Maximum Likelihood Estimation of Generalized Linear Models with Covariate Measurement Error

摘要: Generalized linear models with covariate measurement error can be estimated by maximum likelihood using gllamm, a program that fits large class of multilevel latent variable (Rabe-Hesketh, Skrondal, and Pickles 2004). The uses adaptive quadrature to evaluate the log likelihood, produc- ing more reliable results than many other methods 2002). For single measured (assuming classical model), we describe "wrapper" command, cme, calls gllamm estimate model. wrapper makes life easy for user accepting simple syntax data structure producing extended easily interpretable output. commands preparing running also obtained from cme. We first discuss case where several measurements are avail- able subsequently consider estimation when variance is instead assumed known. latter approach useful sensitivity analy- sis assessing impact assuming perfectly covariates in generalized models. An advantage directly co- variate model various ways. instance, use nonparametric (NPMLE) relax normality assumption true covariate. specify congeneric which relaxes unit exchangeable replicates allowing different scales variances.