topic Non-paramentric: reporting Wilcoxon Two-sample tests in Statistical Procedures

Non-paramentric: reporting Wilcoxon Two-sample tests

deleted_user — Mon, 25 May 2009 02:51:08 GMT

Hi. I am doing a lot of non-parametric testing. I am running exact Wilcoxon two-sample tests. However, I am not sure what to report from the SAS output, specifically which "statistic" to report. The outprint provides an "S" statistic and a "Z" statistic, and I'm not sure which one should be reported alng with the p-value. I don't even know what each means. Any advice you can give would be appreciated.

As well, why is the Wilcoxon p-value different than the Kruskal Wallis test p-value? I wlways thought these two tests were comparable with a 2-level independent variable.

Thanks again.

Re: Non-paramentric: reporting Wilcoxon Two-sample tests

Doc_Duke — Mon, 25 May 2009 14:02:19 GMT

First, the standard Wilcoxon (or Mann-Whitney) two sample test is equivalent to the Kruskal-Wallis test when there are just two groups. The p-values are not the same because the normal approximation includes a continuity correction. Form the documentation: "The normal approximation includes a continuity correction. To remove this, you can specify the CORRECT=NO option. "

The "S" statistic is pretty much useless for external reports because it's magnitude is a function of the sample size. There is a one-to-one correspondence between the uncorrected Z statistic and the p-value, so I generally just report the p-value.

Doc Muhlbaier
Duke