Modelling aggregate travel behaviour over time in the absence of panel data: methods and a case study from Colombia

Most travel behaviour studies rely on cross-sectional data. The absence of a temporal dimension in such data hinders analysts from establishing causal relationships and understanding travel patterns over time, which are key to developing accurate analyses and forecasting tools. Studies aiming to capture choices over time are often constrained by the limited availability of panel data, especially for long-term evaluations and in low- and middle-income countries, where, at best, repeated cross-sections are available. This paper applies a novel Latent Class (LC) choice model to accommodate heterogeneity in mode choices over time when panel data is unavailable, exploiting four waves from a repeated cross-sectional dataset collected in Bogotá, Colombia, between 2011 and 2023. Instead of modelling individual choices, we track the behaviour of groups of people over time. To do so, the data is spatially aggregated to link observations across waves, thereby constructing a pseudo-panel at the zonal level, and modal market shares between origin–destination pairs are modelled using a Fractional Multinomial Logit Model (FMNL). The LC structure introduces temporal variation in preferences by allowing zonal groups to switch classes over time, in contrast to traditional LC, in which class membership is fixed. Our findings suggest that two main modality styles (i.e., classes) explain mode choices in Bogotá, the first is characterised by a lower value of time and a higher willingness to cycle and walk longer distances, and the other is more time sensitive. We find that the probability of belonging to the first class increases over time, a finding enabled by the novel approach we propose.