The Hidden Erosion of Trust in Science

When survey participation depends on the attitude being measured, survey estimates are biased by design. Trust in science is especially vulnerable because scientists field the surveys. We analysed five waves from 2019-2020 through 2023-2024 in the New Zealand Attitudes and Values Study, a probability panel, restricting analyses to the 2019 cohort (N = 42,681), and compared observed responses with multiple-imputation estimates that adjust differential attrition. Correction changes the substantive conclusion for both facets: for confidence in scientists, an apparent rise becomes a pattern of increasing low trust and contracting high trust; for attitudes toward science's social value, an apparently strong increase shrinks to a level consistent with no change. Because initial nonresponse to a science survey may be selective and is unobserved, the corrected estimates are conservative lower bounds. Longitudinal panels can diagnose and partially bound this bias; cross-sectional surveys cannot. Low trust in science may be systematically underestimated.