(Mis)measuring Uncertainty: Bunching in Probabilistic Expectations

Probabilistic expectation questions are often used to measure subjective uncertainty, but respondents frequently assign all probability to one outcome. We show that this bunching is partly a survey artifact rather than genuine certainty. In the (online) ECB Consumer Expectations Survey and the (in person) Spanish Survey of Household Finances, bunching is higher among less financially literate respondents and increases with panel tenure. Two survey-design changes provide stronger evidence: enforcing neutral interviewer protocols reduces bunching by about 15 percentage points, and a later questionnaire redesign reduces it by a further 18 percentage points. Interviewer-level variation also falls after standardization. These findings imply that degenerate responses in probabilistic expectations can reflect task burden and survey administration, causing standard measures to overstate certainty and understate uncertainty.