Date: Thu, 30 Jun 2005 09:57:48 -0400 cbautista@hivresearch.org "SPSSX(r) Discussion" Christian Bautista Re: Mann-Whitney U-test To: Marco van de Ven <20050630125432.12733.qmail@web25201.mail.ukl.yahoo.com> text/plain; charset=us-ascii

Hi Marco,

As far as I know, you have to take into account before applying this test the following:

- The presence of ties. This should small relative to the total number of observations - MW tests two populations equivalent in location (central tendency). If you assume that the two distributions are symetric, then you can apply this test, - When N increases the power of this test increases, as well.

In you case you can apply this test.

/Christian

|---------+----------------------------> | | Marco van de Ven | | | <mvenus82@yahoo.c| | | o.uk> | | | Sent by: | | | "SPSSX(r) | | | Discussion" | | | <SPSSX-L@LISTSERV| | | .UGA.EDU> | | | | | | | | | 06/30/2005 08:54 | | | AM | | | Please respond to| | | Marco van de Ven | |---------+----------------------------> >------------------------------------------------------------------------------------------------------------------------------| | | | To: SPSSX-L@LISTSERV.UGA.EDU | | cc: | | Subject: Re: Mann-Whitney U-test | >------------------------------------------------------------------------------------------------------------------------------|

Dear all,

Thank you for providing me with the assumptions of the Mann-Whitney U test. Yet, the responses I got were contradictory; one of you said that distributions do not matter, while the other one said they do...

Let me give you the design I want to apply it to:

It contains two groups with large, yet unequal, sample sizes; group one: 297, group two: 165. Variances are unequal, but the distributions (when looking at them with the bare eye) are roughly equal.

Do you think I can safely apply it here? In order to make the testing more accurate, I opted for the Monte Carlo approach of the MW test; with 10.000 times resampling. I hope you agree with me that this is a solid way of testing the differences between these independent samples.