Journal of the Histol)' of the Behm·ioral Sciences: Vol. 32(4) 354-378 October 1996 © 1996 John Wiley & Sons. Inc. DURKHEIM AMONG THE STATISTICIANS STEPHEN TURNER CCC Emile Durkheim 's working notion of causality, the relevance of statistics to it, and his various methodological utterances on these subjects have long puzzled commentators. War- ren Schmaus, in his recent book, Durkheim' s Philosophy of Science and the Sociology of Knowledge, wrestles with these subjects in the light of the tradition of the philosophy of sci- ence.' He concludes that Durkheim was a kind of hypothetico-deductivist who sought cru- cial experiments and used Mill's method of difference together with statistical evidence to produce the kind of experiment analogues that deductive hypothesis testing demands. Schmaus does an impressive job of normalizing Durkheim 's methodology. and he does so by going in a different direction from much of the literature produced by sociologists, 2 which has generally tried to make Durkheim into a normal part of the tradition of twentieth-cen- tury, variable analytic correlational sociology. Neither approach is satisfactory, for Durkheim was anomalous in both the statistical and the philosophical tradition. Schmaus's argument depends on seeing Durkheim as em- ploying Mill's method of difference in his statistical work, and using differences in rates as evidence of causal influence. There are two major things wrong with this argument. The first is that it conflicts with Durkheim 's own sometimes mysterious methodological sayings, such as his comment that sometimes a few cases can prove a law, 3 and with his insistence that only Mill's method of concomitant variation is applicable in sociology. The latter suggests that Durkheim can't have meant his tables as tests of "method of difference" hypotheses. The former raises the question of how statistical analysis is relevant at all for Durkheim. The former suggests that he meant statistics to prove laws, which lends itself to the idea that he might have construed his statistical work as crucial experiments. But it leaves open the ques- tion of how he could have regarded concomitant variation as the means of conducting exper- iments, for it is not a method of experiment. 4 The literature that surrounds the interpretation of Durkheim 's statistical practice is not very helpful in figuring out what Durkheim did have in mind. An influential series of papers by Selvin, 5 some of which are still cited in the literature, 6 argued that if Mill had only had Yule's methods, he would have concluded, contrary to what Durkheim said in Le Suicide, 7 that cosmic factors and others factors as well were associated with suicide. Selvin 's reason- ing takes it for granted that Durkheim was making the same kind of error that Yule attrib- uted, in his classic papers on correlation, 8 to Booth; namely, of misreading the significance of tabular data in showing associations. But this is completely wrong. Durkheim was simply restating the basic findings of previous statistical studies of suicide, which concluded, on the basis of their own methods, that there was a relationship between these facts. Nothing in the subsequent literature, using different methods, came to different conclusions. Durkheim knew perfectly well that there were statistical relationships, associations, between the vari- ables he discusses in the first part of Le Suicide and suicide rates. He rejects the claim that they are causal relationships. And this means that he quite self-consciously rejected the STEPHEN TuRNER teaches in the Philosophy Department, University of South Florida, Tampa, FL 33620. 354 cl Ol e\ id es he bl· wl tic th co of de Sht of she tha of Du ver stat tid: stat diti trer me; spe und pre1 Ant suit stat oft the atio "c01 stati