PREDICTION OF THE FUTURE VERSUS AN UNDERSTANDING OF THE PAST - A BASIC ASYMMETRY

Most of our knowledge in psychology and allied social sciences is based on observing consequences and seeking antecedents. The statistical analysis of such retrospective knowledge thus involves conditioning on consequences. This article demonstrates that given the common conditions of investigating ''unusual'' consequences, the degree of statistical contingency between a single consequence and a single antecedent is greater when conditioning on the consequence than when conditioning on the antecedent-which is, of course, necessary for prediction. Moreover, this asymmetry is exacerbated when the investigator is free to search for antecedents in a situation involving multiple potential antecedents. This asymmetry is exacerbated to an even greater extent when the investigator relies on memory rather than recorded observations in this search. Thus, if we identify the degree of statistical contingency in prediction with the degree we find in retrospection, we seriously overestimate. Our subsequent disappointment in our deficient predictive abilities can easily lead to rejecting variables and analyses that are in fact predictive in favor of those of unknown validity.