Streetscape perception conditions the causal impacts of low-stress bicycle infrastructure

Abstract Perceived streetscapes (people's visual evaluations of the surrounding street environment) influence how people respond to transportation infrastructure, yet their role in shaping cycling behavior remains underexplored. Here, we combine street-view imagery, spatial AI techniques and causal inference to examine how neighborhood perceptions condition the effectiveness of bicycle infrastructure. We derive six types of perceptions from panoramic images and link them to built environment features. Leveraging the staggered rollout of protected and painted bike lanes in New York City, we implement a quasi-experimental, causal inferential framework using propensity score matching, Difference-in-Differences, and causal forest models on bikeshare origin–destination data. We find that bike lanes significantly increase ridership, with no detectable pre-treatment effects. Protected lanes increase ridership by ∼ 32% (β = 0.2705, p-value < 0.001), while painted lanes increase ridership by ∼ 10% (β = 0.0967, p-value < 0.05). Treatment effects vary systematically across perceived environments: painted lanes perform best on already safe and lively streets, whereas protected lanes remain effective across a wider range of contexts and generate the strongest gains in stressful or lower-quality environments. Street level counterfactual predictions provide the possibility to integrate human perception into infrastructure planning and design.