作者: W. Bentley MacLeod
DOI: 10.1111/CAJE.12249
关键词:
摘要: The purpose of this essay is to discuss two approaches inference, and how "human capital" can provide a way combine them. first approach, ubiquitous in economics, based upon the Rubin/Holland potential outcomes model relies randomized treatment measure causal effect choice. second widely used pattern recognition machine learning literatures, assumes that choice conditional current information optimal (or at least high quality), then provides techniques generalize observed new cases. approach combines these methods by using decisions experts reduce demensionality feature space allow categorization their propensity score. fact human capital heterogeneous implies errors decision making are inevitable. Moreover, under appropriate conditions, random This turn allows us identify average for wider class situations than would be possible with control trials. point illustrated data from medical context treating depression, heart disease, adverse childbirth events.