A discussion of the notion that “past performance is the best predictor of future performance” needs to acknowledge that it is based on the behavioral consistency philosophy. Although it may not be widely recognized as such, selection tests and other entry-level screening systems are really substitutes for the ideal selection model, where in the presence of unlimited resources, there would be no need to use tests to make inferences or predictions about future job performance. Instead, we would use probationary periods where performance during said period provides the basis for making inferences about future performance.
Thus in the ideal world, given unlimited resources (and zero liability for risks, injuries, and such), we would not need to use tests to predict or make inferences about future performance; instead, we would let everyone in, have them perform on the job for a specified period time to obtain reliable measures of their performance, and then make subsequent retention and rejection decisions on the basis of this past performance.
Of course, we do not have unlimited resources (and zero risk and liability) and so this ideal probationary selection model rarely exists. Nevertheless, the axiom-”the best predictor of future performance is past performance”-serves as the conceptual basis for this ideal selection model.
Although again it may not be explicitly or widely recognized as such, this axiom also serves as the basis for ongoing performance-related retention and termination decisions. We retain high performers because on the basis of their past performance we expect them to be or remain high performers in the future, and we release weak performers because based on their past performance we expect them to remain weak performers.
In summary, from both a theoretical and empirical perspective, I do not think it is unreasonable to conclude that past behavior predicts future behavior and that all things being equal (including reasonable good measurement systems), past job performance should be the best predictor of future job performance, since this would be a case where the predictor and the criterion are the same.
