SPSS-FORUM.DE

[BorisBB]

I have a question asked "Which class to students?" and get answers like 1,2,3,4 and a another question asked with answer yes and no. Which test to choose to see the significant difference according to the classes? According to me first question is ordinal independent second one is nominal dependent if I use chi square I have to look linear by linear line am I right? 1-4Y/16N, 2-4Y/14N, 3-4Y/8N, 4-7Y/6N (Yes No with classes)

[ponderstibbons]

Mann-Whitney U-test (a.k.a. Wilcoxon rank sum test).

Mit freundlichen Grüßen

PonderStibbons

[BorisBB]

While using the mann whitney u test than I will choose grouping variable yes/no and test variable classes (1,2,3,4).
(I thought the reverse, I have 4 grouping variable and cant use mann whitney).
When I applied the test I got Exact Sig (1-tailed) and Exact Sig (2-tailed) then I have to look 1-tailed am I right?

[ponderstibbons]

I have a question asked "Which class to students?" and get answers like 1,2,3,4

1, 2, 3, 4 = categorical? or ordinal?

Mit freundlichen Grüßen

PonderStibbons

[BorisBB]

Class numbers in a department like 1st class 20 sudents, 2nd class 18 student etc. I thought ordinal. (Classes also grade 1st grade 2nd grade etc.)

[ponderstibbons]

The U test tells you whether those who say "yes" tend to be in higher (or lower, respectively) classes
than those who say "no". The advantage of this is, that you retain the ordinal level of "class". Treating
"class" as categorical (as by using a Chi2 test) destroys statistical information. Nevertheless, a 4x2 table
with a Chi² test would not be wrong (just not the best solution).

With kind regards

Karabiner

[BorisBB]

Thanks for your help. I choose Mann Whitney U test.
With these classes I have another question independent with nominal answers (4 choice). I am looking for significant difference in answers according to the classes? Then I have to use Kruskal Wallis (Test variable classes, grouping variable answers). Am I on the right track?

[ponderstibbons]

If the response variable is categorical, then cross-tabulation with Chi² test.
If the response variable is ordinal, then Kruskal-Wallis.

Regards

PonderStibbons

[BorisBB]

The response variable is categorical but when I use chi square 12 cells have expected count less than 5. (%75)

[ponderstibbons]

So possibly you should consider if you can leave out a sparsely
used category, or if you collapse categories, or if you have SPSS
exact tests installed, you can use an exact Chi² test
which does not have the assumption of expected cell frequencies
>= 5.

Regards

Karabiner