st: RE: RE: t values using Scheffe

[Nick Cox <n.j.cox@durham.ac.uk>]

David's miniature review will be useful to many. One addition -- there may well be many more -- would be stuff under a false discovery rate banner. 

Those of us somewhat bemused by the whole activity might see this as yet another example showing that the great thing about standards is that there are so many to choose from. 

Nick 
n.j.cox@durham.ac.uk 

Airey, David C

Some sources (Kirk, experimental design, 3rd ed.) recommend Scheffe's correction when you are also entertaining non-pairwise contrasts. It is a procedure that allows multiple test correction for all kinds of post-hoc hypotheses, unlike some others that are for pairwise tests only. Its flexibility comes at a cost of power, Kirk says. The fact that it is implemented in Stata in a situation where you are only doing pairwise comparisons with oneway then seems odd. The relationship between the t statistic from a t-test of two group means, and the critical F statistic used by the Scheffe test is given in the manual for oneway.

There are other canned solutions besides manual computation available:

net describe sg101, from(http://www.stata.com/stb/stb47) 
net describe tukeyhsd, from(http://www.ats.ucla.edu/stat/stata/ado/analysis) 
net describe tkcomp, from(http://www.ats.ucla.edu/stat/stata/ado/analysis) 
net describe fhcomp, from(http://www.ats.ucla.edu/stat/stata/ado/analysis) 
net describe dunnett, from(http://www.stata.com/users/tboswell/)

Nick Cox 

> Multiple comparisons are a branch of theology. The true way is unknown, but many people have tried to work out what it might be, and most of those are confident that they are very right and the other people very wrong. 
> 
> Be that as it may: 
> 
> As I understand it, and I'm open to correction: 
> 
> 1. No t statistics are saved after -oneway-. 
> 
> 2. The Scheff\'e adjustment doesn't work with a different definition of t. It just provides its own P-values as a correction for multiple comparisons. 
> 
> 3. Thus there is no definition of t-statistic here independent of what -ttest- would provide pairwise, and manifestly any call of -ttest- knows nothing of whatever other inferences you are making, simultaneously, sequentially or otherwise. 

Claude Francoeur
 
>> I'm using the Scheffe option which I find very useful to compare 
>> multiple group means :
>> 
>> oneway func lsi,tabulate scheffe
>> 
>> This command generates differences in means and p-values. Is there 
>> anyway to get the t values also?

*
*   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/