Fast computing of some generalized linear mixed pseudo-models with temporal autocorrelation (Q626250)

Generalized linear regression models with mixed effects are considered. Iterative algorithms are discussed for maximum pseudo-log-likelihood estimation of the parameters. It is proposed to use the Sherman-Morrison-Woodbury matrix identity, the analytical inverse of exponential matrices, and sparse matrix operations to increase the computational speed of these algorithms. Implementation in R software is considered. An example is presented where a generalized linear model is fitted to a binary time series with over 150 000 observations, 50 random effects and over 100 fixed effects.