A Comparison of Clinical Prediction with Several Different Methods of Statistical Prediction
Year: 1975
Type: article
Abstract: A comparison was made of unaided clinical prediction with an actuarially based linear regression equation as well as with linear statistical systems utilizing variable weights obtained from expert judges.The statistical methods considered were bootstrapping, sub jective importance weighting, a subjective correlation matrix ap proach, and unit weighting.Three different tasks were employed.In all three, advanced clinical psychology graduate students predicted judgments (regarding the prognosis of hypothetical children) made by a faculty member with whom they had had considerable professional con tact.In one condition the student judges made predictions on the basis of j highly valid cues and in the other they were provided with these plus 5 additional cues of relatively low incremental validity.The actuarial method proved to be highly valid across environ ments, This method, however, is often difficult to implement in ap plied settings.The importance weighting method, and. the average pre diction of the importance weighting algorithms of several student judges, were also valid across environments.Unaided clinical predic tion proved risky; i.e., some judges were highly accurate but others did poorly.The average prediction of several different judges was less variable and, for certain environments, more accurate than even the actuarial approach.The bootstrapping method generally provided a slight increase in achievement over the unaided clinicians Like the unaided clinical approach, however, it proved somewhat risky.On tasks where all of the cues were of similar validity, the unit weighting method was highly valid, '/¿hen cues were differentially valid, however, the unit weighting algorithm achieved a considerably lower level of accuracy.The method based upon correlation matrices supplied by the • student judges demonstrated little utility for these data, resulting in validity coefficients that were well below zero for some judges.It was concluded that the choice of a prediction method for a given task should be dependent upon such factors as financial considerations and upon considerations regarding the formal characteristics of the en vironment of interest.
Author: Mertz, William H.
Language: English
Subfield: Family Practice
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W6980641840