Cautions regarding the physical interpretation of statistically based structure-activity relationships.

Abstract The distinction between the use of multiple regression analysis as a predictive tool and as a means of investigating controlling physical characteristics in structure‐activity studies often is unrecognized. Three examples of complications that can arise with either of these goals in mind are discussed. The first is an illustration of a “false” parabolic dependence of activity on lipophilicity; the second deals with unrecognized interrelationships between certain physical parameters; and the third is a situation where a number of statistically significant correlations can be presented, each of which may be given a differing physical interpretation.