Heterogeneity in decision processes: Embedding extremeness aversion, risk attitude and perceptual conditioning in multiple heuristics choice making

Abstract

There is an increasing interest, in the discrete choice modelling literature, in alternative behavioural paradigms that represent ways in which individuals make choices when faced with a choice set of alternatives, under conditions defined by revealed preference, stated choice or a mixture of both data sources. Attribute processing has come of age, and we see many studies using process heuristics such as attribute non-attendance (ANA), relative advantage maximisation (RAM), extremeness aversion (EA) and value learning (VL). With some exceptions (e.g., papers by Hensher, Hess, Scarpa, Campbell and colleagues, and Balbontin et el. 2017, 2017a), the study of each heuristic has been undertaken in isolation from other candidate heuristics; the exceptions being a joint investigation into a fully compensatory model defined by a linear additive in attributes and parameters specification and one process heuristic, commonly using latent class models (reinterpreted as probabilistic decision processing). Within the set of more than one candidate heuristic, limited account has been taken of the possibility that attributes are being processed under varying levels of risk attitude (instead assuming risk neutrality), and where multiple levels of an attribute might be observed in real markets (such as travel time over repeated trips with associated occurrences) and/or designed into stated choice experiments, no account is taken of perceptual conditioning. This paper investigates the role that two behaviourally appealing heuristics or decision rules play jointly in explaining choice making, both of which reflect risk attitude in different ways, where each heuristic contributes up to a probability within a sampled population both within and between respondents’ selection of a relevant multiple-heuristic utility expression. We jointly estimate a model that accounts for (i) extremeness aversion and (ii) an extended expected utility transformation for an attribute that accounts for risk attitude and perceptual conditioning. We use a stated choice experiment associated with a commuter car choice between tolled and nontolled roads in Australia, and compare the key behavioural output, the value of travel time savings (VTTS), obtained from the joint model and two stand-alone models. The findings suggest, after accounting for the probability of choosing each heuristic by each individual, in their construction of an empirical utility expression representing each alternative tolled road, that the mean VTTS from the multiple-heuristic model ($24.32/person hour) lies between the mean estimates obtained from the stand alone models ($21.45/person hour under extremeness aversion, and $29.19 when accounting for risk attitude and perceptual conditioning). The extremeness aversion heuristic has, on average, a 0.63 probability of relevance compared to a 0.27 probability of relevance for the other heuristic. Extremeness aversion (or seeking) is an appealing way of handling degrees of attribute risk that are not explicitly conditioned on the more traditionally identified risk parameter.