Are There Really Any Constructive Alternatives to Causal Modeling?

While I find myself in close agreement with much of what Freedman has to say about much of the empirical research using regression-type analyses, I find his comments contain nothing really new in comparison to the numerous cautionary warnings and detailed discussions of pitfalls and required assumptions that have appeared in the sociological literature as far back as the 1950s. Much the same applies to relatively recent critiques by Lieberson (1985) and by Marini and Singer (1988) cited by Freedman. Many of us have long been concerned about the same kinds of abuses by empirical investigators, and I always welcome such notes of caution. However, the basic problem has been with us for a very long time, and there have been extensive discussions that, apparently, have not come to Freedman's attention. And by the same token, neither have empirically inclined sociologists paid serious attention to the methodological literature, and I regard this as a much more serious problem that continually deserves our attention. Let me begin by stressing the major points made by Freedman, and many others before him, with which I am in basic agreement. Certainly, statistical models cannot stand alone and must be supplemented by a series of assumptions, many of which cannot be tested with the data that one has in hand. Technical fixes alone, therefore, cannot be relied on to resolve one's theoretical problems,