Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | rgutierrez@stata.com (Roberto G. Gutierrez, StataCorp) |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: Deriving Bayes estimates from xtmelogit |
Date | Thu, 09 Sep 2010 11:11:34 -0500 |
Jamie Fagg <j.fagg@qmul.ac.uk> asks: > I am trying to derive Bayes estimates from a 3-level logistic regression > model in Stata version 10.1. > The model has the following structure, where psychometric items are nested > in individuals, nested in geographic areas. > Level 1: psychometric items > Level 2: individuals > Level 3: geographic areas > I read in Rabe-Hesketh and Skrondal (Multilevel and Longitudinal modelling > using Stata, 2008, p.162) that "Empirical Bayes predictions of the random > intercepts ... can be obtained" from xtmixed using -predict- with the > reffects option > I couldn't find any analogous advice in the section about multilevel > logistic regression models so I ran the model and ran the predict eb, > reffects (see code below). > Is this the correct way to derive Bayes predictions from xtmelogit? Yes this would be correct, with the one caveat that what you obtain are empirical Bayes _modal_ predictions, rather than empirical Bayes mean predictions. In a logistic regression setting, the posterior distribution of the random effects is no longer symmetric, and thus the posterior modes and the posterior means, while very similar for most data, are not strictly equal. Posterior mean predictions after multilevel logistic regression are not available in current official Stata, but you could obtain these by alternatively fitting your model using -gllamm-; see -ssc describe gllamm- for details. --Bobby rgutierrez@stata.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/