Abstract Body: Purpose: OpenStreetMap (OSM) serves as an open-and-crowd-sourced geospatial database widely utilized as a data-source in walkability studies, yet its sidewalk data accuracy remains unverified.

Methods: An OSM base-map of Maryland was generated, excluding roads inaccessible to pedestrians, then clipped to 141 randomly selected census tracts (~10% of total census tracts), stratified by population density tertiles. Fifty points within each census tract were randomly distributed across qualifying roads. The OSM base-map and Google Street View (GSV) at each point were reviewed visually by trained researchers to determine sidewalk presence (one or both sides of the street) or absence. Points with unavailable GSV were excluded, and unclear points were re-reviewed for consensus. Percent agreement in sidewalk data between GSV and OSM was calculated for all tracts and for a subset excluding tracts with 0 sidewalks in GSV. ANOVA and post-hoc pairwise tests compared mean accuracy scores by population density tertiles.

Results: Of the 7,050 points reviewed, across 141 census tracts, 6103 points were available in GSV, with 3674 points having sidewalks in GSV. When points without sidewalks were excluded from analysis, 11 census tracts were removed, showing that they had no points with sidewalks; all were from the low-density tertile. Agreement at all points and points with sidewalks in GSV are presented in Table 1.
ANOVA analysis conducted at all points showed significant differences based on population density tertiles. Pairwise comparison found the low-density tertile had significantly (p>0.001) higher agreeance from mid and high-density tertiles. ANOVA and pairwise comparison for points with sidewalks in GSV showed no difference (p>0.05) between population density tertiles.

Conclusions: OSM sidewalk data exhibits limited accuracy, with results showing underreporting of sidewalk presence in OSM. In 72% of points with sidewalks in GSV, OSM did not report sidewalks. The higher agreement at all points shows a lack of walkability, especially in lower density census tracts, combined with the default reporting setting of OSM, which is no sidewalks. These findings highlight the need to evaluate sidewalk presence when updating OSM content. Variations between census tracts may be caused by demographic factors, however, additional research is necessary. Potential limitations include availability and date of GSV images, and issues with generalizability outside study location.