The impact of omitting key built environment factors on the potential health outcomes of active travel to school

Introduction: Active travel to school (ATS) programs promote walking and cycling because of the proven health benefits, which underpin the economic and social justifications for these initiatives. ATS programs often focus on improving the built environment (BE) around schools to encourage active accessibility. However, BE variables are frequently intercorrelated, making it challenging to include all relevant factors in choice models. Additionally, they are strongly associated with home-to-school distance, the primary determinant of ATS. Overlooking these interactions increases the risk of omitted variable bias (OVB) in travel mode choice models, reducing their effectiveness in accurately evaluating and informing ATS policies. Methods: Using a sample of 6269 students from 86 schools in New South Wales, Australia, we evaluate OVB in multinomial logit models of active, public transit, and car travel. Factor analysis condenses eleven built environment variables into five factors. Incremental models assess the impact of excluding BE factors on ATS participation, while elasticity comparisons illustrate the practical effects of OVB. We also evaluate hypothetical BE scenarios to demonstrate how OVB can distort ATS policy decisions. Results: We find that omitting key BE factors results in biased and inconsistent parameter estimates. The elasticity analysis of BE variables indicates a variation ranging from 1 % to 36 %. Scenario analyses further reveal that projected walking distance gains from BE enhancements can differ by up to 55 % across models. Conclusions: Differences in elasticity and projected scenario benefits highlight the impact of OVB in ATS evaluations, showing that inadequate BE representation can lead to misleading policy decisions. Bias direction varies, making it difficult for policymakers to determine whether benefits are over- or underestimated. This issue is amplified in equity-focused policies, where accurate benefit estimation is essential. Our findings emphasise the need for robust, nuanced, and inclusive model specifications, particularly when precise walking distance projections inform policy decisions.