Mini-course on Mixed-effects Statistical Modelling

Harald Baayen (MPI-Nijmegen)

The course will address the statistical analysis of data with both fixed and random effects. Fixed-effect predictors are factors with just a few possible levels (values) all of which are present in the experiment. An example of a fixed effect is grammatical number, a factor that in English has two levels, singular and plural. Random-effect predictors are factors with many levels, only a few of which are sampled in a given experiment. In (psycho-)linguistic experimental studies, subjects (participants) and items (words, sentences) are random effects. The statistical analysis of data with subjects and items is traditionally carried out either by calculating quasi-F ratios (but only for specific designs) or by two separate analysis, one aggregating over items (the subject analysis) and one aggregating over subjects (the item analysis).

Recent advances in statistics offer much better ways for analysing data with subjects and items. Linear mixed-effects models offer a sophisticated, flexible but at the same time also highly intuitive approach to understanding the quantitative structure in (psycho-)linguistic data sets down to the level of the individual responses of a given subject to a given item. These methods offer the advantage of superior precision, greater power, and enhanced insight in the data. They also offer new opportunities for bringing control variables into the model.

In this course, Baayen will introduce basic concepts and modeling strategies using the open source programming environment R.

Pre-registration for this event is now closed. Those who have pre-registered will be notified by e-mail about the outcome of the enrollment lottery.