Dr. Molllie Brooks talk
Date: 22.03.19 Time: 15.15 - 16.00 Room: ETH HG G 19.1
The diversity of features of generalized linear models that can be fit in R is huge (e.g. zero-inflation, numerous distributions, and random effect correlation structures), but the features were not easy to use in combination because they were available from separate packages. Ignoring these aspects can give biased estimates and inflate the rates of false-positives or false-negatives in hypothesis tests. The R package glmmTMB was developed using the TMB package to do maximum likelihood estimation to fit a diversity of models in a single robust package. Features of TMB that made this possible include automatic differentiation for calculating gradients, Laplace approximation for integrating over random effects, and adding and subtracting on the log-scale to avoid over- and under-flow. In addition to describing glmmTMB, the talk will include ecological examples that address zero-inflation and underdispersion in count data, as well as Tweedie and beta regression for biomass and proportion data, respectively.