Customer Reviews

Matched Sampling for Causal Effects

5 star:		(2)
4 star:		(1)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

An important issue for researchers is the discovery of cause and effect relationships. It is often the case that those not well educated in statistics will think that a simple correlation between two variables is enough to imply causation (perhaps because of temporal order i.e. A comes before B so A causes B). However, determining causation is a much more complicated issue. A common statistical adage is "correlation does not imply causation". Don Rubin is an accomplished author, teacher and one of the leading developers of the statistical theory of causation. the authorities to read on this subject are Don Rubin and Judea Pearl, two of the pioneers who have written texts on this topics. This theory involves new concepts that one does not learn in introductory statistics course. Contrafactuals represent one such concept. Read this book to learn the details.

This is a very good collection of important articles that Rubin and his students and colleagues have written on causal inference.
Another book written by his student P. Rosenbaum (now a Wharton prof) titled "Observtional Studies" is also very important if one wants to learn the details of propensity scoring or matching.

This is a collection of earlier journal articles by Don Rubin and his collaborators, with some interesting personal narrative to situate each substantive section of the volume. Every chapter is worth reading and the volume is a comprehensive collection of his most important work on the area of subclassification, propensity scores, and matching. There is more recent work by other scholars that complicates some of the theoretical and statistical ideas presented here, but Rubin's (and Cochran's and Rosenbaum's) contributions to causal inference is slowly reshaping applied statistics by directing attention at approaching causal inference. Pearl's Causality: Models, Reasoning and Inference offers a completely different view of causal inference, but Rubin's work is more accessible and more practical for most present applications.