Big data, big bias? On factors shaping transit and shared micromobility integration

Understanding what drives shared micromobility use as a first-/last-mile (FM/LM) connection to transit is vital for enhancing multimodal travel. However, reliable data on transit-connecting micromobility trips are limited. Researchers often use buffer-based methods to infer FM/LM trips from large datasets, but such methods may cause bias in study results. This study uses a novel dataset with thousands of user-reported transit-connecting e-scooter trips to build a ground-truth model. Results show that more FM/LM trips are associated with higher transit-stop and road density, lower residential and intersection density, CBD proximity, greater employment rate, and lower shares of Black residents. Also, we find that using transit-connecting trips inferred from buffer-based methods unavoidably lead to biased model outputs. These biases stem from heterogeneity in station characteristics and transit-connecting trip patterns. We highlight trade-offs in inference accuracy and offer guidance for the use of big data in future research on transit and micromobility integration.