Panel Data and Multilevel Models for Categorical Outcomes [DRAFT]
Richard Williams, University of Notre Dame, https://www3.nd.edu/~rwilliam/
Institute for Political Methodology, Taiwan, July 17 & 18, 2018
NOTE: The following special types of files are used on this web page.
Pdf files. Require Adobe Acrobat.
Stata files.
Overview. This course is going to focus on analyzing categorical outcomes in panel data and multilevel models but many of the same ideas will also apply to linear models. Because the results from categorical outcome models can be difficult to interpret, I will also talk about how adjusted predictions and marginal effects can often make the substantive meaning of results clearer.
Course Materials
Fixed effects and conditional logit models
Fixed effects versus random effects models
Hybrid models (NEW)
hybrid.do (NEW)
Random Slopes Models (Forthcoming)
randomslopes.do (Forthcoming)
Discrete Time Methods for the Analysis of Event Histories
Adjusted predictions and marginal effects (general case)
Adjusted predictions and marginal effects (random effects models)
COMPLETE COURSE PACKET (Includes all of the above handouts except those labeled new or revised)
Keynote Address, June 16, 2018
Panel Data and Multilevel Analyses of Academic Publishing Success (Click here for Powerpoint version)
Panel Data and Multilevel Analyses for Academic Publishing Success: Supplemental Handout
Supplementary Materials
Richard Williams' personal web page includes a great deal of additional information on his research and the statistical programs he has written, as well as most of the notes for the statistics courses he has taught. Of special interest may be his class on Categorical Data Analysis, which covers many other CDA-related topics. This handout of his briefly covers Panel Data for Linear Models. My Stata Highlights page includes links to Stata and statistical handouts from my other courses that may interest students.
The free student version of Don Hedeker's Supermix program will estimate many multilevel models, including models that Stata and other programs can't estimate. Hedeker's personal website also has a lot of useful material. He is a great teacher if you ever get a chance to take one of his workshops,
If you use Stata and have questions about it or just want to learn more, the Statalist Forum is a wonderful resource. Before asking anything be sure to read the forum FAQ first (especially pt. 12) so you will know how to post your questions effectively. Unlike some forums, when signing up real names are greatly preferred.
I am a big fan of Paul D. Allison's Fixed Effects Regression Models (Quantitative Applications in the Social Sciences). It is very easy to read and pretty thorough. (The link is to Amazon in the US. If it doesn't for you, just Google the title.)
If you have panel data with continuous dependent variables, you may want to check my xtdpdml support page. Paul Allison, Enrique Moral-Benito, and I argue that a ML-SEM strategy is sometimes superior to the Arallano/Bond Dynamic Modeling approach that is widely used in Economics. We hope to eventually extend our method so it works with categorical variables.
Useful sites for learning about Stata and SPSS