Partially observable latent class analysis (POLCA): An application to serial participation in mosquito control in Madison, Wisconsin

Serial nonparticipation in nonmarket valuation using choice data is a frequently observed pattern of behavior in which an individual always appears to choose the status quo or ‘no program’ alternative. In choice models serial nonparticipation may be viewed as belonging to a class of deterministic choice patterns, other examples of which include serial participation and lexicographic preferences. While common in the context of environmental goods unfamiliar to respondents, logit-based choice models are ill-equipped for identifying such preferences, because predicted choice probabilities cannot take a value of zero or one. We extend latent class analysis (LCA) of preference heterogeneity to address this issue, for each class specifying a subset of alternatives that are avoided with certainty. We are then able to partially observe class membership, knowing with certainty that an individual does not belong to a class if she selects any alternatives excluded by that class. We apply our model to a discrete choice experiment on mosquito control programs to reduce West Nile virus risk and nuisance disamenities in Madison, Wisconsin. We find that partially observable latent class analysis (POLCA) obtains the same goodness of fit as LCA with fewer parameters. Adjusting for the need to re-specify the reference alternative when the status quo is excluded, our relative valuation measures are significantly different than those obtained from LCA. We argue that our model is useful for detecting and addressing alternative-specific nonidentification in a given dataset, thus reducing the risk of invalid inference from discrete choice data.