# Doubt in performing Kruskal Wallis test

#### fskywalker

##### New Member
Hi Folks:

I'm a newbie in using SPSS. I'm trying to find a way to perform a Kruskal Wallis test (non parametric) in SPSS. I have idefined in SPSS data that I transferred from Ms Excel out of my thesis questionaire, which has a 19 question section of a 1-5 likert scale, plus 8 demographic questions, for a total of 131 respondents. (so have a total of 27 variables.) I'm trying to perform the test between 2 sub-groups of data (0-4 years of experience versus 15 + years of experience, both variables are from demographic data) in SPSS and have 2 questions:

1- I have defined the 19 variables of the likert data as ordinal, while the 8 variables for demographic as nominal. Is this the right way to do it?
2- Under non parametric tests, I selected "Tests for several independent samples", where it asks for the test variable list and the grouping variable. I understand the 19 variables of the likert would go on the test variable list, since want to check the outcome of all 19 variables. Then, what should I input as the grouping variable?

Thanks,

Francisco

Last edited:

#### Fierce

##### New Member
1-Correct for the likert-scale variables but I think some demographic variables can be ordinal (or scale, depending on how they were measured).
2-If you want to check the difference between years of experience, the grouping variable should be the "year of experience" one - or a variable based on it (if you have values from 1 to 15, you should compute a new variable like: 0-4=1, 15+=2).
One more thing: If you have only two sub-groups, you can use a 2 independent sample test (Mann-Whitney).

#### fskywalker

##### New Member
1-Correct for the likert-scale variables but I think some demographic variables can be ordinal (or scale, depending on how they were measured).
2-If you want to check the difference between years of experience, the grouping variable should be the "year of experience" one - or a variable based on it (if you have values from 1 to 15, you should compute a new variable like: 0-4=1, 15+=2).
One more thing: If you have only two sub-groups, you can use a 2 independent sample test (Mann-Whitney).
Thanks FIERCE, I guess I'm getting closer to do what I need. I did define the various levels of groups for experience as follows:
1= 0-4 years
2= 5 to 10 years
3= 11 to 15 years
4= More than 15 years

So, under variable list I input the 19 variables, while on the grouping variable I select the number of years. Under the grouping variable I can, for the Mann Whitney test, input the expericne variable as the grouping variable and then on the define group I can enter the 2 groups under the experience variable, for which I use 1 and 4 so that it compares 0-4 years versus 15 + years. However, for the Kruskallis test under the grouping variable it asks for a define range, where it asks for a minimum and a maximum, so don't see how to perform the test among just the groups 1 and 4. How do I perform the comparisson then?

#### Fierce

##### New Member
There's no need for a Kruskall-Wallis test if you want to test only 2 sub-groups. The K-W test is used to compare more than 2 groups (in your case, compare all 4 groups of experience). That's actually the reason why you select range for K-W and values for M-W.
Anyway, if, for some othe reason, you do need a Kruskall-Wallis test, you canchange the order of your new variable so 0-4=1 and 15+=2 OR use DATA>SELECT CASES select only cases in which experince is 0-4 or 15+.

#### fskywalker

##### New Member
Thamks!! One last question

There's no need for a Kruskall-Wallis test if you want to test only 2 sub-groups. The K-W test is used to compare more than 2 groups (in your case, compare all 4 groups of experience). That's actually the reason why you select range for K-W and values for M-W.
Anyway, if, for some othe reason, you do need a Kruskall-Wallis test, you canchange the order of your new variable so 0-4=1 and 15+=2 OR use DATA>SELECT CASES select only cases in which experince is 0-4 or 15+.
Thanks!! One last question: I know p < 0.05 (SIG in SPSS) is the trigger for rejecting the nule hypothesis; what would be null hypothesis rule on this test:
Ho: Medians of 2 groups are the same
Ha: Medians of two groups are not the same

so SIG less than 0.05 means that the medians of the 2 groups are not the same?

Thanks!

Is that correct?

#### Fierce

##### New Member
Correct Ho, Ha and conclusion - but keep in mind that you don't really need to use 0,05 as alpha; you (or the researcher/client/teacher anyway...) can decide if it is significant or not at different alpha levels.

#### fskywalker

##### New Member
independent samples?

Hi FIERCE. I have another basic question. If I did one sample of 131 questionaires out of a group, and am then making sub groups out of those 131 responses based on the 0-4 and 15+ years of experience; are those 2 groups consider independent samples out of the original 131 sample collected? I ask because there are formulas for dependent and independent samples, both Kruskalis and Whitney assume independent samples and need to make sure am analyzing the data the right way.