Application of a Mixed Logit Model with Explicitly Correlated Coefficients for Best‒Worst Scaling Case 1 Data: A Case Study of Preferences for Job Opportunities in Rural Indonesia

Best‒worst scaling (BWS), which is housed under the category of stated preference methods, elicits respondents to select their top choice and their least favorite choice from choice sets. Mixed logit models have been frequently applied for estimating preferences. This study starts with a discussion of the estimation procedures of BWS and shows the necessity of applying a mixed logit model with correlated parameters to account for taste correlation and scale heterogeneity. As a case study, this study investigates latent preferences for new job opportunities in Indonesia, where the release of mercury (Hg) and its components into the atmosphere has had serious impacts on human health. The estimation results indicate that the most highly valued attribute for job opportunities is the provision of employment opportunities for society, followed by the frequency of salary payments and improvements in the environmental quality of the region. The estimation results also reveal that there is a significant degree of preference uncertainty in the choice of options in the middle-evaluated options. The research emphasizes that careful explanations are required to determine whether the choice sets provide reasonable policy menus for respondents or whether the middle-ranked attributes can be guided by the presence of other attributes.