Alex R. de Leon of the Department of Mathematics and Statistics, University of Calgary, is giving the Winter 2016 SAGE/Biostatistics Seminar on Friday, January 29, 2016. Below is the summary of the talk:
Lele et al (2010) recently introduced data cloning (DC) as a convenient alternative approach for obtaining maximum likelihood estimates (MLEs) in complex models without the need for numerical evaluation and maximization of the (marginal) likelihood. By borrowing ideas from Bayesian inference and Markov Chain Monte Carlo (MCMC) computation, DC only requires the computation of sample mean values and variances of repeated draws from a posterior distribution, based on a fixed number of "clones", to approximate the MLEs and their standard errors (SEs). In addition, DC provides a simple graphical check for parameter identifiability, which is a difficult problem for GLMMs. In this talk, I will illustrate the application of DC and compare the results with standard approaches based on numerical integration (via, e.g., Gauss-Hermite quadrature methods) and maximization (via, e.g., pseudo-Newton methods) of the likelihood function. I will discuss a multivariate mixed-effects probit model with crossed random effects and illustrate the computation of MLEs approximated via DC.