Stated and Inferred Attribute Attendance Models: A Comparison with Environmental Choice Experiments

Abstract There is increasing evidence that respondents to choice experiment surveys do not consider all attributes presented in the choice sets. Not accounting for this ‘attribute non‐attendance’ leads to biased parameter estimates, and hence biased estimates of willingness to pay. Various methods exist to account for non‐attendance in the analysis of choice data, with limited agreement as to which method is ‘best’. This paper compares modelling approaches that can account for non‐attendance, based on stated and inferred attribute non‐attendance. Respondents' stated non‐attendance is incorporated in the specification of multinomial and mixed logit models. Inference of non‐attendance is based on equality constrained latent class models. Results show that model fit is significantly improved when attribute non‐attendance is taken into account, and that welfare estimates are lower when incorporating non‐attendance. The inference based on equality constrained latent class models provides the best model fit. There is little concordance between stated and inferred non‐attendance, suggesting that respondents may not answer attendance statements truthfully.