st: gllamm and relative risk

["Hillel Alpert" <HALPERT@hsph.harvard.edu>]

I appreciate your interest in helping.

"Does not work" means the following error message, which returns with link(log) specified and not with link(logit).

pweight not allowed
r(101);

Has anyone used pweight with link(log) or know how to incorporate probability weights in a gllamm model with link(log)?

Hillel Alpert




Stas Kolenikov

To  mailto:statalist%40hsphsun2.harvard.edu )
Subject  Re: st: gllamm and relative risk
Date  Mon, 15 Nov 2010 12:13:17 -0600
"Does not work" meaning what? Does not converge? Says "log(link) is not appropriate with binary data and pweights" (I highly doubt that)? Issues some other sort of invalid syntax message when it processes the options? Produces some other awkward error, like matrix is not invertible at some point? Check what the FAQ says about specific vs vague requests. I know nothing about relative risks, but I suspect that taking logs of probabilities will lead to weird results unless all probabilities are tiny. Otherwise, there's a risk of getting the predicted probability greater than 1. That might be what -gllamm- stumbles upon, but that ain't -gllamm-'s fault, that's your modeling decision. 

On Mon, Nov 15, 2010 at 11:22 AM, Hillel Alpert< wrote:> Thank you for your response and advice.>> I have a single weight variable at level 1. The labeling seems to be correct. The command with link(logit) works with pweight, but still not with link(log).>
*
*   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/