Lessons in disguise: multivariate predictive mistakes in collective choice models

摘要

Many political processes can be characterized as repeated collective decisions, in which individual choices rendered by the same actors are combined to produce salient outcomes (e.g., voting in legislatures). The inherent interdependence among the choices within repeated collective decisions is often missed in statistical analysis. I introduce the joint prediction error (JPE)—a grouping of individual decisions that is poorly predicted by a model. JPEs capture the intersecting information missed by conventional diagnostics. I demonstrate the use of JPEs on data from two published articles—one on U.S. Supreme Court voting and another on international defense alliances.