Abstract
Discrete choice experiments are widely used in a broad area of research fields to capture the preference structure of respondents. The design of such experiments will determine to a large extent the accuracy with which the preference parameters can be estimated. This paper presents a new R package, called idefix, which enables users to generate optimal designs for discrete choice experiments. Besides Bayesian D-efficient designs for the multinomial logit model, the package includes functions to generate Bayesian adaptive designs which can be used to gather data for the mixed logit model. In addition, the package provides the necessary tools to set up actual surveys and collect empirical data. After data collection, idefix can be used to transform the data into the necessary format in order to use existing estimation software in R.
Highlights
Discrete choice experiments (DCE's) are used to gather stated preference data
This family of models is known as random utility maximization (RUM) models, and the most well known members are the multinomial logit model (MNL) (McFadden 1974), and the mixed logit model (MIXL) (Hensher and Greene 2003; McFadden and Train 2000; Train 2003)
The idea behind such models is that people maximize their utility, which is modeled as a function of idefix: optimal designs for DCE’s the preference weights and attribute levels
Summary
Discrete choice experiments (DCE's) are used to gather stated preference data. In a typical DCE, the respondent is presented several choice sets. Where orthogonal designs were mostly used at first, statistically optimal designs have acquired a prominent place in the literature on discrete choice experiments (Johnson et al 2013) The latter aims to select those choice sets that forces the respondent to make trade-offs, hereby maximizing the information gained from each observed choice, or alternatively phrased, to minimize the confidence ellipsoids around the parameter estimates. In order to reduce this sensitivity, Bayesian efficient designs were developed In the latter, the researcher acknowledges his uncertainty about the true parameters by specifying a prior preference distribution. It is limited to effects coded designs, and does not allow the user to specify alternative specific constants Such design packages are still often used, because some linearly optimized designs are optimal for MNL models when the preference parameters are assumed to be zero.
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
