Accounting for the imperfect external validity of discrete choice experiments when predicting demand

This paper proposes a transparent and data-driven method to strengthen predictions from discrete choice experiments (DCEs). Firstly, we conduct a systematic review and meta-analysis to summarise how well DCE predictions reflect real-world choices. We find that DCEs have moderate accuracy when predicting health-related choices, with pooled sensitivity and specificity estimates of 88% (95% CI: 80%, 93%) and 0.31 (95% CI: 0.17, 0.50) respectively. Secondly, we set out a methodology to incorporate observed variation in prediction accuracy by adjusting DCE predictions to account for hypothetical bias. Thirdly, we present a case study applying the approach to a DCE predicting uptake for new HIV prevention products in South Africa.