Correlation as a deceiving measure of fit

Abstract

There has been a continuing controversy over the use of correlations for evaluating fit of a model. Lost in this controversy, however, has been any evidence that correlations have in fact proved to be deceptive. This note discusses three instances in the human judgment/decision making literature where such deception has actually taken place. In each instance, high correlations provided initial support for the models involved. Reanalyses and later evidence, however, have shown that these correlations were deceptive. Thus, correlations not only can be deceptive, as has been argued earlier; these instances illustrate that correlations have been deceptive. © 1977, The Psychonomic Society, Inc.. All rights reserved.