Prediction: Clinical vs. Statistical

Introduction

In psychological assessment, an important step in the process of assessment is the prediction of criteria on the basis of assessment data, the so-called predictors. The processing of assessment data, which yields a prognosis, can follow two different methods of data combination: the statistical method and the clinical method. In the case of the statistical method, the combination of predictors is determined entirely on the basis of known empirical relationships between the predictors and the criteria; that is, on intersubjective knowledge as it arises from well-designed empirical investigations. In the case of the clinical method, no such empirical relationships are available or used; the combination of predictors is done in an ‘intuitive’ way based upon the subjective knowledge of the assessor which arises from his or her personal and professional experience.

The term ‘clinical’ in clinical prediction should not be misunderstood. It does not mean that this method of data combination occurs only in clinical psychology. The reasons for the choice of this term are historical and not systematic ones; and they reflect a long-standing controversy in psychological assessment between the advocates of intersubjective ways of information processing on the one side and the advocates of subjective ways on the other. Wiggins (1973) gives the best available reconstruction of this controversy, which has its origin in clinical psychology, but is in no way confined to this field of application of psychological assessment. Wherever, inside or outside of clinical psychology, criteria are predicted (only) on the basis of subjective knowledge, it is an instance of clinical prediction. And wherever criteria are predicted (only) on the basis of well-confirmed empirical knowledge, it is an instance of statistical prediction, regardless of the concrete statistical procedures used, e.g. [Page 750]actuarial tables, or linear or non-linear regression equations.

Theoretical Issues

The most important issue in the controversy between advocates of clinical and statistical methods of data combination was and is the question: ‘Which method to combine assessment data in the course of predicting criteria is better: the clinical or the statistical one?’ This issue can be treated from a theoretical and an empirical point of view. Meehl (1954) considered both points in his famous book Clinical versus Statistical Prediction: A Theoretical Analysis and a Review of the Evidence. From a theoretical point of view, the central question is: ‘Are there any a priori reasons why one method of data combination should be superior to the other?’

At the beginning of the controversy, the adherents to the statistical approach thought that their method must be better than the clinical one, because their method could be considered as mathematically optimal. The regression weights in a linear regression equation, for example, are determined in such a way that the (sum of squared) deviations of the observed criterion scores from the predicted ones are minimized. Therefore, it is impossible – that was the argument – that clinical prediction could beat statistical prediction. Only in those highly improbable cases where the intuitive estimation of regression weights by an assessor who uses the clinical method comes to the same results as the application of the statistical method will the clinical method be equal to the statistical method; in all other cases, it will be inferior.

...