Attribute attendance in choice experiments: Exploring issues of scale and attribute farming

An increasing number of choice experiment studies has shown that not all respondents consider all attributes when choosing their preferred alternative in the presented choice sets. Not accounting for this so-called ‘attribute non-attendance’ will lead to bias parameter estimates, and can hence increase or reduce estimates of willingness to pay. In this study, three aspects of attribute non-attendance are investigated. First, scale heterogeneity models are specified to test whether attribute attendance affects the variance of the error term. Second, the concordance between stated non-attendance and inferred non-attendance is assessed by comparing answers to supplementary questions with results from equality constrained latent class models. Finally the impacts of varying attribute descriptions or attribute levels on attendance is analysed. Results show that model fit is significantly improved when attribute non-attendance is taken into account, that but welfare estimates are not significantly different. The scale heterogeneity models reveal significant individual heterogeneity in scale, which decreases when incorporating ANA in the model specification. There is little overlap between stated and inferred non-attendance. Finally, describing a rare species attribute in terms of “number of species lost” attracts more attention to that attribute compared to “number of species present”, and that using a higher price vector may reduce attendance to the cost attribute.