Does the number of discrete choice alternatives matter for respondents’ stated preferences? The case of tap water quality improvements

Contingent valuation is among the most widely used techniques for studying consumers’ preferences. Nevertheless, whether respondents reveal their true preferences in contingent valuation surveys is still the subject of academic debate. The existing literature indicates that the truthful disclosure of preferences pivots on the number of alternatives presented in a single choice task. On a theoretical basis, the use of a two-alternative task format has long been recommended because of its incentive-compatible properties, which ensure that respondents’ disclosure of their true preferences constitutes their optimal strategy. However, the empirical literature presents nascent evidence that providing more than two choice alternatives may increase the respondents’ likelihood of finding an option that satisfactorily matches their preferences; consequently, a multiple-alternative task format likely enhances the accuracy of preference disclosures. Furthermore, empirical studies often employ multiple alternatives for a single task because of statistical efficiency gains. The lack of consensus about the impact of the number of alternatives on respondents’ truthfulness when stating their preferences in contingent valuation surveys motivates this study. Using data from a discrete choice experiment, we examine whether willingness-to-pay (WTP) estimates depend on the number of alternatives provided for a single choice task. We employ a split-sample design that uses two- and three-alternative formats in a contingent valuation survey of proposed public policies for the improvement of tap water quality (iron and chorine content, hardness) in Milanówek, a town in the Warsaw agglomeration in Poland. Drawing on a generalized mixed logit model with scale heterogeneity, we find no significant differences in the mean WTP values elicited with two- and three-alternative tasks, while the WTP estimates based on three-alternative tasks appear to have relatively lower standard errors compared with two-alternative tasks. This finding indicates that using three or more alternatives per choice task may offer a way to increase efficiency without biasing the results.