Re: st: RE: Re: pwcompare

[Janet Hill <janethill73@yahoo.co.uk>]

Tony,

Thanks for that. I will go back to West, Welch & Galecki to try and improve my understanding of the underlying theory.

I am now totally confused as to how to compare the groups after using xtmixed in a repeated measures scenario, can anybody offer me any suggestions or advice?

Thank you,
Janet


----- Original Message -----
From: "Lachenbruch, Peter" <Peter.Lachenbruch@oregonstate.edu>
To: "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Cc: 
Sent: Saturday, 26 November 2011, 18:44
Subject: st: RE: Re: pwcompare

I didn't see an answer to this.  I suspect it is related to the fact that the standard errors in the xtmixed model account for the random effects, and so are larger.  Note that the contrasts are the same in both as they are the sample means.

Tonyh

________________________________________
From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] On Behalf Of Janet Hill [janethill73@yahoo.co.uk]
Sent: Friday, November 25, 2011 8:35 AM
To: statalist@hsphsun2.harvard.edu
Subject: st: Re: pwcompare

Why does pwcompare give different results after anova and xtmixed? Using the following syntax:
xtmixed lhist i.group##i.time || dog:, reml cov(id)
or
anova lhist group / dog|group time time#group, repeated(time)
followed by
pwcompare group, emptycells(reweigh) eff asbalanced post

I get the following for xtmixed
------------------------------------------------------------------------------
             |                            Unadjusted           Unadjusted
             |   Contrast   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
lhist        |
       group |
     2 vs 1  |  -.4591643   .5681373    -0.81   0.419    -1.572693    .6543643
     3 vs 1  |   1.200595    .525993     2.28   0.022     .1696677    2.231522
     4 vs 1  |  -.3857097    .525993    -0.73   0.463    -1.416637    .6452175
     3 vs 2  |   1.659759   .5681373     2.92   0.003     .5462306    2.773288
     4 vs 2  |   .0734546   .5681373     0.13   0.897    -1.040074    1.186983
     4 vs 3  |  -1.586305    .525993    -3.02   0.003    -2.617232   -.5553775
------------------------------------------------------------------------------
and for the anova
------------------------------------------------------------------------------
             |                            Unadjusted           Unadjusted
             |   Contrast   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       group |
     2 vs 1  |  -.4591643   .1047385    -4.38   0.000    -.6722564   -.2460723
     3 vs 1  |   1.200595    .096969    12.38   0.000      1.00331     1.39788
     4 vs 1  |  -.3857097    .096969    -3.98   0.000    -.5829947   -.1884248
     3 vs 2  |   1.659759   .1047385    15.85   0.000     1.446667    1.872851
     4 vs 2  |   .0734546   .1047385     0.70   0.488    -.1396375    .2865467
     4 vs 3  |  -1.586305    .096969   -16.36   0.000     -1.78359    -1.38902
------------------------------------------------------------------------------

I can see that the standard errors are different, but the question arises should I use xtmixed or anova for repeated measures?

I am using Stata 12.0, update 10 Nov 2011.

Thanks,
Janet

*
*   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/


*
*   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/