Introduction to Multilevel Models for Social Scientists (Doctoral Seminar)

Traditional methods of analysis often ignore the natural clustering that we see in social science data, as well as more widely in health science and beyond. Pupils are naturally clustered in classes and in schools, electors are nested electoral constituencies, patients are clustered in hospitals and so on. Multilevel analysis is a powerful technique that allows us embrace the real world complexity and to start to account for it in our models incorporating predictors appropriately at all levels of analysis.

In this one-day workshop, we will concentrate on linear models (although these techniques can be applied to binary, count and categorical outcomes). We will look at variance components models, varying intercept models, varying slope models, and three-level models. We will discuss fitting via Maximum Likelihood, model diagnostics and interpretation.

The workshop assumes no prior knowledge of multilevel modelling techniques, although participants should have some familiarity with multiple regression and statistical notation. We will use examples from Education and Political Science, although participants are encouraged to bring their own multilevel data. Throughout the practical sessions the software of instruction will be R using the package lme4. However, instructional materials will also be provided for other commonly used software (e.g. STATA) where possible.

Course convenor: Dr Michael Adkins