Standard introductory courses in multilevel modelling focus on two-level models for continuous outcomes, mostly in clustered cross-sectional data settings. For example, they may cover two-level analyses of student test scores where students are nested within schools, and interest lies in studying the effects of both student and school characteristics, and their interactions. These courses often then show how to extend the analysis to binary outcomes using multilevel logistic regression - for example, to analyze whether students pass or fail their tests.

 

This three-day advanced course in multilevel modelling greatly expands on these foundations.

 

Day 1 is dedicated to modelling more complex data structures. We begin with three-level models, appropriate, for example, when individuals are nested within neighbourhoods, which are themselves nested within regions. We then consider cross-classified models, appropriate, for example, in studies of interviewer effects where respondents are nested within areas, but interviewers work across multiple areas and areas contain multiple interviewers. Finally, we consider multiple-membership models, appropriate, for example, when workers simultaneously belong to multiple teams and interest lies in understanding how those teams and their characteristics influence individual outcomes.

 

Day 2 focuses on the application of multilevel models to longitudinal data. We begin with classic panel-data models, appropriate, for example, when analysing workers' wages, and contrast fixed-effects and random-effects approaches. We then move to lagged dependent variable models that simultaneously capture both state and trait influences on outcomes, appropriate, for example, when analysing relationship happiness. We conclude with growth trajectory models, appropriate when studying predictors not only of initial differences in outcomes, but also of change in outcomes over time.

 

Day 3 explores MAIHDA (Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy), an innovative reimagining of multilevel models for studying intersectional sociodemographic inequalities. This approach is appropriate when analysing how outcomes vary across high-dimensional combinations of characteristics such as gender, ethnicity, and social class. We begin by showing how multilevel models can improve upon simple tables of sample averages. We then examine how they can be used to assess the extent to which inequalities are additively patterned versus genuinely intersectional. Finally, we consider extensions of MAIHDA to settings involving binary outcomes, three-level data, and longitudinal data.

 

Throughout the course, you will develop both advanced theoretical understanding and practical skills through hands-on software exercises in either R or Stata. Emphasis is placed not only on how to fit multilevel models, but also on how to interpret them and identify the types of research questions for which they are best suited. Real-world examples from the social and health sciences are used throughout.

 

The course follows a 2:1 balance of lectures and hands-on practical software sessions, applying the taught methods to real datasets using participants' choice of R or Stata.

 

Each day focuses on a different major topic area within multilevel modelling and will consist of three sessions. Every session begins with a lecture on a new multilevel concept, motivated by a substantive application and emphasizing model building and interpretation, delivered in a software-independent manner.

 

Each lecture will be immediately followed by a practical session, giving you the opportunity to replicate the analyses and consolidate your understanding. Practicals are self-directed, allowing you to work at their own pace. Both basic and advanced versions of each exercise are provided to accommodate varying levels of prior experience with multilevel modelling and software. At the end of each session, the instructor demonstrates the analyses in both R and Stata.

 

You are strongly encouraged to ask questions during lectures, including how the content relates to your own research. During practical sessions, there are further opportunities to discuss methods with the instructor and apply them to your own research data.

You will find the course useful if:

By the end of the course, you will:

 

You may choose to do the practicals in R or in Stata. In R, we will make extensive use of the lme4 package, particularly the lmer function. In Stata, we will focus on the mixed command.