Fixed, Random, and Mixed Effects: Modern Approaches to Dealing with Nested, Clustered, Panel, and Longitudinal Data
2018-11-02 (Creation date: 2018-11-02)
- Main contributor
Researchers often get contradictory advice from professors, colleagues, reviewers, and textbooks on how to deal with clustering across time and space. Economists argue strongly for “fixed effects” models. Psychologists and statisticians more typically push for “mixed effects” models. Most applied researchers in the social sciences are told to use a Hausman test to decide between fixed and random effects. This is complicated by the fact that different disciplines, articles, and books use very different terminology and notation to describe models. This lecture will walk participants through the basic problems of clustered data and translate the solutions from economics, psychology, and statistics into a common language. We will focus on how to make practical decisions on model choices for linear and nonlinear models, what problems can crop up, and how to describe/justify your methods to different audiences.
- Related Item
Dr. John Poe is currently a research methodologist working as a postdoctoral scholar for the Center for Public Health Services and Systems Research at the University of Kentucky. He received his PhD in the Department of Political Science at UK in 2017. He teaches the advanced course on multilevel modeling for the ICPSR summer program at the University of Michigan and the GSERM program in Europe. His methodological training comes mostly from econometrics, psychometrics, statistics, and biostatistics. Dr. Poe's current substantive work is focused on understanding community health systems using network science. In particular, he's focused on understanding how health system structures and interactions affect health disparities in different segments of the population. His past (and future) work was split between research about the determinants of the policy process and understanding how different mechanisms in policy making operate and how people react to the their political and social environments. Methodologically, he is focused on problems of endogneity and model misspecification with clustered, multilevel, longitudinal, and network data structures.