CHAPTER 11: HYPOTHESIS TESTING

Dr. Jitendra Sharma, Research Scholar, University School of Sciences

In this chapter, we will learn:
 The concept of Hypothesis.
 The process of Hypothesis testing.
 Various tests related to Hypothesis testing.

Introduction
Science and research are processes of formulating assumptions around a concept or problem and
then accepting or rejecting them after statistical scrutiny based upon a set methodology. The
assumption in the form of a statement is referred to as a hypothesis (Newey & McFadden, 1994).
A hypothesis is a well-informed guess about anything in the immediate environment. It ought to
be verifiable through experimentation or observation. There are many examples like a prediction
of the functionality of a new machine or concept on the successful teaching process or a fairer
way to administer standardized tests. Each concept or problem can have two logical outcomes –
positive or negative – that are labelled alternate/ experimental hypothesis (H 1) and the null
hypothesis (H0) respectively. The null hypothesis (Anderson, Burnham, & Thompson, 2000)
wishes to disprove and the alternate hypothesis wishes to prove. Inferential statistics is all about
testing hypotheses. Typically, the research hypothesis will be that there is a link between the
independent and dependent variables, or that therapy has a population-wide effect. The null
hypothesis, on the other hand, will typically be that there is no association between the
independent and dependent variables, or that the therapy has no real effect. In other words,
apparent variations in the samples are the result of chance. Many different types of statistical
tests are employed to investigate these possibilities. These tests are chosen based on the variables
being investigated and the types of samples.
The process of hypothesis testing goes through the following stages after the identification
of a research problem/ concept:
1. Framing of hypotheses
2. Ascertaining level of significance
3. Statistical testing for significance – Depending upon the type of research problem, type of
data, and purpose the researcher may want to test for mean, proportion, and/or variance
using appropriate statistical formulae/tests.
4. Critical value
5. Decision making
Framing of the hypothesis should meet the following criteria:
 Testability: the hypothesis should be testable. There must be available parameters that
can be employed with the sample population to obtain data. The researcher must not
frame a hypothesis based on something that cannot be measured e.g. pain levels in
comatose patients.
  The hypothesis should be specific and realistic in terms of the size and scope of what the
researcher aims to accomplish.
 The wording of the hypothesis should be consistent with the research objectives.
 The variables should be lucid, precise, and implicit.

When the hypotheses are being framed, three possibilities exist.

1. H0: μ=μ0 against H1 that μ≠μ0 - This possibility suggests that as per the null hypothesis
there should be no difference between the population mean and the hypothesized mean.
The alternate hypothesis on the other hand would state that the two means are not equal.
The outcome may be less or more.
2. H0: μ≤μ0 against H1that μ>μ0– This possibility suggests that as per the null hypothesis the
population mean should be equal to or less than the hypothesized mean but the alternate
hypothesis states that the population mean is greater than the hypothesized mean.
3. H0: μ≥μ0 against H1 that μ<μ0 - This possibility suggests that as per the null hypothesis
the population mean should be equal to or more than the hypothesized mean but the
alternate hypothesis states that the population mean is less than the hypothesized mean.

Based upon the sign in the alternate hypothesis, for the first possibility, the researcher employs a
two-tailed test. For the other possibilities, one-tailed tests (right or left) are employed. This can
be also comprehended in terms of critical value/ value of significance that the researcher pre-
assigns.
Fig. 1 & 2: Representation of one and two-tailed tests
 Actual Situation
Decision Taken H0 True H0 False
H0 Accepted No error Type-II error

Probability = 1 -  Probability = β

H0 Rejected Type-I error No error

Probability =  Probability = 1 -β

Table 1: Types of errors in the testing of hypothesis

During the process of hypothesis testing, the research may make two types of errors.
Type-I error: This error is deemed to occur when the researcher fails to accept a true null
hypothesis. It is also known as the error of commission because the researcher has taken an
unnecessary decision.
Type-II error: This error is deemed to occur when the researcher accepts a false null hypothesis.
The gravity of this error is more than the Type-I error. It is also known as an error of omission
because the researcher has failed to take the correct decision.
The errors in statistical testing can be comprehended with the help of Table 1.
It is important to test the hypotheses especially for ascertaining the statistical significance and
validity of the research statement. Further, the researcher will be able to understand a qualitative
problem in quantitative terms. This is the inferential part of statistics.

During the process of hypothesis testing, the problems that the researcher faces are:
1. Testing a hypothesis based upon a set of known observations against a new set of
observations.
2. Statistical evidence does not provide irrefutable proof regarding the truth of a hypothesis.
The whole process of hypothesis testing is built on the premise of disproving and
rejecting a hypothesis thereby giving scope to accept the alternate possibility.
3. The hypothesis must find logic in the context of the established existing knowledge.
4. The researcher may construe a positive correlation as a basis for causation, which may
actually be erroneous.
5. Some parametric tests depend upon reliable existing population data. The incompleteness
or variability in the existing population data will adversely impact the reliability of the
outcomes from these tests.
6. The researcher may have an ideological bias that may impel him to manipulate his data
and interpretation(s) to prove or disprove the hypothesis of his interest.

Most of the studies carried out in health sciences are either experimental or correlational by
design and explore a relationship between the variables. The former estimates the qualitative and
quantitative differences between variables whilst the latter observes the pattern(s) among them.
Once the researcher is clear about the design of his research, he can choose and apply the
appropriate statistical test(s). The following steps may be of help for this.
Step 1 – What is the nature of the variable?
The dependent and independent variables (Irlbeck, 2002) may be nominal, ordinal, or
continuous. The combination of variables that the researcher needs to interact with will dictate
the choice of statistical test that can be understood with the help of Table 2 as given below.
 Table 2: Choosing the statistical test on basis of interacting variables.

Step 2- What is the data type?

Having made sure of the study design, the researcher has to decipher whether the data is
parametric or non-parametric. Parametric tests are more sensitive and accurate than non-
parametric tests but essentially require the data to be interval/ ratio type based on parameters of
the population like standard deviation and composition etc. They cannot be used for ordinal or
nominal data – for these non-parametric tests are used. The other conditions for using parametric
tests are not stringent like randomization during sampling, normal distribution of the sample,
and/ or the range of variation should be similar. A small consideration needs to be given to the
sample size. Since small samples may not be truly representative of the population (and not show
normal distribution), it is appropriate to use non-parametric tests (Simar & Wilson, 2002) in
such cases.

Step 3 – What is the analysis type?

The researcher needs to be clear about how many groups have to be tested. The data might be
from the same set of the population at different times (pre-and post-) or from different sets of the
population at the same time or different sets of the population at different times.

Based on these three elements – study design, data type, and analysis type – the appropriate test
is chosen/ applied as illustrated from Figs. 2 and 3.

Fig 2. Flowchart to choose statistical test in Experimental studies

 Fig 3. Flowchart to choose statistical test in Correlation studies

Some of the commonly used statistical tests are:

1. T-test: This is a parametric test that is used for a normally distributed population with an
unknown standard deviation from which a small sample size has been drawn. It compares
the mean difference between two readings of the same subject or difference of means
between two matched groups as is generally seen in pre-and post-test experimental
designs. Also known as Student's T-test (Tawali, 2018), it can also be used to ascertain
the confidence intervals of the population if the sample size is less than 30.
The t-test requires that the differences are larger for the results to be significant. With smaller
samples, the probability of getting respectively large differences increases and so does the
probability of getting significant results.
2. Chi-Square Test: The Chi-Square test (Rao, 2002) is a non-parametric analysis of
frequencies in two groups/ populations that are unmatched and different in experimental
settings. It can test the variance in the sampled population against the actual or estimated
value of the frequency of the general population from which the sample has been drawn.
The Chi-square test does not rely upon estimates like sample mean, variance or
proportion. It is based upon nominal data only that may be dichotomous (two options) or
polytomous (three or more options).
It is ideally used to examine the goodness of fit of data in terms of distribution or difference of
more than two proportions (test of homogeneity) and/or independence of discrete variables (test
of independence).

Fig 4: Flowchart for selecting a simple hypothesis test

1. ANOVA: Analysis of Variance or ANOVA (Square, 2018) is a parametric test that

simultaneously compares multiple data sets (3 or more) from the same individuals or
matched samples or different populations. If applied to 2 data sets, it would give similar
results as the Z-test or the T-test. ANOVA allows the researcher to disregard the possible
multiple combinations of samples and tests thus saving time and resources. The
researcher may also infer from the ANOVA test whether the samples have been drawn
from populations with similar distribution and means. ANOVA tests may be run as one-
way or two-way tests. One-way ANOVA considers a single factor and observes the
variance in three or more groups on its basis. Similarly, two-way ANOVA considers two
factors together while observing for variance in three or more groups.
This test was developed by the British statistician R.A. Fischer and is hence also known as the
F-test. ANOVA is frequently supplemented with Scheffe's test under circumstances when the
F- value is significant. Scheffe's test is a post-hoc test to find which pair combinations that
have been analyzed under ANOVA are significant.
2. ANCOVA: Analysis of Covariance or ANCOVA (Sudret & Caniou, 2013) is another
parametric test that assesses how two variables change w.r.t. each other in experimental
research. If one variable changes and there is a simultaneous change in the second
 variable in the same direction– it is deemed as positive covariance. If the two variables
change in opposite directions, it is deemed as a negative covariance.

3. Wilcoxon Signed-Rank Sum Test – This is a non-parametric test that is used in place of
a one-sample t-test and paired t-test. It also suits continuous but ordinal data where the
researcher can rank the observations. Medians belonging to two data sets from dependent
samples (same or matched subjects)are compared in this analysis but not data from
independent samples (different groups) – for this Mann-Whitney U Test is used.
4. Mann-Whitney U Test – In this non-parametric test, the data from both independent
samples are first clubbed as if belonging to a single population and then ranked
preferably in ascending order. Hence, the null hypothesis essentially assumes that the
samples come from identical populations. For tied/ identical values, the mean of the two
values is given during the ranking process.
5. Z-test- It is a parametric statistical calculation to test that the means of two population
samples are different when their standard deviation is known. For a Z-test, the sample has
to be large (more than 30) and normally distributed. When the sample is small, the
standard deviation cannot be used reliably, and hence, the t-test is more suitable. Further,
as the sample size increases, it becomes distributed normally and better represents the
population from where the sample was drawn.

It is important to note that the t-test compares means using standard error while the z-test
compares means using standard deviation.
6. Karl Pearson’s coefficient of correlation: It is a parametric test that measures the
strength of association between two variables (same subject or matched group) that are
continuous data by nature. The actual values of the variable are known.
However, if the two data are ordinal (nominal/ categorical), Spearman rank correlation
needs to be applied. It is a non-parametric test in which the actual values of variables are not
known but ranks are available.
7. Linear Regression: This is a parametric analysis whereby the researcher predicts a
variable based on another variable. It works when the researcher has two sets of related
variables (Aalen, 1989). The interaction of variables is charted on a graph called
scattergram and the grouping of the dots (points of interaction) gives an estimate of the
correlation between the two variables and based on which further predictions can be
made.

Chapter Summery
Science and research are processes of formulating assumptions around a concept or problem. The
assumption in the form of a statement is referred to as a hypothesis. A hypothesis is a well-
informed guess about anything in the immediate environment. It ought to be verifiable through
experimentation or observation. Hypothesis testing is one of the most integral parts of research
paper/ thesis writing. The above mentioned tests can be used as per the requirement.

Key Terms
Hypothesis testing: Hypothesis testing is a type of statistical inference that employs sample data
to conclude a population parameter or probability distribution.

T-test: The one-sample t-test is a statistical hypothesis test used to see if an unknown population
mean differs from a given value.
ANOVA: An ANOVA test is used to determine whether or not survey or experiment results are
significant. In other words, they assist you in determining whether you should reject the null
hypothesis or accept the alternate hypothesis.

ANCOVA: The method of analysis of covariance is used to test the main and interaction effects
of categorical variables on a continuous dependent variable while controlling for the effects of
other continuous variables that co-vary with the dependent.

Questions:
1. What is meant by hypothesis testing?
2. What is hypothesis testing and examples?
3. What are the different steps in hypothesis testing?
4. Explain ANOVA and ANCOVA with suitable example?
5. What is T –test and Chi-Square Test ?
References:
Aalen, O. O. (1989). A linear regression model for the analysis of life times. Statistics in
medicine , 8 (8), 907--925.
Anderson, D. R., Burnham, K. P., & Thompson, W. L. (2000). Null hypothesis testing:
problems, prevalence, and an alternative. The journal of wildlife management , 912--923.
Irlbeck, S. A. (2002). Human performance technology: An examination of definitions through
dependent and independent variables. Performance Improvement Quarterly , 15 (2), 84--95.
Newey, W. K., & McFadden, D. (1994). Large sample estimation and hypothesis testing.
Handbook of econometrics , 4, 2111--2245.
Rao, C. (2002). Karl Pearson chi-square test the dawn of statistical inference. Goodness-of-fit
tests and model validity , 9--24.
Simar, L., & Wilson, P. W. (2002). Non-parametric tests of returns to scale. European Journal of
Operational Research , 139 (1), 115--132.
Square, M. (2018). Appendix A ANOVA Definitions. Laser Fabrication of Microstructured
Polymer-based Ultra Thin Layer Chromatography Platforms , 149.
Sudret, B., & Caniou, Y. (2013). Analysis of covariance (ANCOVA) using polynomial chaos
expansions. CRC Press/Balkema.
Tawali, T. (2018). THE EFFECT OF DEFINITIONS TOWARD STUDENTS’VOCABULARY.
Journal of Languages and Language Teaching , 6 (1), 34--41.