Population-Level Route Diversity Varies with Origin–Destination Structure: Evidence from Mobile Phone Signaling Data in Shenzhen, China

Route-choice research has treated route diversity mainly as an individual behavioral property, but population-scale route signatures make it possible to examine diversity as an attribute of origin–destination (OD) flows. Using September 2025 mobile phone signaling data from the DaaS BI platform (Zhihuizuji, China Unicom), we analyze map-matched route signatures for 1.61 million stable commuters in Shenzhen, China. The primary outcome is the route diversity index (RDI), computed as the ratio of distinct route signatures to total route observations within an analytical unit. OD flows whose endpoints occupy the same geohash-7 cell are more diverse than flows whose endpoints occupy different cells (RDI = 0.83 vs. 0.49; Cohen’s d = 1.61), and route diversity follows a nonlinear distance profile, with the lowest values in the 5–10 km range. The full OD-level model achieves high in-sample fit, while a model without the endpoint-grid indicator has nearly the same fit. The near-equivalent fits are consistent with shared information among flow scale, distance, and OD structure rather than being unique contributions from the endpoint-grid indicator. Social gradients are present but smaller than the OD contrast: the same-cell vs. different-cell gap is 1.6 times the full age-gradient range and 5.4 times the ARPU range. The local road-node density measure adds little independent information after OD-level controls. The largest observed differences occur at the OD-pair level. All the results refer to DaaS-derived, map-matched route signatures, not to direct observations of every trip.