Scientific Coordination

This five-day course provides comprehensive training in modern multilevel regression methods for clustered cross-sectional and longitudinal data. Participants will develop both advanced theoretical understanding and practical skills, with hands-on software exercises in their choice of R or Stata.

We focus on multilevel models for continuous and binary dependent variables. These models extend conventional linear and logistic regression by explicitly accounting for clustering in the data, enabling researchers to draw deeper insights from hierarchical structures.

Applications include analyzing student test scores within schools, job satisfaction of employees within firms, and health outcomes of patients within hospitals. A central theme is disentangling social processes that operate at different levels of analysis by distinguishing within-cluster from between-cluster effects of explanatory variables. Another is estimating organizational and area effects on individual outcomes, relevant for monitoring, accountability, and policy choice.

Longitudinal data are also naturally clustered, with repeated measures nested within individuals or multiple panel waves within respondents. Here, interest often lies in modelling change over time and individual trajectories.

We also consider more complex three-level cross-sectional and longitudinal models involving multiple sources of clustering, such as voters within counties within states or respondents within survey waves across countries. Hierarchical structures may also break down-for instance, students from the same neighborhood may attend different schools, requiring cross-classified models, or they may change schools over time, requiring multiple membership models.

The course further explores the use of survey weights and concludes with an introduction to MAIHDA (Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy), an innovative reimagining of multilevel models for studying intersectional sociodemographic inequalities.

Throughout, we emphasize not only how to fit multilevel models, but also how to interpret them, and the types of research questions they are best suited to address. Real-world examples from the social and health sciences are used throughout.

The complete syllabus for this course will shortly be available for download here.

Organizational structure of the course

The course will follow 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 will focus 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 is immediately followed by a practical session, giving participants the opportunity to replicate the analyses and consolidate their understanding. Practicals are self-directed, allowing participants 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.

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

Target group

You will find the course useful if:

You are a PhD student, postdoctoral researcher, or other academic/non-academic researcher working in the quantitative social or health sciences.

You need to analyze survey, cohort, administrative, or other data where individuals are nested within organizations or geographical areas, including where interest focuses on differences in mean outcomes across these units.

You need to analyze longitudinal data where individuals, organizations, or countries are tracked over time, especially where interest lies in describing change over time or differential outcome trajectories.

Learning objectives

By the end of the course, you will:

Have a strong overview of modern multilevel modelling methods.

Understand the strengths and limitations of multilevel models in comparison with alternative approaches.

Accurately interpret results from multilevel models.

Specify, estimate, and visualize multilevel models in R or Stata, including generating plots.

Identify potential applications of multilevel modelling in research.

Purposefully apply the methods in the context of your own work.

Prerequisites

Strong experience with linear regression, including confidence interpreting model equations and working with dummy variables and interaction terms.

Some experience with logistic regression, including an awareness of log-odds, odds, odds ratios, and probabilities.

A basic familiarity with, and ideally some first-hand experience of, multilevel or panel data models.

Regular experience using R or Stata.

Software and hardware requirements

As a participant, you should bring your own laptop for use in the course. Please install R/RStudio and/or Stata before the course. The latest versions of R and RStudio are available for free at https://cran.r-project.org/ and https://www.rstudio.com/. You may receive a Stata short term license provided by GESIS for the duration of the course if needed.

Please make sure that you are able to download files from the internet-free Wifi is provided by GESIS-and that you have the rights to install R packages or Stata ado files on your laptop during the course.

In R, we will make extensive use of the lme4 package, particularly the lmer and glmer functions. In Stata, we will focus on the mixed and melogit commands.

Schedule

Day 1: Review of Two-level models

Session 1: Overview of Multilevel Modelling

Session 2: Review of random intercept models

Session 3: Review of random slope models

Concepts reading (optional):

Leckie (2019) provides a brief introduction to two-level models with applications to education.

Steenbergen and Jones (2002) go heavier on the algebra with applications to political science.

Hox et al. (2017, Chapters 1, 2 and 3) give a longer but gentle textbook coverage.

Snijders. & Bosker (2012, Chapters 4 and 5) gives an intermediate level textbook coverage.

Raudenbush & Bryk (2002, Chapters 1, 2, 3, 4, and 5) gives a comprehensive advanced treatment of two-level models.

R reading (optional):

Huang (2022, Chapters 1, 2, 3, and 4).

Stata reading (optional):

Rabe-Hesketh & Skrondal (2021, Chapters 1, 2, 3, and 4).

Day 2: Three-level models and beyond

Session 4: Three-level models

Session 5: Cross-classified models

Session 6: Multiple membership models

Concepts reading (optional):

Steenbergen and Jones (2002) cover three-level models with applications to political science.

Hox et al. (2017, Chapter 9) give a gentle textbook treatment of three-level models.

Snijders & Bosker (2012, Chapter 4) covers three-level models in more detail.

Raudenbush & Bryk (2002, Chapters 8) gives an advanced treatment of three-level models.

Snijders & Bosker (2012, Chapter 13) covers cross-classified and multiple membership models.

Raudenbush & Bryk (2002, Chapters 12) covers cross-classified models with some treatment of multiple membership models.

Cafri et al. (2015) covers cross-classified and multiple membership models.

Leckie (2009) presents an application combining cross-classified and multiple membership models.

R reading (optional):

Huang (2022, Chapter 9) covers three-level models.

Stata reading (optional):

Rabe-Hesketh & Skrondal (2021, Chapter 8) covers three-level models.

Rabe-Hesketh & Skrondal (2021, Chapter 9) covers cross-classified models.

Leckie (2013) covers multiple membership models.

Day 3: Longitudinal data analysis using multilevel models

Session 7: Fixed effects vs. random effects panel data models

Session 8: More panel data models

Session 9: Growth-curve (trajectory) models

Concepts reading (optional):

Bell and Jones (2015) discuss the fixed effects vs random effects debate with applications in political science.

Hedeker (2004) provides an introduction to growth-curve (trajectory) models.

Hox (2017, Chapter 5) covers growth-curve (trajectory models) at an introductory level.

Snijders & Bosker (2012, Chapter 15) covers growth-curve (trajectory models) at an intermediate level.

Raudenbush & Bryk (2002, Chapter 6) covers growth-curve (trajectory models) at an advanced level.

Heck et al. (2022) explores how to fuse growth-curve models with individuals changing organizations over time.

R reading (optional):

Huang (2022, Chapter 5) covers growth-curve (trajectory models).

Stata reading (optional):

Rabe-Hesketh & Skrondal (2021, Chapter 5) covers fixed effects vs random effects.

Rabe-Hesketh & Skrondal (2021, Chapter 7) covers growth-curve (trajectory models).

Day 4: Multilevel logistic regression

Session 10: Review of conventional logistic regression

Session 11: Two-level logistic regression for cross-sectional data

Session 12: Two-level logistic regression for longitudinal data

Concepts reading (optional):

Austin & Merlo (2017) covers two-level logistic regression with applications in population health.

Hox et al. (2017, Chapter 6) give a gentler slower treatment.

Snijders & Bosker (2012, Chapter 17) gives intermediate level treatment.

Raudenbush & Bryk (2002, Chapter 10) gives more advanced treatment.

R reading (optional):

Huang (2022). Chapter 8.

Stata reading (optional):

Rabe-Hesketh & Skrondal (2021, Chapter 10) covers two- and three-level logistic regression.

Day 5: Further topics in multilevel modelling

Session 13: Multilevel survey weights

Session 14: MAIHDA for studying intersectional sociodemographic inequalities.

Session 15: Further resources on multilevel modelling

Concepts reading (optional):

Snijders & Bosker (2012, Chapter 14) gives an intermediate level coverage of multilevel survey weights.

Carle (2009), provides an accessible introduction to multilevel weights.

Rabe-Hesketh & Skrondal (2006), discusses weights, including for binary outcome models.

Asparouhov, T. (2006), presents an advanced simulation study on the properties of multilevel weights

Evans et al. (2018) proposed MAIHDA and presents the first application.

Evans et al. (2024) presents a recent how-to tutorial on MAIHDA.

R reading (optional):

Atasever et al. (2025) provide R syntax on Huang's website to replicate their multilevel weights analysis.

Evans et al. (2024) provide annotated R syntax in their supplemental materials to fully replicate their MAIHDA tutorial.

Stata reading (optional):

Rabe-Hesketh & Skrondal (2006) provide brief Stata syntax in the appendix to their article.

Evans et al. (2024) provide annotated Stata syntax in their supplemental materials to fully replicate their MAIHDA tutorial.

Scientific Coordination

Administrative Coordination

Advanced Problems in Multilevel and Longitudinal Modeling

About the lecturer - George Leckie

About the lecturer - George Leckie

Course description

Target group

Learning objectives

Prerequisites

Schedule

Schedule

Recommended readings

Recommended readings