Influence of rudimentary attribute non-attendance (ANA) on choice experiment parameter estimates and design efficiency: A Monte Carlo Simulation analysis

The issue of attribute non-attendance (ANA) has been gaining increasing attention in the field of choice modeling. While the modeling issues, effects on parameter estimation, and, to a lesser degree, causes of ANA have been the main concern of research in this area, to date few studies have produced generalizable results about the effects of ANA on parameter estimates and little attention has been paid to the efficiency of experimental design in the face of ANA. This paper looks at these issues and also introduces a distinction between random and systematic ANA, which is defined to be ANA that is persistent in the face of choice task and/or attribute order randomization. As part of this study, Monte Carlo simulations are run to examine the effects of ANA on parameter estimation, under the conditions of random and systematic ANA. Simulations with respondent heterogeneity are also carried out to test the efficiency of latent class model estimations. The models perform well, but it is argued that the underlying assumption of serial ANA is indistinguishable from zero preferences with respondent heterogeneity, and such ANA is inconsequential to the choice made (i.e. the same choice is made whether or not the attribute is being attended to). In contrast, when a non-zero preference attribute is ignored, the latent model does not pick up the effects of ANA and additional data is required. Not incorporating ANA data significantly biases estimates of all parameters, especially when the marginal effects of the ignored attribute are relatively large. Finally, it is shown that orthogonal design is significantly disturbed by systematic ANA, and there is scope to improve it by using a D-efficient design.