Ranking Test (non-parametric)

Question 1

When are non-parametric tests used?

Answer 1

Tests that are used when assumptions of symmetry are violated. The data does not have the normal ‘bell-shaped’ distribution

Question 2

How do we analyse data in non-parametric tests?

Answer 2

We rank the raw data then carry out the relevant significance tests for the data.

Question 3

Why do we do ranking with non-parametric tests?

Answer 3

Because it gives us a standard distribution of scores with standard characteristics. It is also familiar and intuitive to us.

Question 4

What is the non-parametric tests for the two types of t-test?

Answer 4

Paired - Wilcoxon Matched Pairs Test

Independent - Mann-Whitney U-Test

Question 5

How does the Wilcoxon Matched Pairs Test work?

Answer 5

Done similarly to related t-test.
Get data, find the D score.
Then we rank the D scores and create a new column on our table called the Rank of Difference ignoring sign during ranking and enter this data here.
When ranking the D scores, we ignore the +/- signs. Once it has been ranked we will reattach these at the end.

To conduct the test we:
Sum positive ranks
Sum negative ranks
Select lowest summed rank (this is the T score)
Check significance table

Question 6

Do we get given a T score or Z score in SPSS when conducting Wilcoxon?

Answer 6

We are given a Z score.

Question 6

What is the Z score in Wilcoxon?

Answer 6

This is the score that is calculated when working with more than 25 pairs. If it is 25 or less pairs then we just use the T score. The Z score is derived from the T score.

Question 7

How does the Mann-Whitney U-Test work?

Answer 7

First we rank all scores irrespective of their group.
Then we select the group with most participants and sum the ranks. If the groups have the same amount of participants then we can use either.
Carry out the formula and check U score against significance tables.

Question 8

How do we conduct these tests on SPSS?

Answer 8

Analyze -> nonparametric tests -> legacy dialogs -> 2 independent samples

Question 9

What do we do when we have 3 or more groups in non-parametric testing?

Answer 9

Friedman test for related data (equivalent of related ANOVA)

Kruskal-Wallis test for unrelated data (equivalent of unrelated ANOVA)