Predicting the Outcome of Psychotherapy: Findings of the Penn Psychotherapy Project

Our study of predictability of outcomes of psychotherapy used predictions of two kinds: (1) direct predictions by patients, therapists, and clinical observers; and (2) predictive measures derived from the same sources. Seventy-three nonpsychotic patients were treated in psychoanalytically oriented psychotherapy (mean, 44 sessions). Two thirds of the therapists were residents in psychiatry; one third were more experienced. The two main composite outcome measures, measured at termination, were Raw Gain (residualized) and Rated Benefits, which intercorrelated at 76. Most patients improved and showed a considerable range of benefits. The clinical observers' direct predictions of Rated Benefits were highest (.27, P <.05). The success of the predictive measures were generally insignificant, and the best of them were in the 2 to 3 range, meaning that only 5% to 10% of the outcome variance was predicted. The prognostic Index Interview variables did the best (eg, emotional freedom composite, 30; a crossvalidation for 30 patients was³⁹ (P <.05). Neither the therapist measures nor the early psychotherapy session measures predicted significantly. Reanalysis of the similar Chicago Counseling Center study, in our terms, showed a similar low level of prediction success, eg, adequacy of functioning, marital status match, and length of treatment predicted significantly in both studies.