Multiple Regression in Small-N Comparisons
暂无分享,去创建一个
Michael Shalev has turned his attention, once again, to the bad methodological habits that social scientists – like myself – often adopt. As always, he presents us with thoughtful, rigorous, and penetrating criticism, but also with a generous dose of constructive prescription. His target is the widespread use of regression techniques in cross-national comparative research. The gist of the argument is that multiple regression (MR) is a far too blunt instrument if our aim is to arrive at a robust identification of crucial causal mechanisms. MR, as he puts it (p. 42), renders the cases invisible and, hence, precludes researchers from having any dialogue with them. The case becomes a set of scores; the causal mechanisms are reduced to correlation coefficients. As a result, analytical power is sacrificed rather than gained. Shalev advocates simpler ‘low-tech’ approaches such as tabular representations, tree diagrams, or clustering techniques either as substitutes for, or as companions to, regression analysis.
[1] M. Sobel,et al. Identification Problems in the Social Sciences , 1996 .
[2] James D. Fearon,et al. Counterfactuals and Hypothesis Testing in Political Science , 1991, World Politics.
[3] A. Przeworski,et al. Quantitative cross-national research methods , 2001 .