Attribute non-attendance (ANA) in choice experiments has recently been the subject of much investigation. There is no general agreement about the magnitude and direction of the effects of non-attendance but most studies conclude that ANA matters in applied choice analysis for policy purposes. Two approaches have been used so far to identify and quantify attribute non attendance, namely respondent-reported non-attendance and analytical (or inferred) non-attendance. The consistency between self-reported ANA and ANA inferred by statistical models is not yet resolved.

The present study aims at testing the internal consistency of inferred non-attendance from different stages of a rank-ordered best/worst elicitation method and at teasing out drivers of non-attendance in full rank-ordered data.

The empirical dataset comes from a multi-attribute stated choice survey carried out in 2008 in the province of Trento (North East Italy) aiming at estimating willingness to pay for access to the Alpine grazing commons. These areas provide many interesting features to visitors but they are nowadays under threat of abandonment. Visitors’ WTP for access may contribute to their up-keep as public funding dwindles.

With regards to the effects of ANA on implied welfare estimates, we focus on interval estimates and we find that accounting for ANA does affect interval estimates of welfare, especially the values of at the upper range of the confidence intervals, and this might be of consequence in the sensitivity of benefit-cost analysis.

ANA models were estimated at the various ranks and inferred ANA at the individual level was investigated by using posteriors from the various models. A multivariate probit analysis of the individual respondent’s pattern of inferred ANA behaviour finds important determinants that support the notion of some ANA coherence across ranks in the sequence of choice by the same respondent. Particularly, we find that framing the question around the identification of “guiding” attributes gives significant effects with plausible signs, while statements about ignored attributes do not.