edit(wnv) #browse data frame
fix(wnv) #browse with ability to edit (be careful!!!)

Convert non-standard dates to Julian dates

wnv$date.onset2 <- as.Date(wnv$date.onset, format="%m/%d/%Y")

wnv$date.tested2 <- as.Date(wnv$date.tested, format="%m/%d/%Y")

Display histogram of onset dates (epidemic curve)

hist(wnv$date.onset2, breaks= 26, freq=TRUE, col="slategray1")

Describe a continuous variable (e.g., age)

summary(wnv$age) # no standard deviation provided

249
250 B Outbreak analysis template in R

range(wnv$age, na.rm=TRUE); mean(wnv$age, na.rm=TRUE)

median(wnv$age, na.rm=TRUE); sd(wnv$age, na.rm=TRUE)

Describe continuous variable, stratified by a categorical variable

tapply(wnv$age, wnv$sex, mean, na.rm = TRUE)

tapply(wnv$age, wnv$county, mean, na.rm = TRUE)

Display a continuous variable

hist(wnv$age, xlab="x", ylab="y", main="title", col="skyblue")

Describe a categorical variable (e.g., sex)

sex.tab <- xtabs(~sex, data = wnv)

sex.dist <- prop.table(sex.tab)
cbind(sex.tab, sex.dist)

Display a categorical variable (e.g. sex)

barplot(sex.tab, col="pink", ylab="Frequency", main="title")

Re-code continuous variable to categorical (e.g., age)

wnv$age3 <- cut(wnv$age, breaks=c(0,45,65,100), right=FALSE)

age3.tab <- xtabs(~age3, data = wnv)
age3.dist <- prop.table(age3.tab)
cbind(age3.tab, age3.dist)

Describe two categorical variables (e.g. sex and age)

sexage <- xtabs(~sex + age3, data = wnv)

sexage
prop.table(sexage) #joint distribution
prop.table(sexage, 1) #row distribution
prop.table(sexage, 2) #column distribution

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

B Outbreak analysis template in R 251

Plot age vs sex distribution

barplot(sexage, legend.text=TRUE,
xlab="Age", ylab="Frequency", main="title")
barplot(sexage, legend.text=TRUE, beside=TRUE,
xlab="Age", ylab="Frequency", main="title")
barplot(t(sexage), legend.text=TRUE, ylim=c(0, 650),
xlab="Sex", ylab="Frequency", main="title")
barplot(t(sexage), legend.text=TRUE, beside=TRUE, ylim=c(0, 300),
xlab="Sex", ylab="Frequency", main="title")

Hypothesis testing using 2-way contingency tables

From the main menu select Packages > Install Package(s). Select CRAN
mirror near you. Select epitools package.
library(epitools) #load ’epitools’; only needed once per session
tab.age3 <- xtabs(~age3 + death, data = wnv)
epitab(tab.age3) #default is odds ratio
epitab(tab.age3, method = "riskratio")
prop.table(tab.age3, 1) #display row distribution (2=column)
prop.test(tab.age3[,2:1]) #remember to reverse columns
chisq.test(tab.age3) #Chi-square test
fisher.test(tab.age3) #Fisher exact test

Graphical display of epidemiologic data

Histogram (continuous numbers or date objects)

hist(wnv$age, xlab="x", ylab="y", main="title", col="skyblue")

hist(wnv$date.onset2, breaks= 26, freq=TRUE, col="slategray1")

Bar chart (categorical variable)

barplot(table(wnv$sex), col="skyblue", xlab="Sex", ylab="Freq",

main="title", legend = TRUE, ylim=c(0,600))

Stacked bar chart (2 or more categorical variables)

barplot(table(wnv$sex, wnv$age3), col=c("blue","green"),

xlab="Sex", ylab="Freq", main="WNV Disease, Sex by Age",
legend = TRUE, ylim=c(0,400))

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

252 B Outbreak analysis template in R

Group bar chart (2 or more categorical variables):

barplot(table(wnv$sex, wnv$age3), beside=TRUE, xlab="Sex",

ylab="Freq", main="Sex by Age", col=c("blue","green"),
legend = TRUE, ylim=c(0,250))

Proportion bar chart (2 or more categorical variables)

sexage <- xtabs(~sex + age3, data = wnv)

barplot(prop.table(sexage, 2), xlab="Sex", ylab="Proportion",
main="WNV Disease, Sex by Age", col=c("blue","green"),
legend = TRUE, ylim=c(0,1.2))

Time series (single x values vs. single y values)

United States measles surveillance data

measles <- read.table("http://www.medepi.net/data/measles.txt",

sep="", header=TRUE)
str(measles); head(measles)
plot(measles$year, measles$cases, type="l", lwd=2, col="navy")
plot(measles$year, measles$cases, type="l", lwd=2, log="y")

Time series (multiple x values vs multiple y values)

United State AIDS and hepatitis B surveillance data

aids <- read.table("http://www.medepi.net/data/aids.txt",

sep="", header=TRUE, na.strings=".")
hepb <- read.table("http://www.medepi.net/data/hepb.txt",
sep="", header=TRUE)
years <- cbind(aids$year, hepb$year)
cases <- cbind(aids$cases, hepb$cases)
matplot(years, cases, type="l", lwd=2, col=1:2, main="title")
legend(x=1980, y=80000, legend= c("AIDS","Hepatitis B"),
lty=1:2, col=1:2, lwd=2)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

B Outbreak analysis template in R 253

Working with dates and times

Convert non-standard dates to standard Julian dates

dates <- c("11-02-1959","1959Nov02","November 2, 1959")

jdates<-as.Date(dates, format=c("%m-%d-%Y","%Y%b%d","%B %d, %Y"))
jdates; julian(jdates)

Converting non-standard dates and times to R date-time object:

dtim <- c("4/19/1940 12:30 AM", "4/18/1940 9:45 PM")

std.dt <- strptime(dtim, format="%m/%d/%Y %I:%M %p")
std.dt
Try ?strptime to see all format options.

Manually creating an epidemic curve

Single variable

labs <- c("Sun", "Mon","Tue","Wed","Thu","Fri", "Sat")

cases <- c(0, 25, 15, 5, 10, 20, 0)
names(cases) <- labs
barplot(cases, space=0, col="skyblue", xlab="Day", ylab="Cases",
main="Title")

Single variable—Change x-axis labels to perpendicular

xv <- barplot(cases, space=0, col="red", xlab="Day",

ylab="Cases", main="Title", axisnames=FALSE)
axis(side=1, at=xv, labels=labs, las=2)

Stratified by second variable

male.cases <- c(0, 15, 10, 3, 5, 5, 0)

female.cases <- c(0, 10, 5, 2, 5, 15, 0)
cases2 <- rbind(Male = male.cases, Female = female.cases)
colnames(cases2) <- labs
xv <- barplot(cases2, space=0, col=c("blue", "green"),
xlab="Day", ylab="Cases", main="Title",
axisnames=FALSE, legend.text=TRUE, ylim=c(0, 30))
axis(side=1, at=xv, labels=labs, las=2)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

254 B Outbreak analysis template in R

Running batch jobs

source("c:/myoutbreak/job01.R") #run program file called job01.R

Creating output log files

From within job01.R program file

x <- 1:5
y <- x^2

Sink printed objects to log file

sink("c:/temp/job.log")
print(x)
sink()

Capture output without requiring print command

capture.output(cbind(x, y),
file="c:/temp/job.log", append=TRUE)

Multivariable analysis

Logistic regression (binomial data: cohort, case-control)

Using WNV data with age3 variable created previously:

mod1 <- glm(death ~ age3, family=binomial, data=wnv)
summary(mod1) #full results
exp(mod1$coef) #calculate odds ratio
mod2 <- glm(death ~ age3 + sex, family=binomial, data=wnv)
summary(mod2) #full results
exp(mod2$coef) #calculate odds ratio

Conditional logistic regression (matched case-control)

Here is a case-control study of myocardial infarction (Kleinbaum 2002): one

case was matched to two controls on age, race, and sex.
library(survival) #load survival package
chd <- read.table("http://www.medepi.net/data/chd.txt", sep=",",

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

B Outbreak analysis template in R 255

header=TRUE)
head(chd)
chd$mi2 <- ifelse(chd$mi=="Yes", 1, 0) #re-code case status
mod1 <- clogit(mi2~smk+strata(match), data=chd)
summary(mod1)
mod2 <- clogit(mi2~smk+sbp+strata(match), data=chd)
summary(mod2)
mod3 <- clogit(mi2~smk+sbp+ecg+strata(match), data=chd)
summary(mod3)
anova(mod1,mod2,mod3, test="Chisq") #compare nested models

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Appendix C
Programming and creating R functions

“Good programmers write good code, great programmers borrow good code.”

R is a comprehensive and powerful programming language. In this section

we briefly summarize how to use R for introductory programming, including
writing and executing functions.

C.1 Basic programming

Basic epidemiologic programming in R is just a list of R expressions, that

are collected and executed as a batch job. The list of R expressions in an
ASCII text file with a .R extension. In RStudio, from the main menu, select
File → New → R Script. This script file will be saved with an .R extension.
This script file can be edited and executed within RStudio. Alternatively, we
can edit this file using our favorite text editor (e.g., GNU Emacs).
What are the characteristics of good R programming?
• Use a good text editor (or RStudio) for programming
• Organize batch jobs into numbered sequential files (e.g., job01.R)
• Avoid graphical menu-driven approaches
First, use a good text editor (or RStudio) for programming. Each R ex-
pression will span one or more lines. Although one could write and submit
each line at the R console, this approach is inefficient and not recommended.
Instead, type the expressions into your favorite text editor and save with a
.R extension. Then, selected expressions or the whole file can be executed in
R. Use the text editor that comes with R, or text editors customized to work
with R (e.g., RStudio, Emacs with ESS).
Second, we organize batch jobs into sequential files. Data analysis is a se-
ries of tasks involving data entry, checking, cleaning, analysis, and reporting.
Although data analysts are primarily involved in analysis and reporting, they

257
258 C Programming and creating R functions

may be involved in earlier phases of data preparation. Regardless of stage of

involvement, data analysts should organize, conduct, and document their
analytics tasks and batch jobs in chronological order. For example, batch
jobs might be named as follows: job01-cleaning.R, job02-recoding.R, job03-
descriptive.R, job04-logistic.R, etc. Naming the program file has two compo-
nents: ’jobs’ represent major tasks and are always numbered in chronological
order (job01-*.R, job02-*.R, etc.); and a brief descriptor can be appended
to the first component of the file name (job01-recode-data.R, job02-bivariate-
analysis.R).
If one needs to repeat parts of a previous job, then add new jobs, not
edit old jobs. This way our analysis can always be reviewed, replicated, and
audited exactly in order it was conducted. We avoid editing earlier jobs. If we
edit previous jobs, then we must rerun all subsequent jobs in chronological
order.
Third, we avoid graphical, menu-driven approaches. While this is a tempt-
ing approach, our work cannot be documented, replicated, and audited. The
best approach is to collect R expressions into batch jobs and run them using
the source function.

C.2 Intermediate programming

The next level of R programming involves (1) implementing control flow

(decision points); (2) implementing dependencies in calculations or data ma-
nipulation; and (3) improving execution efficiency,

C.2.1 Control statements

Control flow involves one or more decision points. A simplest decision point
goes like this: if a condition is TRUE, do {this} and then continue; if it is
FALSE, do not do {this} and then continue. When R continues, the next R
expression can be any valid expression, including another decision point.

C.2.1.1 The if function

We use the if function to implement single decision points.

if(TRUE) {execute these R expressions }
If the condition if false, R skips the bracketed expression and continues exe-
cuting subsequent lines. Study this example:
> x <- c(1, 2, NA, 4, 5)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

C.2 Intermediate programming 259

> y <- c(1, 2, 3, 4, 5)

> if(any(is.na(x))) {x[is.na(x)] <- 999}
> x
[1] 1 2 999 4 5
> if(any(is.na(y))) {y[is.na(y)] <- 999}
> y
[1] 1 2 3 4 5
The first if condition evaluated to TRUE and the missing value was replaced.
The second if condition evaluated to FALSE and the bracketed expressions
were not evaluated.

C.2.1.2 The else functions

Up to now the if condition had only one possible response. If there are two,
mutually exclusive possible responses, add one else statement:
if(TRUE) {
execute these R expressions
} else {
execute these R expressions
}
Here is an example:
> x <- c(1, 2, NA, 4, 5);
> y <- c(1, 2, 3, 4, 5)
> if(any(is.na(x))) {
+ x[is.na(x)] <- 999; cat("NAs replaced\n")
+ } else {cat("No missing values; no replacement\n")}
NAs replaced
> if(any(is.na(y))) {
+ y[is.na(y)] <- 999; cat("NAs replaced\n")
+ } else {cat("No missing values; no replacement\n")}
No missing values; no replacement
> x
[1] 1 2 999 4 5
> y
[1] 1 2 3 4 5
> y
[1] 1 2 3 4 5
Therefore, use the if and else combination if one needs to evaluat of one of
two possible collection of R expressions.
If one needs to evaluate possibly one of two possible collection of R expres-
sions then use the following pattern:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

260 C Programming and creating R functions

if(TRUE) {
execute these R expressions
} else if(TRUE) {
execute these R expressions
}
The if and else functions can be combined to achieve any desired control
flow.

C.2.1.3 The “short circuit” logical operators

The “short circuit” && and || logical operators are used for control flow in if
functions. If logical vectors are provided, only the first element of each vector
is used. Therefore, for element-wise comparisons of 2 or more vectors, use the
& and | operators but not the && and || operators (discussed in Chapter 2).
For if function comparisons use the && and || operators.
Suppose we want to square the elements of a numeric vector but not if it
is a matrix.
> x <- 1:5
> y <- matrix(1:4, 2, 2)
> if(is.numeric(x) && !is.matrix(x)) {
+ x^2
+ } else cat("Either not numeric or is a matrix\n")
[1] 1 4 9 16 25
> if(!is.matrix(y) && is.numeric(y)) {
+ y^2
+ } else cat("Either not numeric or is a matrix\n")
Either not numeric or is a matrix
The && and || operators are called “short circuit” operators because not
all its arguments may be evaluated: moving from left to right, only sufficient
arguments are evaluated to determine if the if function should return TRUE
or FALSE. This can save considerable time if some the arguments are complex
functions that require significant computing time to evaluate to either TRUE
or FALSE. In the previous example, because !is.matrix(y) evaluates to
FALSE, it was not necessary to evaluate is.numeric(y).

C.2.2 Vectorized approach

An important advantage of R is the availability of functions that perform

vectorized calculations. For example, suppose we wish to add to columns of
a matrix. Here is one approach:

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

C.2 Intermediate programming 261

> tab <- matrix(1:12, 3, 4)

> tab
[,1] [,2] [,3] [,4]
[1,] 1 4 7 10
[2,] 2 5 8 11
[3,] 3 6 9 12
> colsum <- tab[,1]+tab[,2]+tab[,3]+tab[,4]
> colsum
[1] 22 26 30
However, this can be accomplished more efficiently using the apply function:
> colsum2 <- apply(tab, 1, sum)
> colsum2
[1] 22 26 30
In general, we want to use these types of functions (e.g., tapply, sweep,
outer, mean, etc.) because they have been optimized to performed vectorized
calculations.

C.2.2.1 The ifelse function

The ifelse function is a vectorized element-wise implementation of the if

and else functions. We demonstrate using the practical example of recoding
a 2-level variable.
> sex <- c("M", NA, "F", "F", NA, "M", "F", "M")
> sex2 <- ifelse(sex=="M", "Male", "Female")
> sex2
[1] "Male" NA "Female" "Female" NA "Male" "Female" "Male"
If an element of sex contains ”M” (TRUE), it is recoded to ”Male” (in sex2),
and otherwise (FALSE) it is recoded to ”Female”. This assumes that there
are only ”M”s and ”F”s in the data vector.

C.2.3 Looping

Looping is a common programming approach that is discouraged in R because

it is inefficient. It is much better to conduct vectorized calculations using
existing functions. For example, suppose we want to sum a numeric vector.
> x <- 1:10
> xsum <- 0
> for (i in 1:10){
+ xsum <- xsum + x[i]

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

262 C Programming and creating R functions

+ }
> xsum
[1] 55
A much better approach is to use the sum function:
> sum(x)
[1] 55
Unless it is absolutely necessary, we avoid looping.
Looping is necessary when (1) there is no R function to conduct a vector-
ized calculation, and (2) when the result of an element depends on the result
of a preceding element which may not be known beforehand (e.g., when it is
the result of a random process).

C.2.3.1 The for function

The previous example was a for loop. Here is the syntax:

for (i in somevector {
do some calcuation with ith element of somevector
}
In the for function R loops and uses the ith element of somevector either
directly or indirectly (e.g., indexing another vector). Here is using the vector
directly:
> for(i in 1:3){
+ cat(i^2,"\n")
+ }
1
4
9
The letters contain the American English alphabet. Here we use an integer
vector for indexing letters:
> for(i in 1:3){
+ cat(letters[i],"\n")
+ }
a
b
c
Somevector can be any vector:
> kids <- c("Tomasito", "Luisito", "Angela")
> for (i in kids) {print(i)}
[1] "Tomasito"

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

C.2 Intermediate programming 263

[1] "Luisito"
[1] "Angela"

C.2.3.2 The while function

The while function will continue to evaluate a collection of R expressions

while a condition is true. Here is the syntax:
while(TRUE) {
execute these R expressions
}
Here is a trivial example:
> x <- 1; z <- 0
> while(z < 5){
+ show(z)
+ z <- z + x
+ }
[1] 0
[1] 1
[1] 2
[1] 3
[1] 4
The while function is used for optimization functions that are converging
to a numerical value.

C.2.3.3 The break and next functions

The break expression will break out of a for or while loop if a condition
is met, and transfers control to the first statement outside of the inner-most
loop. Here is the general syntax:
for (i in somevector {
do some calcuation with ith element of somevector
if(TRUE) break
}
The next expression halts the processing of the current iteration and ad-
vances the looping index. Here is the general syntax:
for (i in somevector {
do some calcuation with ith element of somevector
if(TRUE) next
}
Both break and next apply only to the innermost of nested loops.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

264 C Programming and creating R functions

C.2.3.4 double for

In the next example we nest two for loop to generate a multiplication table
for the integers 6 to 10:
> x <- 6:10
> mtab <- matrix(NA, 5, 5)
> rownames(mtab) <- x
> colnames(mtab) <- x
> for(i in 1:5){
+ for(j in 1:5){
+ mtab[i, j] <- x[i]*x[j]
+ }
+ }
> mtab
6 7 8 9 10
6 36 42 48 54 60
7 42 49 56 63 70
8 48 56 64 72 80
9 54 63 72 81 90
10 60 70 80 90 100

C.3 Writing R functions

Writing R functions involves three steps:

• Prepare inputs
• Do calculations
• Collect results
The best way to learn these steps is to incorporate them into our regular
R programming. For example, suppose we are writing R code to calculate the
odds ratio from a 2 × 2 table with the appropriate format. For this we will
use the Oswego data set available from the epitools package.

> ## Prepare inputs

> library(epitools)
> data(oswego)
> tab1 = xtabs(~ ill + spinach, data = oswego)
> tab1
spinach
ill N Y
N 12 17
Y 20 26
> aa = tab1[1, 1]

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

C.3 Writing R functions 265

> bb = tab1[1, 2]
> cc = tab1[2, 1]
> dd = tab1[2, 2]
>
> ## Do calculations
> crossprod.OR = (aa*dd)/(bb*cc)
>
> ## Collect results
> list(data = tab1, odds.ratio = crossprod.OR)
$data
spinach
ill N Y
N 12 17
Y 20 26

$odds.ratio
[1] 0.9176471

Now that we are familiar of what it takes to calculate an odds ratio from
a 2-way table we can convert the code into a function and load it at the R
console. Here is new function:
myOR = function(x){
## Prepare input
## x = 2x2 table amenable to cross-product
aa = x[1, 1]
bb = x[1, 2]
cc = x[2, 1]
dd = x[2, 2]

## Do calculations
crossprod.OR = (aa*dd)/(bb*cc)

## Collect results
list(data = x, odds.ratio = crossprod.OR)
}
Now we can test the function:
> tab.test = xtabs(~ ill + spinach, data = oswego)
> myOR(tab.test)
$data
spinach
ill N Y
N 12 17
Y 20 26

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

266 C Programming and creating R functions

$odds.ratio
[1] 0.9176471

C.3.1 Arguments default values

Now suppose we wish to add calculating a 95% confidence interval to this

function. We will use the following normal approximation standard error
formula for an odds ratio:
r
1 1 1 1
SE[log(OR)] = + + +
A1 B1 A0 B0

And here is the (1 − α)% confidence interval:

ORL , ORU = exp{log(OR) ± Zα/2 SE[log(OR)]}

Here is the improved function:

myOR2 = function(x, conf.level){
## Prepare input
## x = 2x2 table amenable to cross-product
aa = x[1, 1]
bb = x[1, 2]
cc = x[2, 1]
dd = x[2, 2]
if(missing(conf.level)) stop("Must specify confidence level")
Z <- qnorm((1 + conf.level)/2)

## Do calculations
logOR <- log((aa*dd)/(bb*cc))
SE.logOR <- sqrt(1/aa + 1/bb + 1/cc + 1/dd)
OR <- exp(logOR)
CI <- exp(logOR + c(-1, 1)*Z*SE.logOR)

## Collect results
list(data = x, odds.ratio = OR, conf.int = CI)
}
Notice that conf.level is a new argument, but with no default value. If a
user forgets to specify a default value, the following line handles this possi-
bility:
if(missing(conf.level)) stop("Must specify confidence level")
Now we test this function:
> tab.test = xtabs(~ ill + spinach, data = oswego)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

C.3 Writing R functions 267

> myOR2(tab.test)
Error in myOR2(tab.test) : Must specify confidence level
> myOR2(tab.test, 0.95)
$data
spinach
ill N Y
N 12 17
Y 20 26

$odds.ratio
[1] 0.9176471

$conf.int
[1] 0.3580184 2.3520471
If an argument has a usual value, then specify this as an argument default
value:
myOR3 = function(x, conf.level = 0.95){
## Prepare input
## x = 2x2 table amenable to cross-product
aa = x[1, 1]
bb = x[1, 2]
cc = x[2, 1]
dd = x[2, 2]
Z <- qnorm((1 + conf.level)/2)

## Do calculations
logOR <- log((aa*dd)/(bb*cc))
SE.logOR <- sqrt(1/aa + 1/bb + 1/cc + 1/dd)
OR <- exp(logOR)
CI <- exp(logOR + c(-1, 1)*Z*SE.logOR)

## Collect results
list(data = x, odds.ratio = OR, conf.int = CI)
}
We test our new function:
> tab.test = xtabs(~ ill + spinach, data = oswego)
> myOR3(tab.test)
$data
spinach
ill N Y
N 12 17
Y 20 26

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

268 C Programming and creating R functions

$odds.ratio
[1] 0.9176471

$conf.int
[1] 0.3580184 2.3520471

> myOR3(tab.test, 0.90)

$data
spinach
ill N Y
N 12 17
Y 20 26

$odds.ratio
[1] 0.9176471

$conf.int
[1] 0.4165094 2.0217459

C.3.2 Passing optional arguments using the ...

function

On occasion we will have a function nested inside one of our functions and we
need to be able to pass optional arguments to this nested function. This com-
monly occurs when we write functions for customized graphics but only wish
to specify some arguments for the nested function and leave the remaining
arguments optional. For example, consider this function:
myplot = function(x, y, type = "b", ...){
plot(x, y, type = type, ...)
}
When using myplot one only needs to provide x and y arguments. The type
option has been set to a default value of ”b”. The ... function will pass
any optional arguments to the nested plot function. Of course, they optional
arguments must be valid options for plot function.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

C.4 Advanced topics 269

C.4 Advanced topics

C.4.1 Lexical scoping

The variables which occur in the body of a function can be divided into
three classes; formal parameters, local variables and free variables. The formal
parameters of a function are those occurring in the argument list of the
function. Their values are determined by the process of binding the actual
function arguments to the formal parameters. Local variables are those whose
values are determined by the evaluation of expressions in the body of the
functions. Variables which are not formal parameters or local variables are
called free variables. Free variables become local variables if they are assigned
to. Consider the following function definition.
f <- function(x){
y <- 2*x
print(x)
print(y)
print(z)
}
In this function, x is a formal parameter, y is a local variable and z is a free
variable. In R the free variable bindings are resolved by first looking in the
environment in which the function was created. This is called lexical scope.
If the free variable is not defined there, R looks in the enclosing environment.
For the function f this would be the global environment (workspace).
To understand the implications of lexical scope consider the following:
> rm(list = ls())
> ls()
character(0)
> f <- function(x){
+ y <- 2*x
+ print(x)
+ print(y)
+ print(z)
+ }
> f(5)
[1] 5
[1] 10
Error in print(z) : object ’z’ not found
> z = 99
> f(5)
[1] 5
[1] 10
[1] 99

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

270 C Programming and creating R functions

In the f function z is a free variable. The first time f is executed z is not

defined in the function. R looks in the enclosing environment and does not
find a value for z and reports an error. However, when a object z is created
in the global environment, R is able to find it and uses it.
Lexical scoping is convenient because it allows nested functions with free
variables to run provided the variable has been defined in an enclosing envi-
ronment. This convenience becomes obvious when one writes many programs.
However, there is a danger: an unintended free variable many find an unin-
tended value in an enclosing environment. This may go undetected because
no error is reported. This can happen when there are many objects in the
workspace from previous sessions. A good habit is to clear the workspace of
all objects at the beginning of every session.
Here is another example from the R introductory manual1 . Consider a
function called cube.
cube <- function(n) {
sq <- function() n*n
n*sq()
}
The variable n in the function sq is not an argument to that function. There-
fore it is a free variable and the scoping rules must be used to ascertain the
value that is to be associated with it. Under static scope (S-Plus) the value
is that associated with a global variable named n. Under lexical scope (R) it
is the parameter to the function cube since that is the active binding for the
variable n at the time the function sq was defined. The difference between
evaluation in R and evaluation in S-Plus is that S-Plus looks for a global
variable called n while R first looks for a variable called n in the environment
created when cube was invoked.
## first evaluation in S
S> cube(2)
Error in sq(): Object "n" not found
Dumped
S> n <- 3
S> cube(2)
[1] 18

## then the same function evaluated in R

R> cube(2)
[1] 8

1 http://cran.r-project.org/doc/manuals/r-release/R-intro.pdf

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Solutions

Problems of Chapter 1

1.1 To download R go to http://cran.r-project.org/ and follow the

instructions for your operating system. Once installed, start R. The computer
file path to the workspace file, .RData, is obtained using the getwd function
(see Table 1.3 on page 12 for more useful functions).
> getwd()
[1] "/home/tja/Data/R/home"
This displayed in R the file path on the computer. To see the actual .RData
file, we must enable our computer system to view hidden files and then use
the computer’s program for viewing files. This is useful to know in case we
want to physically move the file to another location.
1.2 To list the R packages currently loaded, use the search function.
> search()
[1] ".GlobalEnv" "package:stats" "package:graphics"
[4] "package:grDevices" "package:utils" "package:datasets"
[7] "package:methods" "Autoloads" "package:base"
Alternatively, use the searchpaths function to see the file paths.
> searchpaths()
[1] ".GlobalEnv" "/usr/lib64/R/library/stats"
[3] "/usr/lib64/R/library/graphics" "/usr/lib64/R/library/grDevices"
[5] "/usr/lib64/R/library/utils" "/usr/lib64/R/library/datasets"
[7] "/usr/lib64/R/library/methods" "Autoloads"
[9] "/usr/lib64/R/library/base"
1.3
ls()
rm(list=ls())

271
272 Solutions

The ls function returns a character vector of object names. This character

vector can be used as the list argument in the rm function to remove all the
objects. Instead of ls we could have used the objects function.
1.4
inches <- 1:12
centimeters <- inches*2.54
cbind(inches, centimeters)
1.5
celsius <- c(0, 100)
fahrenheit <- ((9/5)*celsius) + 32
fahrenheit
Notice that we used a numeric vector so that the calculation requires fewer
steps. This is an example of a vectorized (or spreadsheat-like) operation.
1.6
celsius <- seq(0, 100, 5)
celsius
fahrenheit <- ((9/5)*celsius) + 32
fahrenheit
Notice that we used a numeric vector so that the calculation requires fewer
steps. This is an example of a vectorized (or spreadsheat-like) operation.
1.7 Suppose you weigh 150 lbs, what is your weight in kilograms? Hint:
Remember dimensional analysis?
1 kg 150 kg
150 lb × =
2.2 lb 2.2
Suppose your height is 5 feet 8 inches, What is your height in meters? (5’ 8”
= 5’ + 8/12” = 5.75 feet)
1m 5.75 m
5.75 ft × =
3.3 ft 3.3
mywt.lb <- 150
myht.ft <- 5.75
mywt.kg <- mywt.lb/2.2
myht.m <- myht.ft/3.3
bmi <- mywt.kg/myht.m^2
bmi
1.8
> 7/2 #divide
[1] 3.5

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Solutions 273

1
0
log(x)

−2 −1

0 1 2 3 4 5 6

Fig. C.1 Plot of y = loge (x)

> 7%/%2 #integer divide

[1] 3
> 7%%2 #modulus = remainder
[1] 1
1.9 See Figure C.1. The number e is a very special number. When we take
the logarithm of a number using base e we map the values [0, +∞) into (−∞,
+∞), where the loge (1) = 0. More specifically, the number range [0,1] maps
into (−∞, 0], and [1, +∞) maps into [0, +∞). Of note, the loge (e) = 1.
In epidemiology, disease counts and physical measurements (e.g., weight)
have the asymmetric range [0, +∞). The natural logarithm transformation
allows us to work with values between 0 and 1, in a sense unbounding the
left tail distribution.
1.10 See Figure C.2. The logit transformation is a double transformation.
First, the odds tranformation (R/(1 − R)) unbounds the probabilities near
1; second, the natural logarithm of the log-odds, or the logit transforma-
tion, (log(odds)) unbounds the probabilities near 0. In other words, the logit
transformation maps the numeric range [0, 1] into (−∞, +∞), where the
log(0.5/(1 − 0.5)) = 0. The logit transformation allows us to work with prob-
abilities that, of course, have the range [0, 1].

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

274 Solutions

risk odds = R/(1−R) log(risk odds) = logit

log(R/(1−R))
4

4
R/(1−R)

0
−4

−4
0.0 0.4 0.8 0.0 0.4 0.8

R R

Fig. C.2 The logit transformation is a double transformation. First, the odds tranfor-
mation (R/(1 − R)) unbounds the probabilities near 1; second, the logit transformation
(log(odds)) additionally unbounds the probabilities near 0.

1.11
n <- 365
per.act.risk <- c(0.5, 1, 5, 6.5, 10, 30, 50, 67)/10000
risks <- 1-(1-per.act.risk)^n
risks
##label risks (optional)
act <- c("IOI", "ROI", "IPVI", "IAI", "RPVI", "PNS", "RAI", "IDU")
names(risks) <- act
risks
1.12 For this problem I put the following code into an ASCII text file named
job01.R:
n <- 365
per.act.risk <- c(0.5, 1, 5, 6.5, 10, 30, 50, 67)/10000
risks <- 1-(1-per.act.risk)^n
risks
##label risks (optional)
act <- c("IOI", "ROI", "IPVI", "IAI", "RPVI", "PNS", "RAI", "IDU")
names(risks) <- act
risks
Here is what happened when I sourced it:
> source("/home/tja/Documents/courses/ph251d/jobs/job01.R")

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Solutions 275

> source("/home/tja/Documents/courses/ph251d/jobs/ph251d-chp1-job01.R", echo = TRUE)

> n <- 365
> per.act.risk <- c(0.5, 1, 5, 6.5, 10, 30, 50, 67)/10000
> risks <- 1-(1-per.act.risk)^n
> risks
[1] 0.01808493 0.03584367 0.16685338 0.21126678 0.30593011
[6] 0.66601052 0.83951869 0.91402762

> ##label risks (optional)

> act <- c("IOI", "ROI", "IPVI", "IAI", "RPVI", "PNS", "RAI", "IDU")
> names(risks) <- act
> risks
IOI ROI IPVI IAI RPVI PNS
0.01808493 0.03584367 0.16685338 0.21126678 0.30593011 0.66601052
RAI IDU
0.83951869 0.91402762
Conclusion: running source alone runs R commands in a source file but does
not echo the input and output to the screen unless echo = TRUE.
1.13 I ran this code in R (Linux):
sink("/home/tja/Documents/courses/ph251d/jobs/job01.log1a")
source("/home/tja/Documents/courses/ph251d/jobs/job01.R")
sink() #closes connection

sink("/home/tja/Documents/courses/ph251d/jobs/job01.log1b")
source("/home/tja/Documents/courses/ph251d/jobs/job01.R", echo = TRUE)
sink() #closes connection
The job01.log1a is empty. Here are the contents of job01.log1b:
> n <- 365
> per.act.risk <- c(0.5, 1, 5, 6.5, 10, 30, 50, 67)/10000
> risks <- 1-(1-per.act.risk)^n
> risks
[1] 0.01808493 0.03584367 0.16685338 0.21126678 0.30593011
[6] 0.66601052 0.83951869 0.91402762

> ##label risks (optional)

> act <- c("IOI", "ROI", "IPVI", "IAI", "RPVI", "PNS", "RAI", "IDU")
> names(risks) <- act
> risks
IOI ROI IPVI IAI RPVI PNS
0.01808493 0.03584367 0.16685338 0.21126678 0.30593011 0.66601052
RAI IDU
0.83951869 0.91402762

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

276 Solutions

Conclusion: running the sink function sends what would normally go to the
screen to a log file.
1.14 Sourcing job02.R at the R command line looks like this:
> source("/home/tja/Documents/courses/ph251d/jobs/job02.R")
[1] 0.01808493 0.03584367 0.16685338 0.21126678 0.30593011
[6] 0.66601052 0.83951869 0.91402762
Conclusion: The source function, without echo = TRUE, will not return
anything to the screen unless the show (or print) function is used to “show”
an R object. This make complete sense. If one is sourcing a file with thousands
of R expressions, we do not want to see all those expressions, we only want to
selected data objects with relevant results. Sinking a file only directs anything
that would appear on the screen to a log file.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Solutions 277

Problems of Chapter 2

2.1 n/a
2.2 See Table 2.1 on page 28.
2.3 We can index by position, by logical, and by name—if it exists.
2.4 Any R object component(s) that can be indexed, can be replaced.
2.5 Study and practice the following R code.
tab <- matrix(c(139, 443, 230, 502), nrow = 2, ncol = 2,
dimnames = list("Vital Status" = c("Dead", "Alive"),
Smoking = c("Yes", "No")))
tab

# equivalent
tab <- matrix(c(139, 443, 230, 502), 2, 2)
dimnames(tab) <- list("Vital Status" = c("Dead", "Alive"),
Smoking = c("Yes", "No"))
tab

# equivalent
tab <- matrix(c(139, 443, 230, 502), 2, 2)
rownames(tab) <- c("Dead", "Alive")
colnames(tab) <- c("Yes", "No")
names(dimnames(tab)) <- c("Vital Status", "Smoking")
tab
2.6 Using the tab object from Solution 2.5, study and practice the following
R code to recreate Table 2.38 on page 103.
rowt <- apply(tab, 1, sum)
tab2 <- cbind(tab, Total = rowt)
colt <- apply(tab2, 2, sum)
tab2 <- rbind(tab2, Total = colt)
names(dimnames(tab2)) <- c("Vital Status", "Smoking")
tab2
2.7 Using the tab object from Solution 2.5, study and practice the following
R code to calculate row, column, and joint distributions.
# row distrib
rowt <- apply(tab, 1, sum)
rowd <- sweep(tab, 1, rowt, "/")
rowd

# col distrib

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

278 Solutions

colt <- apply(tab, 2, sum)

cold <- sweep(tab, 2, colt, "/")
cold

# joint distrib
jtd <- tab/sum(tab); jtd
distr <- list(row.distribution = rowd,
col.distribution = cold,
joint.distribution = jtd)
distr
2.8 Using the tab2 object from Solution 2.6, study and practice the following
R code to recreate Table 2.39 on page 103. Note that the column distributions
from Solution 2.7 can also be used.
risk = tab2[1,1:2]/tab2[3,1:2]
risk.ratio <- risk/risk[2]
odds <- risk/(1-risk)
odds.ratio <- odds/odds[2]
ratios <- rbind(risk, risk.ratio, odds, odds.ratio)
ratios
Interpretation: The risk of death among non-smokers is higher than the risk
of death among smokers, suggesting that there may be some confounding.
2.9 Implement analysis below.
wdat = read.table("http://www.medepi.net/data/whickham-engl.txt",
sep = ",", header = TRUE)
str(wdat)
wdat.vas = xtabs(~Vital.Status + Age + Smoking, data = wdat)
wdat.vas
wdat.tol.vas = apply(wdat.vas, c(2, 3), sum)
wdat.risk.vas = sweep(wdat.vas, c(2, 3), wdat.tot.vas, "/")
round(wdat.risk.vas, 2)
Here are the final results:
> round(wdat.risk.vas, 2)
, , Smoking = No

Age
Vital.Status 18-24 25-34 35-44 45-54 55-64 65-74 75+
Alive 0.98 0.97 0.94 0.85 0.67 0.22 0.00
Dead 0.02 0.03 0.06 0.15 0.33 0.78 1.00

, , Smoking = Yes

Age

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Solutions 279

Vital.Status 18-24 25-34 35-44 45-54 55-64 65-74 75+

Alive 0.96 0.98 0.87 0.79 0.56 0.19 0.00
Dead 0.04 0.02 0.13 0.21 0.44 0.81 1.00
Interpretation: The risk of death is not larger in non-smokers, in fact it is
larger among smokers in older age groups..
2.10 First, look at the data set at http://www.medepi.net/data/syphilis89c.
txt. Then read in.
std <- read.table("http://www.medepi.net/data/syphilis89c.txt",
head = TRUE, sep = ",")
str(std)
head(std)
lapply(std, table)
Creating 3-D array without attaching std data frame.
table(std$Race, std$Age, std$Sex)
xtabs(~ Race + Age + Sex, data = std)
Now repeat attaching std data frame using attach function. Study the dif-
ferences.
attach(std)
table(Race, Age, Sex)
xtabs(~ Race + Age + Sex)
detach(std)
2.11
tab.ars <- table(std$Age, std$Race, std$Sex)
# 2-D tables
tab.ar <- apply(tab.ars, c(1, 2), sum); tab.ar
tab.as <- apply(tab.ars, c(1, 3), sum); tab.as
tab.rs <- apply(tab.ars, c(2, 3), sum); tab.rs

# 1-D tables
tab.a <- apply(tab.ars, 1, sum); tab.a
tab.r <- apply(tab.ars, 2, sum); tab.r
tab.s <- apply(tab.ars, 3, sum); tab.s
2.12 For this example, we’ll choose one 3-D array.
tab.ars <- table(std$Age, std$Race, std$Sex)
# row distrib
rowt <- apply(tab.ars, c(1, 3), sum)
rowd <- sweep(tab.ars, c(1, 3), rowt, "/"); rowd
#confirm
apply(rowd, c(1, 3), sum)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

 280 Solutions

# col distrib
colt <- apply(tab.ars, c(2, 3), sum)
cold <- sweep(tab.ars, c(2, 3), colt, "/"); cold
#confirm
apply(cold, c(2, 3), sum)

# joint distrib
jtt <- apply(tab.ars, 3, sum)
jtd <- sweep(tab.ars, 3, jtt, "/"); jtd
#confirm
apply(jtd, 3, sum)

distr <- list(row.distribution = rowd,

col.distribution = cold,
joint.distribution = jtd)
distr
2.13 It is a good idea to understand how the rep function works with two
vectors:
> rep(4:6, 1:3)
[1] 4 5 5 6 6 6
We can see that the second vector determines the frequency of the first vector
elements. Now use this understanding with the syphilis data.
sdat89b <- read.csv("http://www.medepi.net/data/syphilis89b.txt")
str(sdat89b)
Sex <- rep(sdat89b$Sex, sdat89b$Freq)
Race <- rep(sdat89b$Race, sdat89b$Freq)
Age <- rep(sdat89b$Age, sdat89b$Freq)
sdat89.df <- data.frame(Sex, Race, Age)
str(sdat89.df)
Discuss in class 2.14 Pending

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Solutions 281

Problems of Chapter 3

3.1 First, we recognize that this data frame contains aggregrate-level data,
not individual-level data. Each row represents a unique covariate pattern,
and the last field is the frequency of that pattern. Because the data frame
only has a few rows here is one way:
Status <- rep(c("Dead", "Survived"), 4)
Treatment <- rep(c("Tobutamide", "Tobutamide",
"Placebo", "Placebo"), 2)
Agegrp <- c(rep("<55", 4), rep("55+", 4))
Freq <- c(8, 98, 5, 115, 22, 76, 16, 69)
dat <- data.frame(Status, Treatment, Agegrp, Freq)
dat
An alternative, and better way, is to create an array that reproduce the
core data from Table 3.1 on page 107. Then we use the data.frame and
as.table functions. Here we show a few ways to create this array object.

#answer 1
udat <- array(c(8, 98, 5, 115, 22, 76, 16, 69), dim =
c(2, 2, 2),
dimnames = list(Status = c("Dead", "Survived"),
Treatment = c("Tolbutamide", "Placebo"),
Agegrp = c("<55", "55+")))
dat <- data.frame(as.table(udat))
dat

#answer 2
Status <- rep(c("Dead", "Survived"), 4)
Treatment <- rep(rep(c("Tolbutamide", "Placebo"),
c(2, 2)), 2)
Agegrp <- rep(c("<55", "55+"), c(4, 4))
Freq <- c(8, 98, 5, 115, 22, 76, 16, 69)
dat <- data.frame(Status, Treatment, Agegrp, Freq)
dat

#answer 2b, equivalent to 2a

dat <- data.frame(
Status = rep(c("Dead", "Survived"), 4),
Treatment = rep(rep(c("Tolbutamide", "Placebo"),
c(2, 2)), 2),
Agegrp = rep(c("<55", "55+"), c(4, 4)),
Freq = c(8, 98, 5, 115, 22, 76, 16, 69)
)
dat

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

282 Solutions

3.2 See Chapter.

3.3
adat <- read.table("http://www.medepi.net/data/aids.txt", header=TRUE,
sep="", na.strings=".")
head(adat)
plot(adat$year, adat$cases, type = "l", xlab = "Year", lwd = 2,
ylab = "Cases", main = "Reported AIDS Cases in United States, 1980--2003")
3.4
mdat <- read.table("http://www.medepi.net/data/measles.txt",
header=TRUE, sep="")
head(mdat)
plot(mdat$year, mdat$cases, type = "l", xlab = "Year", lwd = 2,
ylab = "Cases",
main = "Reported Measles Cases in United States, 1980--2003")
plot(mdat$year, mdat$cases, type = "l", xlab = "Year", lwd = 2,
log = "y", ylab = "Cases",
main = "Reported Measles Cases in United States, 1980--2003")
3.5
aids <- read.table("http://www.medepi.net/data/aids.txt", sep="",
header = TRUE, na.strings=".")
hepb <- read.table("http://www.medepi.net/data/hepb.txt", sep="",
header = TRUE)
matplot(hepb$year, cbind(hepb$cases,aids$cases),
type = "l", lwd = 2, xlab = "Year", ylab = "Cases",
main = "Reported cases of Hepatitis B and AIDS,
United States, 1980-2003")
legend(1980, 100000, legend = c("Hepatitis B", "AIDS"),
lwd = 2, lty = 1:2, col = 1:2)
3.6 Answer to (a):
edat <- read.table("http://www.medepi.net/data/evans.txt",
header = TRUE, sep="")
str(edat)
#
table(edat$chd)
edat$chd2 <- factor(edat$chd, levels = 0:1,
labels = c("No", "Yes"))
table(edat$chd2)
#
table(edat$cat)
edat$cat2 <- factor(edat$cat, levels = 0:1,
labels = c("Normal", "High"))

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Solutions 283

table(edat$cat2)
#
table(edat$smk)
edat$smk2 <- factor(edat$smk, levels = 0:1,
labels = c("Never", "Ever"))
table(edat$smk2)
#
table(edat$ecg)
edat$ecg2 <- factor(edat$ecg, levels = 0:1,
labels = c("Normal", "Abnormal"))
table(edat$ecg2)
#
table(edat$hpt)
edat$hpt2 <- factor(edat$hpt, levels = 0:1,
labels = c("No", "Yes"))
table(edat$hpt2)
Answer to (b):
quantile(edat$age)
edat$age4 <- cut(edat$age, quantile(edat$age),
right = FALSE, include.lowest = TRUE)
table(edat$age4)
Answer to (c):
hptnew <- rep(NA, nrow(edat))
normal <- edat$sbp<120 & edat$dbp<80
hptnew[normal] <- 1
prehyp <- (edat$sbp>=120 & edat$sbp<140) |
(edat$dbp>=80 & edat$dbp<90)
hptnew[prehyp] <- 2
stage1 <- (edat$sbp>=140 & edat$sbp<160) |
(edat$dbp>=90 & edat$dbp<100)
hptnew[stage1] <- 3
stage2 <- edat$sbp>=160 | edat$dbp>=100
hptnew[stage2] <- 4
edat$hpt4 <- factor(hptnew, levels=1:4,
labels=c("Normal", "PreHTN", "HTN.Stage1", "HTN.Stage2"))
table(edat$hpt4)
Answer to (d):
table("Old HTN"=edat$hpt2, "New HTN"=edat$hpt4)
3.7
wdat <- read.table("http://www.medepi.net/data/wnv/wnv2004raw.txt",
header=TRUE, sep=",", as.is=TRUE, na.strings=c(".","Unknown"))

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

284 Solutions

str(wdat)
wdat$date.onset2 <- as.Date(wdat$date.onset, format="%m/%d/%Y")
wdat$date.tested2 <- as.Date(wdat$date.tested, format="%m/%d/%Y")
write.table(wdat, "c:/temp/wnvdat.txt", sep=",", row.names=FALSE)
3.8 See Appendix A.2 on page 244 for Oswego data dictionary.
a. Using RStudio plot the cases by time of onset of illness (include appropriate
labels and title). What does this graph tell you? (Hint: Process the text
data and then use the hist function.)
Plotting an epidemic curve with this data has special challenges because we
have dates and times to process. To do this in R, we will create date objects
that contain both the date and time for each primary event of interest: meal
time, and onset time of illness. From this we can plot the distribution of
onset times (epidemic curve). An epidemic curve is the distribution of illness
onset times and can be displayed with a histogram. First, carefully study the
Oswego data set at http://www.medepi.net/data/oswego.txt. We need to
do some data preparation in order to work with dates and times. Our initial
goal is to get the date/time data to a form that can be passed to R’s strptime
function for conversion in a date-time R object. To construct the following
curve, study, and implement the R code that follows:
odat <- read.table("http://www.medepi.net/data/oswego.txt",
sep = "", header = TRUE, na.strings = ".")
str(odat)
head(odat)
## create vector with meal date and time
mdt <- paste("4/18/1940", odat$meal.time)
## convert into standard date and time
meal.dt <- strptime(mdt, "%m/%d/%Y %I:%M %p")
## create vector with onset date and time
odt <- paste(paste(odat$onset.date,"/1940",sep = ""), odat$onset.time)
## convert into standard date and time
onset.dt <- strptime(odt, "%m/%d/%Y %I:%M %p")
hist(onset.dt, breaks = 30, freq = TRUE)
b. Are there any cases for which the times of onset are inconsistent with the
general experience? How might they be explained?
Now that we have our data frame in R, we can identify those subjects that
correspond to minimum and maximum onset times. We will implement R
code that can be interpreted as “which positions in vector Y correspond to
the minimum values in Y?” We then use these position numbers to indexing
the corresponding rows in the data frame.
##Generate logical vectors and identify ’which’ position
min.obs.pos <- which(onset.dt==min(onset.dt,na.rm=T))
min.obs.pos

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Solutions 285

max.obs.pos <- which(onset.dt==max(onset.dt,na.rm=T))

max.obs.pos
##index data frame to display outliers
odat[min.obs.pos,]
odat[max.obs.pos,]

c. How could the data be sorted by illness onset times?

We can sort the data frame based values of one or more fields. Suppose
we want to sort on illness status and illness onset times. We will use our
onset.times vector we created earlier; however, we will need to convert it to
“continuous time” in seconds to sort this vector. Study and implement the
R code below.
onset.ct <- as.POSIXct(onset.dt)
odat2 <- odat[order(odat$ill, onset.ct), ]
odat2
d. Where possible, calculate incubation periods and illustrate their distribu-
tion with an appropriate graph. Use the truehist function in the MASS
package. Determine the mean, median, and range of the incubation period.

##Calculate incubation periods

incub.dt <- onset.dt - meal.dt
library(MASS) #load MASS package
truehist(as.numeric(incub.dt), nbins = 7, prob = FALSE,
col = "skyblue", xlab = "Incubation Period (hours)")

##Calculate mean, median, range; remember to remove NAs

mean(incub.dt, na.rm = TRUE)
median(incub.dt, na.rm = TRUE)
range(incub.dt, na.rm = TRUE)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)

References

1. Centers for Disease Control and Prevention. Antiretroviral postexposure prophylaxis

after sexual, injection-drug use, or other nonoccupational exposure to HIV in the
United States: recommendations from the U.S. Department of Health and Human
Services. MMWR Recomm Rep. 2005 Jan;54(RR-2):1–20. Available from: http:
//www.cdc.gov/mmwr/preview/mmwrhtml/rr5402a1.htm.
2. Olsen SJ, Chang HL, Cheung TYY, Tang AFY, Fisk TL, Ooi SPL, et al. Trans-
mission of the severe acute respiratory syndrome on aircraft. N Engl J Med. 2003
Dec;349(25):2416–2422. Available from: http://dx.doi.org/10.1056/NEJMoa031349.
3. Rothman KJ. Epidemiology: An Introduction. 1st ed. Oxford University Press, USA;
2002.
4. Aragón TJ, Lichtensztajn DY, Katcher BS, Reiter R, Katz MH. Calculating ex-
pected years of life lost for assessing local ethnic disparities in causes of premature
death. BMC Public Health. 2008;8:116. Available from: http://dx.doi.org/10.1186/
1471-2458-8-116.
5. Glass RI, Svennerholm AM, Stoll BJ, Khan MR, Hossain KM, Huq MI, et al. Pro-
tection against cholera in breast-fed children by antibodies in breast milk. N Engl
J Med. 1983 Jun;308(23):1389–1392. Available from: http://dx.doi.org/10.1056/
NEJM198306093082304.
6. Jewell NP. Statistics for Epidemiology. 1st ed. Chapman and Hall/CRC; 2003.
7. Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. Third edition ed. Lip-
pincott Williams & Wilkins; 2008.
8. Centers for Disease Control & Prevention. Prevention and control of meningococ-
cal disease. Recommendations of the Advisory Committee on Immunization Prac-
tices (ACIP). MMWR Recomm Rep. 2005 May;54(RR-7):1–21. Available from:
http://www.cdc.gov/mmwr/PDF/rr/rr5407.pdf.
9. Selvin S. Statistical Tools for Epidemiologic Research. 1st ed. Oxford University Press,
USA; 2011.
10. Selvin S. Survival Analysis for Epidemiologic and Medical Research (Practical Guides
to Biostatistics and Epidemiology). 1st ed. Cambridge University Press; 2008.
11. Selvin S. Statistical Analysis of Epidemiologic Data (Monographs in Epidemiology
and Biostatistics). 3rd ed. Oxford University Press, USA; 2004.
12. Selvin S. Epidemiologic Analysis: A Case-Oriented Approach. 1st ed. Oxford Univer-
sity Press, USA; 2001.
13. Abrams B, Carmichael S, Selvin S. Factors associated with the pattern of maternal
weight gain during pregnancy. Obstet Gynecol. 1995 Aug;86(2):170–176.
14. Dalgaard P. Introductory Statistics with R (Statistics and Computing). 2nd ed.
Springer; 2008.

287
288 References

15. Kleinbaum DG, Klein M, Pryor ER. Logistic Regression: A self-learning text. 2nd ed.
Springer; 2002.

Applied Epidemiology Using R 25-Nov-2012 c Tomás J. Aragón (www.medepi.com)