Estimating Daily Trip Distributions from Aggregate Trip Data

Abstract

Travel demand models typically produce single point estimates of trip flows, discarding the uncertainty inherent in behavioral heterogeneity and external factors. This paper extends the utilitybased Bayesian MCMC framework of Scheffer et al. (2021) in three ways. First, the full MCMC posterior is propagated through the forward model via posterior predictive sampling, yielding a distribution of hourly trips — characterised by a mean and 95% credible band — rather than a single point estimate. Second, the aggregate trip distribution is decomposed into activity-specific components estimated from aggregate trip counts and activity participation data, without individual travel records. Third, scalability is evaluated on random subsamples (5%, 10%, 25%) and demographic strata, with outputs linearly rescaled to population size. Applied to a dataset synthetically derived from the Luxembourg national travel survey (27,490 respondents, 13,714 qualifying tours, four activity types), the posterior predictive mean closely matches observed demand (R2 = 0.963). A 5% random sample rescaled by a factor of 20 achieves R2 = 0.969. Scalability is bounded by sample representativeness, not data volume.