when to carry out a t-test and when to use mann whitney U?

Debate and discussion of any biological questions not pertaining to a particular topic.

Moderators: honeev, Leonid, amiradm, BioTeam

Post Reply
not.k8e
Garter
Garter
Posts: 4
Joined: Fri Apr 27, 2007 4:13 am

when to carry out a t-test and when to use mann whitney U?

Post by not.k8e » Sat Apr 28, 2007 3:29 pm

Im doing some work on exam questions for the Edexcel alternative paper and im a little confused with the stats tests (mathsis not my strong point).when is it appropriate to use the Mann whinty U test in place of the T-test or calculating correlation coefficents?

Any help would be much appreciated. :D

blcr11
Viper
Viper
Posts: 672
Joined: Fri Mar 30, 2007 4:23 am

Post by blcr11 » Sat Apr 28, 2007 10:40 pm

I guess the short answer is you can always use Mann-Whitney while t-tests are strictly valid only when the underlying distribution is Normal. Both test the hypothesis that the medians (Mann-Whitney) or means (t-test) of two samples are different against the null hypothesis that they are equal. Mann-Whitney belongs to the class of non-parametric statistics. Since there is no assumption about the nature of the underlying distribution, non-parametric statistics can be used in cases where the assumption of normality doesn’t hold. When the assumption of normality holds, both U and t should lead to similar conclusions.

As with testing for the equality of medians/means, there is a parametric (Pearson) and non-parametric (Spearman) method to test the significance of the correlation coefficient. The same considerations apply: the non-parametric Spearman method should always be valid, while the parametric Pearson method is valid when the distribution is Normal.

Grant it, most people assume normality whether it’s true or not. Fortunately, t-tests and F-tests, anyway, are fairly robust under deviations from normality.

Post Reply

Who is online

Users browsing this forum: No registered users and 16 guests