Incorporating Certainty and Attribute Non-Attendance in Choice Experiments : An Application to Valuation of Coastal Habitat

Due to their non-market characteristics, coastal habitats are often valued using stated preference methods, such as discrete choice experiments. An extensive literature cautions the confounding influence of both choice complexity and hypothetical bias in these types of welfare analyses. In response, researchers have developed a variety of ex post controls to reduce the influence of potential biases. This study utilizes a discrete choice experiment with the goal of measuring residents and visitors' willingness-to-pay for estuarine habitat and access in the Sarasota Bay Estuary watershed. Our application examines model calibration to address two important issues in the empirical literature: 1) reported choice certainty and 2) stated attribute non-attendance. We examine choice certainty by focusing on recalibrating data to account for uncertain choices by jointly modeling choice and certainty. We also account for stated attribute attendance and non-attendance and compare results with more standard models. Our results indicate that accounting for attribute non-attendance and choice certainty have minor effects on coefficient estimates, generally leading to a small reduction in the marginal utilities for coastal habitat and a small increase in the marginal utility of income. The more apparent influence of these approaches comes via smaller standard errors. These approaches result in slightly lower estimates of marginal-willingness-to-pay (WTP), suggesting that accounting for uncertainty may ameliorate hypothetical bias but they also provide more precise estimates. Results from our preferred model indicate respondents are willing to pay between $1.50 to $2.06 for an additional acre of wetlands, $0.54 to $4.55 for an acre of oyster habitat, $0.57 to $2.28 for an acre of seagrass, $0.09 to $ 0.32 for an additional artificial reef dome, and $ 2.87 to $ 7.18 for an additional ecological park. Uncertain respondents exhibit WTP that are not statistically different than $0