The Predictive Value of Discrete Choice Experiments in Public Health: An Exploratory Application

Abstract

The objective of this study was to assess the predictive value of a discrete choice experiment (DCE) in public health by comparing stated preferences to actual behavior. 780 Type 2 diabetes mellitus (T2DM) patients received a questionnaire, containing a DCE with five attributes related to T2DM patients' willingness to participate in a combined lifestyle intervention. Panel mixed-multinomial-logit models were used to estimate the stated preferences based on 206 completed DCE questionnaires. Actual participation status was retrieved for 54 respondents based on patients' medical records and a second questionnaire. Predicted and actual behavior data were compared at population level and at individual level. Based on the estimated utility function, 81.8 % of all answers that individual respondents provided on the choice tasks were predicted correctly. The actual participation rate at the aggregated population level was minimally underestimated (70.1 vs. 75.9 %). Of all individual choices, 74.1 % were predicted correctly with a positive predictive value of 0.80 and a negative predictive value of 0.44. Stated preferences derived from a DCE can adequately predict actual behavior in a public health setting.