Chapter 1 Models of Bounded Rationality under Certainty

Soora Rasouli and Harry Timmermans

Abstract

Purpose – This chapter reviews models of decision-making and choice under conditions of certainty. It allows readers to position the contribution of the other chapters in this book in the historical development of the topic area.

Theory – Bounded rationality is defined in terms of a strategy to simplify the decision-making process. Based on this definition, different models are reviewed. These models have assumed that individuals simplify the decision-making process by considering a subset of attributes, and/or a subset of choice alternatives and/or by disregarding small differences between attribute differences.

Findings – A body of empirical evidence has accumulated showing that under some circumstances the principle of bounded rationality better explains observed choices than the principle of utility maximization. Differences in predictive performance with utility-maximizing models are however small.

Originality and value – The chapter provides a detailed account of the different models, based on the principle of bounded rationality, that have been suggested over the years in travel behaviour analysis. The potential relevance of these models is articulated, model specifications are discussed and a selection of empirical evidence is presented. Aspects of an agenda of future research are identified.

The study of travel behaviour concerns the description, analyses and modelling of decision processes related to multi-faceted travel behaviour. It aims at better understanding and predicting travel choices and how these co-vary with the decision- makers' personal traits and characteristics, attributes of the choice alternatives and context. The focus may be on cross-sectional analysis or on dynamics as it relates to scripts, variability in activity-travel patterns and adaptation. Decision-makers can be individuals, households, friends, business partners, etc. Facets concern departure time, destination, day of the week, travel party, transportation mode(s) and route. Generally, the attributes depend on the facet or combination of facets being modelled, while context may relate to the urban setting, economic conditions, weather, time pressure, etc. The study of travel behaviour ranges from a focus on a single facet to the modelling of dynamic comprehensive activity-travel patterns.

Forecasting travel behaviour is an essential component of feasibility and impact studies. The purpose of a feasibility study is to assess whether a project can achieve (financially or otherwise) a targeted performance. Performance measures often require a forecast of the number of people using the planned infrastructure or facility. Unless the new project attracts more people or expenditure from a larger market area than the minimum required to achieve feasibility, it will compete with the existing facilities in the market area. In that case, it is relevant to assess not only the feasibility, but also the impact the project will have on each of the competing facilities. Although under such conditions, the impact will be negative, the policy question is how negative and how the effects are distributed across the existing facilities.

In this forecasting setting, if one is not satisfied with a statistical analysis only, but it is felt important to base the model on a theory of decision-making, preference or choice, then the researcher has to (i) select or formulate a theory of decisionmaking that is assumed valid for the problem at hand, (ii) translate the general theory into an operational model that mathematically expresses the functional relationship between decision outcomes and the set of individual and household characteristics, attributes of competing choice alternatives and context conditions. An adequate reproduction of observed decision outcomes is then seen as a validation of the behavioural postulates underlying the model.

In turn, models based on behavioural theories are often implicitly or explicitly perceived as being superior to aggregate statistical models. A good example is the shift from aggregate spatial interaction models, embedded in four-step travel demand forecasting models, to activity-based travel demand models, which are based on behavioural postulates and mechanisms (Rasouli & Timmermans, 2014a). While we generally agree with the contention that models capturing behavioural processes are better capable of predicting the effects of policies than statistical models that are only based the outcomes of decision processes than on decision processes themselves, particularly if the policies violate the antecedent conditions that have led to observed aggregate patterns, the issue is more complicated than often suggested in the literature.

Firstly, it is common practice to estimate variations of a model that are based on the same underlying behavioural postulate. For example, one may be satisfied with estimating a multinomial logit model only because the literature has suggested it to be robust. Alternatively, one may compare the performance of a multinomial model against a model allowing for different variances and/or covariances, but both are based on the same behavioural postulate of utility-maximizing behaviour. The more complex versions of the model may better reproduce the observed choices, but this approach does not give any guidance whether the assumed utility-maximizing decision process is the best representation of the decision-making process.

Secondly, one should realize that the same mathematical expression, depicting the functional relationship between the dependent and the set of independent variables, can often be derived from different conflicting behavioural theories. For example, the multinomial logit model can be deducted from Luce's choice theorem and random utility theory, which fundamentally differ with respect to the nature of preferences (deterministic vs. stochastic) and the nature of the decision process (probabilistic vs. deterministic). Regret-based choice models which define regret as a linear function of attribute difference between the best non-chosen and the chosen alternative are mathematically equivalent to the multinomial logit model; yet the principle of regret-minimization is fundamentally different from the principle of utility-maximizing behaviour. To make matters even more complicated, the mathematical expression of the multinomial logit model can also be derived from the quantum response model, which is a theory of decision-making under uncertainty rather than a theory of riskless choice. This equivalence implies that any satisfactory fit of the model to the data is just a necessary but not a sufficient condition for validating the behavioural principles and mechanisms underlying the mathematical model.

When developing behavioural models, it is important to critically consider which theory seems most valid for the decision-making process under investigation. Unfortunately, the travel behaviour community, unlike for example the marketing community, does not have a rich tradition of developing, let alone systematically comparing, alternate theories of choice and decision-making. The vast majority of studies on various facets of travel behaviour has been based on discrete choice models, which in turn have been interpreted as representations of random utility theory. This theory can be seen as an example of a theory of rational decision-making. Individuals are assumed engaged in a high involvement decision process in which they have full information about the set of attributes, characterizing the choice alternatives in their choice set, from which they derive a utility. The behavioural principle of utility-maximizing behaviour then leads to a set of probabilities of choosing the alternatives in an individual's choice set. The approach negates any emotional considerations.

Although the literature in travel behaviour research on models of bounded rationality is relatively small, travel behaviour researchers have occasionally explored the formulation and application of such models. The purpose of this chapter is to provide an overview of these models, allowing readers to better understand the contribution of the specific papers, included in this book.

This chapter is organized as follows. Firstly, we will present a general framework for positioning various models and theories of choice and decision-making. Based on this framework, we will continue the conditions under which we would consider the decision-making process evidencing bounded rationality. These conditions are used in the remainder of the chapter to organize existing research in mainly transportation and urban planning research on bounded rationality. More specifically, we will first discuss models that do not necessarily lead to an optimal choice. Next, we will discuss models that involve simplifications of the decision process by not considering all relevant attributes. This is followed by a discussion of models, which assume that individuals ignore small attribute or alternative differences, and therefore are indifferent between those choice alternatives that only differ marginally. Finally, we will discuss modelling attempts aimed at mimicking how individuals ignore choice options to reduce their consideration set. The chapter is completed with a conclusion, discussion and agenda of further research.

Found a mistake? Please highlight the word and press Shift + Enter