Presentation at the 2015 Active Living Research Annual Conference.
Land use mix reflects the availability of diverse destinations providing opportunities for active transportation. Land uses can be measured by in-person field audits or virtual audits. Google mapping platforms show promise for measuring neighborhood features due to their ease of use and accessibility to the public1. Street View offers a panoramic view of the street and local establishments. Aerial View offers a “bird’s eye view” with the option of quickly searching for neighborhood destinations. These two virtual perspectives have yet to be compared empirically to field audits.
To evaluate the validity of Google Aerial View (AV), Street View (SV) and the sum of non-overlapping land uses compared to field-observed (criterion) uses using an adapted version of the Microscale Audit of Pedestrian Streetscapes tool for use with Google (GMAPS). Agreement was also explored after stratifying neighborhoods by low/high socio-economic status.
Block groups in San Diego, CA and Phoenix, AZ regions were categorized as low versus high on GIS-measured walkability and median income (SES) to create a 2X2 matrix of four quadrants to ensure variance in neighborhood features. Pre-determined quarter mile routes (N=120, 60 per region, 30 per quadrant) were selected starting in residential areas moving towards commercial destinations. Raters in each region audited routes in person while the other region’s team completed virtual audits using the route section of the GMAPS tool. Field auditors tallied establishments and land uses on both sides of the street. Raters conducted similar assessments via SV by travelling the route and rotating perspective 180 degrees approximately every 100 feet. AV audits were conducted from approximately 2000 feet above ground level by searching for establishments. Raters audited 30 items on a scale ranging from 0 (none) or 1, or ≥ 2 land use establishments. Land uses were tallied along with the method of collection (field, AV or SV). A non-overlapping combination was calculated creating a sum of unique establishments collected by virtual audit (Total). Related items were grouped into scales with valence scores reflecting positive and negative influence on physical activity. An overall score was calculated from the difference between positive and negative valence scores2. Validity was explored using percent agreement, weighted kappa statistics (κ) for individual items, and intra-class correlation coefficients (ICCs) for scales. Kappa statistics were not possible when field audits revealed 95% or more of the routes contained no establishments. Agreement statistics were conducted after stratifying by low and high SES.
Percent agreement between field and virtual audit approaches (Table 1) were good to perfect3 across all items (74.2-100.0%). Field and virtual audits showed moderate to substantial agreement across items (κ=0.28–0.88), with the majority of items classified as substantial. SV or AV showed no distinct pattern favoring one method over the other compared to field audits. Total was not substantially better (κ>0.05) than either SV or AV. Most item agreement statistics were qualified similarly across low and high SES with the exceptions of Specialty Food Stores and Unmaintained Lot/Field, which showed better agreement in high SES neighborhoods (κ=0.39-0.52 vs 0.55-0.81, κ=0.34 vs 0.57 respectively), and Public Parks, which showed better agreement in low SES neighborhoods (κ=0.70-0.75 vs 0.40-0.46). Subscales, valence scores, and the overall land use score showed moderate to perfect agreement between the field observation and AV, SV, and Total observations (Table 2; ICC=0.48–0.93). In low SES neighborhoods, field audits agreed better with AV, SV, and Total for the Public Recreation item (ICC=0.60, 0.62, 0.51). In high SES neighborhoods, field audits agreed better with AV, SV, and Total observations for the Government Service subscale (ICC=0.39, 0.57, 0.39) and Negative Destination Land Use valence score (ICC=0.60, 0.54, 0.60).
In two US regions virtual audits provided an acceptable alternative for evaluating land uses compared to the time consuming and costly field audits. Few appreciable differences were found among SV and AV methods, suggesting either approach is acceptable. The improved validity of the non-overlapping total was minimal. Agreement between field and virtual audits showed no systematic differences across items after stratification by SES, suggesting that validity of virtual audits was not dependent on SES but rather on the nature of the land use. The exceptions were Government Service and Public Recreation subscales and Negative Destination Land Use valence score.
Results suggest that auditing neighborhoods virtually for various land uses is a valid approach and neighborhood socio-economic status does not influence these audits. The GMAPS tool is acceptable to use virtually for the assessment of land uses.
Lefer TB, Anderson MR, Fornari A, Lambert A, Fletcher J, Baquero M. Using Google Earth as an innovative tool for community mapping. Public Health Rep Wash DC 1974. 2008;123(4):474-480.
Millstein RA, Cain KL, Sallis JF, et al. Development, scoring, and reliability of the Microscale Audit of Pedestrian Streetscapes (MAPS). BMC Public Health. 2013;13:403. doi:10.1186/1471-2458-13-403.
Landis JR, Koch GG. The Measurement of Observer Agreement for Categorical Data. Biometrics. 1977;33(1):159. doi:10.2307/2529310.