What's the Question to That Answer? Measures and Marginals in Crosstabulation

In the context of a practitioner's view of statistics, this paper attempts to revive and clarify G. Udny Yule's concern with the effects of marginal distributions upon summary measures of association. Ignoring these effects may give you the right answer to the wrong question. In particular, when the hypothesis is causal, varying independentvariable distributions can artifactually distort association measures. The distortions of some common measures are shown in three kinds of tables. A cure, equiweighting, is explained. Reasons are given for not also equiweighting the dependent variable. A socialization example illustrates the use of equiweighting in a complex problem.