Multilevel Models II: Advanced Topics

Instructor(s):

  • John Poe, University of Michigan

This workshop will be offered in an online video format.

This course is designed to provide participants with the practical skills and workflow to deal with almost any kind of multilevel modeling problem no matter how complicated. We begin with a review of the basic toolkit taught in standard multilevel, longitudinal, and panel data courses. The same set of models introduced here will be referenced repeatedly throughout the course. After the general overview of tools, we spend the rest of the first week on the key theme of the class: that missing data, missing variables, and missing dependency structures can all lead to biased inferences.

In the second week, we step back and focus on practical implementation and interpretation. We begin with a general analysis workflow for how to think about dependency structures, omitted variable bias, and missing data in a practical sense. We then walk through debugging, diagnostics, and actually getting code to work from both the MLE and Bayesian perspective. We spend a day on model comparison and how to tell if you actually need a multilevel model or if an alternate approach will work better. Finally, we go through how to explain and interpret these models.

In the last two weeks we put theory and coding skills into practice. Each day is focused on a practical research problem. We spend the first hour working through the logic of the analysis with a checklist of issues we need to consider from the first two weeks. We then spend an hour building a roadmap (with code in either Stata or R) to execute the analysis.

Prerequisites: Prior exposure to maximum likelihood or basic categorical models as well as some previous background in either multilevel or longitudinal modeling/panel data econometrics is a prerequisite. Ideally, students will have had courses like the Maximum Likelihood Estimation I: Generalized Linear Models and Multilevel Models I: Introduction and Application classes at ICPSR prior to attending. The course covers Bayesian applications of hierarchical models so some exposure to Bayes would be beneficial but is neither required nor assumed. The primary goal of the course will be to allow students to use advanced models and to understand how and when more sophisticated techniques will provide practical benefits.

Software: We provide detailed code and notes in both R and Stata. Not all models from the course can be used in both programs so some flexibility would be beneficial. Anyone who wishes to use other software (e.g. SAS, Mplus, HLM) is free to do so but the instructor and TAs will be of extremely limited help.

Fee and Registration: This course is part of the second four-week session. Please see our fee chart on our Registration page for the cost of attending one (or both) four-week session(s). Participants who enroll in a four-week session may take as many courses (workshops and lectures) as desired during the session for which they are enrolled. Participants in the four-week sessions are also welcome to attend all of the lectures and discussions offered in the Blalock Lecture Series.

Tags: multilevel models, hierarchical linear models

Course Sections

Section 1

Location: Online -- Video format,

Date(s): July 20 - August 14

Time: 1:00 PM - 3:00 PM

Instructor(s):

  • John Poe, University of Michigan