Efficient Conjoint Choice Designs in the Presence of Respondent Heterogeneity

Abstract

Random effects or mixed logit models are often used to model differences in consumer preferences. Data from choice experiments are needed to estimate the mean vector and the variances of the multivariate heterogeneity distribution involved. In this paper, an efficient algorithm is proposed to construct semi-Bayesian -optimal mixed logit designs that take into account the uncertainty about the mean vector of the distribution. These designs are compared to locally -optimal mixed logit designs, Bayesian and locally -optimal designs for the multinomial logit model and to nearly orthogonal designs (Sawtooth (CBC)) for a wide range of parameter values. It is found that the semi-Bayesian mixed logit designs outperform the competing designs not only in terms of estimation efficiency but also in terms of prediction accuracy. In particular, it is shown that assuming large prior values for the variance parameters for constructing semi-Bayesian mixed logit designs is most robust against the misspecification of the prior mean vector. In addition, the semi-Bayesian mixed logit designs are compared to the fully Bayesian mixed logit designs, which take also into account the uncertainty about the variances in the heterogeneity distribution and which can be constructed only using prohibitively large computing power. The differences in estimation and prediction accuracy turn out to be rather small in most cases, which indicates that the semi-Bayesian approach is currently the most appropriate one if one needs to estimate mixed logit models.semi-Bayesian mixed logit design, fully Bayesian mixed logit design, heterogeneity, prediction accuracy, multinomial logit design, model-robust design, -optimality, design construction algorithm

The Impact of the Soccer Schedule on TV Viewership and Stadium Attendance: Evidence from the Belgian Pro League

“…The primary advantage of this criterion over other optimality criteria is that the optimal design is independent of the scale or coding of the attributes (Zwerina et al 1996;Goos 2002;Kessels et al 2006;Yu et al 2009). In line with the previous notation, the D-error is defined as…”

“…However, these mean values were estimated on a small sample (60 respondents), therefore we used a relatively large prior variance of 1 for each parameter. With large prior variances, the design will be relatively efficient for a larger region around the mean values (Zwerina et al 1996;Sandor & Wedel 2001;Kessels et al 2006;Yu et al 2009). …”

Improving the efficiency of individualized designs for the mixed logit choice model by including covariates

“…Sándor and Wedel (2002) and Yu et al (2009) construct locally and Bayesian D-efficient designs respectively for a mixed logit choice model not taking the correlation between individuals' successive choices into account. Bliemer and Rose (2010) on the other hand do account for this correlation in the design setup but only manage to compute locally efficient designs.…”

Optimal Paired Choice Block Designs

“…D-optimal designs have been obtained theoretically under the utility-neutral setup, for example, see Graßhoff et al (2003), Graßhoff et al (2004), Street and Burgess (2007), Street and Burgess (2012), Demirkale, Donovan, and Street (2013), Bush (2014), Großmann and Schwabe (2015) and Singh, Chai, and Das (2015). In contrast, in the locally-optimal and the Bayesian approach, D-optimal designs have been obtained using computer algorithms (see, Huber and Zwerina (1996), Sándor and Wedel (2001), Sándor and Wedel (2002), Sándor and Wedel (2005), Kessels, Goos, and Vandebroek (2006), , , Kessels et al (2009), Yu, Goos, and Vandebroek (2009)). In this paper, we follow the utility-neutral approach.…”