Causes and consequences of lexicographic choices in stated choice studies

Abstract

Stated Choice (SC) methods, and other choice experiments, are now becoming increasingly popular for the valuation of environmental goods. This paper shows that lexicographic choices (LCs) in an SC task, or other types of choice experiments, do not imply that the respondent has lexicographic preferences. LCs may be a result of (i) study designs where differences between the alternatives are too great and (ii) simplification of the choice task. A traditional approach that investigates motives for lexicographic choices mainly as a result of lexicographic preferences may therefore fail to reveal the whole motive range. Study designs that cause LCs, whether such choices actually are based on non-compensatory choices (i.e. genuine lexicographic preferences or strategic behaviour) or compensatory choices, provide less information about preferences. However, providing they are not totally dominant, such choices are not normally a serious modeling problem. It is argued that simplification is a consequence of respondents' differing abilities to choose. Such LCs contribute to the larger variances of SC data relative to less cognitively demanding valuation methods and might therefore have a significant impact on the implied valuation of non-market goods if they are not handled in the analysis. In data collection, we should use survey designs that collect more signals and less noise, and in data analysis we should use models that can separate signals from noise. However, there is obviously a limit to the amount of noise which can be separated from signals in a model and therefore how much noise such models can tolerate and still produce useful valuation estimates for the goods under consideration. For further investigation of such limits, a direct test procedure that can investigate whether each respondent has completed the task in a compensatory manner, like the one presented in the current study, is valuable.