RE: st: ST: postestimation test

["Bontempo, Daniel E" <deb193@ku.edu>]

Thanks Maarten this reply was helpful. I am now acquainted with "margins, post" and the subsequent use of test, as well as example #10, testing Margins, in the user guide entry for margins.

I have two followup questions.

First, if I understand the point about "test" only being available for model coefficients (or combinations), I am unsure what it was doing when I specified some combinations that were not estimated coefficients. For example:


. test 0.condition#0.ageGrp = 1.condition#1.ageGrp

 ( 1)  [condCorr_pct]0b.condition#0b.ageGrp - [condCorr_pct]1.condition#1.ageGrp = 0

           chi2(  1) =    1.38
         Prob > chi2 =    0.2402

. test 0.condition#1.ageGrp = 1.condition#1.ageGrp

 ( 1)  [condCorr_pct]0b.condition#1o.ageGrp - [condCorr_pct]1.condition#1.ageGrp = 0

           chi2(  1) =    1.38
         Prob > chi2 =    0.2402


In the 1st test, I compare the 11 interaction (which was a coefficient) to the 00 reference condition. Butin the 2nd test I tried to specify the 10 group to the 00 reference, and this was not a coefficient - yet it appeared to do something. Further what it did is identical to the result of the 1st test. What was Stata doing here?


Second, I am doing analyses for a colleague who writes for a literature where reporting mean differences, SDs, t/F stats, and p-values is the accepted practice. Is there some way to translate into t or F metric the chisq from the Wald test performed by the test command after I post the margins?

Also, can SE be sample size adjusted to yield SD with multilevel models? Would the number of level-2 groups be used as the sample size?

I know this is getting complex, but any further discussion or reference would be appreciated.

Thanks

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Maarten buis
Sent: Friday, August 06, 2010 10:17 AM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: ST: postestimation test

--- On Fri, 6/8/10, Bontempo, Daniel E wrote:
> The model showed the interaction of two factors was
> significant, and I thought I could use "test" to probe
> which pairs of means were actually different. But some
> comparisons involving the reference condition do not
> work, and I am not sure why the test is giving chisq.
> -----
> 
> . margins condition#ageGrp
<snip>
> . test 0.condition#1.ageGrp=1.condition#1.ageGrp

this test refers to the parameters in your model, not
to the table you get from margins. To test the differences
in predicted outcomes, you must first specify the -post- 
option in -margins-. 

-test- gives you a Chi square statistic  because it
produces a Wald test of possibly multiple constraints, in
which case it cannot use the normal distribution as the 
sampling distribution of the test statistic.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------


      

*
*   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/

*
*   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/