Abstract

Correlational analysis is the basis for much explanation in contemporary sociology. Mainstream sociologists often infer causation through the use of quantitative techniques that depend, in one way or another, on the existence of bivariate correlations. Even sophisticated multivariate statistical methods that allow for the parceling of variables and the estimation of average net causal effects ultimately rely on such correlations. Correlational analysis so is central to explanation that many sociologists simply take it for granted, assuming that the existence of a correlation is a basic component of causality. The two books reviewed in this essay, however, argue that correlational analysis is by itself an inadequate mode of causal assessment. The problems identified are not simply familiar themes such as a correlation fails to represent causation because it might be spurious or because the time order of correlated variables may not be clear. Rather, these works suggest that even nonspurious correlations in which the time order of variables is well-established may be inherently limited representations of causal processes.