Machine Learning Who to Nudge - Causal vs Predictive Targeting in a Field Experiment on Student Financial Aid

Explore causal versus predictive targeting strategies in machine learning through a comprehensive analysis of a large-scale field experiment involving over 53,000 college students. Examine how interventions can be more effectively targeted to specific individuals by analyzing a study that used "nudges" to encourage students to renew their financial-aid applications before deadlines. Learn about baseline targeting approaches including causal forest methods that assign students to treatment based on estimated treatment effects, and compare alternative targeting policies that focus on students with either low or high predicted probability of renewing financial aid without treatment. Discover why predicted baseline outcomes may not be the ideal targeting criterion and understand the challenges of determining whether to prioritize low, high, or intermediate predicted probabilities. Investigate hybrid approaches that combine the accurate estimation strengths of predictive methods with the correct criterion focus of causal approaches. Analyze findings showing that targeting intermediate baseline outcomes proves most effective while targeting based on low baseline outcomes can be detrimental, with practical implications demonstrating that nudging all students improved early filing by 6.4 percentage points over a 37% baseline average, while targeted approaches using preferred policies can achieve approximately 75% of this benefit by treating only half the students.