A Taboo Trade Off model for Discrete Choice Analysis

Abstract

The idea that decision makers have to, and are willing to, make tradeoffs between the attributes of choice alternatives lies at the heart of microeconomic consumer theory (e.g., Lancaster, 1966; Keeney & Raiffa, 1976). This notion is embedded in the overwhelming majority of choice models, including but not limited to those which contain a conventional linear additive utility function (e.g., Ben-Akiva & Lerman, 1985; Train, 2009; Hensher et al., 2015). Even the majority of discrete choice models based on alternative, semi-compensatory decision rules such as loss aversion or regret minimization implicitly presume that decision makers are still willing to trade off one attribute (such as price) against another one (such as quality) to arrive at a choice that is optimal for them (e.g., Leong & Hensher, 2012; Chorus, 2014).

However, a large and growing body of literature in psychology, based on the seminal work of Philip Tetlock and co-workers (e.g., Fiske & Tetlock, 1997; Tetlock et al., 2000), disputes this viewpoint; it puts forward the idea that decision makers consider some types of trade-offs morally problematic, or taboo. A body of theoretical considerations and empirical findings suggest that there are indeed situations where decision makers refuse to make a tradeoff between different attributes and even become upset (express moral outrage) when being asked to consider such tradeoffs. These studies find that tradeoffs are considered taboo when the two attributes belong to different ‘spheres’; usually one attribute belongs to the sphere of market transactions (e.g. a price attribute), while another attribute belongs to, for example, the sphere of social relations (e.g. friendship) or another sphere in which market transactions are frowned upon (e.g. healthcare, or matters of war and peace). To consider just one example: while a decision maker, e.g. a traveler, may have no trouble trading off money and travel time, she is likely to consider a tradeoff between money and traffic safety (of herself and others) more morally problematic.

While the notion of taboo tradeoffs is by now firmly established in the behavioral sciences, its incorporation in an operational econometric model for discrete choice analysis is lacking. This is problematic, in light of the fact that many domains of application of discrete choice theory are likely to contain taboo tradeoffs. Think of health decision making, studies into the value of nature, the value of statistical life, or political decision making. The application of misspecified compensatory choice models in situations featuring taboo tradeoffs is likely to lead to biased estimates and misleading behavioral forecasts.

This study derives an operational, discrete choice theory based model which accommodates taboo tradeoffs. In a crucial departure from existing choice models including lexicographic choice models (Saelensminde, 2006), the taboo trade off model features an estimable penalty which is associated with the mere act of trading off two attributes; that is, our model separates the taste of an individual for a particular attribute, from the penalty she may assign to trading off the attribute with another one. Our model specification is econometrically tractable (and estimable using conventional software). In a crucial departure from transitivity axioms, our model allows for the potential situation where an individual is willing to trade off attribute A (e.g., travel time) with attribute B (e.g. non-fatal traffic accidents), and attribute B with attribute C (e.g. fatal traffic accidents), but hesitates to trade off attribute A with attribute C. The model allows for two different behavioral responses to taboo tradeoffs: a strong preference for the status quo alternative and a strong inclination to opt out of the choice situation altogether.

After introducing the model, we derive and interpret its econometric and behavioral properties. Then, we proceed with an empirical application in the context of traffic safety, based on a data set specifically collected for the purpose of testing our model. We compare our taboo trade off model with conventional linear additive models as well as with more recently proposed, semi- and non-compensatory choice models. To conclude, we provide an extensive discussion of the potential and limitations of our taboo trade off model for discrete choice analysis; and of the implications of (modeling) taboo tradeoffs for policy design in a variety of application domains.

Note: at the moment of submitting this abstract, the model has been derived, and a selection of its properties have been studied. The model has been successfully tested using synthetic data. Stated Choice data collection is underway. Our empirical analyses should be ready, well before the start of the conference, where we would like to present results ‘hot from the press’.