Statistics Seminar, Dr. Walter Stroup, Department of Statistics, University of Nebraska-Lincoln
Abstract title: PSEUDO-LIKELIHOOD OR QUADRATURE? WHAT WE THOUGHT WE KNEW, WHAT WE THINK WE KNOW, AND WHAT WE STILL DO NOT KNOW
Abstract: Two predominant computing methods for generalized linear mixed models (GLMMs) are linearization, e.g. pseudo-likelihood (PL) and penalized quasi-likelihood (PQL), and integral approximation, e.g. Gauss-Hermite quadrature and Laplace. The primary GLMM package in R, LME4, uses a quadrature algorithm. SAS® software’s primary GLMM procedure, PROC GLIMMIX, uses PL as the default algorithm, and has Laplace and quadrature options. The choice of methods presents a dilemma for GLMM users. Which approximation for GLMM estimation and inference should one use, and why? Linearization is more versatile and can handle both conditional and marginal GLMMs. On the other hand, GLMM software documentation and the literature on which it is based tend to focus on the limitations of linearization. Stroup (2013) reiterated this theme in his GLMM textbook. As a result, prevailing “conventional wisdom” holds that integral approximation – quadrature when possible – is always best. However, accumulating experience with GLMMs, and on-going research about its small sample behavior, has made it clear that this “conventional wisdom” is often misleading – or simply wrong. This seminar presents an updated look at what we now know about quadrature and PL, and offers a “30,000 foot view” of some general operating principles for making an informed choice between the two.
In addition, this seminar will begin with a brief “autobiography” featuring the path that led me to a career in Statistics and where that path has led.
Thursday, October 10 at 4:00pm to 5:00pm