Clinical versus Actuarial Prediction
2001; Elsevier BV; Linguagem: Inglês
10.1016/b0-08-043076-7/01296-1
Autores Tópico(s)Meta-analysis and systematic reviews
ResumoMany statistical models have been developed for making predictions, according to some criterion of optimal fit (e.g., minimizing mean square error). Paul Meehl's influential 1954 book surveyed about 20 studies comparing such predictions—of important human outcomes (e.g., parole violation)—to predictions of clinical experts who had access to exactly the same information on which the statistical prediction was based. In no case was the human expert superior. Later work examined predictions where the information base for the statistical model and the clinical judge were not equivalent, or where predictions involved models—such as unit weighted ones—not based on optimality principles. In business and medical contexts, clinical judges sometimes make superior predictions, but only when they have access to information not included in the statistical models, or incorporate the models' predictions in their own. In psychology, statistical models are superior even when based on a subset of the information available to clinicians (who might, for example, also interview people). Moreover, statistical models need not be optimal to be superior but may even be 'improper'—as when a linear combination is based on ad hoc (but directionally correct) or intuitive weights. By the year 2000, roughly 150 studies supported these conclusions.
Referência(s)