Why the linear utility function is a risky choice in discrete-choice experiments

This article assesses how the form of the utility function in discrete-choice experiments (DCEs) affects estimates of willingness-to-pay (WTP). The utility function is usually assumed to be linear in its attributes. Non-linearities, in the guise of interactions and higher-order terms, are applied only rather ad hoc. This paper sheds some light on this issue by showing that the linear utility function can be a risky choice in DCEs. For this purpose, a DCE conducted in Switzerland to assess preferences for statutory social health insurance is estimated in two ways: first, using a linear utility function; and second, using a non-linear utility function specified according to model specification rules from the econometrics and statistics literature. The results show that not only does the non-linear function outperform the linear specification with regard to goodness-of-fit, but it also generates significantly different WTP. Hence, the functional form of the utility function may have significant impact on estimated WTP. In order to produce unbiased estimates of preferences and to make adequate decisions based on DCEs, the form of the utility function should become more prominent in future experiments.