Personalizing probabilistic survey scales

Abstract

Central bank surveys frequently elicit households’ probabilistic beliefs about future inflation by employing response scales centered around zero inflation. Analyzing data from the high-inflation period of 2022/2023, we demonstrate how this practice leads to distortions in households’ responses, causing inconsistencies and resulting in biased estimates of both mean inflation expectations and uncertainty. In two large-scale randomized experiments, we show that shifting the response scale to better align with households’ current expectations improves response quality. The greatest improvements are observed when the response scale is centered on each household’s point forecast. This personalized approach significantly increases the quality of responses by reducing inconsistencies, decreasing the use of unbounded intervals, and lowering the frequency of bi-modal responses. Centering the response scale on point forecasts removes the need to adjust the scale when inflation falls outside the originally intended range and helps mitigate panel conditioning effects.