Some questions (and a few answers) about multilevel models

Here are the slides of the talk I gave at the CDC last week. And here’s the abstract:

Multilevel (hierarchical) models are increasingly popular for data with hierarchical, longitudinal, and cross-classified structures. We consider several questions that arise in the application of multilevel models, including (in no particular order): How many groups do you need to fit a multilevel model? When to use fixed or random effects? Can multilevel models be used for nonnested data? Is it true that Anova is just a special case of linear regression? Is there such a thing as R-squared for multilevel models? Can predictive error be used to compare models? How to handle correlations between individual-level predictors and group-level errors? How to model varying slopes and intercepts? How can I get my model to converge faster on the computer? How to summarize and display the estimates from a model with a zillion coefficients? How to check the fit of a multilevel model? How to choose among the thousands of possible interactions? What to do when a realistic model is so complex that I can’t understand it???

2 thoughts on “Some questions (and a few answers) about multilevel models

  1. Bernard,

    Thanks for the interest. CDC made a videotape so perhaps you can get it from them; you could ask Dr. Michael Kashon at [email protected], who is the coordinator for the CDC seminars.

Comments are closed.