st: Re: Contrasts and repeated measures

[David Airey <david.airey@Vanderbilt.Edu>]

.

It might be easier to use the amusingly written advice on this page:

http://www.uvm.edu/~dhowell/StatPages/More_Stuff/RepMeasMultComp/RepMeasMultComp.html

even though SPSS is the software used.

You say you have 4 treatments, with subjects nested in treatments (I'm  
assuming, but you don't make this clear), and each subject is  
repeatedly measured (5 times), and you want to test for treatment  
differences at two time points.

Ignoring polynomial contrasts for trend analysis, which might get you  
out of lots of post-hoc comparisons, you could use paired t-tests (if  
interested in time 1 versus time 2 in treatment 1 for example) and or  
independent sample t-tests (if interested in treatment 1 vs treatment  
2 at time 1 for example), and separately compute a critical statistic  
for the family of tests as suggested.

One point is that with repeated measures designs, comparisons of the  
repeated measures may benefit from just using the error associated  
with the data used in the comparison rather than the model error.

The -test- command following anova for a split plot like "anova y b /  
subject|b w b*w /, repeated(w)" or more complicated ANOVA may not be  
using the error term you really want. You may have to pick that up  
separately from another anova on a data subset if its not left behind  
in the -ereturn list-, in a manner similar to described in another FAQ  
(http://www.stata.com/support/faqs/stat/singledof.html).

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/