Graduate sociology student Adam Visokay, along with a group of his peers at the University of Washington*, have investigated how the recency and frequency of GSV street-level imagery correlates with socioeconomic conditions at the neighborhood level in their recent publication Street View for Whom? An Initial Examination of Google Street View’s Urban Coverage and Socioeconomic Indicators in the U.S.
Street-level imagery from services like Google StreetView (GSV) have become foundational to urban informatics, supporting diverse research in urban planning, infrastructure monitoring, neighborhood audits, and real estate analytics.
With over 220 billion images across more than 100 countries, GSV has become particularly influential due to its geographic coverage, accessibility, and high-resolution panoramic images. Researchers leverage GSV imagery to investigate urban environments, perform virtual audits, and extract detailed environmental and infrastructural features at scale. Despite GSV’s widespread adoption and practical utility, disparities in how frequently or recently these images are updated across neighborhoods and cities remain largely unexamined. These disparities can reflect and exacerbate deeper inequalities in the digital representation of urban spaces, potentially biasing downstream analyses and influencing policy decisions.
In this work, Visokay and his colleagues introduce a novel pipeline for assessing coverage bias in the recency of publicly available Google Street View (GSV) and Census APIs across socioeconomically varied neighborhoods. Their approach computes Pearson correlations between key socioeconomic indicators (e.g., median income, population density) and image staleness using high-quality North American Census APIs.
Their findings have important implications for researchers relying upon SVI data instead of in-person audits, as bias can be propagated into downstream analysis without correction. In addition, they also developed GSVantage, an interactive dashboard that maps spatial staleness patterns, enables synchronized scatterplot comparisons, and allows filtering based on user-selected demographic thresholds. Their case studies reveal statistically significant evidence of “digital redlining”: higher-income, higher-value suburbs tend to have older imagery compared to denser, more pedestrian-oriented areas. Interestingly, this contrasts with some findings from Latin American cities, suggesting contingencies that warrant future research. The modular framework can be extended to other cities with minimal input adjustments; however, researchers outside the U.S. may need to identify analogous census-data APIs as alternatives. Their code and datasets are publicly available on GitHub.
*Zeyu Wang, UW Urban Design & Planning, Yingchao Jian UW Electrical & Computer Engineering, Don MacKenzie, UW Civil & Environmental Engineer, and Jon E. Froehlich, UW Allen School of Computer Science.
This work was supported by NSF grant #2411222. Visokay would also like to acknowledge the resources provided by the International Max Planck Research School for Population, Health and Data Science (IMPRS-PHDS).