Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: STATA multilevel modelling

From   Steve Samuels <>
Subject   Re: st: STATA multilevel modelling
Date   Fri, 20 Jul 2012 11:57:04 -0400

On Jul 11, 2012, at 9:12 AM, Kate Xu wrote:

> I am new to STATA and I would like to check a few things about the
> possibility under STATA to model multilevel data with sampling
> weights. I am mainly interested  inoutcomes that are continuous,
> categorical, and count. So basically multilevel models for linear,
> poisson, logistic, ordinal and multinomial regressions with random
> intercepts and random slopes.
Welcome to Statalist!

Small point: correct spelling is "Stata", whereas GLLAMM is also correct
spelling for that command. "GLLAMM" is an acronym; "Stata" is not. See the FAQ.

> 1. Sampling weight. It seems that the xtmixed can model continuous
> outcomes with possibility of controlling for sampling weight, but not
> for xtmelogit (logistic) and xtmepoisson (poisson)?

Correct, and you can tell this from each command's -help- file, since [weight] is not mentioned as an option. If it had been, you would need to check if "pweight" was a possibility.

> 2. Is it possible to use multilevel model for ordinal and multinomial
> data under STATA, with the weighting taken into account?

With GLLAMM.  Note that sampling weights are not the only issue: you must
decide if you want to base standard errors on the model or on the survey design.

> 3. What is the difference of these STATA based commands and GLLAMM? I
> understand that GLLAMM is a user written package of STATA that can
> also model multilevel data. What is the advantage of it. And is GLLAMM
> able to model the ordinal and count data with random slopes and
> weighting?

Just answered.

Some people report that GLLAMM can be
slow.  But if you run -xtmelogit- first, you can get good starting values, which
should speed things up.  Note that GLLAMM requires a different sampling  weight
 for each level.

See: for some references.

You can calculate the two weights from contributed commands -pwigls- or -mpml_wt

found at:


*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index