Document Type


Date of Original Version



Decision-theoretic criteria are presented for optimizing the information gathered from a series of interviews over time. It is shown that the optimum interviewing strategy depends strongly on assumptions about the covariation of behavior over time. Standard interviewing strategies, including the major-problem/target-complaints approach, are optimal only under extreme assumptions about behavior. An interviewing strategy based on dynamic programming is presented that will provide optimal information return from a series of interviews under assumptions that are realistic for mental health applications. A system using this approach can tailor its interviewing strategy to adapt to differences in interview content, item importance, and individual response patterns, selecting the optimally informative questions to ask each subject at each point in time. Simulation results show that this approach achieves a 34% reduction in the false negatives obtained with the major-problem/target-complaints method, and, depending on the acceptable error rate, a reduction of 47 % or more in the questions that are needed in standard interviewing.