Workshop: Introduction to Multilevel Modelling

Students in classrooms, regions in countries, and cross-national surveys are just a few examples of multilevel data structures. All of them have one problem in common. The observations are not independent of each other because of the clear clustering within the data causing potential problems with autocorrelation. Researchers need to do something about this and control the potential heterogeneity across the clusters otherwise their regression results will be biased. Multilevel modeling comes to the rescue, though it is not the best solution in all instances. In this workshop, you will learn the most common solutions to the problem including an in-depth introduction to how multilevel models work.

Target Group

This workshop is targeting scientists from different disciplines and researchers at all levels of their career who have a solid foundation in regression and those who would like to extend their knowledge to include the basics and application of multilevel modelling.

(A solid foundation is one that goes beyond knowing what to click in the software and what numbers to copy from the output into the table. Let’s say, you know why heteroskedasticity and autocorrelation are problematic in the regression framework, and you want to do something about it, you are ready for your first multilevel modeling course. If you have conducted regression before but are unsure about this, an in-depth course in regression is what your trainer would recommend before this workshop.)

Goals

• Participants will learn why unmodeled heterogeneity is a problem and go through multiple ways of modeling such heterogeneity.
• Models covered will include:
  • fixed effects regression models, modeling the cluster-specific variation in slopes and intercepts in regressions.
  • hierarchical linear, mixed effects, and/or multilevel models
• Continuous and limited dependent variable models such as binary-logistic, ordered, and count regressions.
• Finally, participants will explore modeling multiple sources of heterogeneity simultaneously.

Content

Day 1.

• What are Multilevel Models?
• Heterogeneity in the Regression Framework
• Building Blocks of Multilevel Models
• Navigation of the Jargon and Terminology
• Estimation
• Model Building Approaches
• Model Fit
• Centering and separating impact across the levels of analysis

Day 2.

• Review
• What we didn’t finish from day 1
• General Linear Mixed Models (Binary-logistic, ordered and count dependent variables)
• Cross-classification
• What else is out there (where you can go from this point forward)?

Software R and RStudio

The software is mostly needed for practice purposes outside of the in-person sessions. To be able to work through results and discuss them in class please bring your own device and make sure that the software R and RStudio (an integrated development environment for R) is installed.

R is an open-source software and if you have not installed it yet on your device, here you can get the software:

For a WU laptop:

• You can download R and RStudio from the IT-SERVICES Software Center

If you use your own device:

• The software R can be downloaded here: https://cran.r-project.org./
• R Studio can be downloaded here: https://rstudio.com/products/rstudio/download/.
• More information on installing R and RStudio can be found at e.g. https://rstudio-education.github.io/hopr/starting.html.

In cooperation with

Competence Center for Empirical Research Methods

• What are Multilevel Models?
• Heterogeneity in the regression Framework
• Building Blocks of Multilevel Models
• Model Building Approaches and Model Fit
• Centering and separating impact across the levels of analysis
• General Linear Mixed Models
• Cross-classification