This paper has two objectives. The first is to estimate the value of implementing new marine conservation measures in Vietnam, focussing on the relative benefits of water quality improvements, coral conservation and control of marine plastics pollution. The second is to explicitly model heterogeneity in choice set formation alongside preference heterogeneity. The study examines any tendency of respondents to give consideration only, or not at all, to the "opt-out" option in our choice experiment, due to social and cultural norms to be willing to make some contributions to the public good. It further assesses the impact on welfare estimates of failing to account for differences in choice set formation (i.e. processing strategies). Results show that model fit substantially improves traditional multinomial, random parameters and latent class logit models when probabilistic choice set formation is accounted for. The analysis finds that not accounting for different choice set formation across respondents leads to biased welfare estimates.