An alternative method for predicting attrition from an alcohol treatment programme.

Aims: To test the predictive validity of a vignette methodology based on a Signal Detection model by examining treatment attrition within an alcohol clinic.

Methods: Participants were asked to categorize vignettes that described individuals drinking alcohol as problem or nonproblem alcohol use at the beginning of a 4-week intensive course of treatment. These participants were divided retrospectively into two groups: those who completed treatment and those who dropped-out of treatment. A matched post-treatment long-term abstainer group was also tested.

Results: Signal Detection analyses demonstrated that response bias scores predicted who would drop out of treatment (P = 0.01).

Conclusions: The vignette methodology provided useful levels of prediction in an applied clinical setting. It was argued that verbal reports from problem alcohol users may be more usefully conceptualized in terms of sensitivity and response bias than in terms of memory or 'truth'.