Question of 'Variables.xlsx' for group comparison

Submitted by qtwing@nate.com on

Dear Johann,

I just tested group comprison results with different input in 'Variables.xlsx'.

One case is input with numbers (e.g. group 1 / group 2), the other is input with letters (e.g. group a / group b).

The results were quite different each other.

For group comparison, should I input only letters in 'Variables.xlsx' as the attached figure ?

Best regard,

Saske

Johann Kruschwitz

Tue, 03/17/2015 - 09:41

Hi Saske,

for group comparison it does not matter if you put in "letters", "group label names", or "numbers". The (parametric) results will always be the same. I belief that you may have used permutation testing to test for group differences. As this is based on random permutations it can be the case that the non-parametric p-values may differ from "run to run" (the actual group difference and the parametric p-values will remain the same as this is independent of the permutation testing). If  non-parametric p-values differ substantially, it is an indicator that you did not perform sufficient permutations to obtain a reliable estimate. The same thing holds for network based statistics.

Best,
Johann

P.S.: I just substituted the group labels "male, female" in the variables sheet from the sample workspace with "0, 1" and parametric results are identical - please try this out if you have any doubts.

First of all, thanks for your kind reply.

I tried what you mentioned.

I run with same parameter except for group labellings, but I got different results as the attached pdf file.

In addition to this problem, I wonder how I could get one-sample t-test result without any covariates(e.g., age, IQ).

I just only get a raw matrix of one specific group.

Best regard,

Saske

 

Attachment Size
figure1.JPG 35.85 KB
Forums