Beyond Regression? Predictive vs. Postdictive Models

Abstract

Linear regression (OLS, probit, logit, etc.) is the standard workhorse for data analysis in political science (and in social sciences more broadly), along with methods like factor analysis that also use mainly linear relationships. Such approaches account for some 75% of all numerical data analysis in journals like American Political Science Review. While products of two variables are sometimes included, division seems unknown. The format of the resulting equations is very different from the multiplication-division format most prevalent in physics. Once published, coefficient values in such postdictive "empirical models" are rarely used by other researchers. Could such methods uncover basic relationships of the type that prevail in physics, if they existed in politics? Or do we need to expand the use of predictive quantitative models based on logical considerations, prior to applying statistical analysis?

Title	Details
1. Mathematical Models: From Physics to Economics, and to Political Science?	View Paper Details
2. Testing Theories with Quantitative and Qualitative Predictions	View Paper Details
3. Predictive vs. Postdictive Models	View Paper Details