Conference Agenda

In stated preference methods, researchers often rely on surveys and typically aim to obtain a random sample of the target population. However, participation is rarely random, and sample selection can bias welfare estimates. The problem compounds in multi-wave designs, where there is an initial participation stage into the first wave, an agreement to be re-invited stage into a second wave and a follow-through stage into the second wave. Many stated preference studies with longitudinal data do not model these layered selection processes, potentially leading to bias in welfare estimates. To address this research gap, we extend the Heckman selection framework commonly used for cross-sectional data to a dataset combining respondents who participated in one survey wave with those who participated in both waves. This extended framework corrects for (i) selection into the initial survey, (ii) selection into re-invitation, and (iii) selection into completing the second wave, each linked to willingness-to-pay (WTP) elicited through double-bounded dichotomous choice contingent valuation. We apply the model to a Danish survey study on public WTP for Carbon Capture and Storage (CCS) conducted in 2022 and 2023. Ignoring selection bias leads to an underestimation of household WTP by 43%, with the bias driven primarily by systematic differences between respondents who participated in both waves and respondents who participated in one wave Respondents who participated in both waves report higher WTP that declines over time, while respondents who participated in one wave report lower WTP that rises modestly. Aggregate WTP for CCS is otherwise temporally stable.