The Use of Alternative Preference Elicitation Methods in Complex Discrete Choice Experiments

Abstract

The full-text may be used and/or reproduced, and given to third parties in any format or medium, without prior permission or charge, for personal research or study, educational, or not-for-prot purposes provided that:• a full bibliographic reference is made to the original source • a link is made to the metadata record in DRO • the full-text is not changed in any way The full-text must not be sold in any format or medium without the formal permission of the copyright holders.Please consult the full DRO policy for further details. We analyse stated preference data over nursing jobs collected from two different discrete choice experiments: a multi-profile case best-worst scaling experiment (BWS) prompting selection of the best and worst among alternative jobs, and a profile case BWS wherein the respondents choose the best and worst job attributes. The latter allows identification of additional utility parameters and is believed to be cognitively easier. Results suggest that respondents place greater value on pecuniary over non-pecuniary gains in the multi-profile case. There is little evidence that this discrepancy is induced by the extra cognitive burden of processing several profiles at once in the multi-profile case. We offer thoughts on other likely mechanisms.JEL classification: C23, C25, C81, J44 Key words: discrete choice experiment, preference elicitation, rank-ordered data, latent class logit, best-worst scaling, maximum-difference model Highlights:• We compare preferences on nursing jobs elicited by profile and multiprofile case DCEs.• The paper is the first to contrast the two types of DCEs using monetary and nonmonetary attributes.• Preferences are comparable across the DCEs but only for non-monetary attributes.• Respondents value salary gains relatively more in the multi-profile DCEs.• The evidence suggests that this discrepancy is not due to the variation in cognitive difficulty.

The use of alternative preference elicitation methods in complex discrete choice experiments

“…For a sensitivity check, we have also estimated a latent class variant of the sequential best-worst model (Marley and Louviere, 2005) which postulates a simpler choice behaviour: a respondent looks for the best of 12 attribute-levels, and then the worst among the other 11 attribute-levels in two statistically independent steps. This alternative model, however, has a very similar likelihood as the workhorse max-diff model, and leads to almost identical estimates; see our earlier draft (p.17, Yoo and Doiron, 2012) for further comments. Our findings are in line with Flynn et al (2008) who also find the two behavioural models empirically comparable.…”

“…12 Our earlier draft (p. 21, Yoo and Doiron, 2012) presents the results from the simple max-diff and HROL models that ignore unobserved heterogeneity. The results from these simple models closely resemble those in Figure 3; the main difference is a decrease in scale which is expected since omitted preference heterogeneity increases unexplained variances (Revelt and Train, 1998 Figure 3 also shows that the relative utility gains from two different non-salary characteristics is much more robust across data sets than relative gains involving salary and non-monetary characteristics.…”

A Latent Class Nested Logit Model for Rank-Ordered Data with Application to Cork Oak Reforestation

“…Following the common practice (Train, 2008;Claassen, Hellerstein and Kim, 2013;Keane and Wasi, 2013;Shulz, Breustedt and Latacz-Lohman, 2013;Yoo and Doiron, 2013), we use the Bayesian Information Criterion (BIC) to choose the optimal number of classes.…”

“…It is therefore important to consider developing econometric methods for rank-ordered data, to complement the continual effort to understand better and improve ranking survey designs (Caparrós, Oviedo and Campos, 2008;Chang, Lusk and Norwood, 2009;Scarpa et al, 2011;Akaich, Nayga, and Gil, 2013;Louviere, Flynn and Marley, 2015). The existing approach to analyzing rank-ordered data usually exploits extensions and variants of the exploded logit (Chapman and Staelin, 1982), both within (Chang, Lusk and Norwood, 2009;Scarpa et al, 2011;Resano, Sanjuan and Albisu, 2012;Othman and Rahajeng, 2013;Varela et al, 2014) and outside (Fok, Paap and Van Dijk, 2012;Yoo and Doiron, 2013) the environmental valuation literature. Our strategy adds to the empirical practitioner's toolkit an approach building on the nested rank-ordered logit of Dagsvik and Liu (2009) that allows for more plausible substitution patterns.…”

“…Building on the idea of Heckman and Singer (1984), our usage of the latent class approach is to specify a type of mixed logit specification which is called, inter alia, finite mixture logit, discrete mixture logit, or non-parametric mixed logit (Train, 2008;Claassen, Hellerstein and Kim, 2013;Keane and Wasi, 2013;Yoo and Doiron, 2013). As the last name, due to Train (2008), makes explicit, this specification uses a discrete distribution with C support points as a tool to obtain a non-parametric approximation to the unknown population distribution of random utility parameters, without committing to a particular parametric form (e.g.…”