Thinking about a summer vacation project? Academic prices for Maximizing Predictive Accuracy, and for MegaODA and CTA software, is now available to everyone: please see the Resources page.
Paul R. Yarnold & Ariel Linden
Optimal Data Analysis, LLC & Linden Consulting Group, LLC
The use of CTA to construct propensity score weights is complicated by division by zero in models having any perfectly predicted endpoints: omitting undefined propensity scores yields a degenerate solution. This note presents an algorithmic remedy to this situation.