Accounting for Spatial Heterogeneity Using Crowdsourced Data

Alattar, Mohammad Anwar; Cottrill, Caitlin; Beecroft, Mark · 2021 · Crossref

DOI: 10.32866/001c.22495

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the need to improve the predictive performance of active travel modeling by accounting for spatial heterogeneity using high-resolution crowdsourced data. While human-powered transportation offers significant health and resilience benefits, previous research has largely relied on traditional data sources and non-spatial regression techniques, such as Ordinary Least Squares (OLS), which often fail to capture the spatial dependence inherent in cycling patterns. The authors argue that crowdsourced datasets, characterized by their volume, velocity, and fine spatial granularity, can better incorporate these spatial components. The primary objectives were to examine the spatial dependence of cycling within Glasgow and to compare the performance of OLS against two spatial regression models: the Spatial Lag Model (SLM) and the Spatial Error Model (SEM). The methodology utilized two crowdsourced datasets: Strava Metro data from 2018, providing counts of cycling trips at street intersections, and street network data derived from OpenStreetMap using the OSMnx Python toolkit. The researchers quantified four street network centralities—degree, betweenness, closeness, and eigenvector—to serve as independent variables. Data preparation involved integrating trip counts with centrality metrics using QGIS, logarithmically transforming variables to reduce skewness, and defining spatial neighbors via Thiessen polygons and a Queen’s contiguity matrix. Spatial dependence was assessed using Univariate Moran’s I analysis. The study then implemented OLS, SLM, and SEM using GeoDa, performing multicollinearity checks and residual diagnostics for heteroskedasticity and spatial dependence to evaluate model adequacy. The findings revealed significant spatial autocorrelation in cycling trips (Moran’s I = 0.481, p < 0.05), indicating that high or low cycling volumes cluster geographically. This clustering is attributed to factors such as proximity to infrastructure, area affluence, and the "safety-in-numbers" effect. Diagnostic tests confirmed that OLS violated key assumptions due to residual heteroskedasticity and spatial dependence, rendering it inadequate for this dataset. In contrast, the SEM demonstrated superior predictive performance compared to both SLM and OLS. The SEM achieved a higher R² (0.43 vs. 0.408 for SLM and 0.165 for OLS) and better fit according to Log-likelihood, AIC, and Schwarz criteria. A bootstrapping analysis on a subset of data further validated these results, confirming that accounting for spatial spillover effects yields more accurate models. The study concludes that crowdsourced data enables the implementation of robust spatial modeling techniques that traditional data sources cannot support. By demonstrating the inadequacy of OLS and the superiority of SEM in capturing spatial variation, the research highlights the potential of high-resolution crowdsourced data to provide deeper insights into active travel behaviors. These findings suggest that incorporating spatial heterogeneity in transport modeling can lead to more informed interventions and improved decision-making for promoting active travel.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• exposure measurement