A common trick is to translate xvar1 and xvar2 such that the value 0 is
meaningful, e.g. by substracting the mean or the minimum observed
value. Different translations can lead to very different estimated
variances and covariances of the random effects, and might lead to
convergence.
-- Maarten
--- Glenn Goldsmith <[email protected]> wrote:
> I am trying to fit a linear mixed model using xtmixed with two random
> coefficients and a random intercept, using the following syntax:
>
> xtmixed depvar xvar1 xvar2 [otherxvars] if touse || groupid : xvar1
> xvar2,
> covariance(unstruct)
>
> The model appears to converge, but two of the standard errors are
> missing,
> as shown below (all other standard errors are present):
>
>
----------------------------------------------------------------------------
> --
> Random-effects Parameters | Estimate Std. Err. [95% Conf.
> Interval]
>
-----------------------------+----------------------------------------------
> --
> groupid: Unstructured |
> sd(xvar1) | .5210501 .0146728 .4930711
> .5506168
> sd(xvar2) | .7152157 .0202753 .6765606
> .7560793
> sd(_cons) | 3.777054 .1089148 3.569505
> 3.996671
> corr(xvar1,xvar2) | -.8403815 . .
> .
> corr(xvar1,_cons) | .2627643 .0004142 .2619524
> .2635758
> corr(xvar2,_cons) | -.739374 . .
> .
>
-----------------------------+----------------------------------------------
> --
> sd(Residual) | .8609064 .0026954 .8556397
> .8662056
>
----------------------------------------------------------------------------
> --
> LR test vs. linear regression: chi2(6) = 6855.91 Prob > chi2
> =
> 0.0000
>
> I'm not sure how to interpret this. Does it mean that the model has
> not
> actually converged? There are a number of "(not concave)" error
> messages in
> the iteration log, but no problems are reported in the final 5
> iterations.
>
> Any suggestions for fixing the problem would be very gratefully
> received.
> I'm reluctant to impose a covariance structure, as there's little
> reason to
> think this is theoretically justified (indeed, part of my substantive
> interest is in the unstructured covariance estimates).
>
> Thanks in advance,
>
> Glenn.
>
> *
> * 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/
>
-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands
visiting address:
Buitenveldertselaan 3 (Metropolitan), room N515
+31 20 5986715
http://home.fsw.vu.nl/m.buis/
-----------------------------------------
Send instant messages to your online friends http://uk.messenger.yahoo.com
*
* 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/
