Many of the methods available via POSTHOC in GLM are valid only for one way independent samples designs (and some only with equal sample sizes). Thus, the POSTHOC options are not the general way to handle multiple comparisons in the general linear model in GLM. The place to go is to the EMMEANS COMPARE option. This was added in Release 7.5. It allows you to obtain omnibus tests and all pairwise comparisons among the levels of any main effect, simple main effect, or averaged main effect in any design. In Release 7.5, pairwise comparisons gave confidence intervals that were Bonferroni corrected for between factors, but LSD or uncorrected for within factors. In Release 8.0, the default for both types is LSD, but you have Bonferroni or Sidak options for either type of factor.

In the dialog boxes, click on Options. Click on the within subjects factor name and move it to the right side of the box. Then check the checkbox for Compare main effects. Finally, choose your correction (Sidak's method is more powerful than the Bonferroni and still conservative).

For those working on earlier releases of SPSS without access to EMMEANS COMPARE, there are macros on the SPSS Web site for the one way repeated measures case (http://www.spss.com/tech/macros/).

David Nichols Principal Support Statistician and Manager of Statistical Support SPSS Inc.

-----Original Message----- From: Jeff Mather [SMTP:Jmather@HARTHOSP.ORG] Sent: Friday, September 04, 1998 8:38 AM To: SPSSX-L@UGA.CC.UGA.EDU Subject: one-way repeated measures

I have a simple design with 15 subjects treated with four different regimens. I know if I use GLM repeated measures ANOVA I can get the within subject differences, but I can't get post-hoc comparison's (it's only set up for between subject factors-and I have none).

To obtain multiple comparison's, I've rearranged my data so I can run a one-way ANOVA. Here, I can get post-hoc testing, as the setup now sees the regimen as a factor. My question: I don't believe that any of the post-hoc tests offered in the simple 1-way anova are designed specifically for a paired comparison. Please let me know if I'm wrong here. What I've done is gone back to the original data set up and performed the post-hoc tests using all the combinations for a paired t-test, and manually adjusted the error rate by incorporating a bonferronni correction.

this is pretty combersum. Is my line of thinking correct here, is this the easiest way to get the post-hoc's done in a simple repeated measures design.

thanks jeff