The SPARC framework: Integrating cognitive processes into mathematical modelling of travel behaviour

This paper proposes the SPARC (Systemic–Psychological–Analytical–Reflective–Cognitive) framework for organising how psychological mechanisms are represented within mathematical models of travel behaviour. Rather than introducing a new estimator, it provides a mapping architecture spanning discrete choice, latent-class, dynamic, agent-based and machine-learning approaches, arguing that psychological claims in such studies often depend on data and measurement design, not model structure. Where existing integrative approaches bond psychological constructs to choice models as variables in a utility function, SPARC bonds the process architecture to model structure: which cognitive system is operative becomes structural, subject to an explicit identification condition. We carry one such bond through completely, from dual-process theory to a gated mixture of two behaviourally distinct kernels and the data required to license the interpretation, and sketch a scheme for others. Estimated on the public Swissmetro panel against three conventional comparators, it proves competitive; no predictive superiority is claimed. Its evidence is of a different kind: a latent class model free to respond to attributes in both classes places one at the attribute insensitivity the theory predicts, a psychological prediction confirmed by an estimator with no reason to confirm it. On simulated data it is correctly rejected where no switching exists, and recovers within-person switching where a trigger varies across occasions. In Swissmetro the identification condition is not met; we specify data that would meet it. The contribution is methodological: a scheme for bonding process theory to estimable structure, and the conditions under which such a bond is earned.