in error should diminish our inclination to conceive of associations between two variables as varying in degree only. Clearly, relationships have "shapes" or "forms" (represented in the rules for estimating one variable from another) that may be fully as informative as the degree of association. Since a "P-R-E" interpretation, strictly speaking, cannot be made without specifying the rule for estimating one variable from another, this criterion would focus attention on the "shape" of a relationship and require that it be made explicit. The general notion of a proportional reduction in error appears to be applicable to types of data analysis not ordinarily thought of as requiring a measure of association. Menzel's coefficient of scalability,39 for example, has a proportional-reduction-in-error form. It conforms, in other words, to conditions (a) through (d) outlined in this paper. The entire problem of assessing the "comprehensiveness" of a typology, and not simply of Guttman scale types, may be conceived in proportional-reduction-in-error terms. The P-R-E criterion appears to be as applicable to multiple and partial measures of association as to "zero-order" measures involving only two variables. These points have not been explored here for lack of space, but the general notion of proportional reduction in error does provide a useful approach to a variety of problems in data analysis. Finally, conceiving of measures of association as indices of the proportional reduction in error, and, more specifically, conceiving of such measures as having a common definition but with different "characteristics" to be estimated, different rules of estimation, and different definitions of error, should help clarify the issues at stake in the choice of a measure. Choosing a measure is not simply a matter of finding one that makes no assumptions clearly violated in one's data. The choice is, in large part, a problem of deciding which "characteristic" should be estimated by which estimation rules and with what definition of error. This is a more specific guide than the vague admonition that the measure should be adapted to the nature of the data and to the nature of the problem, and it also permits theoretical considerations to have an impact on the forms in which data are summarized. 39 Herbert Menzel, "A New Coefficient for Scalogram Analysis," Public Opinion Quarterly, 17 (Summer, 1953), pp. 268-280.