Data Analysis for

Scientific Research
BAE 815
Dr. Zifei Liu
 In the old days Modern life

• Heavily relied on • More and more data

experience is available

• Very limited data • Less and less

experience is needed

Experience Data Statistics Big data

How do we make decisions?

2
• Be able to extract value from data
is extremely important in today’s
world (information overload – too
much data! )
• Sort out what is important and
what is not! Make the data tell us a
story!
• Successful professionals are those
who can understand and make
sense of data.
The Data-Information-
• The point is not skill development, Knowledge-Wisdom Hierarchy.
but rather mindset. - Russell Ackoff

Why do I have to analyze data?

3
 Answer Positive
Theory

Observations Questions Hypothesis Testing

Negative, revise hypothesis

Classical scientific research

4
• Very likely, your research will involve data collection and
analysis in order to test your hypothesis.
• Data is a universal language allowing scientists to work
together no matter where they are and when do they live.
• Without careful data analysis to back up your conclusions,
the results of your scientific research won't be taken
seriously by other scientists.

Why do I have to analyze data?

5
Data analysis is a systematic process of utilizing data to
address research questions.

Data collection Data processing Data modeling

Descriptive
Data requirement Predictive
Inspecting
Sampling Exploratory
Cleaning Decision
Experimental design Confirmatory
Transforming making
Numerical Data mining
Integration
Categorical Interpretation
Visualization

Making data tell its story

6
• Problem statement
– What is a researchable question?
• Theory, assumptions, background literature
• Variables and hypotheses
• Research design and methodology
• Instrumentation, sampling
• Data analysis
• Conclusions, interpretations, recommendations,
limitations

Important components of empirical research

7
 Research question

Data/information Elements of Lesson/conclusion

(what?) data analysis (so what?)

Be accurate Be careful and be honest

Working with about limitations
uncertainties Correlation does not
always imply causation.
Scientific reasoning/
argument (how?)

Be objective
Separate facts and opinion
Support argument with data
8
The goal is to obtain usable and useful information.
• To identify and understand patterns in data
• To identify relationships between variables
• To compare variables and identify the difference between
variables
• To explain cause-and-effect phenomena
• To forecast outcomes

Goal of data analysis

9
• Data requirements
– Forms of data: text, numbers, images, audio, video.
– Scales of data: nominal, ordinal, numerical.
– Quantitative, qualitative, or mixed?
• Scope of study: case or sample?
– What is your population of interest? How do you want to
generalize your results?
– How many data points do we need? Do they represent all the
population we want to study?

Before you collect your data

10
• Sampling errors (random, representative, nuisance
factors)
• Validity (systematic error/bias), reliability (random
error/bias)
• Accuracy, precision, reproducibility
– Effective digit
• Standard deviation vs. standard error
• Propagation of errors
• Quantitative vs qualitative
• Statistical significance (P value)

Working with uncertainties

11
• Data collection is the most important step. If the collected
data is wrong, analyses and conclusions can not be right!
• Mode of enquiry: observational or interventionist?
– Observational: The aim is to gather data or information
about the world as it is. You hope the act of studying
doesn't substantially modify the thing you are
interested in. Qualitative methods are often required.
– Interventionist : You do something and see what
happens. You gather data or information before and
after the intervention, then look for changes, or effects
of the treatment on the subject.

Data collection
12
• Check for completeness and accuracy of data, handle
missing values, undetected values, duplicates, outliers, and
correct errors
• Code, clean
• Initial data analysis: check and question the assumptions
required for the following data analysis and hypothesis
testing.
– Linearity
– Normality
– Symmetry
– Effect of uncommon observation
• Make transformations of variables as needed.

Data processing
13
• Descriptive: How can the data be summarized?
• Exploratory/Inferential: focuses on discovering new
features in the data and suggest new hypotheses. How can
we draw inferences from the data?
• Confirmatory: focuses on confirming or falsifying existing
hypotheses.
• Predictive: How can we build predictive models using the
data available?

Levels of data analysis

14
• A hypothesis makes a prediction of the expected
outcome in a given situation

• Probability of research
– Nothing is certain
– Scientific “truth” is usually a statement of what is
most probable given the currently known data

• Statistical techniques help us show to which extent our

data do support the hypothesis

Test of hypotheses
15
• Statement 1: A is a human being. B is a gorilla. Between
and A and B are many similarities, but A has many
superior attributes when compared with B.
• Statement 2: The similarities show that both A and B had a
common origin. The superiorities suggest that A evolved
from B over millions of years.
• Statement 3: The similarities show that both A and B had a
common origin: the creator God. The superior attributes of
A show that God chose to create human beings in His own
image, and this was not the case with the creation of
animals.
Credit: Elaine Kennedy

Data and interpretation: Knowing the difference

16
• Data are usually regarded as facts, and are used as a basis
for reasoning, discussion, or calculation.
• As technology and science progress, “facts” will be
discarded, modified, or replaced with new data
• Numbers do not speak for themselves.
• Interpretation demands fair and careful judgments. Often
the same data can be interpreted in different ways. So it is
helpful to involve others or take time to hear how different
people interpret the same information.
• Much of the controversy that exists in the scientific
literature is generated by a common problem:
interpretations drawn from limited databases.

Data interpretation
17
 The duck-rabbit illusion

Data interpretation
18
• Justifying the methodology; citing agreement with
previous studies
• Offer an interpretation/explanation of the results
• Discussing limitations, pointing out discrepancies
• Commenting on the data; state the implications and
recommend further research

There is some merit in indicating what you did not find,

or what surprised you.

• Demonstrate your integrity and professionalism

• Chance to get useful feedback

The results section of your paper/presentation

19
• How to report results?
– Tables, graphs, drawing, flow charts,
photographs, multimedia presentations …
• Think in terms of harmony, rhythm, flow, balance,
and focus.
• Think creatively to combine these elements
together to convey the essential conclusions to the
audience effectively.

Presentation and visualization of results

20
Thinking like a scientist Thinking like a statistician
• Think in terms of validity and • Think in terms of probabilities
reproducibility. and uncertainties.
• Set up tests that eliminate – Significant level, P-value
alternative explanations in such a • Understand the central
way that any observer would come tendencies, the distributions,
to the same conclusion if they did the correlations, and the
the work themselves.
clusters of the variables
associated with the problem
Thinking like a modeler and its solution.
• Think in terms of variables and
their logic connections. Thinking like a data manager
– Independent variables, • Think in terms of tables and
response variables, nuisance matrix.
factors • Define the rows, columns, and
• Decompose the problem into it cells of the tables; associate
basic components; represent those tables with one another; and
components numerically; and create systems to ingest, store,
combine the components together and retrieve tables.
into an accurate expression of the
problem and its solution. Thinking like a visual artist 21
• Understand the basics first - start from basic data
analysis textbooks. General reading should never
stop.
• Courses related to data analysis
• Read research papers.
– What sort of research is typically conducted in your
discipline and how are studies designed
– What are the procedures, techniques, software and tools
commonly used in your field
– To learn how to be scientific in your field

What should you be reading?

22
• “It is commonly believed that anyone who tabulates
numbers is a statistician. This is like believing that
anyone who owns a scalpel is a surgeon.” - R. Hooke
• “Torture numbers, and they'll confess to anything.” -
Gregg Easterbrook
• How to lie with statistics? - Darrell Huff, 1954
– A most widely read statistics book

How NOT to lie with statistics:

Avoiding common mistakes
23
24
(From “20 Insane Things That Correlate With Each Other”)

25
• Old friend: MS Excel
• Abaqus
• Ansys
• LAMMPS
• Matlab
• Mathematica
• LabView
• SAS, SPSS
• R is available free over internet
• Many more!

Tools for data analysis

26
• Every researcher is going to require data analysis skills at
some point or the other.
• Understand the assumptions and capabilities (and
limitations) of different methods or techniques, select the
right one and use them carefully.
• Be very careful when you want to extrapolate results and
conclusions.

Final word
27
Blind men and an elephant
28
“A data scientist is someone who knows more statistics than a
computer scientist and more computer science than a statistician.”
- Josh Blumenstock

29