The Hidden Decline of Trust in Science

Surveys of trust in science are conducted by scientists, creating a structural selection problem: those who distrust science are less likely to participate. In cross-sectional designs this bias is invisible; in panel studies it appears as differential attrition that inflates apparent trust over time. Using five waves (2019–2024) from a national probability panel in New Zealand (N = 42,681) and multiple imputation to correct attrition bias, we find that apparent gains in the social value of science vanish after correction and that confidence in scientists, after an initial pandemic-era rise, falls to or below baseline by 2024. Simulation analyses indicate that correction methods recover trajectories under standard missing-data assumptions. Categorical estimates reveal hidden erosion of high confidence in scientists and growth in low trust that uncorrected data obscure.