chi-square tests for random effects in growth curve model

[Alice Nee]

Hi,

I am running a unconditional growth curve model with linear age and quadratic age. My outcome is BMI.

In Raudebush and Bryk's book titled "Hierarchical Linear Models," on Page 165, they talk about the chi-square test for random effects. Does any of you know how I should do it in Stata?

Thanks,

Alice

P.S. here are my codes and results:

mixed bmislf_npw ctage1 c.ctage1#c.ctage1 if `f1'==1 [pweight=w1_wc] || aid: ctage1, pweight(schwt1) pwscale(size) nolog cov(un) mle variance

Mixed-effects regression Number of obs = 50,915
Group variable: aid Number of groups = 14,997

Obs per group:
min = 1
avg = 3.4
max = 5

Wald chi2(2) = 7217.89
Log pseudolikelihood = -24603019 Prob > chi2 = 0.0000

(Std. Err. adjusted for 14,997 clusters in aid)
-----------------------------------------------------------------------------------
| Robust
bmislf_npw | Coef. Std. Err. z P>|z| [95% Conf. Interval]
------------------+----------------------------------------------------------------
ctage1 | .6552839 .0144448 45.36 0.000 .6269727 .6835952
|
c.ctage1#c.ctage1 | -.0111753 .0006175 -18.10 0.000 -.0123856 -.009965
|
_cons | 20.44683 .0698669 292.65 0.000 20.30989 20.58377
-----------------------------------------------------------------------------------

------------------------------------------------------------------------------
| Robust
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
aid: Unstructured |
var(ctage1) | .0830208 .0037882 .0759183 .0907878
var(_cons) | 15.67297 .5520139 14.62754 16.79311
cov(ctage1,_cons) | .0888155 .0278265 .0342766 .1433545
-----------------------------+------------------------------------------------
var(Residual) | 4.404893 .1402543 4.138401 4.688545
------------------------------------------------------------------------------

[Clyde Schechter]

I don't have Raudebush & Bryk available here, but if memory serves, what you want to do is run this model as is, store the estimates, and then re-run the model with no random effects specified. Then do a likelihood-ratio test. I'm pretty sure that's what's referred to there.

[Alice Nee]

Thank you Clyde, for your quick response.

I tried the way you described last week. But I am using weights in my xtmixed model, the likelihood-ratio test cannot be done for models with weights.

Do you have a solution for that?

Thanks,

Alice

[Clyde Schechter]

I'm afraid I don't have a solution for that. Perhaps somebody else does and will respond.

[Alice Nee]

OK. Thanks for reading my reply.

Alice