International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Modelling awareness and consideration in mode choice: an application on the Rome – Milan corridor
Mauro Capurso, Stephane Hess, Thijs Dekker

Last modified: 28 March 2017


The conventional assumption that decision makers are aware of all the available alternatives, and that these alternatives are all considered, appears unrealistic, even when their number is limited. Awareness might play an important role especially when individuals need to make an intrinsic effort (i.e. search costs) to identify all available alternatives. From a given awareness set, individuals might actually consider fewer alternatives as feasible ones, and then choose among them. The presence of consideration sets is behaviourally consistent with task-simplifying heuristics driven, for example, by self-imposed thresholds on attributes, captivity, inertia, boredom or fatigue. Methodologically, ignoring  awareness and the presence of consideration sets might result in less precise - or even biased - parameter estimates and forecasts and the violation of the IIA assumption (Williams and Ortúzar 1982, Swait 1984). Also, from a policy and industry perspective, a more comprehensive understanding of the decision making process can provide useful insights in terms of marketing strategies (i.e. in determining the relevant competition), as well as gains in determining the effects of policy actions.

This paper contributes to the discussion on the role of awareness and consideration sets in the estimation of discrete choice models. Awareness has received very little attention in the literature, probably because information on awareness is typically missing in traditional data sources. In many cases, given the limitations deriving from treating it as a truly Boolean indicator, familiarity has been considered a more reliable index. On the other hand, a number of probabilistic models dealing with the presence of consideration sets have been developed in the last forty years. Manski (1977) described decision making as a two-stage process, with individuals creating a nonempty subset of alternatives, from which to choose one (second stage). In terms of modelling, this means that each possible consideration set has a probability of being the true one, and that choice probability is conditional to a specific consideration set. Following this seminal work, other two-stage models have been proposed (Gaudry and Dagenais 1979, Swait and Ben-Akiva 1987a-b, Swait 2001b, Cantillo and Ortúzar 2005), while other authors preferred one-stage formulations discounting the utility for unconsidered alternatives (Cascetta and Papola 2001, Martinez et al. 2009). A number of studies also inferred consideration using processing data on perceived availability/consideration, as well as on the presence of attributes’ thresholds (Ben-Akiva and Boccara 1995, Swait 2001a, Hensher and Ho, 2015). Finally, an alternative stream of research looks at consideration and choice as the result of the same compensatory process (Horowitz and Louviere, 1995), meaning that individuals only consider the alternatives with the higher expected utility.

Although Manski’s formulation is  appealing in the way it deals with consideration, it suffers from a number of limitations. First, it does not recognise the role of awareness. Second, it becomes computationally infeasible as the number of alternatives increases. Third, identification problems arise if the same variables are assumed to influence both consideration and choice. In this paper we aim at addressing these issues, proposing an alternative two-stage probabilistic model, named class-allocation multinomial logit, which makes use of SC data and answers to supplementary questions on task-level consideration, and on the presence of self-imposed thresholds on attributes. In this model, the probabilities for all possible consideration sets are no longer estimated within the choice model, but they are estimated separately, thereby solving the second limitation of the Manski model. However, given that self-reported consideration indicators cannot be used as error-free measures, they are linked to respondents’ awareness of the alternatives and socio-economic characteristics through latent variables.

Empirically, we test the performance of the proposed model against a multinomial logit model (MNL) in a mode choice context, which is highly influenced by awareness and consideration. A sample of travellers on the Rome-Milan corridor, in Italy, were presented with two different SC surveys in the form of an online journey planner. In the first one, depending on the answers on the awareness of the alternatives (collected through unaided recall), respondents were first presented a restricted choice set, and asked to choose among those alternatives. They were then presented with additional choice tasks containing all actually available alternatives. In the second survey respondents were also asked which alternatives they were aware of, but such information was not used to restrict the set of the presented alternatives, so they could choose among all alternatives. In both surveys, at the end of each choice task respondents were asked which other alternatives they considered apart from the one they chose. Moreover, at the end of the survey they were asked to elicit, if present, their self-imposed thresholds on travel time and cost. Preliminary modelling shows that, as expected, models which probabilistically account for individual-specific consideration sets provide significant gains in terms of statistical fit with respect to the MNL model. Moreover, parameters’ estimates are also affected, revealing substantial differences in terms of willingness-to-pay indicators. From a statistical viewpoint, given the non-nested structure of the proposed model with the MNL model, it is acknowledged that traditional model comparisons might be not applicable in this context. Hence, we study the performance on a holdout (validation) sample. On this regards, it has been shown that choice models incorporating any type of choice set generation process forecast choices better than the MNL, given that at the individuals’ level, changes in the attributes of poorly considered alternatives should (in principle) not affect their behaviour.

To conclude, this paper highlights the wide range of benefits from asking additional questions on task-level consideration and self-imposed thresholds. We provide further evidence that such information can contribute to the identification of non-compensatory heuristics, hence suggesting its collection during SC experiments despite the associated costs in terms of the additional burden to respondents.


Ben-Akiva, M., and Boccara, B. (1995). Discrete choice models with latent choice sets. International Journal of Research in Marketing, Vol. 12(1), 9-24.

Cantillo, V., and de Dios Ortúzar, J. (2005). A semi-compensatory discrete choice model with explicit attribute thresholds of perception. Transportation Research Part B: Methodological, Vol. 39(7), 641-657.

Cascetta, E., and Papola, A. (2001). Random utility models with implicit availability/perception of choice alternatives for the simulation of travel demand. Transportation Research Part C: Emerging Technologies, 9(4), 249-263.

Gaundry, M. J., and Dagenais, M. G. (1979). The dogit model. Transportation Research Part B: Methodological, Vol. 13(2), 105-111.

Hensher, D. A., and Ho, C. (2015). The role of perceived acceptability of alternatives in identifying and assessing choice set processing strategies in stated choice settings: The case of road pricing reform. Transportation Research Part E: Logistics and Transportation Review, Vol. 83, 225-237.

Horowitz, J., and J. Louviere. (1995). What is the Role of Consideration Sets in Choice Modeling? International Journal of Research in Marketing, Vol. (12), 39-54.

Manski, C. F. (1977). The structure of random utility models. Theory and decision, 8(3), 229-254.

Martínez, F., Aguila, F., and Hurtubia, R. (2009). The constrained multinomial logit: A semi-compensatory choice model. Transportation Research Part B: Methodological, 43(3), 365-377.

Swait, J. (1984). Probabilistic choice set formation in transportation demand models. Unpublished Ph.D. Dissertation, MIT, Cambridge, MA.

Swait, J. (2001a). A non-compensatory choice model incorporating attribute cutoffs. Transportation Research Part B: Methodological, Vol. 35(10), 903-928.

Swait, J. (2001b). Choice set generation within the generalized extreme value family of discrete choice models. Transportation Research Part B: Methodological, Vol. 35(7), 643-666.

Swait, J., and Ben-Akiva, M. (1987). Incorporating random constraints in discrete models of choice set generation. Transportation Research Part B: Methodological, 21(2), 91-102.

Swait, J., and Ben-Akiva, M. (1987). Empirical test of a constrained choice discrete model: mode choice in Sao Paulo, Brazil. Transportation Research Part B: Methodological, 21(2), 103-115.

Williams, H. and Ortuzar, J. (1982). Behavioral theories of dispersion and misspecification of travel demand models. Transportation Research Part B: Methodological, Vol. 16 (3), 167-219.

Conference registration is required in order to view papers.