The first speaker in this semester’s SWE Colloquia series was Douglas Bates, Professor Emeritus, Department of Statistics, University of Wisconsin, Madison.

Fitting complex mixed-effects models to large datasets

Mixed-effects models are a type of statistical model that incorporate both fixed-effects parameters and random effects.  From the point of view of experimental design, fixed-effects are associated with experimental factors (e.g. priming or not) or with covariates that have a fixed, reproducible set of levels (e.g. sex of the subject, socio-economic status).  Random-effects are associated with blocking factors – a known source of variability for which we wish to control.  The most common such blocking factor is “Subject”.  In many studies “Item” will be another blocking factor.  Mixed-effects models provide a way of taking into account these different sources of variability in the analysis of the data, but only in the last decade or so has software been available to fit complex mixed-effects models, especially those with crossed random effects such as “Subject” and “Item”, to large data sets.  As usually happens, the models that researchers wish to fit are becoming more and more complex and the data sets are becoming larger and larger, straining the capabilities of some of the software used to fit these models.  Dr Bates’ presentation on 20 June included a discussion of these models, some of the software used to fit them, and future directions in R/lme4 and the recently developed Julia language.