Multilevel Modeling of Categorical Outcomes

A 4-Day Remote Seminar Taught by
Donald Hedeker, Ph.D.

Read reviews of this course

To see a sample of the course materials, click here.


Multilevel models are increasingly used for analysis of clustered and longitudinal data, and methods for continuous outcomes are commonly used and applied. However, many research studies have non-normal outcomes, for example, outcomes that are dichotomous, ordinal, or nominal. Although methods for such non-normal outcomes have been available for quite some time, they are perhaps not as routinely applied as models for continuous outcomes.

This workshop will focus on analysis of dichotomous, ordinal and nominal multilevel outcomes. Both clustered and longitudinal data will be considered, and the following models will be described: multilevel logistic regression for dichotomous outcomes, multilevel logistic regression for nominal outcomes, and multilevel proportional odds and non-proportional odds models for ordinal outcomes. The latter models are useful because the proportional odds assumption of equal covariate effects across the cumulative logits of the model is often inconsistent with the data.

Starting June 16, we are offering this seminar as a 4-day synchronous*, remote workshop for the first time. Each day will consist of a 3-hour, live morning lecture held via the free video-conferencing software Zoom. Participants are encouraged to join the lecture live, but will have the opportunity to view the recorded session later in the day if they are unable to attend at the scheduled time. Each lecture session will conclude with a hands-on exercise reviewing the content covered, to be completed on one’s own that afternoon. A final session will be held each evening as an “office hour”, where participants can review the exercise results with the instructor and ask any questions.

*We understand that scheduling is difficult during this unpredictable time. If you prefer, you may take all or part of the course asynchronously. The video recordings will be made available within 24 hours of each session, meaning that you will get all of the class discussion and exercise solutions even if you cannot participate synchronously.


COMPUTING

This remote seminar is held via Zoom, a free video conferencing application. Instructions for joining a session via Zoom are available here. Prior to each session, participants will receive an email with the meeting code you must use to join.  

In all cases, methods will be illustrated using software, with SAS, Stata, and SuperMix examples and syntax. Some familiarity with reading in data and performing basic statistical analyses in either SAS or Stata is recommended.

This is a hands-on class involving several structured and supervised assignments. To do the exercises, you will need your own laptop computer with a recent version of SAS or Stata and the free student version of SuperMix installed. SuperMix can be downloaded here

NOTE: If you have Windows 7 or older, please run SuperMix as administrator. Windows 8 and more recent should run as a regular user.


WHO SHOULD Register? 

If you plan on analyzing multilevel data (either clustered or longitudinal) and have categorical outcomes of interest, this course is for you. Analyzing multilevel categorical data is necessary in many research fields where normally distributed continuous outcomes are not obtained or relevant.    

Participants should be thoroughly familiar with multiple linear regression, and have some knowledge of logistic regression.


SEMINAR OUTLINE

Day 1
I. Introduction to Multilevel Data: 
   – What are multilevel data?
   – Why do multilevel data analysis? 
   – Overview of multilevel models for categorical data.

II. Multilevel analysis for clustered binary outcomes
   – Multilevel logistic regression model.
   – Scaling of regression coefficients.
   – Students in classrooms and schools example.
   – Using Stata, SAS, and Supermix for multilevel analysis.
   – Interpreting output, testing for cluster effects, creating plots of cluster effects.
   – Fit of models to observed proportions.

III. Clustered binary data assignment distributed

Day 2
IV. Multilevel models for clustered ordinal outcomes
   – Multilevel cumulative logistic regression model.
   – Proportional odds assumption.
   – Students in classrooms and schools example.
   – Using Stata, SAS, and Supermix for multilevel ordinal analysis.
   – Interpreting output, and testing for cluster effects.
   – Fit of models to observed proportions.
   – Non-proportional odds multilevel models.

V. Introduction to multilevel models for longitudinal binary outcomes: 
   – Multilevel logistic regression model.
   – Random intercept and trend models.
   – Subject-specific and population-averaged estimates.
   – Longitudinal psychiatric clinical trials example – descriptive statistics and plots.

VI. Clustered ordinal data assignment distributed

Day 3

VII. Multilevel for longitudinal binary outcomes (continued):
   – Longitudinal psychiatric clinical trials example – random intercept and trend models.
   – Using Stata, SAS, and Supermix for multilevel analysis.
   – Conditional and marginalized estimates.
   – Interpreting output, testing for random subject effects, creating plots.
   – Fit of models to observed proportions.

VIII. Multilevel models for longitudinal ordinal outcomes: 
   – Multilevel cumulative logistic regression model.
   – Proportional odds assumption
   – Random intercept and trend models.
   – Longitudinal psychiatric clinical trials example – descriptive statistics and plots.
   – Using Stata, SAS, and Supermix for multilevel analysis.
   – Conditional and marginalized estimates.
   – Interpreting output, testing for random subject effects, creating plots.
   – Fit of models to observed proportions.

IX. Longitudinal binary data assignment

Day 4

X. Multilevel models for longitudinal ordinal and nominal outcomes: 
   – Longitudinal homelessness study – descriptive statistics and plots.
   – Using Stata, SAS, and Supermix for multilevel analysis.
   – Non proportional odds models; testing proportional odds assumption.
   – Treating outcome as nominal; reference cell comparisons.  
   – Interpreting output, testing for random subject effects, creating plots.
   – Fit of models to observed proportions.

XI. Multilevel analysis of longitudinal count outcomes
   – Multilevel Poisson regression model.
   – Number of headaches and Aspartame crossover study.    
   – Using Stata, SAS, and Supermix for multilevel Poisson analysis.
   – Interpreting output, and testing for random subject effects.
   – Fit of models.
   – Overdispersion.

XII. Longitudinal count outcome assignment distributed


REVIEWS OF Multilevel Modeling of Categorical Outcomes

“Professor Hedeker is excellent. I really liked his ability to cover a lot of material and do it clearly. This is a weak area of mine and he helped fill in many of the gaps in my own understanding.”
  Richard Williams, University of Notre Dame

“Dr. Hedeker put very rich contents into this class. Students are allowed to raise questions at any time through the class. This helps students to catch the important points in a timely manner and be able to get through the learning easily.”
  Qin Liu, The Wistar Institute

“This class offers an in-depth discussion of the theory, fitting, and interpretation of multilevel modeling for non-normal data that does not seem to exist elsewhere. Don puts together and delivers arguably the best short course I’ve ever taken. This course is well worth the time and money.”
  Amy Hughes, University of Texas

“I found Don’s course very helpful for advanced as well as intermediate analysis of categorical data in clustered structures. There were many tips, tricks, nuances, and insights communicated from Don’s many years of experience with categorical data problems. I am taking away many helpful strategies for approaching my ongoing projects.”
  Andrea Howard, Carleton University