Latent paths in mode choice models: incorporating street-level environment features for active travel policy

There is growing empirical evidence that walking and cycling are influenced by street-level environment features such as greenness and traffic stress. To assess active travel policy, these features are increasingly incorporated into mode choice models through predictors that aggregate spatial information along the route. However, this requires assumptions about which routes decision-makers evaluate. Simple assumptions (e.g., fastest path) can produce unrealistic routes and therefore inaccurate predictors, potentially biasing estimated street-level environment coefficients. We present a practical estimation process to evaluate plausible paths without prior route choice knowledge. In this proof of concept, we introduce a latent path selection mechanism in which routes are optimised together with mode choice. Implementations in Manchester (England) and Melbourne (Australia) produced plausible estimates comparable to existing mode and route choice literature. For example, cycling on high-stress paths was utility-equivalent to travelling 94–177% longer on low-stress paths for males, and this was higher for females (up to 389%) and children (up to 768%). Furthermore, using latent paths improved log-likelihoods and produced larger street-level environment coefficients versus fastest paths. For scenario forecasts, our methodology produced similar mode shifts at the population level, but with substantial variations at the trip-level by evaluating more decision-relevant routes. Latent path estimation could therefore be advantageous for forecasting applications where reliable chosen route data is unavailable. With growing availability of micro-spatial data and policy focus towards sustainable transport, this methodology can enhance how street-level features are represented in empirical literature and forecasting models.