CHAPTER 7

Inferential Statistics
 Parametric vs. non-parametric tests
The assumptions for parametric test
The assumptions for Non-parametric test
1. Equal interval or ratio level data (scale • If any of assumptions 1, 2, and 4 are
data in SPSS terminology).
not met then non-parametric tests
2. Normal distribution or closely so. should be used,
3. Homogeneity of variance. That is, the
variance should be similar in each • But because non-parametric tests
group. Variance is the standard require ordinal or nominal data they
deviation squared. SPSS will provide a are less powerful and less sensitive
measure of this in Levine’s Test in to real differences between groups
producing the t test output. We will than parametric tests.
demonstrate this in the SPSS practice
activity below.
4. Samples randomly drawn from the
population.
Examples of parametric tests
 One-sample t-test
• The one-sample t-test is used to compare the mean of a sample with a specified test
value. For example, it could be used to see whether the mean of the scores on the variable
att1 is significantly different from 3, the neutral point of the scale for att1 which ran from 1
to 5.
• Obtain the one-sample t-test by selecting it from the Analyze /Compare Means menu. Select
the variable to be tested and type into the text box the test value with which the mean is to
be compared.
 Within-subjects (paired-samples) t-test
• used when testing for significant differences between two samples which
are ‘related’.
• it will be assumed that the scores are obtained from the same subjects. Example:
• In a before-and-after type of study where changes after an intervention are compared with the performance
level before the intervention.
• The use of paired samples permits a more precise analysis because there is greater control of
sources of error stemming from individual differences.
• For example, the head of a government department wants to determine whether typing
speed depends on the word processing software used. If 20 administrative officers use
the Write software package and 20 use the Input software package and the results compared,
this would be an independent groups design.
• But differences in results might be due to individual typing skill anyway and not the packages.
• But if 20 administrative officers were each tested on both packages using counterbalancing then
the design is the related or repeated measures design.
 Within-subjects (paired-samples) t-test-cont’d
• Computing the effect size (d) statistic for the parametric paired-samples t test
• SPSS supplies all the information necessary to compute d.

d= or

• Regardless of sign, d values of 0.2, 0.5, and 0.8 traditionally represent small, medium
and large effect sizes, respectively.
 Within-subjects (paired-samples) t-test-cont’d
• In SPSS, for the paired-samples t-test
Analyze
Compare Means
Paired-Samples T Test

• Indicate the two variables to be compared by clicking on each of them;

• the first one will appear as the Current Selections Variable 1, and
• the second one will appear as Current Selections Variable 2.
• The paired t-test involves taking the difference between the two scores for each
respondent and finding the mean of these difference scores.
• Example: There is a significant difference between attitude1 and attitude2 for
salespersons.
 Within-subjects (paired-samples) t-test-cont’d
• Output from a paired-sample t-test

The value of the t statistic is then shown, with its degrees of freedom (df) and its significance level (Sig. 2-tail): you can
conclude there was a statistically significant difference between the means of the two sets of scores.
 paired-samples t-test-cont’d
• How to report the results:

• ‘A paired samples t test (N = 22) was conducted to evaluate whether

there was a significant difference between attitude 1 and attitude 2 of
salesperson. The mean scores between attitude 1 and attitude 2 differ
significantly (t = 3.196, df= 21, p < .001) with attitude 2 having a
significantly higher mean (M= 3.09) than the mean of attitude1 (M=
2.54).
 Between-subjects (independent-samples) t-test
• The aim is to determine whether two groups differ significantly or not.
• There are only two possibilities.
• One is that the means of both groups only differ by chance sampling variations (i.e. not
statistically significantly – maintain null hypothesis).
• The other is that their means do differ statistically significantly.
 Independent Samples T Test-Cont’d
• You need two variables to calculate an independent t test.
• The first variable consists of the scores of the dependent variable for both groups. The
dependent variables to be compared have to be inserted into the Tests Variable(s) text
box.
• The second represents the independent variable (or grouping variable) which must be
a nominal variable
• has to be inserted into the Grouping Variable text box, and you then have to click the Define
Groups button.
• The normal procedure is to enter into the Group 1 and Group 2 windows the values
on the grouping variable which define the two groups to be compared.
• For example, if you wanted to compare the scores on sales_1
of employer 2 with employer 3, you would insert the variable
(empl) in the Grouping Variable box of the Independent
Samples T Test dialogue box
Independent Samples T Test-Cont’d

Output for an independent t-test

Independent Samples T Test-Cont’d
• The printout consists of two sub-tables.
• The top one summarizes the descriptive statistics for the two groups (Smith & Co and
Tomkins) on the dependent variable of Sales_1. The mean Sales_1 score for Smith & Co is
$4,625.1713 while that for Tomkins is $ 8,207.4150. Is this difference of $3,582.2437
sufficiently large for us to say that the two groups are significantly different in terms of
Sales volume? The bottom table reveals this.
• In the bottom table, the output for the independent groups t-test on
SPSS is somewhat confusing because there are two versions of the
independent samples t-test.
• Which one you should use depends on whether the estimated
variances for the two groups of scores are significantly different or
not.
• Read the columns under Levene’s Test for Equality of Variances. This test tells you whether you have
satisfied the homogeneity of variance assumption. If the Sig-value is statistically significant then your
variances are unequal and you must use the lower line of figures.
• Non-significance indicates that variances are approximately equal so use the upper line.
Independent Samples T Test-Cont’d

How to report the results:

• An independent-samples t test was conducted to evaluate the
hypotheses that Smith & Co and Tomkins differ significantly in their
sales volume levels. The mean sales volume of Tomkins (M =
$8,207.41, sd = 925.18) was statistically significantly different (t
=6.48, df = 12, two-tailed p = .000) from that of Smith & Co (M =
$4,625.17, sd = 1,089.79).
 Analysis of Variance Techniques (ANOVA)
• ANOVA. This is a hypothesis testing procedure used to determine if mean differences exist
for three or more samples or treatments.
• t tests are limited to two treatment conditions.
• When we are looking at a complex situation with more than two conditions of the IV, the
methodology for dealing with this situation is classified under the general family of
analysis of variance (ANOVA) techniques.
The logic of ANOVA:
• ANOVA is built on comparing variances from two sources. These two estimates are:
• The Between Group variance – a measure of the effect of the independent variable plus the error
variance.
• The Within Groups variance – the error variance alone.
• The ratio between them is called the F ratio. The ratio is interpreted in association with the degrees of
freedom associated with sample sizes.
F=
• So the bigger the value for F, the more likely that the difference is significant
and will also be found in the population means.
 Analysis of Variance Techniques (ANOVA)-Cont’d
• Rejecting the null hypothesis indicates that at least one significant
difference exists between the means.
• Analysis of variance can tell you that there is a difference between the
means of the three or more groups of respondents, but it does not tell you
just where the differences occur: is group 1 different from both group 2 and
group 3, does group 2 differ from both group 1 and group 3?

• To answer these questions we need to perform a post hoc analysis (post hoc
is Latin for after the event). The major post hoc tests are Tukey’s HSD,
Scheffe, Bonferroni and the Games-Howell procedure.
• The Bonferroni is best when there are only a few comparisons
• A post hoc test makes pair-wise comparisons.
 Analysis of Variance Techniques (ANOVA)-Cont’d
Different types of analysis of variance:
• one-way analysis of variance
• Repeated measures analysis of variance. Analysis of variance in which each individual is
measured more than once so that the levels of the independent variable are the different
times or types of observations for the same people.
• factorial ANOVA: When two variables (factors) of interest are to be examined at the same
time.
• Which ANOVA do you require? is determined by the design of your study.
How many independent Between or within SPSS procedure
variables?* subjects? required
1 Between One-way ANOVA
1 within GLM

2 or more All between Factorial analysis

 ANOVA-cont’d: One independent variable, between-subjects
• Select from the menu system
Analyze
Compare Means
One-way ANOVA
• Suppose you want to compare the scores on sales_1 for respondents from the three
different employers in the file salesq. This involves the one-way analysis of variance and can
be achieved by following the stages listed here:
1. Click on its name and then on the right-pointing arrow button.
2. Click on the name of factor in the variable list and then click on the right-arrow button
next to the box headed Factor
3. Request the means of the subgroups. If you want the output to show you the means on
the dependent variable (sales_1 in this example) for the various levels of the independent
variable, click on the Options button; yet another dialogue box is revealed, and you select
Descriptive.A plot of the means can also be requested. When you have finished with the
Options window, click on Continue.
4. Request any post hoc tests by pressing the Post Hoc button and completing the dialogue
boxes revealed.
 ANOVA-cont’d: One independent variable, between-subjects
• Output of analysis of variance
One-way ANOVA-cont’d

• How to report the results

‘The effect of types of employer on sales amount was significant overall (F(2,19)
=6.8, p < 0.001). Post hoc tests indicate that the means the Smith and Co group
differs significantly from the Tomkins group; the Tomkins group differs significantly
from both the Jones and Son group and the Smith and Co group. The findings
indicate that there was a significant difference between the employers in the
sales_1 scores.
Non-parametric Analyses:
This part is a Reading
Assignment
Inferential statistics: Parametric test & Non-parametric test
Parametric test Non-parametric test
1. One sample t-test One-sample t-test (chi-square)

2. Paired samples t-test (for two related samples) 2 related samples test (Wilcoxon test)

3. Independent samples t-test (2-indendent samples) 2 independent samples t-test (Mann-Whitney

U test)

4. For more than 2 samples (ANOVA)

4. 1. Paired samples………………….GLM K related samples (Friedman test)

4.2. Independent samples………One-way ANOVA K independent samples (Kruskal Wallis test)

 Exercise:
Determine the type of statistical test to be used for the following situations:
• If we hypothesize that there is a significant difference in self-concept between those
who smoke and those who do not.

• If we wanted to test the hypothesis that there is a significant difference between

starting salary and current salary of the employees.

• For a study on whether there are significant differences in current salary between
employees who hold different academic qualifications (DIPLOMA, First Degree, Masters
Degree, Doctorate Degree.
• Do female employees earn less than male employees?’