
As a result, Chapter 5 now focuses on propensity score matching methods alone, including greedy matching and optimal matching.

Given these considerations, we treated dosage analysis in the second edition as a separate chapter. There is an increasing need in social behavioral and health research to model treatment dosage and to extend the propensity score approach from the binary treatment conditions context to categorical and/or continuous treatment conditions contexts. Because subclassification and weighting methods have been widely applied in recent research and have become recommended models for addressing challenging data issues (Imbens & Wooldridge, 2009), we decided to give each topic a separate treatment. A propensity score is the probability that a subject will be assigned to a condition or group, based on conditions that exist at the time of the group assignment. A propensity score is simply a probability a number rangingfrom 0 to 1. These methods are closely related to the Rosenbaum and Rubin’s (1983) seminal study of the development of propensity scores-it is for this reason that Chapter 5 of the first edition pooled these methods together. implementing propensity score matching with SAS is relatively straightforward. The most significant change of the second edition is discussion of propensity score subclassification, propensity score weighting, and dosage analysis from Chapter 5 to separate chapters. New statistical approaches necessary for understanding the seven evaluation methods are included.However, GBM and other machine learning methods have been shown in some cases to perform better than logistic regression in simulation studies (Luellen et. Data simulation is used to illustrate key points. First, to our knowledge, estimating propensity scores by GBM or CART is not possible in standard SAS, which is why we focused our demonstration on logistic regression-based propensity scores.Examples in every chapter demonstrate real challenges found in social and health sciences research.Two conceptual frameworks-the Neyman-Rubin counterfactual framework and the Heckman econometric model of causality-provide a foundation for understanding key topics.Each method, and its empirical examples, is linked to specific Stata programs for seamless integration of learning and application.The authors present key information on model derivations and summarize complex statistical arguments-omitting their proofs to challenge readers to apply their learning.
SAS UNIVERSITY EDITION PROPENSITY SCORE SOFTWARE
The authors demonstrate new software and include clear illustrations for analyzing treatment dosage with GPS.The principles and issues related to running propensity score models with sub-classification and weighting are covered in depth.Expanded content on propensity score analysis with multilevel data includes new discussions of four multilevel models for estimating propensity scores and two strategies for controlling clustering effects in outcome analysis.New coverage of modeling heterogeneous treatment effects includes two nonparametric tests and a discussion of modeling issues to ensure students are on the cutting edge.Newly expanded coverage of analyzing treatment dosage in the context of propensity score modeling broadens the scope of application for readers.

