International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Beyond a two-way hybrid model: maximize utility or minimize regret, or mix both
Mara Thiene, Juergen Meyerhoff

Last modified: 28 March 2017

Abstract


Random regret minimization (RRM) hast recently established itself as a complement to the random utility maximization (RUM) model. Generally, the RRM paradigm bases on the notion that when choosing, people are rather keen on minimizing future regret than aiming at maximizing future utility. Regret is here defined as what a decision maker experiences when a non-chosen alternative performs better than a chosen one. Chorus et al. (2009) describe this experience anecdotally as what a person might experience when standing in line at a cash register and observing that, while the own line making no progress, an adjacent one does. This example points directly to the crucial difference between the RUM and the RRM paradigm. While in the former only the chosen alternative counts, in the latter the decision maker is said to bilaterally compare the considered alternative – the chosen line at a cash register – with the other available alternatives – the other line.

A large number of studies demonstrate that RRM is a valid rational for individual choice behavior. Moreover, many studies have compared the random utility maximization and random regret minimization in order to find out which approach performs better. While the jury is still out, it looks like as if both approaches are close to each other with the RRM performing slightly better in some cases and the RUM slightly performing better in other cases. Despite generally small (although significant) differences in model fit, choice probabilities generated by estimates from the two models (as well as elasticities and parameter ratios) differ substantially, thereby implying different policy implications.

Pushing the debate about the accommodation of decision rules in discrete choice experiments further and trying to integrate both choice paradigms, a couple of studies have investigated whether it is reasonable to assume that only one decision rules is applied by all decision makers, i.e. utility maximization or regret minimization. One branch of literature has therefore used latent class models that allow for both pure RUM and pure RRM driven choices. Among the first investigating whether both paradigms are present among interviewees were Hess et al. (2012). They accommodate regret minimization and utility maximization by means of latent classes. In a following paper Hess and Stathopoulos (2013) linked a latent class model comprising a pure RUM and a pure RRM class with a latent variable that influences class membership. Boeri et al. (2014) also used a two class latent class model, again with a pure RUM and a pure RRM class, but they entered covariates directly in the latent class membership function and allowed for taste heterogeneity within both classes. All find evidence for a mixture of both rationales among respondents. Chorus et al. (2013) employ a different kind of hybrid model. They assume that decision makers do not process all choice attributes following one or the other paradigm but mix utility maximization and regret minimization. Their analysis shows that a hybrid model that contained regret-based and utility-based attribute decision rules outperformed, although only marginally, models where all attributes are assumed to be processed by means of one and the same decision rule, although differences in performance were rather small.

This paper adds to the literature by investigating whether a combination of the so far existing hybrid models with regard to RUM and RRM can provide further insights into the integration of both paradigms. We integrate the so far present hybrid models into a more comprehensive model that allows for three different processing strategies: pure random utility maximization, pure regret minimization, and a mixture of both. For our analysis we use a latent class model that simultaneously predicts membership to the three classes. Following the literature that used equality constrained latent class models to avoid confounding of taste heterogeneity and heterogeneity in processing strategies, we also constrain the attributes to be equal when the decision rules are the same across classes. The data are from a survey conducted to elicit visitors’ preferences of a protected area located in the Eastern Alps in Italy, the Dolomiti Bellunesi National Park.

Overall, the results indicate that in the present case the majority of respondents are likely to be a member of the class in which attributes are processed differently (mixture class). Most people do not seem to process all attributes in the same way. This is in line with the findings presented by Chorus et al. (2013) that it depends on the attribute. However, we also find evidence that some respondents indeed process all attributes according to one processing rule (Boeri et al. 2014). This indicates that an extension of the so far presented hybrid models could be an attempt to better capture processing rules applied by respondents while answering discrete choice experiment surveys. Moreover, this research also adds to the literature as it provides further evidence of the usefulness of the RRM approach in the field of environmental and resource economics.

Boeri et al. (2014) Stated choices and benefit estimates in the context of traffic calming schemes: Utility maximization, regret minimization, or both? Transportation Research Part A: Policy and Practice, 61, pp. 121-35.

Chorus et al. (2009) Spatial choice: a matter of utility or regret? Environment and Planning B: Planning and Design, 36(3), pp. 538-51.

Chorus et al. (2013) Regret minimization or utility maximization: it depends on the attribute, Environment and Planning B: Planning and Design 40(1), 154-69.

Hess et al (2012) Allowing for heterogeneous decision rules in discrete choice models: an approach and four case studies. Transportation 39 (3), 565–591.

Hess & Stathopoulos (2013) A mixed random utility — Random regret model linking the choice of decision rule to latent character traits, Journal of Choice Modelling, 9, pp. 27-38.


Conference registration is required in order to view papers.