Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

Re: st: gllamm and weight


From   Steve Samuels <sjsamuels@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: gllamm and weight
Date   Fri, 26 Apr 2013 16:40:53 -0400

Lionel:

Five countries is not enough for a random-effects analysis or for
inference based on countries as clusters (e.g. Moineddin et al., 2007)
A random-effects analysis is also questionable because of the
non-random selection of countries: what population would an estimated
between-country variance component describe?

So treat countries as fixed factors; you won't need -gllamm-. If
individuals were selected in sample surveys, you can use -svy:logit-
with countries as design strata as well as analysis factors.
otherwise plain -logit- will do
.
Reference:

Moineddin, Rahim, Flora Matheson, and Richard Glazier. 2007. A simulation study of sample size for multilevel logistic regression models. BMC Medical Research Methodology 7, no. 1: 34. https://www.biomedcentral.com/1471-2288/7/34)

Steve


Dear stata users,

I have some questions about gllamm and multilevel model with weight
and any help about it will be appreciated. First, to better
understand, I describe my analysis. We have realized a survey in 5
countries (countries have not been choosen randomly). In each country,
300 people have responded to survey and in each country, to insure
that the results were representative of the target population, the
data were weighted according to age, gender. Until now, I have studied
each database separatly. Today, I have merged the five database to
obtain  a weighting single database of 1500 people. I would like to
study a binary outcome so I think a multilevel logistic regression is
the most appropriated. There are two levels : first is individual and
second is country. So I try to use xtlogit but unfortunately, I can't
use weight. So I look for gllamm and it seems possible to use weight.
The syntax code is :
gllamm outcome explanatory_variables, i(country) fam(binom)
link(logit) nip(30) pweight(poids)

I read help about gllamm and I see there are two sort of weights :
pweight (just for individuals) and weight (but integer are required).
So I think, in my analysis, i must use "pweight". I have to be careful
with suffix option (in my database weight is named "poids1" so in the
stata syntax i just write "poids"). I have no weight for level 2 i.e
country. So, if I well understand, automatically a weight of 1 is
created. I also read there might be a problem about rescaling for
weight and I read the paper of Rabe-Hesketh (2006) about it but I am
not sure to have understood everything.

So, to be sure that I don't make some mistakes about weight use,
syntax, methodology etc..., I post it on statalist. Any help would be
greatly appreciated.

Thank you very much for your response and sorry for my english (I'm french...).

Lionel FUGON
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index