Date: Fri, 4 Sep 1998 10:52:50 -0500
Reply-To: "Nichols, David" <nichols@SPSS.COM>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: "Nichols, David" <nichols@SPSS.COM>
Subject: Re: Non-par tests
NPAR TESTS does not have any ability to incorporate covariates.
Siegel's classic though now very old text indicates that the K-S test is
apparently slightly more efficient in small samples, but the M-W is better
in larger samples. No definitions of exactly what that means or the
crossover point are given. I assume much work has been done since the 1950s,
so I wouldn't take this as definitive.
Principal Support Statistician and
Manager of Statistical Support
From: Alex Walker [SMTP:A.Walker@unsw.EDU.AU]
Sent: Thursday, September 03, 1998 12:41 AM
Subject: Non-par tests
When comparing groups using non-parametric tests, is there any way
controlling for a covariate type variable, as in ANCOVA?
Secondly, having found a group difference (3 levels) with a
Kruskall-Wallis, I compared 2 pairs of groups based on a priori
assumptions and used both Mann-Whitney and Kolmogorov-Smirnov (not
knowing which was appropriate). The Kol-Smi gave less significant
results on several variables than the Mann-W. On what basis should I
make a decision about which test is more appropriate?? The help
talks about location and shape with the Kol-Smi- is it better for
samples? I have 30 subjects in each group.
Thankying you in anticipation.