Paul Allison
President, Statistical Horizons LLC
Paul Allison, Ph.D., is professor emeritus of the University of Pennsylvania where he taught graduate courses in methods and statistics for more than 35 years. He is widely recognized as an extraordinarily effective teacher of statistical methods who can reach students with highly diverse backgrounds and expertise.
Dr. Allison completed his doctorate in sociology at the University of Wisconsin. He then went on and did postdoctoral study in statistics at the University of Chicago and the University of Pennsylvania. He has published eight books and more than 75 articles on topics that include linear regression, log-linear analysis, logistic regression, structural equation models, inequality measures, missing data, and survival analysis.
Much of his early research focused on career patterns of academic scientists. Currently, his principal methodological research is on the analysis of longitudinal data. This focus is to determine the causes and consequences of events, and on methods for handling missing data.
A former Guggenheim Fellow, Allison received the 2001 Lazarsfeld Award for distinguished contributions to sociological methodology. In 2010, he was named a Fellow of the American Statistical Association and is also two-time winner of the American Statistical Association’s award for “Excellence in Continuing Education.”
Research Interests
HOW CAN WE MAKE BETTER CAUSAL INFERENCES FROM SURVEY DATA? Paul’s main goal of research is to improve statistical methods for making causal inferences using non-experimental data. Although randomized experiments are the best methods for demonstrating causal relationships, they are usually not ethical or practical for answering the kinds of questions that most social scientists ask. That said, we have to settle for “second best” methods that have lots of potential pitfalls. Dr. Allison has been studying a collection of methods that enable us to get much closer to an experimental design. Known as “fixed effects methods”, these statistical techniques enable one to control for all stable characteristics of persons, regardless of whether we can measure those characteristics. They accomplishes this by using each person as his or her own control.
In some versions, these methods answer questions about directions of causality: does X cause Y or does Y cause X? Allison’s other major research interest is statistical methods for handling missing data. There are two new methods that are far superior to traditional missing-data methods: multiple imputation and maximum likelihood. Despite this being known, they are still not widely used by social scientists. Paul hopes to change that by making these methods easier to use and better understood.