ECC 802– Educational Statistics

INTRODUCTION TO THE UNIT

Module Objectives

By the end of this module, you should be able to:

 Appreciate the value of effective descriptive statistics for the teacher when
presenting results for his students.
 Produce a variety of descriptive statistics designed to describe and summarize
data.

There are many ways of presenting and interpreting data from social sciences
investigations. All of them involve the use of statistics, and this topic is about helping to
make you familiar with a range of statistical techniques that can be used to describe and
analyse data.

Many students that are not mathematical oriented approach this course with a fair amount
of fear and misgivings. Hence, this topic will also help you realise the following:
 Statistics is a very useful tool, which can help you describe and make sense
of the data you collect in your practical work and particularly in your
teaching profession
 It is not necessary for you to be a mathematical genius in order to use a wide
range of statistical procedures. The basic techniques for displaying your data
are straightforward and easily used. Most of them you will already be using
them at school and in everyday life.
Statistics that are used to analyze and assess your data in school do not involve
complicated mathematical procedures of committing long, involved formulae to memory.

The first half of this module basically presents the student teacher with basic descriptive
statistics essential for handling and interpreting the students’ scores at any level of
education.

The second half is evaluation, and here issues in testing, including basic skills in test
construction, scoring and interpretation, are discussed.

Remember that descriptive statistics aims to describe or summarise the data you may
have collected in a practical exercise. If descriptive statistics are used well, they will
present your results clearly and in a way that can be readily understood by anybody
reading your work. The fig. 1 shows the range of descriptive statistics that will be
discussed in this module.
Tables measures Graphs
Frequency tables Measures of Central tendency Bar charts
(mode, median,and mean)
Tables of raw scores Measures of dispersion/variability Histogram
(range, mean-deviation, Variance
and standard deviation
Tables of summary scores Measures of relationship Pie charts
Frequencydistribution tables Frequency polygons
Tables of ranked scores Scattergrams
Fig. 1: Kinds of Descriptive Statistics.

It will be helpful to introduce you to some of the symbols that you are likely to meet in
the module from time to time.

Some of the symbols used in the module

 = Summation / add together
N = Number of items/ score/ cases (customary N is used to show total number in the
population as n is used for total number in the sample)
F = Frequency
D = Difference __
x= Deviation score from the mean =X  X)
X = Raw scores
X = Mean, pronounced as bar X
f = N = Total frequencies, total number of items
Example:
X /N or X/f means add all the raw scores and divide by the number of scores
rxy = Pearson product moment correlation coefficient.
rho = Spearman rank order correlation coefficient
rxx = Reliability coefficient
CONTENT
INTRODUCTION.............................................................................................................1
LECTURE 1 DEFINITIONS OF CONCEPTS..............................................................6
LECTURE 2 MEASUREMENT SCALES (LEVELS)..................................................9
2:1 Definition of measurement..................................................................................9
2:2 Scales (levels) of measurement............................................................................9
2:2:1Nominal scale of measurement..............................................................................10
2:2:2 Ordinal scale of measurement..............................................................................10
2:2:3Interval scale of measurement...............................................................................11
2:2:4Ratio scale of measurement....................................................................................11
LECTURE 3 TABULATION AND GRAPHICAL REPRESANTATION OF DATA
...........................................................................................................................................13
3:1Data organization.......................................................................................................13
3:2 Graphical Representation of Frequency Distribution...........................................16
4:4 Choice of Graph.......................................................................................................22
3:5Shapes of Frequency Distributions...........................................................................22
LECTURE 4 Measures of Central Tendency...............................................................26
4:1The mode (Mo)............................................................................................................26
4:2 The Median(Mdn).....................................................................................................29
4:3The Mean (X)..............................................................................................................34
4:4 Comparison of the measures of central tendency...................................................38
4:5Choice of the measure of central tendency..............................................................39
LECTURE 5 MEASURES OF VARIABILITY...........................................................42
5:1 The range.............................................................................................................42
5:2 The Quartile deviation (Q)................................................................................43
5:4 The mean deviation (MD)..................................................................................44
5:4 Variance (S2)......................................................................................................45
5:5 Standard deviation (Sd)............................................................................................46
LECTURE 6 Score interpretation – measures of Relative standing..........................51
6:1 Percentiles..........................................................................................................52
6:2 Standard Scores..................................................................................................58
6:3 Standardized scores............................................................................................60
6:3 Normalized standard scores..............................................................................61
LECTURE 7 THE NORMAL DISTRIBUTION..........................................................64
7:1 Characteristics of the normal curve.................................................................64
7:2 Simple application of the normal curve...........................................................65
LECTURE 8.....................................................................................................................69
8:1 Definition.............................................................................................................69
8:2 Methods of Estimating Correlation.................................................................70
8:2:1Logical Examination...............................................................................................70
8:2:2Scatter Diagram......................................................................................................71
8:2:3Pearson’s product-moment correlation coefficient (rxy)...................................74
8:2:4The Spearman’s (rho) correlation coefficient.......................................................77
8:3 Interpretation of correlation coefficient index................................................80
8:3:1Coefficient of determination( r2)............................................................................80
8:4 Significance Test..........................................................................................................................80
LECTURE 9 REGRESSION ANALYSIS.....................................................................83
 LECTURE 1
DEFINITIONS OF CONCEPTS

Objectives:
By the end of this lecture you should be conversant with the basic concepts commonly
used in statistics.

 Statistics: a variety of methods and techniques that are used for organizing,
summarizing, analyzing and interpreting data.
There are two main branches of statistics:
(i) Descriptive statistics: a group of methods and technique that attempt to summarize
and describe the data that have been collected e.g. calculating the mean or standard
deviation, producing a histogram or frequency polygon to describe a distribution.
(ii) Inferential statistics: a group of methods or techniques that allow us to infer
characteristics of a larger group from the data collected from a smaller group.
They help us to draw conclusions-make inferences-about the data collected in a
study.
 Population: a complete set of individuals cases or objects with some common
observable characteristics e.g. all primary school children in Kenya, all married
women above the age of 20 years, etc.
 Sample: a smaller group of individuals, cases or objects obtained from the accessible
population. A subject of the population. A sample must be carefully selected so as to
be a representative of the whole population with the relevant characteristics. Each
member or case in the sample is referred to as aSample
subject.
subset of
Population: the the
complete
Population
Set (parameter)
(statistics).

Fig. 1(a): The relationship between the population and the sample

 Statistic-sample measures e.g. sample mean, sample variance.

 Parameters- population measures e.g. population mean, population variance

We estimate population parameters using sample statistics. The two are differentiated by
the symbols used. Roman symbols are used in sample statistics while Greek symbols are
used to represent population parameters e.g.
statistic Sample statistic Population
symbol parameter symbol
Mean X 
Standard deviation Sd 

 Data: all the information a researcher gathers for his study.

 Primary data- information a researcher obtains from the field i.e. from the
subjects in the sample.
 Secondary data- information a researcher obtains from research articles, books,
casual interviews, etc.
 Quantitative data- numerical
 Qualitative data- words or phrases
 Parametric data – data that is based on the following assumptions: must be at
interval or ratio scale of measurement; its drawn from a population of scores
which is normally distributed – assumes normality: if two or more sets of data
are being analyzed, they ought to have been drawn from populations of scores
which have similar variances – homogeneity of variances
 Non – parametric data - data that is not based on the above sophisticated
mathematical properties of interval or ratio scales, normality, or equal variance.
Measurement of different variables will yield different types of data (measurement
scales – Lecture 2).

 Variable: any characteristic in individuals, objects or events that can assume

different values
There are different classifications of variables, but for the purpose of this topic, we
differentiate among discrete and continuous variables, quantitative and qualitative
variables, categorical variables, independent and dependent variables.
 Discrete variables: those variables that increase or decrease by whole numbers
and not by fractional amounts. These are variables that come in whole units and
have a finite number of values between any two points. Families, for example,
cannot have half a child. They have 1, 2, or more children in their family and
only one value can exist between the values 1 and 3. This value is two children.
Therefore, any value that cannot be conceptualized as less than a whole unit must
be treated s a discrete variable.
 Continuous variables: those variables that can theoretically assume an infinite
number of values between any two points. Therefore, continuous variables can
assume numbers that represent any fraction of a whole number e.g. height. We
could measure someone’s height in inches. However, if we wanted to be more
precise, we could measure that person’s height to the nearest hundredth or
thousandth of an inch. The important point to remember about measuring
continuous variables is that the measurement will always be approximate. We
can never provide an exact measurement of a continuous variable because we can
always develop a more precise measurement.
  Qualitative variables: those variables where actual numerical measurement is not
possible e.g. gender, blood group, color of the skin, etc.
 Quantitative variables: those variables where actual and accurate measurement is
possible using appropriate tools e.g. weight, height, time, etc.
 Categorical variable: variables that are distinctly classified or grouped e.g.
categories of patients observed in the hospital may be: malaria patients,
pneumonia patients, common cold patients.
 Independent variables: predictor variables
 Dependent variable: Outcome variables (see full definition for these two

Activity:
Differentiate between each of the following concepts:
1. descriptive versus inferential statistics,
2. sample versus population,
3. statistic versus statistics,
4. statistic versus parameter,
5. continuos versus discrete variable,
6. qualitative versus quantitative variable,
7. independent versus dependent variable,
8. Parametric versus non-parametric data.

Further Readings
1. Ingule, F. and Gatumu, H. Essentials of Educational Statistics. Nairobi: East African Educational
Publishers 1996
2. Kerlinger, F N. Foundations of Behavioural Research. 2nd Edition. New York: Holt, Rinehart and
Winston Inc. 1973
3. Malim, T. and Birch, A. Research Methods: London,Macmillan Press 1997
4. Mugenda O.M. and Mugenda A.G. Research Methods, Quantitative and Qualitative Approaches.
Nairobi: Acts Press 1999.
5. Siegel, S. and N. John C. Non-Parametric Statistics for the Behavioural Sciences.2nd Ed. New York:Mc
Graw Hill 1988
 LECTURE 2
MEASUREMENT SCALES (LEVELS)

Objectives

By the end of this topic, you should be able to:

 Define measurement
 Explain why measurement is necessary in social research
 Describe the scales of measurement

2:1 Definition of measurement

Measurement refers to the assignment of numbers to objects or events or individuals

according to a specific set if rules. It is a process we all engage in e.g. any time you
determine how tall you are or how many books you have or what position the child was
you engage in measurement because you are assigning numbers according to a given set
of rules

? Why do we measure?

Generally we measure to account for the things we observe. In the physical sciences we
measure mainly to answer the question “How much?” in other words, to quantify things.
In the social sciences we use numbers (measure) for essentially three purposes:
 To classify
 To order and
 To quantify things

2:2 Scales (levels) of measurement

When carrying out measurement of physical characteristics such as height or length, we
do not stop to think because the rules for assigning the numbers here are standardized,
known, and used in the same manner by everyone. However when we turn to measuring
many of the variables of interest to social and behavioral scientists, we find that
standardized measurement procedures do not exist. For example, in the behavioral and
social sciences we deal with variables such as personality, attitudes, aggressiveness in
children and whether people do or do not vote. Such variables have properties and vary
just as height and weight vary from time to time. To communicate these measures we
must measure them as precisely as possible. This means that we must assign them a
number according to a given set of rules. However the type of information that is
communicated and the arithmetic operations that can be calculated on the numbers
assigned to such variables depend on the scale of measurement being used.
Four different scales of measurement can be identified:
Nominal, Ordinal, Interval and Ratio
Each scale represents a different way of assigning numbers, communicates a different
type of information and allows us to use different arithmetic operations.

The numbers in the above scales could have the properties of magnitude, equal interval
or absolute zero.

2:2:1 Nominal scale of measurement

The nominal scale of measurement is a scale in which numbers are used to label,
classify, or identify the people or objects of interest. Numbers in this scale have none of
the above properties (magnitude, equal interval or absolute zero). They have the property
of equivalence. They have no numerical meaning of what we know numbers are hence
any mathematical operations like addition, subtraction, multiplication, divisions are not
permissible with numbers at this level of measurement. The only arithmetic operation
that can be performed with nominal scale data is counting.
Example 1: If you were interested in measuring smoking behavior of some group of
people, your first impulse might be to assign people to the two categories: those who
smoke and those who do not smoke. If you give smokers the label 01 and non-smokers
the label 02, you are using numbers at the nominal scale of measurement.
Example 2: You are interested in finding out gender differences in mathematics
performance. For the variable gender, a label 01 is assigned to boys and 02 to girls.

Activity: Give other examples where numbers are assigned at the nominal scale level of
measurement.

Answer: religious affiliation, race, blood group, tribe, nationality, makes of cars, subjects offered etc.

Note: The nominal scale of measurement is the simplest. It is useful in the social sciences
because some observations exist only at this level.

2:2:2 Ordinal scale of measurement

This is a rank order scale. An ordinal scale of measurement with which most of us are
familiar with is the class-positions of pupils in primary or high school showing the order
of performance in class from the best (position 1) to the (position 40). The ordinal
measurement includes information about numbers in each category and adds to this a
rank order.
Ordinal scale of measurement allows you to make ordinal judgements; that is, it allows
you to determine which is better or worse than the other. However, an ordinal scale of
measurement does not allow you to assess anything other than rank ordering of the
variables.

Example:
You may want to know how much better a pupil in position 5 performed than a pupil in
position 6, or whether the difference between the performance of pupils in position 5 and
6 is the same as the difference in performance between pupils in position 6 and 7. To
make such an assessment you would have to know the difference in scores among these
pupils. In other words you would have to be able to identify the difference between the
pupils in position 5 and 6 and do the same for pupils in position 6 and 7. An ordinal scale
does not provide such information; it only gives you the rank order of the position.
Numbers at the ordinal scale of measurement have the property of magnitude but lack the
property of equal intervals and absolute zero. Still, they do not have the numerical
meaning of numbers. We know hence that addition, subtraction, multiplication and
division are not permitted. Allowed functions at this level of measurement are greater
than (>), less than (<), and equal to (=).

2:2:3 Interval scale of measurement

The interval scale of measurement has equal interval distance between adjacent numbers
ass well as ordinality. Therefore, with an interval scale of measurement you gain ability
to specify the distance that exists between people, objects, or events of interest for the
variable being measured. Numbers used at the interval scale for the first time have
numerical meaning of what we know numbers to be. They have the properties of
magnitude and equal intervals between adjacent numbers. However, they lack the
property of zero. The zero point on the interval scale is arbitrary. Consequently, it doesn ’t
represent an absence of something as the term zero denotes.

Example:
The Celsius temperature scale represents an interval scale of measurement because all the
points on the scale are equal distance from each other. In other words, the difference in
temperature between 0c and 50c is the same as the difference between 50c and 100c.
In addition, the points can be rank ordered from the highest to the lowest (ordinality). The
zero point on the Celsius scale denotes the point at which water freezes at sea level, but
an absolute zero point with respect to temperature refers to an absence of heat. This point
has been determined to be approximately -273c not the 0c. Therefore, zero points on
interval scales of measurement represent arbitrarily defined points and do not refer to an
absence of an attribute.

At the interval level of measurement, numbers can be added, subtracted, multiplied and
divided. However, due to the absence of zero, ratio statements are prohibited. E.g. if the
temperatures of two objects were taken in Celsius scale such that A was 50c and B was
100c, we cannot conclude that B has twice as much heat as A. A simple conversion of
these temperatures to another scale like Fahrenheit will show that the ratio of B: A is not
2: 1. Other examples of measurement at the interval scale include marks students obtain
in tests and IQ scores.

2:2:4 Ratio scale of measurement

The ratio scale of measurement allows us to make ratio as well as ordinal statements
because it has equal intervals between adjacent points on the scale and an absolute zero
point.
The ratio scale of measurement represents the strongest and most desirable scale of
measurement.
Numbers in this scale of measurement can be used to identify, order and quantify. All
mathematical operations can be carried out with numbers at this level of measurement.
Zero means total absence of the property measured. In other words, the zero point is real
and fixed. In this scale, therefore, it is meaningful to say that A has three or four times ass
much property as B. this is not possible with interval data.
Examples: Weight and height

Summary on scales or levels of measurement

We have looked at what measurement is and why we measure in the social sciences, the
four scales/levels of measurement with their properties each level building on the
characteristics of the preceding one. That is, each succeeding level permits us to represent
more information about the world.

Activity:
Identify the scale (or level) of measurement and type of variable (discrete or continous,
Qualitative or quantitative) of each of the following:
a) Registration numbers of the students of open learning.
b) Number of women registered for the examinations.
c) Types of vehicles driven by the members of parliament.
d) Categories of patients admitted in a hospital if classified according to the seriousness
of the disease being suffered from.
e) Proficiency in Kiswahili speaking if graded on a scale of 0-100 but with intervals
within NOT known.
f) Leadership qualities of aspiring candidates as rated by the audience after listening to
their speeches.
g) Job groups of teachers on staff.
h) Classification of the animal kingdom.
i) Social class
j) Amount of faith Christans have in a given church

Answers: a) nominal, b) ratio, c) nominal, d) ratio, e) ordinal, f) ordinal, g) ordinal, h) nominal, I i)

nominal, j) ordinal

Further Readings
6. Ingule, F. and Gatumu, H. Essentials of Educational Statistics. Nairobi: East African Educational
Publishers 1996
7. Kerlinger, F N. Foundations of Behavioural Research. 2nd Edition. New York: Holt, Rinehart and
Winston Inc. 1973
8. Malim, T. and Birch, A. Research Methods: London,Macmillan Press 1997
9. Mugenda O.M. and Mugenda A.G. Research Methods, Quantitative and Qualitative Approaches.
Nairobi: Acts Press 1999.
10. Siegel, S. and N. John C. Non-Parametric Statistics for the Behavioural Sciences.2nd Ed. New York:Mc
Graw Hill 1988
 LECTURE 3
TABULATION AND GRAPHICAL REPRESANTATION OF DATA

Objectives
By the end of this lecture you should be able to:
 Organize data in frequency distributions, both ungrouped and grouped.
 Present data using different types of graphs, histograms, bar graphs, frequency
polygons and cumulative frequency curves.
 Select the most appropriate type of graph for her data.
 Describe different shapes of distributions, normal, skewed, kurtosis, multi-modal etc.

3:1 Data organization

When given a list of raw data it is necessary to give them some order so that you can
make meaning out of them

Example: below is a list of raw scores obtained from 20 students.

72 78 68 69 82 81 76 64 73 74 66 72 70 75 76 80 72 70 83 66
? How can you organize these data to make it meaningful?

(i) Rank Order Distribution

This is convenient when N  20. Here the scores are arranged in descending order of
magnitude. The arrangement enables us to see clearly the position of the scores relative to
others. For the data above a rank distribution will be as shown in Table 3:1 below.

Table 3:1 A rank order distribution

Participant Position Score
1. 83
2. 82
3. 81
4. 80
5. 78
6. 76
6. 76
8. 75
9. 74
10. 73
11. 72
11. 72
11. 72
14. 70
14. 70
16. 69
17. 68
18. 66
18. 66
20. 64
With this kind of arrangement we are able to see clearly the position of the score relative
to others straight away.

(ii) Frequency Distribution Tables

When the entries are many, more than 20 and are repeated, then a list of individual marks
as in rank distribution becomes cumbersome. Here we use a table that simply illustrates
the number of times a particular event occurred. This is known as a frequency
distribution table.

Example 1: Given the following list of marks:

10 3 3 9 8 8 7 7 6 5 4 5 7 6 5 6 4 8 7 8 9 7 6 7 9 7 9 7 8 8 8 7 8 6 8 6
7 9
Frequency distribution tables give a record of the number of times a score or a response
occurs. For the above data:

Table 3:2 Frequency distribution of scores

Score Tallies Frequencies Cumm. Cum
Freq Freq
(x) (f) (cf) (cf)
10 / 1 38 100.00
9 //// 5 37 97.37%
8 //// //// 9 32 84.21%
7 //// //// 10 23 60.53%
6 //// / 6 13 34.21%
5 /// 3 7 18.42%
4 // 2 4 10.53%
3 // 2 2 5.26%
f=38


Activity: Make a frequency distribution for the following list of scores:

7 8 6 5 2 10 9 3 2 1 6 7 5 4 5 0 2 5 6 4 3 4 6 8 5 7 4 3 5 3 4 9 5 7 6 5
3 1 N = 38

The above arrangement of scores is referred to as Frequency Distribution of Individual

Scores or for Ungrouped Data

(iii) Grouped Frequency Distribution Tables

When the range of scores is greater than 15 (>15), an individual frequency distribution of
scores becomes too long and cumbersome or difficult to interpret. In such cases there is
need to group the scores. Other reasons why we may need to group scores are:
 When the scores are distributed in such a way that certain scores are not
obtained by any subjects and

 When information being sought is sensitive, such as annual income, and the
response categories in the questionnaire are given in intervals.

To group data you need to determine the class interval (ci).

Steps to Determining the Class Interval

As a rule in a given distribution, number of classes should be > 10  20. Based on this:

Step 1: Determine the inclusive range of your data: Highest Score  Lowest Score + 1
For instance when given the following data:
30 60 25 45 78 36 50 60 75 40 54 85 80 36 48 50 90 60 32

For this data inclusive range = 90  25 + 1 = 66

Step 2: Divide the inclusive range by 12. This gives the largest possible class interval
e.g. 66  12 = 5.5, round this value down to 5.

Divide the inclusive range by 15 This gives the smallest possible class interval 66 15 =
4.4, round this one up to give 5.

Step 3: Using the two values obtained in step 2 decide on the class interval. For our data
it is 5.
Note:
i) There is no fixed rule of choice of step or class interval. But for the purpose of
convenient calculations, it is usually desirable to have odd numbered rather than
even numbered class intervals, because as odd numbers they yield a class
midpoint which is a whole number rather than one ending in 0.5.
ii) It is also desirable to choose as the beginning of the first step in a frequency table
a number which is a multiple of the step interval. This is simply to make your
work easier.
A complete grouped frequency distribution will be as shown in Table 3:3
Table 3:3 A complete grouped frequency distribution of scores
Class Actual Limits Mid pts Tallies Fre C %CF
q F
90-945 89.5-94.5 92 / 1 20 100%
85-89 84.5-89.5 87 / 1 19 95%
80-84 79.5-84.5 82 / 1 18 90%
75-79 74.5-79.5 77 // 2 17 85%
70-74 69.5-74.5 72 - 0 15 75%
65-69 64.5-69.5 67 - 0 15 75%
60-64 59.5-64.5 62 //// 4 15 75%
55-59 54.5-59.5 57 - 0 11 55%
50-54 49.5-54.5 52 /// 3 11 55%
45-49 44.5-49.5 47 // 2 8 40%
40-44 39.5-44.5 42 / 1 6 30%
35-39 34.5-39.5 37 // 2 5 25%
30-34 29.5-34.5 32 // 2 3 15%
25-29 24.5-29.5 27 / 1 1 5
%
Note: the above actual limits (exact limits) column is only relevant for data that is
continuous.

Activity:
Given the following data :
61 60 50 52 58 60 51 55 68 55 62 47 39 58 42 47 42 48 48 46 55 51 58 65
45 35 43 54 52 56 46 53 34 48 50 39 59 53 52

a) Determine a convenient class interval for the data..

b) Make a grouped frequency distribution table showing the seven columns for the data.
Answer: suitable ci = 3.

3:2 Graphical Representation of Frequency Distribution

A graph is often a more effective method of clarifying a point. Graphs enable the reader
to see the trend of the distribution more easily than is possible by simply looking at
numbers in a frequency distribution.

A graph has two axes, a horizontal axis and frequencies along the vertical axis. The
intersection of the two axes represents the zero point on both the horizontal and vertical
scales.
Note:
i) To avoid a graph being too flat or too steep, it is advisable to arrange scales on the
vertical and horizontal axis so that the width of the graph is about one and two
thirds times its height, that is a ratio of 3:5.
ii) A graph should be well labelled on both the vertical and horizontal axis.
iii) A graph should also have a title.

There are three common types of graphs commonly used to represent data in research
reports. These are:
 Histograms
 Frequency Polygons and
 Bar Charts

1. Histograms
A histogram is the “Sky Line” of the figure formed by erecting a rectangular column on
each step in a way that the width of each column is equal to the class interval and the
height is equal to the frequency. The rectangular should be constructed in such a way
that they start from the lower limit to the upper limit of the class. The column is centred
on the mid-point of the score class interval.

Class interval (scores) on the X-axis and frequencies on the Y-axis.

Example:
Table 3:4 Distribution of scores from a maths test

Scores Mid-points (X) Frequencies (f)

33-35 34 2
30-32 31 4
27-29 28 4
24-26 25 8
21-23 22 8
18-20 19 5
15-17 16 2
12-14 13 1

f= 34

To convert the data from Table 3:4 into a histogram, we do the following:
 Draw the vertical and horizontal axis
 Mark the vertical axis with numbers to represent the frequency of the
scores
 Draw columns above each score to indicate the number of times it occurs
 Ensure that all parts of the histograms are clearly labelled and that there is
an overall title.

Fig. 3(a): Histogram derived from scores in Table 3:4

It is now possible to see the trend taken by the data at a glance.
Histograms are used to present data tat is continuous at the interval and ratio scales of
measurement.

2. Frequency Polygons
Another way of graphically representing frequency distributions is by use of frequency
polygons. A polygon is a many sided closed figure. In plotting frequency polygons
points are marked at the points of interaction between mid points of the class and the
corresponding frequency. The points are then joined using straight lines. Hence to draw
a frequency polygon:

 Draw vertical and horizontal axis

 Mark the vertical axis with numbers to represent the frequency of
scores
 Mark the horizontal axis in sections to represent the scores.
 Mark the point of intersection between the frequencies and the
corresponding mid points – A point above the mid point corresponding
to the frequency.
 Join the points using straight lines.
 Ensure that all parts of the frequency polygon are labelled and there is
an overall title.

Fig. 3(b) Frequency Polygon derived from the data in Table 10:3
Note: Rather than leave the polygon suspended about the x-axis, we anchor it at the
lower end and upper end by extending the line on both sides to the midpoints of the
previous/next class intervals at the end and top respectively as shown by the broken lines
(these sections are not part of the data)
Frequency polygons are especially useful where we want to compare two or more sets of
data, since we can plot them on the same pair of axis. This is not possible with
histograms.

Example: Given the following 2 data sets showing frequencies of scores on nonsense
syllables correctly recalled in experiments under two different conditions A and B.

Table 3:5 Frequency of scores (nonsense syllables correctly recalled).

Condition A (with the aid) Condition B (without the aid)

Score (x) Frequency (f) Score (x) Frequency (f)
5 - 5 1
6 - 6 1
7 – 7 1
8 – 8 2
9 2 9 3
10 2 10 4
11 3 11 2
12 5 12 3
13 4 13 2
14 3 14 1
15 1 15 -
 
N=20 N=20
Fig.3 (c) Frequency Polygon of the scores from conditions A (with the aid) and
B (without the aid) in the memory experiment.

The frequency polygon can also be super imposed onto the histogram as shown below in
Figure 3(d)

Fig. 3 (d) Histogram and Frequency Polygon derived from the data in Table 3:4

3. Bar Charts
Bar Charts are preferred when data is discrete or categorical. They are appropriate only
for nominal or ordinal scaled variables. They are used to illustrate the frequency of
observations in a given category. The Bar Chart is very much like a histogram it also has
a column graph, except that the bars do not touch. This is a characteristic of this graph.
Bars that touch would imply that there is some continuity between the various categories,
which in discrete or categorical data, there obviously is not. Therefore the bars in the bar
graph should be separated.

Example: The following table illustrates the enrolment of students in 8 primary schools
in Butula Division of Busia District Kenya.

Table 3:5 School Enrolment

School No. of Students % Frequency

Buhuyi 420 19.20

Tingolo 320 12.80
Siribo 220 8.80
Bumala 260 10.40
Emauko 128 5.12
Butula Boys 560 22.40
Butula Girls 532 21.21
 
N=2,500 100.00

Fig 3 (e) Bar Graph based on the data in Table 3:5

3:3 Choice of Graph

The choice of graph should depend on the following:

a) Data to be represented
If discrete variable- bar graph
If continuous- Histogram or frequency polygon.
b) Do we want to make comparisons among various distributions? If yes,
then frequency polygon is best, because several polygons can be
Superimposed with less crossing of lines. In this form, comparisons
Among various distributions can be made more easily.

If sample sizes are different and we want to compare the distributions,

It is important to change frequencies to relative frequencies, that is, in
Terms of proportions or percentages. This conversion should be done
Before graphing and used on the vertical axis (y) in place of
frequency. Frequency polygons also give a much better conception of the
trend of the distribution than histograms.

c) Histograms on the other hand are easier to understand especially when only one
distribution is being represented. However they give an impression that subjects fitting
within each interval are evenly distributed over the interval, which is not always the case.

3:4 Shapes of Frequency Distributions

Frequency distributions can occur in an unlimited number of shapes and forms. Since
form or shape is often an important bit of information about a distribution, it is useful to
know some of the descriptive terms used to indicate various types.

Figure 6:4 shows three most common shapes of frequency distributions. They all have
high frequencies in the centre of the distribution and low frequencies at the two extremes,
which are called the upper and lower tails of the distribution.

Fig. 3(f) Different Shapes of Frequency Distributions.

(i) Normal Distribution

- Symmetrical
- Bell Shaped
- Cone Peak
Mode = median =mean
(all located at the centre)
 (ii) Positively Skewed
-Asymmetrical-
Many low scores and a few
high scores.

Mode<median<mean

(iii) Negatively Skewed

-Asymmetrical

-Many high
scores and few low
scores

Mode>median>mean

The distribution in fig.3f (i) is symmetrical about the centre, its bell shaped and has a
cone like (rounded peak). The tails of the distribution continually approach but never
touch the horizontal axis. This is a normal distribution.

The two other distributions (ii) and (iii) are asymmetrical or skewed. A distribution is
skewed if the scores “ trail off” in one direction. A distribution is said to be skewed if it is
not symmetrical. However, a non-symmetrical distribution can be skewed positively or
negatively. A positively skewed distribution has many low scores and few high scores;
in other words, this distribution has the “ tail” in the positive (upper) end of the scale.
See fig.3f (ii)

A negatively skewed distribution has few low scores and many high scores; in other
words the distribution has the “tail” in the negative (lower) end of the scale. See fig3f
(iii).

All the above three distributions are unimodal i.e. they only have one peak. Fig.3 (g) i
show a bimodal frequency distribution, that is a distribution with two peaks. This is
occasionally seen and usually indicates that the data are a mixture of two separate
distributions. Also shown on fig 3(g) ii, iii and iv are three other distributions that are
sometimes found; the reverse J-Shaped, the rectangular or uniform distributions and a V-
Shaped distribution.
Fig 3 (g) i Bimodal e.g. hormone Fig 3(g) ii Reversed J-Shaped e.g
levels of males and females – it survival time after diagnosis of lung
reflects two different categories or cancer
data points that occur most frequently

Fig. 3 (g) iii Rectangular e.g. Fig 3 (g) iv U-Shaped distribution e.g.
organised group work scores, each
score is obtained by same number of
people

iv Kurtosis

This is another slightly different way of describing the shapes of frequency distributions.
Kurtosis is the flatness or peakedness of a distribution in relation to the normal curves’
peak. Description in terms of Kurtosis; Leptokurtic; Mesokurtic and Platykurtic.

a) Leptokurtic: a distribution characterised by having most of its scores piled up

around the midpoint. Therefore represents a thin distribution. The range of scores or
variability is relatively small compared to the normal distribution- has a pointed
peak.
b) Platykurtic: distribution characterised by a flat curve (plateau), meaning that the
scores are more evenly distributed throughout the curve. Very close to rectangular
distribution- has a flat peak. Almost same number of students obtains every score, as
it may be the case of group work.

c) Mesokurtic: this one conforms to the ideal shape of the symmetrical distribution that
is, the normal curve- has cone shaped peak see fig.3 (h).

Fig. 10(h) Kurtosis.

Symmetrical versus non-symmetrical distributions.

Symmetrical and bell shaped Symmetrical but non-bell shaped Non-symmetrical
Normal /mesokurtic, Platykurtic/rectangular, bimodal, +ely and –vely
leptokurtic u-shaped skewed

Activity:
Given the following scores: 22, 27, 33, 26, 16, 25, 24, 31, 28, 28, 16, 34, 22, 24, 31, 30,
26, 18, 26, 27, 22, 20, 19, 21, 14, 19, 23, 23, 20, 25, 21, 21, 25, 23, 32, 17.
Make a grouped frequency distribution, using a class interval of 3 with the lowest step
interval as 12 – 14 (include actual limits, tallies, frequencies, midpoints, and the two
cumulative frequency columns)
Draw a histogram for the grouped data
Super-impose the frequency polygon over the histogram drawn in 2.
Describe the performance of the class.
Answer:Slightly positively skewed but close to normal. A little more on the poor performance.
 LECTURE 4
MEASURES OF CENTRAL TENDENCY

MEASURES OF CENTRAL TENDENCY

Objectives
By the end of this lecture you should be able to:
 Describe what measure of central tendency are.
 Describe the three commonly used measures of central tendency, mode, median and
mean
 Find the mode, calculate the median, and mean for a distribution of scores.
 Choose the most appropriate measure of central tendency for any type of data and
any distribution of scores.

Introduction

Graphs as seen in the preceding section provide convenient ways to summarise data. But
when there are many numbers to be considered and information is required quickly then
measures of central tendency may prove to be more useful.

Definition: Measures of Central Tendency describe the typical size of a set of scores.
They are single measures or scores – typically or representative score- which can
summarise a set of scores.
Three key measures of central tendency commonly used are: the mode, median and the
arithmetic mean
4: 1 The Mode (mo)
It the most frequently occurring value. It is useful to gain general impression of the
average where there is a large set of numbers, e.g.

(a) Mode for ungrouped data – the most frequent score

5 4 5 9 8 5 7 5 8 6 3 4 5 mode = 5

(b) When two adjacent scores have same frequency and this common frequency is
greater than that of any other score, the mode is the average of the two adjacent
scores e.g.

01 1 2 2 2 3 3 3 4 5 2 and 3 have a common frequency which is also highest in

the distribution but they are adjacent. It’s a unimodal distribution with the mode at

2 + 3/2 = 2.5

(c) If in a group of scores two non-adjacent scores have the same frequency and this
common frequency is greater than that of any other score, then two modes exist and
it’s a bimodal distribution. Thus:20, 21, 21, 21, 22, 22, 23, 24, 24, 24 and 27 both 21
and 24 are modes (non-adjacent)

Note:

(i) If 2 or more non-adjacent scores appear with equally greater frequency, the
distribution has mote than one mode.

(ii) Larger sets of scores are often referred to as bimodal when they present a
frequency polygon that looks like a bacterian (two humped camels back) even
though the frequencies at the two peaks are not strictly the same e.g.

Fig 4 (a) Fig 4 (b)

This is differentiated from the normal bimodal by specifying that it is a bi-modal
distribution with a minor mode at x 1 and a major mode at x2. The normal bimodal
distribution is simply described as “ a bimodal distribution at x1 and x2 fig 6:5 (b)

Implications of a Bimodal Distribution

2 distinct groups can be separated on the variable being observed e.g. the differences
could be as a result of religious background, cultural background or educational level, for
resources within a community.

(d) When all the scores in a group occur with the same frequency we say the distribution
has no mode. E.g. 7, 8, 9, 10, 11, 12.

(e) The mode for grouped data, the midpoint of the modal class is taken to be the mode
of the distribution. The step/class with the greatest frequency is taken to be the modal
step/class e.g.

Table 4:1 Mode for grouped data

Class Frequency
20-24 4

15-19 8

10-14 10 modal step= 10-14

5-9 6 mo= 12

0-4 2

Note: The following formula could also be used for calculating mode for grouped data:

Mo = Lmo + ( ) Ci

Where Mo = mode

Lmo = lower limit of the modal class

d1 = difference between the frequency of the modal class and the frequency of the
class immediately preceding it.

d2 = difference between the frequency of the modal class and the frequency of the
class immediately following it in the distribution

Ci = the class interval of the distribution.

 Fi = frequency within the modal class.

For the above data using the formula

Mo=9.5 + ( ) 3=

9.5+2.14= 11.64

Advantages of the Mode

 Its easily identified

 Can be used with any level of measurement including the nominal level of
measurement.

Disadvantages of the Mode

 Its not affected by scores outside the modal interval i.e. it does not indicate the
frequency of occurrence of any of the other scores in the distribution.

Uses of the Mode

 When the data is nominal, it’s the only measure of central tendency applicable.

 For multi-modal distributions.

4:2 The Median (mdn)

It is a measure of central tendency that divides a distribution of scores exactly in half so
that 50% of the scores fall below it and 50% fall above it. It is also known as the 50 th
percentile score, In other words, the median is a point in the distribution above which
50% of the scores fall.

Note; The scores must be considered when in order of magnitude.

Median for Ungrouped Data

Case 1; When n is odd the median will be at the place e.g.

5, 9, 21, 35, 40, n=5

Median will be at 5 + 1 place i.e. 3rd position. The score in this position 21-
2

The middle score, is the median

Case 2 ; When n is even, there are 2 middle scores e.g. 23,26, 29, 30,31, 32, 33, 35,
37,40, n = 10 The two middle scores are in the 5 th and 6th position i.e. 31 and 32 median
is the average of the two scores. 31 + 32 / 2 =31.5

Case 3: When several scores have the same value as the mid score

(a) if the repetition is symmetrical around the median score e.g.

6, 7, 8, 11, 11, 11, 23, 13, 15 we accept the mid score as our median i.e. 11.

(b) if the repetition is unsymmetrical around the median score e.g.

1, 2, 3, 4, 5, 5, 5, 6, 7, 8, 9, 10. The 6 th and 7th score averaged should give the median i.e.
5 +5/2=5

But 5 does not divide the distribution symmetrical into two. Hence it cannot be the
median.

To get the exact median for the data above follow the procedure for working out median
for grouped data taking the step interval of 1.

Table 4:2 Calculation of the median

Score (x) f cf

10 1 12

9 1 11

8 1 10

7 1 9

6 1 8

5 3 7

4 1 4

3 1 3

2 1 2

1 1 1
n = 12

There are 12 scores, therefore the median should be a point below which 50% of 12 = 6
of the scores fall. When we count we find the 6 th score falls within the 5’s. If the 5 is
taken as the median we observe 4 scores below 5 and 5 scores above 5 this is proof that 5
is not the median.

We therefore, assume that the three 5’s are evenly distributed within the interval 4.5 –
5.5. Hence the exact median for the distribution will be,
4.5 + (2/3) ci = 4.5 + (2/3) 1 =
4.5 + 0.67 = 5.17 ans

Example

Table 4:3 Median for grouped data

Class Exact Limits f cf

33-35 32.5-35.5 2 34

30-32 29.5-32.5 4 32

27-29 26.5-29.5 4 28

24-26 23.5-26.5 8 24

21-23 20.5-23.5 8 16

18-20 17.5-20.5 5 8

15-17 14.5-17.5 2 3

12-14 11.5-14.5 1 1
n = 34

a) by interpolation method

Step 1; find 50% of the cases i.e. n

2

In the case above 34/2 = 17

Hence, you need the point below which just 17 scores fall and it is that same point above
which 17 scores will fall.

Step 2; Count up from the bottom of the frequency column – or make use of the
cumulative frequency column and note the step interval within which the 17 th score will
fall. In our case above, it is the step 24 – 26, limits 23.5 –26.5.
We know that we have 16 cases below 23.5 and if all the cases in 23.5 – 26.5 are added
then we get 24. Since we only need 17 then this point should be.

Step 3: Lower limit of the class containing the17 th score + some correction. The
correction is determined by interpolation.

Having used 16 cases by reaching 23.5, we need only one case to complete our quota, in
other words we need 1/8 of the cases in 23.5 – 26.5 class. Since the step represents a
score range of 3 (i.e. the class interval is 3) it is logical to take 1/8 of 3 as our correction
to add on our 23.5.

Hence the exact median is:

= 23.5+(1/8)ci

=23.5+0.375=23.875

=23.88 ans

b) Use of Formula

Mdn= L + ( )Ci

Where L = the exact lower limit of the class interval containing the median

N = total number of cases in the distribution

Cfb = cummulative frequency below the interval containing the

Median.

fw = the frequency of scores within the interval containing the

median.

Ci = class interval of the distribution.

Using the formula to find the median of the above scores:

Mdn = L + ( ) Ci

= 23.5 + ( )3

= 23.5 + ( )3
 =23.5 + (1/8) 3

=23.88

Characteristics of the Median

The median is the desirable measure of central tendency in some instances because of
two characteristics:

 Its insensitivity to extreme scores

 Its insensitivity to missing scores

(a) Insensitivity to extreme scores

The median is unaffected by extreme scores (outliers) e.g. in the following array of
scores; 44, 46, 48, 49, 50, 52, 54 the median is 49. However if the last score in the array
were changed to 540 as follows; 44, 46, 48, 49, 50,52,540 the median would still be
unchanged. With such a figure incorporated into a measure that was trying to describe
the average score of that group of people, we would get a distorted figure one that was far
too high e.g. the mean would be 118.4, a value that is very unrepresentative of the group.
However, when the median is reported, as it typically is, we get a figure that tells us the
income that corresponds to the 50th percentile or the mid point of the scores of the group.

(b) Insensitivity to missing scores

The second desirable characteristic of the median is that It can be computed even if some
of the extreme scores are missing e.g. when compiling the weight of a group of ladies, we
might encounter some extremely heavy ladies who do not disclose their weight.
However, based on the knowledge of her size we could ascertain that her weight has to be
he highest weight among the group observed. Once we know on which side of the median
the missing score or scores lie, we can proceed to compute the median value. E.g. assume
that we asked 8 people in an assertiveness training group to give us their age. Six of
these people gave their ages as follows;

22, 26, 25, 20, 29, 28. And two people merely stated that they were more than 30 years of
age. We could still compute the median age of these people. We could rank order the
ages of the above as follows;

20, 22, 25, 28, 29, >30, > 30

Since the median is the point that divides the ages into two groups of equal size, then the
median has to lie midway between the ages of 26 and 28. The midpoint between these
ages is 27, so the median age of the group is 27.

When to Use the Median

 When we have extreme scores as in example (a)

 When we have missing values as in example (b)

 When the distributions are skewed

 When the data is ordinal level

4:3 The Mean (X)

The arithmetic mean symbolised by x refers to the arithmetic average of a group of
numbers. Consequently it is the measure of central tendency that is most commonly
referred to when the term average is used. Unlike the other two previous measures, that
is, the mode and median, the mean takes into consideration each score in the distribution,
it is therefore, very sensitive to any change mode in the distribution.

The mean is computed by summing up all the scores and dividing them up by the total
frequency n.

If we let xi stand for the individual score, then the mean (x) could be expressed as
follows:

That is:

= (X1+ X2+ X3+ + Xn)

This could be simplified as: =

Where  is Sigma (Greek symbol for capital S ) stands for summation or the sum of.

Examples for Calculating the Mean

Case 1; Mean for ungrouped individual scores E.g. 10, 20, 30, 40 =

=100/4 = 25

Case 2; Mean for ungrouped frequency distributions e.g

Marks f fx

60 2 120

45 3 135
30 3 90

20 2 40
f =10 fx =385 Note:
f = N

x= = = 38.5

Case 3: Mean for grouped frequency distribution

When marks are grouped in class intervals, it is assumed that the marks are scattered
uniformly throughout the interval.

Example:
Table 4:5 mean for the following distribution.
class midpoints f fx
66-68 67 1 67
63-65 64 2 128
60-62 61 4 244
57-59 58 4 232
54-56 55 5 275
51-53 52 3 156
48-50 49 1__ _49
f=20 fx=1151

= where x in this case are midpoints

Hence = 1151/20 = 57.55.

Case 4: Finding the mean of means (pooled mean)

If we know the modes, median and means of three separate classes in a school and wish
to find the same measures of all the classes combined, for the median and mode you will
need the raw scores (data) and average it once more. For the means we don ’t need the raw
scores but the values of n for each class.
Example: The means and total number of scores are given below as follows from three
classes.
XA=51 nA=34
XB=54 nB=40
XC=50 nC=38
Find the combined mean for the group.
 Total n =nA+ nB+ nC
=34+40+38
Mean for the combined group is simply the sum of all 112 scores divided by 112.
 XA= nA X xA =34 X 51 =1734
 XB =nB X xB =54 X 40 =2160
 XC= nC xC =50 X 38 =1900
X

 X=5794

Combined mean ( ) =
_
X= (nA X xA + nB X xB + nC X xC … nn xn)/(nA+ nB +nC + nn)
X

__ _

For the above example: =5794 =51.73

Characteristics of the mean

Note: the arithmetic mean is the most frequently used measure of central tendency. The
following are some of its characteristics:

a) Sensitivity to extreme scores: this is because computation of the scores uses all the
scores in the distribution.
Example: Suppose you were interested in computing the mean number of words
spelt wrongly on a test with a hundred words by five standard 2 pupils. If words spelt
wrongly were as follows: 1,3, 0, 2, 4.
The mean number of words spelt wrongly for the five pupils would be
1+3+0+2+4 =2
5
Now assume that a new pupil joins the group and you want to include his data in
your computation. However, because this pupil is too new in the class, he made an
extremely large number of mistakes (38). When his data value is included, the
average (arithmetic mean) number of words spelt wrongly becomes 8 as shown by
the following:
__
X= 1+3+0+2+2+4+38=8
6
This mean of 8 words is not a good measure of central tendency for the group
performance because 5 of the 6 received much fewer than 8.
Hence the above is an illustration that the mean is influenced by extreme scores.
Hence, when we have extreme scores, the mean is not the most appropriate measure
of central tendency.
b) Sensitivity to any changes anywhere in the distribution: the value of the mean
depends on the numerical value of each score in the distribution. Hence any slight
change on the scores will affect the mean respectively.
Examples:
i. If a constant (c) is added or subtracted from or added to any one of the scores in
the distribution the mean is going change accordingly e.g.

X New X
5 5
5 5
4 4
4 4
2 2+1
__
Old X = 20 = 4
__ 5 __
New X = old X + C
N
=4+ = 4.2
ii. If a constant (c) is subtracted from or added to each score in a set of scores with
a mean of X, how will this affect the mean?
Let c= 3

X X+c __
5 8- The resulting new X will be:
5 8 New X = old X + C
4 7 =20+3
4 7
2 5

__
iii. If each score in a set with the mean X is multiplied or divided by a constant c, the
mean of the resulting scores will be:
New X = Old X x C or
__ __
New X = Old X / C

c) Deviation from the arithmetic mean __

i. The difference between a score value and a mean i.e. X i- X is called a deviation
score. The deviation score is symbolized by small X (x), it could be negative or
positive. The sum of deviation scores in a distribution is always equal to zero e.g.
Xi Xi -X
 9 3
8 2
6 0
4 -2
3 -3
x=30 ( Xi -X)=0
__
X =30=6
5

Hence, the mean is seen as the balance point or fulcrum of a distribution because
the weight of the scores above the mean is exactly equal to the weight of the scores
below the mean.

ii. The sum of squares of deviations from the mean is less than the sum of
deviations from any other score. This characteristic implies that the sum of
squares of deviations from the mean will add up to a smaller number than would
be obtained by using deviations from any score but the mean. Let us use these
scores to demonstrate this: 9, 8, 6, 4, and 3. The mean of these scores is 6. The
demonstration appears in Table 10:7

Table 4:6 The sum of squared deviations about the mean is smallest
__ __
Xi X -X (X –X)2 X -8 (X –8)2 X -3 (X –3)2 (X –9)2 (X –4)2
9 3 9 1 1 6 36 0 25
8 2 4 0 0 5 25 1 16
6 0 0 -2 4 3 9 9 4
4 -2 4 -4 16 1 1 25 0
3 -3 _ 9 _ -9 81 0 0 36 1
(X -X) (X–X)2 = (X-8)2
(X-3) 2
(X-9) 2
(X-4)2
=0 26 =102 =71 =71 =46

From the above table, note that the sum of squares of deviations is smallest in
column 3, where deviations are taken from the mean.
This demonstration indicates that the mean of a group of scores could also be
located by identifying the score that produces the minimum sum of squares of
deviations. When you locate the mean this way, you are using the least-squares
method, a method that is of considerable value when using more sophisticated
statistical techniques such as curve fitting.

4:4 Comparison of the measures of central tendency

The preceding sections have presented three different measures of central tendency. As
observed, each measures slightly different aspects of the central tendency of a
distribution.
 The mode is sensitive to only the most frequently occurring score
  The median is only sensitive to the midpoint of the distribution of scores
 The mean is sensitive to the magnitude of each score within the distribution

Whether the three measures agree or differ depends on the shape of the distribution of
scores. See fig. 10 (l)I –iii

Figure 4:2 Relationship among the three measures of central tendency in different
distributions

__
X=mdn=mo
Fig 4:2i a normal distribution1

__
Mo<mdn<x
Fig 4:2 ii Positively skewed distribution

X<mdn<mo
Fig 4:2 iii Negatively skewed distribution

4:5 Choice of the measure of central tendency

Note: The mean is the most representative and most frequently used measure of central
tendency. This is so for 2 reasons:
 The mean is a statistical index that is used in many other statistical
procedures especially in higher statistics (inferential statistics).
  It is the most stable, or reliable measure of central tendency, that is, it shows
less fluctuations from sample to sample compared to the other two.

Hence, when faced with a situation where all are applicable, we choose mean. The
choice, therefore should be based on:
a) Scale of measurement/type of variable
b) Shape of the distribution

a) Scale (level) of measurement/ type of variable

For nominal data, the only measure that could be computed is the mode.
For ordinal date, median is best but mode could also be computed.
For data at interval and ratio scales of measurement, we could choose the mean
provided that the variables describe a symmetrical distribution.
b) Shape of the distribution
For unimodal symmetrical distributions (normal) all the three measures are equal
therefore any could do.
For skewed distributions, the median is the most representative.
If the distribution does not tend centrally e.g. in bimodal distributions, U and J-
distributions, no measure of central tendency would be able to describe the scores
adequately. Hence the description of the distribution as it appears is crucial.

Activity:
1. What is the most appropriate measure of central tendency for the following set of
scores:
a) 20,21,21,21,23,29,48,48,48,53,86,104,104,104,104,104,117,117,120.
b) 10, 23, 32, 40, 40, 46, 48, 53, 72, 312.
Explain why it is the most appropriate measure.

2. Dr. Oyiro conducted a weight reduction experiment in which he weighed his subjects
prior to and after the study. The weight of his subjects on each occasion are as shown
in the table.
Answers:1a) mode, b) median,

Before After
102 93
52 65
123 92
94 89
46 23

a) Compute the mean weight before and after the experiment. What is the difference
in the mean scores
 b) Now compute the weight difference before and after for each subject and then
compute the mean of these difference scores. Compare this mean with the mean
found in part (a). Explain their relationship.

3. Compute the mode, mean and median for the following distribution of scores.
Comment on the nature of the distribution.

Class F
92-98 1
85-91 0
78-84 2
71-77 7
64-70 13
57-63 14
50-56 17
43-49 3
36-42 0
29-35 3
Answer: Q2a) mean before 83.4kg, mean after = 72.4, difference in mean scores =11
Q2b) mean of differences in weights = 11. The two means are equal.
Q3a) Mo =53, Mdn =60, Mean =60.35. Positively skewed distribution

Further Readings
1. Ingule, F. and Gatumu, H. Essentials of Educational Statistics. Nairobi: East African Educational
Publishers 1996
2. Kerlinger, F N. Foundations of Behavioural Research. 2nd Edition. New York: Holt, Rinehart and
Winston Inc. 1973
3. Malim, T. and Birch, A. Research Methods: London,Macmillan Press 1997
4. Mugenda O.M. and Mugenda A.G. Research Methods, Quantitative and Qualitative Approaches.
Nairobi: Acts Press 1999.
5. Siegel, S. and N. John C. Non-Parametric Statistics for the Behavioural Sciences.2nd Ed. New York:Mc
Graw Hill 1988
 LECTURE 5
MEASURES OF VARIABILITY

Lecture Objectives
By the end of this lecture you should be able to:
 Explain what measures of variability are
 Compute the range, quartile deviation, mean deviation, variance and standard
deviation
 Choose an appropriate measure of variability suitable to summarise a given set of
scores.

Introduction: in the foregoing section we have seen that the measures of centrally
tendency, i.e. mode, median and mean are used to summarise sets of scores by indicating
a number which is most representative of the set. However, this does not give all the
information we need when describing and comparing sets of scores. In order to give a
more complete picture, we need to know how spread out or dispersed or varied the scores
are in a given distribution. Measures of variability are important in bringing out this.

Definition: Measures of variability are those statistical techniques that show the extent
of individual differences or variations of scores within a given distribution.
Different types of measures of variability that will be discussed in this module include:
 The range
 The quartile deviation or semi-interquartile range (Q)
 The mean deviation(MD)
 The variance (S2)
 The standard deviation(S) also symbolised by Sd
5:1 The range
This is the simplest and crudest measure of variability or dispersion. As an index of
variability, the range is merely the distance between the highest and lowest scores in any
distribution of scores.
Two types of the range are normally identified. The one described above is the exclusive
range – the common range- it is simply a measure of the distance between the highest
and lowest score in the distribution.
Exclusive range = Hs – Ls
Example: 48, 49, 50, 53, 59, 63
Exclusive range = 63-48 =15

Inclusive range used for sample data measures distance between exact limit of interval
containing highest score and lower exact limit of the interval containing the lowest score.

This can be simplified to:

Inclusive range = exact upper limit of Hs – exact lower limit of Ls OR Hs – Ls +1
For the above example, the inclusive range would be:
63.5 – 47.5 = 16
OR
63 – 48 + 1 = 16

A small value of the range shows little spread or variability in the score. Large value
indicates a distribution that is highly spread out.

The major advantage of the range is that it gives a quick rough estimate of the
variability.
However, it is seldom used because it is very unstable. This is due to the fact that it
depends on only 2 extreme scores in the distribution, and if these two extreme scores
change, the range changes drastically. Hence, the range should be used with caution as it
is subject to large fluctuations. See examples below:
a) 31, 29, 32, 28, 35, 31. Mean =31 Range =17
b) 15, 33, 43, 16, 42, 37 Mean =31 Range =28
It only gives a good description if the scores are fairly bunched together. If there are one
or two extreme scores as in set (b), use of the range can be quite misleading.

5:2 The Quartile deviation (Q)

A quartile is a point on the scale, which divides the scores in a distribution into a number
of equal parts. E.g. If the entire range of scores is divided into 4 quarters then this can be
accomplished by 3 points on the scale called quartiles Q1, Q2, Q3.

When the entire set of scores is divided into 100 sets of points, its done by percentiles
(see lecture 6 section 6:1) P1, P2, P3, P99. Similarly, deciles, D1 for P10 , D2 for P20 , D3 for
P30 ,, D9 for P90 .

The Quartile deviation also known, as the semi-interquartile range is superior to the
range because it provides a more detailed description of the spread of scores. It is one
half the distance between the two-quartile points Q3 and Q1.
Its calculated as follows:

Q = (Q3 - Q1) / 2
Where: Q = the quartile deviation
Q1 = P25 = it’s the first quartile, and it’s a point in the distribution below
which 25%of the scores lie.
 Q3 = P75 = it’s the third quartile, and it’s a point in the distribution below
which 75% of the scores lie.

For any distribution we can calculate 3 quartiles: Q1, Q2, Q3. These points would divide
the distribution into 4 quarters. Q2, is the median or P50 and that is a point in the
distribution below which we have 50% of the scores.

Calculation of the Q1 and Q3 is very similar to that of the median, except this time instead
of dividing by 2 to see how many scores fall each side of the mid point, we divide by 4 or
take 25% of N and 75% of N. Hence given the following distribution, calculate the
quartile deviation.

F cf
33 – 35 2 36
30 – 32 4 34
27 – 29 6 30
24 – 26 8 24
21 – 23 8 16
18 – 20 5 8
15 – 17 2 3
12 – 14 1 1
N=36

Q1, N/4 = 9 Q3 , 3N/4 = 27 Q2, N/2 =18

Hence,
Q1 = 20.5 + (1/8)3 Q3 = 26.5 + (3/6)3 Q2 = 23.5 + (2/8)3
= 20.875. = 28 24.25
Therefore, Q = (28 – 20.875) / 2 = 3.5625 ans

Significance of the quartile deviation

Assuming that we have a given distribution for which we have worked out the values of
the median (mdn) and Quartile deviation (Q), and the distribution is approximately
normal and therefore symmetrical, we are able to state the limits within which the middle
50% of the cases actually fall by Mdn + Q.

Hence we simply subtract Q from the median to get the lower limit (24.5 – 3.5625 =
20.9375) and add Q to the median to get the upper limit (24.5 + 3.5625 = 28.0625). In our
case therefore, the middle 50% of the cases lie between 20.9375 – 28.0625.

If the distribution is skewed, the limits Mdn + Q will also contain 50% of the cases, but
they will not be located in the centre of the baseline.

The knowledge of these limits gives us a concrete idea of just how concentrated about the
median, or how dispersed about the median the scores are. e.g. if its known that 200
scores which are approximately symmetrically distributed around the median of 65 have
a Q of 15, then approximately 50% of 150 i.e. 100 of the scores lie between the points 50
– 80.

If the distribution is very asymmetrical about the median then as many as 70% of the
scores may lie within the range Mdn + Q.

5:3 The mean deviation (MD)

It is also referred to as the Average Deviation (AD). It is definitely a better measure than
the preceding two because it gives the dispersion throughout the whole distribution unlike
the rage that only deals with two extreme values and the median that is simply a measure
of variability in the middle of the distribution.

MD is the average of the absolute deviation scores. As a measure of variability it

describes the absolute amount by which a distribution of scores deviates from its mean.
The mean deviation is given by the following formula.
__
MD= |Xi – X|_ OR MD = |x|_
N N
Where MD = mean absolute deviation
| | = the symbol for the absolute value (for example, |-3| =3,)thus a +3 or –3
becomes 3 in absolute values.

Example: calculate the mean deviation for the following scores.

__ __
Xi Xi – X |Xi – X|
12 4 4
11 3 3
10 2 2
6 -2 2
5 -3 3
4 -4 4 __
 Xi=48 |Xi – X|= 18

N=6 Hence MD= |Xi – X|= 18 =3

X=8 N 6

For frequency distributions and grouped data:

MD= f | Xi – X|_ _OR MD = f | x |_

N N

The larger the MD, the larger the dispersion or variability of scores around the mean.
However, the MD is NOT frequently used because of:
 This erroneous assumption that is made, which is not mathematically correct i.e. –3
+3.
 There is no specific relation between the MD value and the location of specific
scores within a distribution. For example, the value of the MD cannot be used to
identify within a distribution the score that corresponds to the 75th percentile.

The MD therefore cannot be used in higher, inferential statistical procedures. However, it

is an adequate measure of variability as it has some value for comparison for different
distribution of scores.

5:4 Variance (S2)

The variance is the average of the sum of squared deviations of the scores about their
mean.
The definitional formula for variance is:
__
(S ) = (Xi – X) 2
2
or S 2 = X 2
N N

This method of computing variance is very time consuming when many scores are
involved. It requires that we obtain the mean of the scores, then get a deviation score
from the mean for each of the scores, square each one of these deviation squares and sum
these squared deviations before you could compute the variance. Fortunately, the
following raw-score formula makes the computation of variance much easier.

S2 =Xi2 –(Xi)2
______N___
N

Example: Calculate the variance for this distribution.

a) By use of definitional formula:
__ __ ___
X X Xi – X (Xi – X) 2= x2
12 8 4 16
11 8 3 9
10 8 2 4
6 8 -2 4
5 8 -3 9
4 8 -4 16
x2= 58

S2 = x 2 = 58
N 6
= 9.67

b) by use of raw score formula:

x X2
12 144
 11 121
10 100
6 36
5 25
4 16
x=48 x2=442

S2 =Xi2 –(Xi)2
______N___
N

x2=442
(xi)2 =(48)2=2304

Hence,

S2 =442 –2304
______6___
6
=442- 384
6
=58/6 =9.67

The variance as a measure of variability does not suffer from most of the limitations
inherent in the previous measures of variability. Consequently, it is a good measure. The
larger the variance the larger the variation of scores in the distribution about the mean.
However, the value of the variance is normally too large in comparison to the data and so
appears unrealistic. The other problem about variance as a measure of variability is that,
its units of measurement are square units of the original data units. Getting the square
root of the variance obtained solves these two problems. When we compute the square
root of variance, we get the standard deviation.

5:5 Standard deviation (Sd).

The standard deviation of a distribution of scores is the square root of the average of the
squared deviation of scores from the mean. It gives the extent to which scores in a
distribution deviate from the mean.
Standard deviation is merely the square root of the variance, which in algebraic form is as
follows:

By use of definitional formula:

__
Sd = (Xi – X) 2 or Sd = x 2
N N
__
Where x = Xi – X = deviation score
The raw score formula is as follows:
Sd = Xi2 –(Xi) 2
______N___
N
Hence, for these scores:
__ __
X X-X (Xi – X) 2
12 4 16
11 3 9
10 2 4
6 -2 4
5 -3 9
4 -4 16
__
(Xi – X) 2 = 58
S2 =(Xi – X) 2 = 58/6
N

Sd = 52 = = 3.11

Note: The formulas presented in this section yield the variance and standard deviation of
a sample. The same formula could be used to compute the variance and standard
deviation of a population if all population values were known. However, most of the
times we do not have access to the population. We are concerned with using the sample
variance and standard deviation to estimate the population variance and standard
deviation. To provide for an unbiased estimate of the population variance, the
denominator in the formula for variance must be (n - 1) rather than just n. Hence, for
instance, the definitional formula will become:

S2 = (Xi – ) 2 or S2 = X 2
n-1 n-1

__
 Sd = (Xi – X) 2
n-1

Where:sd = sample standard deviation

Xi = individual raw score
X = sample mean
S2 = sample variance
. n = sample size
n – 1= degrees of freedom

Interpretation of the standard deviation

The standard deviation is the best and most popularly used measure of variability. It gives
some index of how much the scores within a distribution vary their mean score. For
example, if the standard deviation for a group of scores was 12, you would know the
scores varied a lot and deviated quite a bit from their mean.

Properties of standard deviation

1. Standard deviation is very sensitive to extreme scores or the outliers. For example:
6, 66, 68, 74, 76.
6 is an extreme score or an outlier. This extreme will affect or distort the
Standard deviation. The standard deviation of this score will be artificially large
because of the outlier. If the score 6 is omitted, the standard deviation of the scores
will be smaller and more realistic in relation to the new scores

2. It is also very sensitive to changes in the distribution, such that:

a) If any score is shifted further from the mean, the standard deviation increases
accordingly

b) If a constant (c) is multiplied to all the scores in the distribution with the original
standard deviation (OS2), the resultant new standard deviation(NS2) will be
NS2 = OS2 X C2
c) If a constant (c) is divided through all the scores in a distribution with a standard
deviation (OS2) the resulting standard deviation for the new
distribution (NS2) will be:
NS2 = OS2/C2.
However, addition or subtraction of a constant has no effect on the standard
deviation.

Activity
1. Given the following scores: 5, 16, 13, 18, 19, 9, 12, 8, 21, 34, calculate the
following:
 Range
 Mean deviation
 Standard deviation.
Answers: range =29, MD =6.1, Sd = 7.86

2. Class F
48 – 50 1
45 – 47 0
42 – 44 1
39 – 41 1
36 – 38 2
33 – 35 3
30 – 32 8
27 – 29 7
24 – 26 6
 21 – 23 4
18 – 20 3
15 – 17 2
12 – 14 1
9 – 11 1
N = 40

k) Compute the following measures: mode, median, mean, range, and quartile deviation,
mean deviation, variance and the standard deviation.
ii) What can you say about the performance of the class?

Answer: Q2i) Mo =31, Mdn =28.786, mean =27.55, range = 39 or 42,Q = 4.5 MD = 5.92, S2 =60.55,
Sd =7.78.
Q2i) Negatively skewed with moderate variability

Further Readings
11. Ingule, F. and Gatumu, H. Essentials of Educational Statistics. Nairobi: East African Educational
Publishers 1996
12. Kerlinger, F N. Foundations of Behavioural Research. 2nd Edition. New York: Holt, Rinehart and
Winston Inc. 1973
13. Malim, T. and Birch, A. Research Methods: London,Macmillan Press 1997
14. Mugenda O.M. and Mugenda A.G. Research Methods, Quantitative and Qualitative Approaches.
Nairobi: Acts Press 1999.
15. Siegel, S. and N. John C. Non-Parametric Statistics for the Behavioural Sciences.2nd Ed. New York:Mc
Graw Hill 1988
 LECTURE 6
SCORE INTERPRETATION – MEASURES OF RELATIVE
STANDING.

Objectives
By the end of this lecture you should be able to:
 Compare scores from different distributions.
 Calculate percentiles, z cores, T scores and stanines.
 Covert scores from any distribution of raw scores to any of the standard scores
 Identify the percentile rank of any score in a given distribution.
 Normalize any distribution of raw scores.

Introduction

So far we have seen how we describe the general characteristics of a group of scores by
use of frequency distribution and different types of graphs. These methods are very useful
to somebody interested in the whole distribution of data. However for an individual who
is concerned with his own score or measure he has obtained in whatever trait that is being
measured the above methods may not be of much use to him. For example if one has just
taken a test and he has received a score of 62, what does this mean? Did he pass the test
or did he fail.

A raw scores like 62 above on its own may be meaningless (also see lecture 20 on score
interpretation). But it becomes meaningful when it is compared with some standard or
base such as the performance of the rest of the class.

If your score of 62 corresponds to a grade B, You would probably still want to know how
the grade compares with the other grades given in the class. Hence, comparison of a score
or a measure to the other scores in the distribution is an issue of concern to the students,
researchers, parents and many others. This information about how an individual score
compares with the others or the relative standing of a score in the distribution can be
communicated by finding out the corresponding percentile rank or the z score of the
individual score of interest.

When students receive a test score, there are two things they are concerned with mainly.
 A letter grade corresponding to their test score
 How their score compares with the rest of the scores in the class. Their
relative standing in class.
This topic deals with ways of comparing scores obtained in different tests. The main tools
of score interpretation and comparison are percentiles ranks, standard scores and
standardized scores.

6:1 Percentiles
A popular technique for comparing different individuals’ test scores is to express them as
percentile
percentile is a point in the distribution below which a certain percent of sores fall. The
nth percentile is that score value below which n% of the cases in the distribution fall.
Percentile is also referred to as percentile point-score. It is symbolized by Pp
The score value of the variable is called the percentile point-score while its
corresponding percentage value is known as is percentile rank. Hence if a score of 70
was at the 50th percentile rank it means that 50% of the people in the distribution scored
at and/or less then the percentile point-score of 70.

A percentile rank is a number that indicates the percentage of scores that are equal to or
less than a specified score. In other words the percentage of scores that fall at or below a
given score. Pr symbolizes it,

Example
If your score in statistics and evaluation course of 86 corresponds to a percentile rank of
60, it means that 60 of the people taking the test with you made a score equal to or less
than yours of 86. It also means that 40 of the people made a score greater than yours In
this way the percentile rank of a given score tells you how that score compares with the
rest of scores in the distribution.

As seen in lecture 5 section 5:2, the 25th percentile is often referred to as the first quartile,
that is the point in the distribution below which 25 % of the distributions scores fall. The
50th percentile, which also is the median, is occasionally referred to as the second
quartile. The 75th percentile is also called the third quartile.

Some special percentiles

1. The upper quartile – P75 is determined by the score value separating the top 25% from
the remainder of the distribution.
2. The lower quartile - P25 determined by the score value separating the bottom 25% of
the distribution.
3. The deciles - P10, P20, P30, P40 , P100.

Calculations of percentiles and percentile ranks.

The method of calculation of either percentiles or percentile point-scores will depend on

the type of information available to us, that is whether we have the whole distribution of
raw scores or just scanty information, the mean and the standard deviation of the
distribution from which the score has come.
A) when given the whole distribution
Calculation of the percentile rank

Let us use the data in Table 6:1

Hypothetical data to work out percentiles.

Test scores Frequencies cf

73 1 40
71 2 39
70 2 37
68 1 35
65 3 34
64 4 31
61 7 27
60 2 20
59 1 18
55 6 17
52 8 11
50 3 3

Suppose you made a score of 60, what was your percentile rank?

To compute the percentile rank of this score we must deal with real limits of the score 60
which are 59.5 to 60.5.The cumulative frequency below a score 60 is 18 and that 2 people
made a score exactly 60. When several people make identical scores we must make the
very important assumption that the cases within a given interval are equally distributed
throughout the interval. Since there are 2 cases within the interval of 59.5 to 60.5 and the
score 60 lies 0.5 units above the lower real limit, we can calculate that the test score of 60
is 0.5x 2, or the 1.0 case within the interval when this one is added to 18 cases below the
real limits of the score of 60, we have a cumulative frequency of 19. We can now
compute the percentile rank as follows.

Percentile Rank = Cumulative frequency x 100

N

19x100 =47.5
40

This means that 47.5 of all the people taking the test received a score equal to or less
than 60. As it is customary percentile ranks are always expressed as whole numbers.
Hence, the percentile rank for the score 60 is 48, written as P48.

An alternative procedure is to use this formula;

Pr = Fw (X- LRL) + Ci (cfb) x 100
NxCi

Where Pr = Percentile rank

Fw = Frequency within the interval
X = Score whose percentile is being determined.
LRL = Lower real limit of the interval
Ci = Class interval
Cfb = Cumulative frequency below the LRL.

Using the formula for the above example that is percentile rank for the score 60’

Pr = 2(60 – 59.5) + 1(18) x 100

40 x 1

= 2 (0.5) + 18 x 100
40

= 19 x 100 = 47.5
40
= 48. Hence, P48
Table 6:2
Class F CF
90 – 99 5 40
80 – 89 15 35
70 – 79 12 20
60 – 69 4 8
50 – 59 2 4
40 – 49 1 2
30 – 39 1 1
20 – 29 0 0

For the above grouped data, determine the percentile rank of the score 52.

Pr = Fw (X- LRL) + Ci (cfb) x 100

NxCi
2(52 – 49.5) + 10 (2) X 100 =
40x10
2(2.5) + 20 x 100 = 25 x 100 = 6.25
400 400
Since percentiles must be expressed as whole numbers then the score 52 is at the 6 th
percentile rank, expressed as P6
Calculation of the percentile point-scores Pp
Test scores Frequencies cf
73 1 40
71 2 39
70 2 37
68 1 35
65 3 34
64 4 31
61 7 27
60 2 20
59 1 18
55 6 17
52 8 11
50 3 3

What about if you know your percentile rank and you would like to find out the score that
corresponds to the percentile rank, that is the percentile point score.

Example 1
Lets calculate the score at the percentile rank 90,or the 90th percentile in table 6:1

We first of all must determine the cumulative frequency corresponding to that value as
follows:

Cumulative frequency = Pr x N
100
= 90 x 40
100
= 36

This means that 36 scores fall at or below the point that corresponds to the 90 th percentile.
From the table we see that the cumulative frequency 36 corresponds to the score 70
because 2 people made this test score and these two test scores represent the 36 th and 37th
observations. This means that the score of 36 is 1 observation into the class interval with
real limits 69.5 to 70.5,because the cumulative frequency below this class interval is
35.Thus the frequency of 36 is ½ of the way through the class interval. Or it extends 0.5
of the way through 2 score units. This means that you have to go 0.5 units of the way
through the interval to find 36. Finally by adding 0.5 to 69.5 we obtain a score of
70,which is the 90th percentile.

Use of formula.
Pp = LRL+ Ci (PxN – Cfb)
Fw
Where:
Pp = Percentile Point score.
 LRL = Lower Real Limit
Ci = Class interval.
P = percentile rank expressed as a proportion.
N = Total number of scores.
Cfb = cumulative frequency below.
fw = Frequency within.

For the above example using formula;

69.5 + 1(0.90 x 40 – 35)

2
69.5 + ½ = 70
Example 2
For the data in table 6:2 calculate the score at P7
Class F CF
90 – 99 5 40
80 – 89 15 35
70 – 79 12 20
60 – 69 4 8
50 – 59 2 4
40 – 49 1 2
30 – 39 1 1
20 – 29 0 0

First get what 7% of N is: in our case, 7%of 40 = 7 x 40 = 2.8

100

Hence we need a point in the distribution below which 2.8 of the scores fall.
Checking on the commutative frequency column, this must fall in the class 50 – 59.
Up to the point 49.5 we have exactly 2 scores, to get the point at which our 2.8 scores fall
We say it’s the lower limit of this class plus some correction i.e
49.5 + some correction.
To get the correction, ask yourself how many scores do we need to add to 2 to give us
2.8? 0.8. The 0.8 out of the 2 scores within the class. These 0.8 out of 2 is evenly
distributed within a class interval (10). Therefore the correction is (0.8/2) 10.

The P7 then = 49.5 + (0.8/2) 10 = 49.5 + 4 = 53.5

Using the formula.
Pp = LRL+ Ci (PxN – Cfb)
Fw
49.5 + 10(0.07x40 –2) = 49.5 + 10 (2.8 – 2) = 49.5 + 10 (0.8)
2 2 2

=49.5 + 4 = 53.5.
Percentile rank is commonly used in reporting performance in the United States, where a
person’s performance in a test is compared with others who took the same test, regardless
of the score obtained or the number of people sitting the test. It is thus assumed to be a
fair way of reporting, but yet has one draw back, being a mere rank, it is amenable to
further analysis.

B) Calculations of percentiles and percentile ranks when only the mean and
standard deviation of the distribution is known we use the normal distribution
table, (See appendix 1).

Example 3
Given that the mean of the distribution was 70 and the standard deviation 10,
1. Calculate the percentile rank of the score 80 on this distribution
2. What will be the score at the 45th percentile rank

Answer for 1. Since we know that percentile rank is the % ge of the distribution below
the given score,
1st step: get the equivalent z score for the score 80, 80 – 70 = 1
10
nd
2 step: get the proportion of scores below z = 1 from the normal distribution table. And
this is 84%.
Hence the percentile rank for that score of 80 in that distribution is 84, expressed as P84
Answer to 2.
1st step: From the distribution table, find the z score below which only 45% of the scores
fall. It is about –0.10.
2nd step : convert the z score value into a score value on the original distribution.
_ _
We know that, Z = Xi – X, Xi, therefore will = Z(Sd) + X
Sd
For the above, our at P45 will be: -0.10 (10) + 70 = - 1. 00 + 70 = 69.

Activity
Given the following distribution of scores:

Class F
46 – 50 2
41 – 45 2
36 – 40 4
31 – 35 9
26 – 30 11
21 – 25 4
16 – 20 5
11 – 15 1
6 - 10 2

1. Calculate the percentile ranks for the scores: 9, 15, 22, 38, and 55.5.
2. Calculate the percentile point scores for: P5, P25, P50, P60, P80, P96
Given that the mean of the distribution as 60 and the standard deviation as 12,
1. Calculate the percentile ranks of the following scores, 48, 30, 50, 60, 75, and 84.
2. Calculate the percentile point scores for the following, P10, P20, P45, P50, P70, P85, P90,

6:2 Standard Scores

Another useful technique for comparing students’ test scores is referred to as standard
score. A standard score tells us in standard deviation units where a student’s score is with
respect to the mean of the distribution. There are several varieties of standard scores but
in this module we shall just look at a few of them, namely the z score, T score, and the
normalized standard score-stanine.

Standard z-score
In a given distribution it is possible to express any score as a number of standard
deviations above or below the mean. For example, if you have a set of scores with a mean
of 50 and a standard deviation 10, the mean score (50) would be zero, one of 55 would be
+0.5 and one of 45 would be –0.5, etc. these are known as z-scores.

Definition: The standard z-score is a score that states the position of a given score in
relation to the mean of the distribution using the standard deviation as the unit of
measurement. It is given by the formula:
__
Z i= Xi – X
Sx

Where:Zi = Z-score
Xi =individual raw score
X =mean of the distribution from which Xi has come
Sx =the standard deviation of the distribution from which Xi has come
from

Any score above the mean of the distribution will have a positive Z value. Any score
below the mean of the distribution will have a negative z value. A score equivalent to
mean value of the distribution will have a z value of zero.
Example: If you had a score of 88 on an exam, the class mean was 79 and standard
deviation was 6, the corresponding z-score to Xi =88 will be given by 88-79 = 1.50
6
Hence, the score 88 is 1.5 standard deviations above the mean.

Activity: Given the following sets of scores:

X= (10, 36, 38, 40, 42, 44, 70) X = 40, = 16
Y= (10, 20, 30, 40, 50, 60, 70) X = 40, = 20
a) Convert each of the scores in each of the distributions to z-scores.
b) Calculate the means and standard deviations for your new z-distributions i.e. Z x and
Zy .
c) What can you say about the mean and standard deviation for the new z-scores for
both distributions?
Answers: b) Mean for each resultant Z distribution =0, Sd = 1 for both.

Characteristics of the z-score

The z-score has three characteristics that make it a very useful score:
 Every distribution of z-scores has a mean of zero.
 Every distribution of z-scores has a standard deviation of one
 The shape of the z- distribution is identical to the shape of the original
distribution, that is, the z-scores are a linear transformation of the raw scores

Advantages of the z-scores over raw scores.

Transformation of scores to z-scores helps us identify the position of a raw score in
standard deviation units. This is important because:

a) It enables us to compare scores from different tests, skills, aptitudes, etc.

Example: Suppose you took an intelligence and aptitude test and made a score of 125 on
each test. If the means and standard deviations in the 2 tests are same, then we may say
that your relative standing on the 2 tests is the same. But if you had the following
information:
X S
Intelligence test 100 15
Aptitude test 115 14

Then, the z-score for the intelligence test:

Z= =1.670

For aptitude: Z= =0.710

The Z for intelligence is much higher than for aptitude. So his IQ score was relatively
better than the aptitude score.

b) The z-scores that conform to the normal curve can be used to interpret scores. That is,
it is possible to work out the probability (p) of any particular z-score occurring; a
table of the probability of z-scores is included at the end of this manual (Appendix A
Table1). The figures in the table identify the proportions of scores falling between the
mean and any z-score.

Problems of z-scores
Interpretation of z-scores is not realistic because of:
a) Negative scores: any score below the mean will have a negative value. Tendency to
forget writing the negative sign is very high, and this would change the value of the
score very significantly.
b) Zero = average: very difficult for people to comprehend that 0 = average score. Any
square value in the distribution equivalent to the mean score will have a z-score of
zero.
c) Decimal fractions: majority of the z-scores are small fractions. The values of z-scores,
irrespective of the size of n, tend to lie between -3 and 3.

These problems can be overcome by converting z scores to standardised scores.

Note: To move from any distribution to any other score normally, we require that we
convert the raw score to the z score first and then from z score we can move to any score
of interest as shown: _ _
We know that, Z = Xi – X, Xi, therefore will = Z(Sd) + X
Sd

6:3 Standardized scores

Standardized scores like the standard z score indicate the relative position of a score in a
distribution in terms of the standard deviations from the mean, but they always must be
expressed in whole numbers.

Note: all standard scores have a fixed mean and a fixed standard deviation e.g.

Score mean sd
IQ 100 15
T 50 10
Stanine** 5 2

** See more about stanines in section 6:4

Scores can be converted to any of these standard scores so long as we know the z- score
value of the score of interest by applying the formula:
Standardized score = Z (sd of the standardized score of interest) + mean of the standardized score of interest
The standardized T score
It is a distribution with a mean of 50 and standard deviation of 10. It is easier to interpret
this score in comparison to the Z score. The formula for converting the Z score to the T
score is T= 10z + 50.

Suppose you have a distribution of test scores in which the mean is 64.6 and the standard
deviation is 8.4. If on that test a student earned a score of 60.4, what would be the T score
for that raw score?
T =10z +50:
Z = (60.4 – 64.6)/8.4 = -0.5
T = 10(-0.5) + 50.
=45
This score of 45 means that the raw score was one-half standard deviation below the
mean of the raw score distribution in which it is located.
This formula may be applied to the stanine with caution because a stanine is not just an
ordinary standard score but a normalized standard score.

6:4 Normalized standard scores

A normalized standard score is a standard score (z or T) that would be equivalent to a raw
score if the distribution had been perfectly normal.

Normalization is process of trying to make a distribution of scores to be as close as

possible to a normal distribution by smoothing out, stretching or condensing irregularities
and departures from normal in the raw distribution.
The process of normalization involves the following steps:
 Convert each raw score in the distribution to its percentile equivalent (as described in
lesson --- section----).
 Consult a table of the normal curve (Appendix A table 1) to discover what the
equivalent z score would be for each percentile.
 Transform these standard z score to standardized scores with a desired mean and
standard deviation
To compute a normalized T score, simply multiply the tabled z score by 10 and add 50.
As with non-normalized z scores, many people prefer to work with normalized T scores
instead of normalized z scores because of the absence of decimals and minus values.

Stanines
A stanine is a normalized standard score based on dividing the normal distribution to nine
intervals along the distribution’s base line. As its name suggests, its formed by combining
the sta of standard and nine. Stanines are normalized in much the same fashion that you
can obtain normalized z ant T scores. But because the stanine scale is a rather gross scale
consisting of only nine values, it is simpler to determine the stanine values that
correspond to given raw scores. The table below the shows the stanine units with the
approximate percentages of the distribution to be included in each stanine.

Stanine %age of cases interpretation

1 4% Excellent – grade A
2 7% Above average - B
3 12% “
4 17% Average – grade C
5 20% “
6 17% “
7 12% Below average - D
8 7% “
9 4% Poor – grade E
The positions of the nine-stanine values of the normal curve are represented in the
diagram below (dig. 6a)

Dig 6a – Stanine units represented in approximate percentages of the normal curve.

Note that, with the exception of the first and ninth stanines, all of the stanines are of an
identical size, namely, one-half standard deviation unit. The middle stanine, therefore
extends plus and minus one-fourth standard deviation unit from the mean. As earlier seen
a distribution of stanine scores has a mean of 5 and standard deviation of approximately 2
stanine score units. Every stanine is expressed as a whole number.

Activity:
1. The PTA of Buhuya High School decided to award Ksh.1000.00 to subject
teachers for every grade ‘A’ their form four candidates obtained in the KCSE
examination. This decision was taken up with great jubilation by teachers in the
Humanities Department, who immediately began budgeting for a shopping trip to
Dubai using projections of the number of ‘A’s they were expecting from the
current candidate class. Teachers in the science Department, however, felt this
decision was unfair, as in the past very few students could obtain even a mere
grade ‘B’. The most hopeless teacher was the Mathematics teacher as the best that
had ever been produced was a ‘B’ plain. The science teachers took their complain
to the headteacher, who summarily turned them away with the remark that it was
an objective way to reward the teachers and all they needed to do was to work
hard on their classes and get ‘A’s like their counterparts in the humanities if they
wanted to benefit from the decision by the PTA. Very demoralized, the science
teachers vowed to boycott teaching until the matter was addressed by the
chairman of the PTA.
Q. As a person with the knowledge of statistics, what advice would you give to the PTA
chairman concerning the comparison of performance in science and Humanities
subjects.
2. On two tests, mathematics and English, the child scored 65 and 85 marks
respectively. Given the mean performance and standard deviations on the two
tests as follows:

Mean Sd
Maths 55 12
English 70 8
In which test did the child perform relatively better?

3. Given a normal distribution of scores with a mean of 40 and standard deviation

of 14, complete the table below:

Raw score(Xi) Z score percentile T score stanine

47 - - - -
- 2 - - -
- - 75 - -
- - - 30 -
- - - - 6

Answers: Q2) Z values in maths =0.83, in English =1.875, He did relatively better in English than in
maths.
Q3
Raw score(Xi) Z score percentile T score stanine
47 -0.500 -69 -55 -6
-68 2 -98 -70 -9
-49.45 -0.675 75 -57 -6
-12 --2.00 -2 30 -1
-47 -0.500 -69 -55 6

Further Readings
1. Ingule, F. and Gatumu, H. Essentials of Educational Statistics. Nairobi: East African Educational
Publishers 1996
2. Kerlinger, F N. Foundations of Behavioural Research. 2nd Edition. New York: Holt, Rinehart and
Winston Inc. 1973
3. Malim, T. and Birch, A. Research Methods: London,Macmillan Press 1997
4. Mugenda O.M. and Mugenda A.G. Research Methods, Quantitative and Qualitative Approaches.
Nairobi: Acts Press 1999.
5. Siegel, S. and N. John C. Non-Parametric Statistics for the Behavioural Sciences.2nd Ed. New York:Mc
Graw Hill 1988
 LECTURE 7
THE NORMAL DISTRIBUTION

Objectives:
By the end of this lecture you should be able to:
 Describe the statistical characteristics of the normal distribution curve
 Use the normal distribution to find the areas under the curve.
 Find the areas under the normal curve for various standard z scores.
 Find the areas under the normal distribution curve for various standard deviations
from the mean.
 Use the normal distribution table to find the areas under the normal curve between
given z score values.
 Use the normal table to find z scores if given the areas under the normal curve.

Definition: it is a theoretical distribution of scores defined by a specific equation. The

type of curve plotted from this equation is commonly referred to as the normal curve. The
normal curve is representative of the theoretical distribution.

Fig. 7a.The normal curve.

7:1 Characteristics and properties of the normal curve

 It is bell-shaped (symmetrical) rising to a rounded peak around the middle and

tapering off symmetrically at both tails.
 The curve is asymptotic to the x-axis (horizontal axis). That is, the tails of the
distribution taper off and come close to touching the abscissa but never do because
they extend from negative infinity to positive infinity (- to +).
 A perpendicular line drawn from the apex of the curve to the base bisects the curve
into two equal parts.
 The curve is based on the z-distribution such that the baseline (horizontal axis) is
marked in standard deviation units.
 Total area under the curve is equal to 1 unit or 100%.
 Area under the curve corresponds to the percentage of cases such that 100% of the
area is under the curve and this corresponds to 100% of the scores or cases.
 A certain percentage of area (or cases) lies between the mean and a certain number
of standard deviation units above or below the mean.

In theory, z can range from zero to negative infinity and zero to positive infinity. But
when we look at the tables, when z=4, area between the mean and 4= 0.49997 (49.997%),
very close to 0.5 (50%) which is the maximum value and only differs by0.00003.
Note: The normal distribution is important not only because it is a good empirical
description of many variables, but because it occupies a central role in the techniques of
statistical analysis. For example, it is the justification for the calculation of the confidence
interval and it also forms the basis of the methodology of significance testing of means in
inferential statistics.

7:2 Simple application of the normal curve

As it has been mentioned earlier in the characteristics, a very important property of the
normal distribution is that it always has the same proportion of scores falling between
particular points of the distribution. As illustrated in figure 10(n), 34.13% of all scores
fall between the mean and one standard deviation above or below the mean; 68.26% of
all scores fall between one standard deviation below the mean (-1 sd) and one standard
deviation above the mean (+1 sd); 95.44% of all scores fall between two standard
deviations below the mean (-2 sd) and two standard deviations above the mean (+2 sd).
In addition, because we know the normal curve of the distribution is symmetrical, we
know that 50% of all the scores fall below and 50% above the mean.

Fig. 7b. The standard deviation and the normal distribution

Example1: suppose the height of adult men in Kenya is normal distributed with a mean
of 169 cm and a standard deviation of 6.5 cm. from the information in figure 7b, it
follows that:
 50% of the Kenyan adult men population would have heights below the mean
and 50% above.
 68.26% (34.13% below the mean and 34.13% above the mean) of the Kenyan
adult men population would have heights between 162.5 cm (-1 sd) and 175.5
(+1 sd); and
  95.44% (47.72% below the mean and 47.72% above the mean) of the Kenyan
adult men population would have heights between 156 cm (-2 sd) and 182 cm
(+2 sd).

Area under the normal curve

A perpendicular line drawn from the base to the crest of the curve bisects the area
under the curve into 2 equal parts.

The table given in appendix 1 shows the area under the normal curve between the
mean and various standard deviations distance from the mean in the B region
expressed as portions of the total area. Below is a small extraction of the table:

Z AreaB
0 .000
0.1 .040
0.2 .079
0.3 .118
0.4 .155
0.5 .192
0.6 .226
1.0 .3413
2.0 .477
3.0 .4990
4.0 .4999

It implies that %age of cases lying between the mean and 1 = 34.13, e.g in any
distribution let’s say IQ tests 34% of the people will always lie between IQ of 100 and IQ
of 115, similarly, 34% of the people will lie between Iqs85 and 100. For the T score,
34% of the population will lie between T score values of 50 and 60, similarly, 34% of the
population will lie between 40 and 50.

When the mean mark is 55 and  is 10, then it means that 34% of the distribution will lie
between 55 and 65, and also 34 % will lie between 45 and 55.

Note:
1. The area values given in the table should be multiplied by 100 to get proportions
in terms of percentages.
2. The values of area given are in the B region, but the normal distribution is
symmetrical, hence through inference area between the mean and 1 = 34.1%
similarly the area between the mean and –1  must be 34.%. Since area does not
have direction, it’s always +ve, we infer area in A region from the values in the B
region.

Exercise
Find the area under the curve between the following z-values.
a) Z = 0 and Z = 1
b) Z = 0and Z = 1.5
c) Z = 0 and Z = -2
d) Z = -1 and Z = 1
e) Z = -1and Z = -2
f) Z = -1 and Z = 1.68
g) Z = 1.5 and Z = 2.5
h) Z = 0.3 and Z =1
i) Z = -3 and Z = 3

Reverse reading - Finding the Z - values when given the area

Find:
a) The z score above which there is 50% of the area
b) The z score below which there is 75% of total area
c) Z scores that bound the middle 68% of the scores
d) Z score below which 16% of scores
e) Z score that bind 50% of the distribution
f) Z score above which 87% of the area.

Example 2: Given a normal distribution with mean = 40 and  = 4.5 what percentage of
the distribution lies between 31 and 49?

Solution: When Xi = 31, Z =31 – 40 = -2

4.5
When Xi = 49, Z = 49 – 40 = 2
4.5
Hence proportion under the curve between -2 and 2 is 47.7% +47.7% =95.4%.

Example 3: If 200 scores on a test are normally distributed with a mean of 80 and
standard deviation of 12. How many scores would lie below the score of 98 in this
distribution?

Solution: The score of 98 is equivalent to Z = 98 –80 = 1.5

12
Between mean and 1.5 area = 43.3%
Total area below 1.5 = 50% + 43.3% = 93.3%
Hence, 93.3% 0f the whole distribution scored below, i.e 93.3 % of 200.
93.3/100 x 200 = 186.6
Lying below score 98 is approximately 187 scores

Reverse calculations
Calculations can also be done the reverse way. Given a percentage of the population it is
possible to calculate the limits of the IQ containing these percentages.
Example 4: Assuming that IQ in a given population is normally distributed,
a) what score cuts off the top 20% of the population
b) What IQ scores cut off the middle 80% of the population
c) If the population in question is 500, and you are told that the bottom 10 require
special attention what is the cut of score for this group of people?

Solutions: Mean for IQ = 100,  = 15.

a) The Z score that cuts off the top 20% is the same as the one that cuts off the
bottom 80%.

Step1 using the normal distribution that we have, check the z score value that
corresponds to the area 30% in the B region because up to the mean we already
have 80%. (Z = 0.84)
Step2 convert Z = 0.84 into the IQ score:
IQ = Z(15) + 100.
= 0.84(15) + 100
= 112.6
IQ values are expressed as whole numbers, Hence the score that cut off the top
20% is about 113.
c) The middle 80% must be symmetrical about the mean, hence:
Step 1: find the z scores both in the and B regions that correspond to the area 40%.
Using the normal distribution table, In A region the score is –1.28. in B region
the score is 1.28.
Step 2:find the IQ scores that correspond to these scores:
When Z = -1.28, IQ = -1.28(15) +100 = 80.8
When Z = 1.28, IQ = 1.28(15) +100 = 119.2.
Hence the IQ scores that cut off the middle 80% of the population are 81 and 119.

d) Step 1: 10 out of the 500 = (10/500 x 100)% = 2% to be taken in for

special classes,
Step 2: From the normal distribution, look up for the z score that cuts off the
bottom 2% of the distribution. This score is about –2.17.
Step 3: what IQ score corresponds to z scopre of –2.17?
IQ = Z(15) + 100 = -2.17x15 + 100 = 67.45, expressed as a whole number, IQ =
67.
Hence for special classes will be anybody with IQ of 67 and below.

Activity.
1) Suppose you take a random sample of 300 adult men with a mean of 165.5 and a
standard deviation of 10:
a) How many of them will fall below 175.5 cm?
b) How many of them will fall between 156cm and 162.5 cm?
c) How many of them will fall above 156 cm?

2) A maths test is given to 400 children. The mean of the resulting scores is 66 and the
 deviation is 12. Calculate:
a) The number of children that scored below 54
b) The number of children that scored above 78
c) The number of children that scored below 54
d) If 360 children passed, calculate the pass mark for the maths test

3) In a class of 200 candidates, the results of a mathematics test had a mean of 50 and a
standard deviation of 10. Calculate:
a) The number of candidates who scored between 40 and 80
b) The number of candidates who were awarded distinction if a mark of 67.5 was
required to earn a distinction
c) The passmark if 10% of the candidates failed
d) What proportion of candidates scored between 20 and 60
Answers:Q1a) 253, b) 64, c) 249. Q2a) 64, b) 64, c) 336. Q3a)168, b)90, c)37.2, d)0.840.

Further Readings
1. Ingule, F. and Gatumu, H. Essentials of Educational Statistics. Nairobi: East African Educational
Publishers 1996
2. Kerlinger, F N. Foundations of Behavioural Research. 2nd Edition. New York: Holt, Rinehart and
Winston Inc. 1973
3. Malim, T. and Birch, A. Research Methods: London,Macmillan Press 1997

LECTURE 8
CORRELATION

Objectives
By the end of this lecture you should be able to:
 Explain what correlation is.
 Describe various methods of detecting presence or absence of relationship.
 Describe various methods for measuring correlation coefficient.
 Construct scatter diagrams to study relationships between variables.
 Calculate the Pearson product-moment correlation coefficient to study the
relationship between two variables x and y
 Calculate the value of Spearman’s rho and use the coefficient to study the
relationship between variable x and y, when measurements of x and y are based on
ordinal data.
 Interpret the values of rxy and rho.
 Choose the most appropriate technique of correlation in any situation
 Use the coefficient of determination to determine the proportion of y which can be
explained by using x
8:1 Definitions
a) Correlation: is a statistical technique that allows us to examine the degree of
relationship between two variables, correlation between two sets of scores can be
positive (+ve), negative (-ve) or zero.
b) A positive correlation: means that increase in one set of scores corresponds to
increase in the other set of scores or decrease in one set of scores corresponds to
decrease in the other set of scores.
c) A negative correlation: means that increase in one set of scores corresponds to
increase in another set of scores.
d) A zero correlation means that there is no relationship between the two sets of scores.
e) Correlation coefficient: numerical index of correlation that shows the strength or
magnitude of relationship between two variables. Correlation coefficient varies
between –1.00 | +1.00
0
f) A perfect positive correlation (+1.00): for example, the number of litres of petrol
put in a car will exactly match the cost in Kshs. This is where an increase in one
variable is always followed by increase in the other or decrease in one is always
followed by decrease in the other with no reversals at all in the data. This type of
correlation has a correlation coefficient of +1.00 an indication of a very strong
relation.
g) A perfect negative correlation (-1.00): For example, if the car was filled at the
beginning of the journey and then at intervals along the way the amount of petrol left
in the car measured as well as the distance travelled there would be a perfect negative
correlation. The more kilometres one has travelled, the less petrol there will be in the
car. Hence a perfect –ve result when a decrease in one variable is always followed by
an increase in the other variable. This relationship has a correlation coefficient of
–1.00 an indication of a very strong relationship.
h) Linear relationship: a relationship in which the points on the scatter diagram can be
summarized by a straight line.
i) Non linear relationship: a relationship in which the points on the scatter diagram are
best summarized by a curved line also referred to as curvilinear relationships.

8:2 Methods of Estimating Correlation

There are various techniques that are used to analyze the degree of relationship. The
choice of which technique to use will depend on;
1) Whether variables of study are discreet or continuos.
2) Whether the scale of measurement used is nominal, ordinal, interval or ratio.
3) Whether a linear or non-linear relationship exists between the two variables.
In this module some of the techniques but not all will be discussed.

There are methods of correlation which describe the presence or absence of relationships;
 Logical examination
 Scatter diagram
Methods that provide indices of correlation;
 Pearson product moment correlation ( rxy)
  Spearman rank order correlation (rho)

8:2:1 Logical Examination

When dealing with only a few scores lets say 10, it is relatively easy to determine the
type of relationship that exists between two sets of the numbers as is the case presented in
table 8:1

Table 8:1 Hypothetical scores attained on different variables

student Geo. exam height IQ Scores No of words
Spelt wrongly
O.O 73 70 100 95
B.N 66 85 105 90
P.N 79 55 110 85
J.K 82 95 115 80
M.K 85 50 120 75
A.M 88 66 125 70
J.A 91 90 130 65
T.O 94 65 135 60
C.W 97 80 140 55
J.N 100 75 145 50

By observation, from the Geography exam score and the IQ scores you can see that as the
exam score improves the IQ score also improves. Such a relationship illustrates a +ve
correlation. Now look at the number of words spelt wrongly and the IQ scores. You can
see that the number of words wrongly spelt decrease as IQ scores increase. This is an
example of a –ve correlation. Now look at the height scores and try to identify the
relationship they hold to IQ scores. You see that larger height sometimes related to high
IQ scores and sometimes to low IQ scores. Similarly, sometimes smaller height
corresponds to high IQ scores and sometimes to low IQ scores. Therefore, there doesn ’t
seem to be any relationship between the height scores and IQ, which means that zero
correlation exists between these two sets of scores.

8:2:2 Scatter Diagram

It is a graphic representation of the degree of relationship between two variables (sets of

scores). Scatter diagrams provide a rapid illustration of the degree of relationship that
exists between two variables. Different types of relationships are shown below:
Fig.8(a) Scatter diagram relating IQ to Geography exam scores (perfect positive
correlation

Fig. 8(b) Scatter diagram relating IQ to words wrongly spelt (perfect negative
correlation)
Fig. 8(c ) Scatter diagram relating IQ scores to height.

No relationship between the two variable and so the points are almost parallel to the
horizontal axis

Other types of Scatter Diagrams

Fig. 8(d) Scatter diagram showing an imperfect +ve correlation

Fig 8 (e) Scatter diagram showing no relationship

Fig 8 (f) Scatter diagram showing an imperfect negative correlation

The above scatter diagrams describe linear relationships. Some data when plotted on a
scatter diagram produce an arched or v-shaped pattern like one of these shown in figures
8(f) – 8(g) These are non linear relationships and are described as curvilinear
relationships. As will be seen later, these types of relationships are not suitable for
Pearson’s product moment test.
Fig 8 (f) Scatter diagrams showing arched relationships

Fig 8(g) Scatter diagram showing U-shaped relationships.

8:2:3 Pearson product-moment correlation coefficient (rxy)

Pearson product-moment correlation coefficient (rxy) is a statistic used to identify the

degree of linear relationship between two variables. It is used when both variables that
the researcher wishes to study are measured at ratio or interval scales and are continuos
e.g. income, height, age, weight, years of education.

The Pearson product moment correlation coefficient (r xy) may be obtained by use of
definitional formula or computational formula.

(i) Definitional formula

rxy =

Where;
rxy is the score of a person on the variable x
Xi is the score of a person on the variable x
Yi is the score of the person on the variable y
 Is the mean of the x distribution?
Is the mean of the y distribution?
Sx is the standard deviation of the x scores
Sy is the standard deviation of the y scores
N is the number of pairs of scores within each distribution
 is the summation sign.

Example 1

Table 8:2 The calculation of rxy using the definitional formula

Student x y x-x (x-x)2 y-y (y-y)2 (x-x)(y-y)

A 57 52 20 400 16 256 320
B 56 57 19 19 21 441 399
C 37 32 0 0 -4 0 0
D 18 17 -19 361 19 361 361
E 17 22 -20 400 14 196 280
185 180 1522 1270 1360

=37 =36
N=5
Sx= (x-x)2 = =17.44
N

Sy= (y-y)2 = =15.9

N

rxy = (X- )(Y- ) = = 0.98

NxSyS

(ii) Computational Formula

The alternative calculation of the definitional formula is called the computational or the
machine formula. It does not require the calculation of the mean or the standard deviation
prior to the computation of rxy but directly utilizes the raw data.

It is given as follows:

rxy = ____NXY-(X)( Y)______

 [NX2- (X)2][ NY2- (Y)2]
All that this formula requires for one to calculate rxy are values of x, y, n (number of pairs
of scores), xy, x, y, x2, y2.

Example 2

Table 8:3 Calculation of rxy using computational formula.

Student x y X2 Y2 Xy
A 57 52 3249 2704 2964
B 56 57 3136 3249 3192
C 37 32 1369 1024 1184
D 18 17 324 289 306
E 17 22 289 484 374
185 180 8367 7750 8350

rxy = ____NXY-(X)( Y)______

 [NX2- (X)2][ NY2- (Y)2]

= ____5(8350)-(185)( 180)______
 [5(8367)- (185)2][ 5(7750)- 1802]

=____407100-33300
 [41835- 34225][38750- 32400]

= 6800
 (7610)(6350)

= 0.98

Assumptions underlying use of rxy;

(i) The relationship between two variables must be linear.
(ii) The two distributions have similar shapes.
(iii) Data should be based on interval or ratio scale of measurement.
(iv) The scatter diagram should be homoscedastic- homoscedasticity or equal variance
of the x and y distributions. In simple terms, the points on the scatter diagram
showing relationship between x and y should be uniformly distributed.
Activity;
Calculate the Pearson product moment correlation coefficient index rxy for the data
below.
1) x y
15 19
16 25
19 22
21 17
25 14
2) x y
15 17
16 14
19 25
21 19
25 22

Answer:Q1, rxy = -0.761 Q2, rxy = 0.589

8:2:4 The Spearman’s Rank order (rho) correlation coefficient.

Definition: The Spearman rho is a statistic used to measure degree of relationship with
ordinal data.
Within the behavioural and the social sciences many variables such as humour, beauty,
performance, leadership, ability, and social or emotional maturity are rather hard to
evaluate quantitatively and at times the best you can do is to rank order the characteristics
from 1st to last or best to worst. At other times it may even be advisable to reduce an
interval-scaled variable to a rank order,

Spearman rho is computed by use of the following formula;

rho =1- ( )

Where;
rho = the spearman correlation coefficient between the rank orders
D = the difference between the rankings.
n = number of pairs of rankings.

Example 3
Given the rankings of 10 contestants in 2 localities below, determine the correlation
between the Kano North and Kano South ranking of the 10 contestants.

Table 8:4 Ranking of 10 contestants.

Contestants Kano North Kano South D D2

J.O 1 3 2 4
L.O 2 1 -1 1
J.M 3 5 2 4
G.O 4 4 0 0
P.S 5 2 -3 9
K.O 6 8 2 4
F.O 7 6 -1 1
T.O 8 10 2 4
R.B 9 7 -2 4
E.T 10 9 -1 1
D2=32

rho =1- ( )

=1- ( )

=1- ( )
=0.81

If n is rather small, less than 30, even for data that is at ordinal, interval or ratio scales
either Pearson or spearman rho may still be calculated since they give the same results.
In such cases Spearman because it is faster and is computationally more convenient to
use so long as tied ranks do nor exist. When many such tied ranks occur, Pearson rxy
and spearman rho can produce different results, when they do, Pearson is the more valid
index of correlation. However, unless the number of tied ranks is quite large, the
difference between rxy and rho is small and can be safely ignored.

Example 4
Table8: 4 illustrate the difference between Pearson r xy and spearman rho for interval data
when the data contains several tied ranks.

Table 8:5 Pearson rxy and Spearman rho for interval data when tied ranks exist
Scores of 10 students on 2 subjects X and Y
student X Y XY X2 Y2 Rx Ry D D2
A 2 3 6 4 90 10 10 0 0
B 4 4 16 16 16 7.5 8.5 -1 1
C 5 4 20 25 16 5.5 8.5 -3 9
D 3 5 15 9 25 9 6.5 2.5 6.25
E 6 5 30 36 25 3.5 6.5 -3 9
F 4 6 24 16 36 7.5 4.5 3 9
G 5 6 30 25 36 5.5 4.5 1 1
H 6 8 48 36 64 3.5 3 0.5 0.25
I 8 9 72 64 81 1 2 -1 1
J 7 10 70 49 100 2 1 1 1
Total X=50 y=60 xy=331  X2 =  Y2 =  D2 =
 280 408 37.5

rxy = ____NXY-(X)( Y)______

rho =1-
 [NX2- (X)2][ NY2- (Y)2]

=1- = ____10(331)-(50)(60)______
 [10(280)- (50)2][10(408)- (60)2]
=1-
=____3310-3000__________
 [2800- 2500][4080- 3600]
=1-0.227
=0.773
= 310
379.47

= 0.817

To compute spearman rho for interval or ratio scale data, the data must be reduced to
ordinal scale through ranking as shown in the above table 8.4 and table 8.5 below.

Example 5
Table 8:6 More Calculation of rho from interval data.
Student x y Rank x Rank y d D2
A 60 65 1 2 -1 1
B 59 70 2 1 1 1
C 40 45 3 3 0 0
D 21 30 4 5 -1 1
E 20 35 5 4 1 1
N=5 D2=4

Rho =1- ( )=1- =1- =1- 1/5 =0.8

Rho is especially useful when the original data are ranks. It is regarded as a quick way of
estimating rxy.

Assumptions of Spearman rank order (Rho)

 Measurements are taken at the ordinal level of measurement. If at higher levels e.g.
interval or ratio they first must be changed to ordinal by giving the respective ranks
see example 4 and example 5.
 The subjects are randomly picked or sampled to remove bias and ensure
independence.

8:3 Interpretation of correlation coefficient index.

Table 8:7 Interpretation of correlation coefficient.
Coefficient Degree or nature of correlation
0.002 to 0.200 Negligible (very weak)
0.212 to 0.400 Relatively weak
0.412 to 0.600 Moderate / average
0.612 to 0.800 Relatively high
0.812 to 1.000 Very high / strong

8:3:0 Coefficient of determination( r2)

The coefficient of determination (r2) indicates the proportion of which y can be accounted
for or explained for by x. It is a more efficient index for expressing relationship between
two variables. To interpret a specific value of Pearson rxy or spearman rho, we square it
to obtain r2. r2 is the proportion variance of y explained by x. The correlation
coefficient is considered to be a descriptive statistic, the size of which indicates the
degree of relationship between two variables x and y.

The Table 8:6 above may be used to give a general interpretation of the correlation
indices and a description of the degree / nature of relationship.

8:4 Significance Test

Correlation can also be tested for significance in the way that the statistic from a test of
difference can. In correlation, what is being examined is the probability that the
relationship between two variables occurred by chance. If this probability i.e. chance
probability is 0.05 or less we have significant correlation and can infer that there is a
relationship between the two variables being correlated.

A t-test is used to test whether r is significantly different from zero or in other words
whether the observed correlation could simply be due to chance.

t=r d.f. =n-2

In the example of table 8:3

r =0.81
n =10

t =0.81

=0.81
=0.81(4.822)
=3.91, df = 8
Using the t distribution table (appendix 2) this is significant at the 0.05 level, confirming
the significance of the apparent relationship in the ranking of the contestants observed in
the two localities.

Note: The significance level is a function of both the size of the correlation coefficient
and the number of observations (n). A weak correlation may therefore be statistically
significant if based on a large number of observations, while a strong correlation may fail
to achieve significance if there are only a few observations.

Activity;

(1) What is the purpose of the statistical technique of correlation?

(2) What is a scatter diagram?
(3) Why is it important to draw up a scattergram when you are assessing the degree of
correlation between two sets of data?
(4) A clinical psychologist was interested in the relationship between number of hours
of daylight and depression. The following data was collected from 10 subjects.

Person Day light hours Depression score

1 10.1 14
2 11.3 12
3 9.8 17
4 10.5 15
5 10.9 10
6 11.8 6
7 11.1 13
8 9.4 15
9 11.6 10
10 9.6 16

(a) What is the appropriate correlational technique to calculate relationship for the data?
Explain.
(b) Draw a scatter diagram for the data.
(c) Compute both the Pearson and Spearman rank correlation coefficients of correlation
between the hours of daylight and level of depression.
(d) What interpretation can you offer for relationship between depression and daylight
from the values of the coefficients obtained in c
Answers: Q4a) Pearson (rxy ), c) (rxy ) = -0.86, rho = -0.85.

Further Readings

1. Hopkins D. K. and Stanley C. J. Educational and Psychologycal Measurement and Evaluation. 6th
edition. New Jersey: Prentice-Hall, Inc., Englewood Cliffs. 1981.
2. Kerlinger, F N. Foundations of Behavioural Research. 2nd Edition. New York: Holt, Rinehart and
Winston Inc. 1973
3. Mehrens A. W and Lehman J. I. Measurement and Evaluation in Educational Psychology. 2nd edition.
New York: Holt, Rinnehart and Winston. 1978.
4. Mugenda O.M. and Mugenda A.G. Research Methods, Quantitative and Qualitative Approaches.
Nairobi: Acts Press 1999.
5. Popham J.W. Modern Educational Measurement, Practical Guidelines for Educational leaders. 3rd ed.
London: Allyn and Bacon 2000.
6. Whitney, F.L. The Elements of Research, 3rd Edition, New York; Prentice-hall. 1950.

LECTURE 9
REGRESSION ANALYSIS

Objectives:
By the end of this lecture you should be able to:
 Explain what regression is and what its purpose in statistics.
 Calculate the value of the slope coefficients for the regression of y on x.
 Calculate the values of slope coefficients for the regression of x on y.
 Calculate the values of the intercepts for the regression of y on x and the regression
of x on y.
 Find the regression equations for the regression of x on y.
 Use the regression equations to make predictions.

Definition
Regression analysis is a statistical analysis used when a researcher is interested in
finding out whether an independent variable predicts a given dependent variable.
Regression analysis could be categorized into;
  Simple regression
 Multiple regression

Simple regression
Simple regression is used when the researcher is dealing with only one independent
variable and one dependent variable.
Example;
A researcher might be interested in predicting the heights of some from those of the
fathers, predicting weights from given heights, predict the earnings of individuals from
their education levels.

Multiple regression; multiple regression attempts to determine whether a group of

variables together predict a given dependent variable.

Example;
A researcher may be interested in finding out whether mothers ’ occupation, fathers ’
occupation, size of family and ordinal position of birth of a child has any influence on
KCPE performance. The five independent variables are considered together (in one
equation) as predictors of the performance (dependent variable)

Regression models;
Simple regression model is given by the equation;
Y1 = 0+1x1+

Where;
0 = a constant y – intercept
1 = the slope of the regression line = change in y divided by change in x
x1 = value of the independent variable – known
y1 = value of the dependent variable to be predicted.
 = Error in prediction.

Fig .9a Scatter diagram showing a simple regression line.

1=

Multiple regression

Y = 0+1x1+2x2+ nxn+

Where;
Y = independent variable
X1 – n = independent variable
0 = constant, y – intercept
1-n = regression coefficients or change induced in y by each x
= Error of prediction

Regression analysis yields a statistic called coefficient of determination or r2

The r2 refers to the amount of variation explained by the independent variable or

variables. For example, if r2 is calculated to be 0.68, it means that 68% of the variation in
a given dependent variable (y) can be explained by or predicted by variables in the
equation – the independent variable (x). The rest 32% cannot be explained by the
variables in the equation.

Test of significance and regression.

For every value of 1 (slope) the computer will give a t-value and the probability level
for each t-value. The significance level of each regression coefficient can also be
determined manually using t-tables. An independent variable is said to be a significant
predictor of the dependent variable if the absolute t-value of the regression coefficient
associated with that dependent variable is greater than the absolute critical t-value.
Regression analysis also yields an F-statistic and its probability level. The F-statistic tells
the researcher whether one or more of the independent variables significantly predict the
dependent variable at the selected probability level. However, the researcher also has to
examine which of the independent variables are significant predictors of the dependent
variable.

Assumptions of regression.

a) Each independent variable is linearly related to the dependent variable.

b) The observations are independent of each other, which implies that the sample was
drawn at random.
c) Homogeneity of variance exists, that is, at each level of xi the variance of y values is
constant.
d) Y values are normalcy distributed around the mean at each level of x in the
population.

Activity:
The following scores were obtained by 5 students in statistics and introduction to
psychology courses:
Statistics intro to psychochology
80 70
70 65
60 60
50 50
40 40
30 35
20 30

a) What is the slope, 1, of predicting the introduction to psychology scores from the
statistics scores?
b) What is the intercept of the regression equation based on this data?
c) What is the regression equation for the above data.
d) What would be the predicted introduction to psychology score for a statistics score of
45

Answers): 0.733, b)15.67, c) Y = 15.67 +0.733X, d) Y= 48.655

Summary activity on statistics

1. Briefly explain the purpose of descriptive statistics. Give two examples
2. What is meant by “measures of central tendency”? explain which is the preferred one
and in what circumstances it might be misleading to use it
3. Explain briefly the terms: “frequency distribution”, “histogram ”, “bar charts ”, and,
“frequency polygon.”
4. In what circumstances would you expect to obtain a normal distribution? What are its
characteristics?
5. Make sketches of “a negatively skewed” and “a positively skewed ” distribution.
Explain in what circumstances you would expect each one to occur.
6. Explain what is meant by “standard deviation”? What is its relationship to a
normal distribution?

Further Readings
6. Chritensen, B.L. and Stoup, C. M. Introduction to Statistics for Social Sciences. Belmont: Brooks
Cole 1991.
7. Dunham P. J. Research Methods in Psychology. New York: Harper and Row. 1988
8. Ingule, F. and Gatumu, H. Essentials of Educational Statistics. Nairobi: East African Educational
Publishers 1996
9. Kerlinger, F N. Foundations of Behavioural Research. 2nd Edition. New York: Holt, Rinehart and
Winston Inc. 1973
10. Malim, T. and Birch, A. Research Methods: London,Macmillan Press 1997
11. Mugenda O.M. and Mugenda A.G. Research Methods, Quantitative and Qualitative Approaches.
Nairobi: Acts Press 1999.
12. Siegel, S. and N. John C. Non-Parametric Statistics for the Behavioural Sciences.2nd Ed. New York:Mc
Graw Hill 1988