Bayes modal estimation

For the first time in gllamm's history (which began in 1998), we have grant funding to support gllamm development. As part of a joint project with Andrew Gelman and Jingchen Liu at Columbia University, gllamm will become a little bit Bayesian. We will allow the user to specify weakly informative priors for the variance-covariance matrices of the random effects (or latent variables) to avoid estimates on the boundary.

Working paper:
Yeojin Chung, Sophia Rabe-Hesketh, Andrew Gelman, Jingchen Liu, and Vincent Dorie, "Avoiding Boundary Estimates in Linear Mixed Models Through Weakly Informative Priors" (June 2011). U.C. Berkeley Division of Biostatistics Working Paper Series. Working Paper 284.

Experimental gllamm code for gamma prior for random-intercept variance/standard deviation

• Install the latest version of gllamm (Sept 7, 2011)
• download init_prior.ado and calc_prior.ado
• copy these files into your working directory (or anywhere in adopath)
• run do-file with 8-schools data to get results from Table 1 in Chung et al. (2011)

See also blme code in R by Vincent Dorie

Grant:
2010-2013 Gelman, A., Rabe-Hesketh, S., and Liu, J. Practical Tools for Multilevel/Hierarchical Modeling in Education Research. U.S. Department of Education, Institute of Education Sciences (IES).