IOP2601/MO001/4/2016

LEARNING UNIT 12: Hypothesis tests applied to means:

t-tests

We have already discussed hypothesis testing in learning unit 11. In this learning unit we will
test a hypothesis empirically in order to determine scientifically whether there is a significant
difference between the means of two sets of scores. Determining the difference between two
sets of data is something that we often have to do in the work situation. Remember the
example of the emotional intelligence of New Stars applicants and contestants given in
learning unit 11?
There are different kinds of test statistics that we can test by means of hypotheses,
depending on the type of data. In this learning unit we look at the different t-tests: the t-test for
independent groups and the t-test for two related groups, as covered by Tredoux and
Durrheim (2013) in Tutorial 9.

12.1
To familiarise yourself with the concepts and some of the assumptions that the data have to
comply with, study Tredoux and Durrheim (2013) from pages 142 to 147. This will also help
you understand the principles underlying hypothesis testing.

Hypothesis tests applied to the means of two independent samples

We are going to talk about hypothesis tests when dealing with two independent samples. We
will go through the same steps of hypothesis testing as discussed in learning unit 11. We will
also look at the adjustments we make when there are differences in homogeneity of variance
and when samples are not of the same size.

160

IOP2601/MO001/4/2016

What is the purpose of applying hypothesis tests to two independent

samples?
The purpose of hypothesis testing with two independent means is to help you decide whether
an observed difference between two sample means is accidental or whether it represents a
real difference between populations. In hypothesis testing terms, the purpose of the t-test is to
help us decide whether or not to reject a null hypothesis postulating no difference between
the means.

What are two independent samples?

Independent samples are samples that are not related in any way, or when there is no
correlation between individual pairs of participants in two samples that is, when participants
are included in only one of the two samples in an experiment.

12.2
Follow the reasoning about sampling distributions of differences between means. (Do you
remember what a sampling distribution is? If not, refresh your memory by going through the
relevant section in learning unit 11 again. As you will see when you read the introductory
part of learning unit 11, we are now dealing with the sampling distribution of the difference
between means.)
1.

Complete the following sentence:

The sampling distribution of the difference between means is
.....
.....
.....
.....
True

2.

Because we sample each population independently,

the sample means will also be independent.

3.

The mean of the sampling distribution will be

1 - 2.

4.

Male and female applicants to New Stars respectively

represent two independent groups.

False

161

IOP2601/MO001/4/2016
1.
2.
3.
4.

The distribution of the differences between means over repeated sampling from the
same population(s).
True
True
True

How do we compute the t-test for independent samples?

Now that you know what independent samples are and why we apply a t-test to them, the
next step is to determine whether there is a significant difference between the means of two
independent samples. This is quite a simple computation when you use the formula that
Tredoux and Durrheim (2013) provide as Equation 9.4 on page 150. Alternatively, you can
use the formula provided in the list of formulas at the back of this study material. You dont
have to know the background; only learn the formula.

12.3
Work through the example which follows and make sure that you know how this test statistic
(in this case the t-test for independent groups) is computed. The t-test is done according to
the nine steps set out in learning unit 11. Did you remember to write down the steps on a
piece of cardboard and stick it above your desk? If not, do so now before you work through
the example. You need to be comfortable with the logical flow of the steps and must be able
to follow the steps in the exam so that you can use the relevant test statistic and solve the
problem statement statistically. These nine steps can be used in the same way for other test
statistics (like the t-test for related groups and the F-test) this is another good reason for
familiarising yourself with the different steps now.

A researcher wants to determine whether there is a significant difference between the overall
entertainment rating allocated to New Stars by male and female viewers. She gets the scores
of 20 randomly selected viewers and decides to test significance at the 5% and 1% levels.

162

IOP2601/MO001/4/2016
Data
Male viewers

Female viewers

X2

Y2

16

25

25

36

16

16

49

25

36

25

X =

37

X = 181
=

3,7

4,88

N=

10

Y =

33

Y = 131
=

3,3

2y = 2,47
N=

10

Step 1: Formulate the null hypothesis

In words
There is no difference between the entertainment scores of male and female viewers of New
Stars.
or
Male and female viewers find New Stars equally entertaining.
In symbols
H0: m f = 0
or
H0: m = f

163

IOP2601/MO001/4/2016

Step 2: Formulate the alternative hypothesis

In words
There is a difference between the entertainment scores of male and female viewers of New
Stars
or
Male and female viewers do not find New Stars equally entertaining.
In symbols
H1: m f 0
or
H1: m f

Step 3: Determine whether the test is one or two-tailed

In this example (like those you will have to handle) the decision whether to use a one-tailed or
a two-tailed test is left up to you. Therefore, you have to study the problem statement to see
whether it is directional or nondirectional. In the example we are working through, we are only
investigating whether there is a difference between the means of the two groups and not
whether the one group does better or worse than the other group. We are therefore working
with a nondirectional hypothesis and so choose the two-tailed test.

Step 4: Determine the level of significance

Fortunately, you will not have to make a decision here because we will always specify at what
level of significance you should test the hypothesis. In this case we have specified two levels,
namely, 0,05 and 0,01.

164

IOP2601/MO001/4/2016

Step 5: Compute the test statistic

Given our problem statement (note that the two groups are independent), we use the t-test for
independent groups and the following formula:






Step 6: Determine the degrees of freedom

df=

 

 

NN



Step 7: Determine the critical value

Use table A1.2 on page 487 in Tredoux and Durrheim (2013) to read the critical values.
t0.05 (18) =

2,1009

t0.01 (18) =

2,8784

165

IOP2601/MO001/4/2016

Step 8: Decide whether the null hypothesis should be rejected/not

rejected
From the computations we have done and the tables, we have determined that
x
x

the value of the test statistic is 0,47

the critical value at the 0,05 level is 2,101 and at the 0,01 level it is 2,878

Use the following decision-making rules:

x
x

reject H0 if the test statistic > the critical value

do not reject H0 if the test statistic < the critical value

For = 0,05

0,47 < 2,1009

For = 0,01

0,47 < 2,8784

? Do not reject H0

? Do not reject H0

(Remember: It is unscientific and wrong to say that we accept the null hypothesis.)

Step 9: Interpret the findings

The researcher concludes with 99% certainty that there is no difference between the
entertainment scores allocated by male and female viewers of the New Stars show.
This is the most basic variation of the t-test for independent groups.

Why do we pool variances?

When we are dealing with samples of unequal size, we have to pool the variances of the two
samples. Tredoux and Durrheim (2013) explain this on page 150. Make sure you can use the
formula for pooled variances by thoroughly studying Equations 9.5 and 9.6 in Tredoux and
Durrheim (2013).

12.4
To familiarise yourself with the various facets of the use of the t-test, it will help you to do an
exercise on your own. Do Exercise 1 on page 158 in Tredoux and Durrheim (2013). Follow
the steps of hypothesis testing, which you should know quite well by now.

166

IOP2601/MO001/4/2016
x
x
x
x

Follow all the steps in the process, including the formulation of null and alternative
hypotheses and the decision on a one-tailed or two-tailed test.
Use your pocket calculator to complete the steps in the computation of the test statistic
and determine the degrees of freedom.
Look up the critical values in the table and apply the decision-making rules. Use the test
significance level of 0.05.
Write down your interpretation concerning the rejection or nonrejection of the null
hypothesis.

You have learnt quite a few new skills and at this stage you should be able to
x
x

know when to use an independent t-test

test a hypothesis by applying the t-test to independent samples and compute a test
statistic, even when sample sizes are not the same or when the variance of one sample
is four times that of the other sample
come to a final conclusion about your research hypothesis

Go through the example below and make sure that you can
identify and interpret the t-value on the printout.

Exercise 2
Data Set 8
For the independent samples t-test, the scores for both groups should be entered in a single
variable. A dummy variable should then be entered as a means of identifying each group. I
called the independent variable laugh, and the independent variable parent. Two-parent
families were labelled 1 and one-parent families were labelled 2.

167

IOP2601/MO001/4/2016
Independent: samples t-test
Select the Independent Samples T-Test option on the Compare means option on the
Analyse menu. In the dialog box, select the test variable (DV) and the grouping variable (IV).
Define the group variable by specifying the lowest and highest value of the dummy variable
in our case the lowest value is 1 (2 parent families) and 1 (one parent families). The outputs
are given below:
































As you know by now, equal variances play a role in a t-test. In the above printout, the first
test (Levenes test) is to test for the assumption of equal variances. If it is significant (less
than 0,05 or 0,01) we read the second row of data. In this case it is not, so we read the first
row. In the column titled sig (two-tailed) we again look to see whether the value is less than
0,05 or 0,01. In this case it is less than 0,05, so the t-test is significant, showing a statistical
significant difference.

168

IOP2601/MO001/4/2016

Nonparametric equivalent of the t-test for independent samples

Having read this heading, you must have noted the new term "nonparametric". You will learn
what a nonparametric statistical test is and how it differs from a parametric statistical test
when you study the introduction to Tutorial 20 on page 385 of Tredoux and Durrheim (2013).
In this section we deal theoretically with the nonparametric equivalent of the t-test for two
independent samples by looking at the Mann-Whitney test. As an industrial psychologist you
have to know what to do when your data are not normally distributed. That is what we explain
in this section.

12.5
There is not a great deal to learn in this section. Study the first paragraph on page 391 in
Tredoux and Durrheim (2013).
Now that you have completed this section you should be able to name the nonparametric
equivalent of the t-test for two independent samples. Remember, you need not be able to
describe it or know the formula.
That is the end of discussion on t-tests for two independent samples. Now we are going to
talk about hypothesis tests when dealing with two related samples.

What are related samples?

What does that mean? Are they the same as independent samples? When do we speak of
related samples and when do we refer to independent samples? This distinction will be clear
to you once you have completed this learning unit. After that we shall look at the other types
of test statistics.

12.6
Do you still remember when we classify samples as independent? The exact opposite
applies to related samples.
First make a list of the characteristics of independent samples. Then complete the table for
related samples. (Read the introduction on page 152 in Tredoux and Durrheim (2013) to help
you with this activity.)

169

IOP2601/MO001/4/2016
1.
Independent samples

x
x

Related samples

Make sure you understand what is meant by related samples.

Use the knowledge you have acquired to do the following exercises:
True

2.
3.
4.
5.

1.
2.
3.
4.
5.

False

Here we are working with two sets of data on the same

participants (the persons who form part of the sample).
We expect the two sets of figures (variables) to correlate.
We take correlation into account when computing the
t-statistic.
The only way to obtain related samples is when the same
participant contributes two scores.

Independent samples are samples that are included in only one of the two samples in
an experiment and related samples are samples that are dependent.
true
true
false
false

Lets explain in greater detail what is meant by related samples, for you have to be able to tell
when a research problem or a research hypothesis involves related or independent samples.

We say that two groups of people or participants are matched according to a variable if for
each member allocated to one group, the other group is assigned a member who
corresponds with the member of group one in respect of the particular variable. Thus
participant pairs are selected in such a way that the two members of each pair are as similar
as possible in respect of the relevant variable(s). Variables like intelligence, age, height, eye
colour and gender can be used to match persons or participants.
To sum up: the only ways of obtaining related sample scores are
x
x

170

by matching
when one participant contributes two scores

IOP2601/MO001/4/2016

How do we compute the test statistic for related samples?

Now that you know what a related sample is, the next step is to determine empirically
whether there is a significant difference between two related samples. This involves a fairly
simple computation by means of a specific formula. But before you can apply the formula you
first have to know how to compute difference scores. To do this you must know how to
) and the standard deviation ( ) of difference scores.
compute the mean (
These are not new formulas. They are the same formulas
that you used to compute the mean in learning unit 5 and
the standard deviation in learning unit 6.

12.7
A practical example always provides a useful learning experience, so it would be a good idea
first to go through Creating the variable D on page 152 in Tredoux and Durrheim (2013) as
an example of how to compute the t-test. If you are not sure how to compute the various
steps of the formula, revise the formula for and in learning unit 5 and learning unit 6
respectively. Tredoux and Durrheim (2013) provide the computed values of D in Table 9.2 on
page 153.
Complete the following:
1.

A difference score is
.....
.....
.....

2.

Check the computation of the difference scores in the example. Can you see when the
scores acquire positive or negative values?
.....
.....
.....

171

IOP2601/MO001/4/2016
3.

Check the formula of the t-test for related samples in the example.
.....
.....
.....

6.

Write down the formula for degrees of freedom.

.....
.....
.....

The answers are on page 153 of Tredoux and Durrheim (2013). Now that you have carefully
read the detailed steps, let us see how well you understand them.

12.8
Before you do an exercise on your own, work through the complete example which follows so
that you can see what we expect you to do in the different steps.

What follows is a fully worked-out practical example of how to do the t-test for related groups.
Work through the example and check all the calculations. Make sure that you understand all
the steps in the process.

EXAMPLE of a t-test for related groups

The research team wants to determine whether there is a difference in the happiness levels
of contestants before and after the apprenticeship programme. They decide to test the
hypothesis on both the 0,05 and 0,01 level of significance.

172

IOP2601/MO001/4/2016
Data
Before
4
3
7
6
5
4
5
4
3
2

After
5
8
6
6
6
4
7
7
6
4

The team follows these steps:

Step 1: Formulate the null hypothesis

In words
There is no difference between the happiness levels of contestants before and after the
apprenticeship programme.
or
Contestants are equally happy before and after the programme.
In symbols
H0: B - A = 0
or
H0: B = A

Step 2: Formulate the alternative hypothesis

In words
There is a difference between the happiness levels of contestants before and after the
apprenticeship programme.
or
Contestants are not equally happy before and after the programme.
In symbols
H1: B - A 0
or
H1: B A

173

IOP2601/MO001/4/2016

Step 3: Determine whether the test is one- or two-tailed

We are working with a nondirectional hypothesis and choose the two-tailed test.

Step 4: Determine the level of significance

We have specified two levels, namely, 0,05 and 0,01.

Step 5: Compute the test statistic

We use the t-test for related groups and the following formula:

For the t-test computation, we first have to compute the difference scores (D), the mean for
D, variance and standard deviation for D scores. We do this by completing the table as
follows:
Before
4
3
7
6
5
4
5
4
3
2

After
5
8
6
6
6
4
7
7
6
4

D
-1
-5
1
0
-1
0
-2
-3
-3
-2
D = -16

D2
1
25
1
0
1
0
4
9
9
4
2
D = 54

The data in column D (difference scores) is a set of X-scores. Therefore, use the formulas for
mean, variance and standard deviation you already know just substitute X for D.
The mean (refer to learning unit 5) = 1,6.
The variance (refer to learning unit 6) = 3,16, therefore the standard deviation = 1,78.
Once you have all the values, substitute the t-test formula like this:

174

IOP2601/MO001/4/2016







Step 6: Determine the degrees of freedom

Using the following formula:
df





Step 7: Determine the critical value

Use Table A1.2 on page 487 in Tredoux and Durrheim (2013) to read the critical values.




175

IOP2601/MO001/4/2016

Step 8: Decide whether the null hypothesis should be rejected/not

rejected
t = -2,86; I t I = 2,86 (Remember that we are working with absolute values and therefore
ignore the minus sign.)

For = 0,05

2,86 > 2,2622

For = 0,01

2,86 < 3,2498

? Reject null hypothesis

? Do not reject the null hypothesis

Step 9: Interpret the findings

The team can say with 95% certainty that there is a difference in the happiness levels of
contestants before and after the apprenticeship programme, but they cannot say so with 99%
certainty.
Is it clear to you what we have done in the different steps? If not, go through the example
again. If you are satisfied that you have a good idea of what you have to do in each step, go
on to the activities which follow.
Do the worked Example 3 on page 154 in Tredoux and Durrheim (2013) on your own, without
looking at the different steps and solutions, and then see if you arrive at the same answer as
theirs.
A further step in the process of hypothesis testing is to determine the number of degrees of
freedom and read their critical value from Table A1.2. When you have managed this once, it
is perfectly simple.

12.9
Check the decision-making rules explained in step 8 in this learning unit and follow the
argument for rejecting or not rejecting the null hypothesis.
Once you have done this, do the following exercises.
1.

Test your ability to find the following critical values in the relevant table:
1.1
1.2
1.3

176

one-tailed test, 29df, 1% significance level

one-tailed test, 16df, 5% significance level
two-tailed test, 8df, 2% significance level

IOP2601/MO001/4/2016
1.4
1.5
2.

Would you reject H0 in the following circumstances? Why?

2.1
2.2
2.3

1.1
1.2
1.3
1.4
1.5
2.1
2.2
2.3

two-tailed test, 29df, 2% significance level

two-tailed test, 23df, 10% significance level

t-test statistic = 2,302 and critical value = 2,508

t-test statistic = 2,682 and critical value = 3,792
t-test statistic = 3,248 and critical value = 2,704

2,4620
1,7459
2,8965
2,4620
1,7139
no
no
yes

You have now gone carefully through detailed examples of hypothesis tests with two
related samples. You should be able to
x
x
x
x

explain what is meant by a related sample

know when to use a t-test for related groups
test a hypothesis by applying the t-test for related groups and computing a test statistic
draw a final conclusion about your research hypothesis

Go through the example below and make sure that you can
identify and interpret the t-value on the printout.

Data Set 7
Enter the data as illustrated alongside.
Related samples t-test
On the Compare means option on the Analyse menu, select the aired samples T-test
option. The dialog box is shown below. You should select both sets of observations, and
enter them into the aired variables window. Click OK to run the procedure.

177

IOP2601/MO001/4/2016

Nonparametric equivalent of the t-test for related samples

In this section we deal theoretically with the nonparametric equivalent of the t-test for related
samples by looking at Wilcoxsons matched-pairs signed-ranks test. As an industrial
psychologist you have to know what to do when your data are not normally distributed. That is
what we explain in this section.
Since you do not need to know the details of nonparametric equivalents, you only have to
study the last paragraph on page 389 in Tredoux and Durrheim (2013).

12.10
For additional exercise, do Exercises 3 and 4 on pages 158 and 159 in Tredoux and
Durrheim (2013).

178

IOP2601/MO001/4/2016
Exercise 3
In this study, our design is to take two measurements from each subject; in other words,
repeated measures. We cannot get two measurements from each person, as some
individuals are unavailable at the time of the second measurement, so we cope with this
situation by using casewise deletion we simply ignore the data from a person if we dont
have two measurements from them. Our table of data looks like this:
Subject
Time 1
Time 2

1
13
9

2
12
10

3
16
NA

4
14
NA

5
13
10

6
15
NA

7
17
11

8
13
10

9
14
NA

10
16
17

11
13
9

12
16
8

13
13
NA

14
19
16

15
12
NA

By excluding the subjects from which we dont have two measurements (those are subjects 3,
4, 6, 9, 13 and 15), our table now looks like this:
Subject

10

11

12

14

Time 1

13

12

13

17

13

16

13

16

19

Time 2

10

10

11

10

17

16

We shall be working from the data provided in the above table, not the data in the original
table.
As this is a repeated measures design, we need to create variable D first of all. Do this by
subtracting time 1 from time 2 for each subject (i.e. D = Time 2 - Time 1). The data for D look
like this:
Subject

10

11

12

14

Time 1

13

12

13

17

13

16

13

16

19

Time 2

10

10

11

10

17

16

-2

From now on, we shall not be using the data from Time 1 or Time 2; only the data from D.
We begin by working out a mean, standard deviation and n for D.

D=

s

N

Our estimate of the standard error uses the following formula:

179

IOP2601/MO001/4/2016
Putting our values into the equation we get:



Now we can calculate t. The formula is:


And we already have all those values, so we can calculate




Now we need to determine whether the difference is statistically significant. To do this we
need the degrees of freedom.
df=N1
Which are
df
df
Now we determine whether it is a one-tailed or two-tailed test. According to the research
hypothesis, we are only interested in seeing if the substantia nigra has become smaller if so
then this is a one-tailed test. Our alternative hypothesis is thus:
H1: Time2 < Time1
We shall use the standard alpha value of 0,05.
We now use this information to look up a critical t-value on our t-table. The critical value for
df = 8, one tailed, alpha = 0,05 is 1,860. Our t-value is 4,257 [we ignore the () sign for this
comparison].
Our value is greater than the critical value, so we reject the null hypothesis, and accept the
alternative. It seems that the average diameter of substantia nigra in psychotic patients
becomes smaller after a period of time.

180

IOP2601/MO001/4/2016

Exercise 4
Here, we are using two measurements from the one basketball team, but this is not a
repeated measures design. The reason is that the conditions were not the same each time
we are looking at the effect of the coach, and not of the team, so the two sets of data are
actually independent.
Although we have data missing (fewer games played with the first coach than with the
second), we do not need to worry about this (it is only a thing to worry about if you are using a
repeated measures design).
To begin, we must set up a null hypothesis. For an independent samples t-test, this is:
H0: coach2- = coach1

The first step is to determine the mean, n, and variance for each group. The variances are:
s2coach1 = 24.214
s2coach2 = 732.62

Xcoach1 = 86.25
Xcoach2 = 79.9167

Ncoach1 = 8
Ncoach2 = 12

Since the sample sizes differ, we need to pool the variances:










  
 

 



 



Now we can insert this value into our independent sample t-test formula. The formula is:

181

IOP2601/MO001/4/2016
We already have all those values, so we can insert them and calculate:



Now that we have a t-value, we must determine whether it is statistically significant. To do
this, we need the degrees of freedom.

  



So we shall use 18 degrees of freedom.
Now, is this a one-tailed or a two-tailed test? According to the research hypothesis, we are
interested in whether the team performed differently under the new coach in other words,
we are interested in either positive or negative chance, so this is a two-tailed test. From this
we can derive our alternative hypothesis, namely:

H
We shall use the standard alpha value of 0,05.
We now use this information to look up a critical t-value on our t-table. The critical value for df
= 18, two-tailed, alpha = 0,05 is 2,1009. Our t-value is 0,65.
Our value is less than the critical value, so we cannot reject the null hypothesis. The average
score with each coach was the same, so it seems that the team is performing the same under
the new coach as it did under the old one.

182

IOP2601/MO001/4/2016

12.11
Consider the following research question from the New Stars programme:
Do male voters give higher scores to female participants than to male participants?
Indicate which analysis technique would be the most appropriate to answer this research
question and explain/substantiate your answer.
In this case the gender of the voters is merely a selection variable (the scores of female
voters will not be considered in this research question). For the male voters, one would
consider the scores that they gave to the participants. Then the scores of the two gender
groups of the participants will be considered to evaluate whether there is a statistically
significant difference in the mean scores of the two groups. This implies a t-test (comparing
two groups) and it would be the t-test for independent groups (as the male and female
participants are unrelated/independent groups).

183