Multilevel/Hierarchical Modeling

Prof. Karreth is such a dedicated lecturer. He was extremely helpful, willing to accommodate individual student's interests/needs, and even spending extra time before and after class to help those who needed additional assistance. — participant from Malaysia

This applied workshop introduces participants to the analysis of multilevel, hierarchical, or structured data. These data are ubiquitous in the social sciences and include observations that are nested in higher-level units, such as groups of survey respondents in different countries, students in different schools, or country-level observations at repeated time points. Participants will learn how to appropriately estimate quantities of interests as effects that vary across units and/or time or how much of a change in an outcome of interest is associated with individual- or group-specific features.

Dates

This one week, the 17.5-hour online course runs Monday-Friday, July 10-14, 2023. The course is scheduled for 9:00am -12:30 pm.

Instructor

Johannes Karreth, Ursinus College

Detailed Description

The course will introduce regression models for multilevel data across a wide variety of contexts, with a focus on data management, estimation, and interpretation. The course is organized in five sections.

In the first section, we explore the characteristics of multilevel data structures and their implications for regression modeling. We also introduce R and the lme4 package as a comprehensive solution to modeling multilevel data structures.

In the second section, we introduce the foundations of the linear multilevel regression model. We clarify the meaning of fixed and random effects in the context of multilevel modeling and explore the use of varying coefficients in different settings.

In the third section, we focus on multilevel regression for non-continuous outcomes. For these outcomes, we introduce a variety of post-estimation techniques to facilitate interpretation and presentation of multilevel modeling results.

In the fourth section, we turn to the use of multilevel modeling for time-series cross-sectional data, such as panel data. We also discuss modern techniques for model assessment and model comparison and preview estimation techniques for thornier multilevel modeling problems, including Bayesian estimation.

In the fifth and final section, we return to data management and discuss common approaches to creating and cleaning multilevel modeling data in R. We conclude with some tips and tricks for presenting and communicating multilevel models.

Upon successful completion of the course, participants will be able to:

Understand how multilevel/hierarchical/structured data challenge the assumptions of pooled, i.e., standard, regression models
Distinguish the concepts of fixed and random effects in the context of multilevel data
Estimate regression models with varying slopes and varying intercepts
Generate such post-estimation quantities as marginal effects, predicted probabilities, etc. from multilevel regression models
Use graphical tools to present results from multilevel regression models

Prerequisites

Participants should have some familiarity with basic statistics and the core concepts of linear regression. Since this course begins with a review of regression as it relates to multilevel data, participants with different levels of prior experience should be able to successfully complete the course. Participants will receive training in the free, open-source software R for data management, estimation, and post-estimation techniques to communicate results from multilevel models to a broad audience. Prior knowledge of R is welcome, but not required.

Requirements

Participants are expected to bring a WiFi-enabled laptop computer. Access to data and installation support will be provided by the Methods School.

Core Readings

The workshop will mainly use the following book:

Gelman, A. and Hill, J. (2007). Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge University Press, New York, NY.

Other readings, ranging from theoretical to examples of applications of multilevel modeling in social science research, will be made available to workshop participants.