re: st: posthoc testing after ANOVA

[David Airey <david.airey@vanderbilt.edu>]

> I have a question related to the FAQ. Can one form _any_ linear 
> combination of any of the cell means from any cell means model ANOVA?
Having read a bit more it seems to be so. And I'm almost getting the 
hang of posthoc testing via Stata's testanova framework.

Below I think I'm being silly and animal doesn't need to be in the cell 
means table which makes things easier to follow from the online 
examples.


> For instance, if I have a split plot design,
>
>                 within-subject factor
>                 ---------------------
>                 stimulus1   stimulus2
>
> drug     animal1  cell 1       2
>          animal2    3          4
>
> control  animal3    5          6
>          animal4    7          8
>
>
> This anova is: anova y drug/animal|drug stimulus drug*stimulus/, 
> repeated(stimulus).
>
> I was interested in the following comparison of cells after finding a 
> drug*stimulus interaction significant:
>
> ((2-1)+(4-3)) / 2 = ((6-5)+(8-7)) / 2, average of differences between 
> stim1 and stim2 by group
>
> This test would be equivalent to transforming the data to difference 
> scores (stimulus2-stimulus1), and then testing for a group difference 
> between drug and control groups. It's easy to do the latter, but I 
> wanted to know if test could do it directly and if the comparison I 
> want is an allowable test.
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/