Content area
Full Text
Statistical Models and Causal Inference: A Dialogue with the Social Sciences, by David Freedman. Edited by David Collier, Jasjeet Sekhon, and Philip Stark. Cambridge, UK: Cambridge University Press, 2010. 399pp. $29.99 paper. ISBN: 9780521123907.
In this posthumously published collection of essays and articles, statistician David Freedman evaluates the work of quantitative social scientists, and gives them a failing grade. According to Freedman, quantitative social scientists routinely estimate models that rely on tenuous assumptions and that bear little resemblance to the manner in which their data were generated in the real world. Testing model assumptions with diagnostic tests is generally unhelpful, since the diagnostic tests are not powerful enough to detect consequential violations. In short, Freedman argues that statistical models have become Procrustean beds into which social data are routinely and ritualistically rammed.
Some methodologists may agree and argue that a solution to the problem can be found through less reliance on canned approaches and more customized models. But Freedman's critique is more far-reaching. While he is sympathetic to model customization as a solution in some cases, Freedman believes that statistical ingenuity has quickly diminishing returns when applied to social data. Social data rarely behave in a way that is congenial for statistical modeling.
In each chapter, Freedman provides an example to illustrate the fragility of a variety of widely used methods and models. Many of these examples are less closely related to the social sciences than one might expect in a book subtitled "a dialogue with the social sciences." For example, chapters are devoted...