Ed Merkle, University of Missouri

Researchers are increasingly amenable to Bayesian statistical methods, but there remains an obstacle preventing their everyday use: the estimation methods remain slow and inefficient for many types of models. With this issue in mind, I will provide an overview and discussion of strategies for Bayesian estimation of psychometric models via Markov chain Monte Carlo. Among the models considered are traditional structural equation models, traditional item response models, and multilevel structural equation models. I will consider the idea that, for some MCMC methods, we can make use of frequentist results to improve sampling. The talk is informed by my recent work on the R package blavaan.

about the speaker

Ed Merkle is a Professor of Quantitative Psychology at the University of Missouri. His recent work has focused on Bayesian estimation of psychometric models, most notably through development of R package blavaan. Other research interests include statistical computing, forecasting, and model comparison