Compare groups: parametric vs non-parametric test

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

remover
Posts: 17
Joined: Wed Dec 01, 2010 4:04 pm

Compare groups: parametric vs non-parametric test

Postby remover » Wed Mar 14, 2012 3:56 pm

I conducted an experiment with two different groups of 8 participants. I want to compare their scores on a survey. Tests showed that the data for both groups are normally distributed and have equality of variance.

Should I conduct a t-test or Mann-Whitney U and why?

Thanks.
StatAdvice
Posts: 43
Joined: Sun Mar 11, 2012 2:26 pm

Re: Compare groups: parametric vs non-parametric test

Postby StatAdvice » Wed Mar 14, 2012 6:47 pm

Normality tests didn't reject the normality hypothesis because of extremely small sample size. In fact the classical assumptions of parametric methods do not hold. No method works with such a small dataset. But nonparametric tests are recommended with small samples. But no matter which method you use, my prediction is that you will reveal no differences between 2 groups because of small sample size.

Bootstraping may help make you analysis more elegant (available in Stata and newer versions of SPSS, for instance).
Eugene
Statistical Consultant
www.StatAdvice.com
info@StatAdvice.com
skype login: StatAdvice
remover
Posts: 17
Joined: Wed Dec 01, 2010 4:04 pm

Re: Compare groups: parametric vs non-parametric test

Postby remover » Wed Mar 14, 2012 10:33 pm

StatAdvice wrote:Normality tests didn't reject the normality hypothesis because of extremely small sample size. In fact the classical assumptions of parametric methods do not hold. No method works with such a small dataset. But nonparametric tests are recommended with small samples. But no matter which method you use, my prediction is that you will reveal no differences between 2 groups because of small sample size.

Bootstraping may help make you analysis more elegant (available in Stata and newer versions of SPSS, for instance).
Thanks for the reply... it's funny because I don't see anything mentioned about that in the Andy Field Discovering Statistics book. And that is supposed to be an introductory book.
StatAdvice
Posts: 43
Joined: Sun Mar 11, 2012 2:26 pm

Re: Compare groups: parametric vs non-parametric test

Postby StatAdvice » Thu Mar 15, 2012 6:20 am

The intuition behind this is that when your sample size is small condifence intervals for the means are wide and they will certainly intersect (in most real-world problems). So statistical tests won't reveal any differences - they are afraid of detecting differences on small samples.
Eugene
Statistical Consultant
www.StatAdvice.com
info@StatAdvice.com
skype login: StatAdvice
remover
Posts: 17
Joined: Wed Dec 01, 2010 4:04 pm

Re: Compare groups: parametric vs non-parametric test

Postby remover » Thu Mar 15, 2012 2:51 pm

StatAdvice wrote:The intuition behind this is that when your sample size is small condifence intervals for the means are wide and they will certainly intersect (in most real-world problems). So statistical tests won't reveal any differences - they are afraid of detecting differences on small samples.

Thanks for that info...

Who is online

Users browsing this forum: No registered users and 1 guest

cron