A straightforward diagnostic tool to identify attribute non-attendance in discrete choice experiments

Abstract

To distinguish between respondents that have attended to/ignored an attribute in discrete choice experiments (DCE), Hess and Hensher (HH) apply the coefficient of variation of the conditional distribution, setting a threshold of 2 as a conservative rule of thumb. This paper develops an analytical framework (piecewise regression analysis — PWRA) to refine the HH approach, offering a flexible method to identify attribute non-attendance (ANA) in highly context-dependent DCE. It is empirically tested on a dataset used to value agricultural public goods. The results suggest that the identification of non-attendance and goodness of fit of different random parameter logit models that accommodate ANA are better when the framework developed in this research is applied. When comparing welfare estimates from the HH and PWRA approach, significant differences are observed. Consequently, the flexibility of the PWRA notably contributes to revealing context-specific ANA patterns that can help to provide more accurate welfare measures and therefore policy recommendations.