Natural disasters disproportionately impact disadvantaged populations, exacerbating pre-existing social inequalities. While there's growing recognition of this issue, quantitative understanding of how these inequalities manifest in large-scale emergency evacuations remains limited. Past research relying on surveys has faced challenges like sample bias and recall issues. This study leverages high-resolution, anonymized GPS data from the Greater Houston Area during Hurricane Harvey to quantitatively analyze evacuation patterns across different neighborhoods categorized by race and wealth. Hurricane Harvey, with its extensive flooding and lack of mandatory evacuation orders in Houston, provided a unique opportunity to observe emergent evacuation behavior driven largely by individual decisions. The study aims to understand how socioeconomic factors influence who evacuates, where they go, when they leave and return, ultimately providing insights for improved emergency management and equitable disaster planning.

Existing research highlights the link between social vulnerability and disaster impacts, showing minority and low-income populations experience disproportionate effects (e.g., Hurricane Katrina). Studies suggest race and class disparities exist throughout all disaster phases, from preparation to recovery. While some research indicates white communities are better prepared, and their residents return at higher rates, large-scale quantitative research on evacuation and return patterns is scarce. Previous studies often relied on retrospective surveys with inherent limitations. Recent advancements in GPS technology and the availability of large-scale mobility data allow for more nuanced and quantitative analyses of human behavior during disasters, including studies on evacuation patterns, probabilistic models, and the role of social networks and socioeconomic factors. However, comprehensive analyses integrating high-resolution mobility data with detailed socioeconomic information are still lacking.

The study utilized over 30 million anonymized GPS records from approximately 150,000 opted-in users in the Greater Houston Area from July 1st to September 30th, 2017. The data, obtained from Cuebiq, were pre-processed to extract meaningful stay points using a stay-point algorithm, filtering out transient locations. To minimize bias from tourists, only users with data from at least 60 unique days and 100 stay points were retained. To address sampling bias, two strategies were employed: (i) a weighting procedure based on the ratio of reporting devices to the population in each census block group (the smallest unit of reporting in the US Census), and (ii) bootstrapping samples from each block group with uniform sampling rates. Home locations were determined by identifying weekly primary locations during weekday evenings. Evacuation detection involved a sliding time window (5 and 7 days) to identify departures of at least three consecutive days from the determined home location, with a 1km distance threshold for evacuation. Net evacuation intensity was estimated using non-parametric kernel density estimation (KDE) to compare pre- and post-disaster densities of home locations. Census block groups were classified into six neighborhood types based on wealth level (below 25% federal poverty level) and majority race: non-poor Black, non-poor Hispanic, non-poor White, poor Black, poor Hispanic, and poor White. Sensitivity analyses were conducted with varying thresholds. The study analyzed who evacuated, where they went, and how long they were gone, relating these patterns to the socioeconomic characteristics of their neighborhoods.

The study found that approximately 6.7% of the population evacuated during Hurricane Harvey, aligning with official reports. However, this evacuation rate varied substantially across neighborhoods. Non-poor White neighborhoods had significantly higher evacuation rates than expected based on their population proportion (19.8% overrepresentation), while poor Hispanic neighborhoods showed the lowest evacuation tendency (12.2% underrepresentation). This disparity held for both long-distance (>41.25 km) and short-distance (<2.71 km) evacuations. There was remarkable socioeconomic homophily in evacuation destinations, with people tending to evacuate to neighborhoods similar to their origin in terms of race and wealth. For example, residents from White neighborhoods had an 88.1% probability of evacuating to similar neighborhoods, compared to 56.8% for Hispanic and 16.7% for Black communities. While most departures occurred within the first week after landfall, return times followed a right-skewed distribution, with some individuals taking much longer to return. Longer evacuation durations (over 30 days) were disproportionately associated with wealthier neighborhoods. Disparities in evacuation and return timing were also observed, with wealthier and white residents overrepresented among early evacuees and early returnees. This suggests that income level strongly influenced evacuation decision-making and recovery times.

The findings demonstrate that despite the seemingly universal pattern of evacuation distance distribution, significant socioeconomic disparities exist in disaster response. The observed disparities cannot be attributed solely to geographical proximity, highlighting the profound impact of race and wealth on evacuation behavior and resilience. Limited access to transportation, unequal regional development, differences in preparedness, and perceived risk likely contribute to these disparities. The lack of a mandatory evacuation order heightened the role of financial resources and social networks in evacuation decisions. The longer displacement durations observed in wealthier communities suggest the substantial financial resources required for extended stays away from home, contrasting with the constraints faced by lower-income populations. The disparities underscore the need for targeted interventions to address social inequalities in disaster preparedness and response.

This study provides compelling evidence of significant race and wealth disparities in Hurricane Harvey evacuation patterns using high-resolution mobility data. The findings highlight the limitations of assuming homogeneous responses to disasters and emphasize the crucial role of socioeconomic factors. Future research should incorporate additional data on factors like home damage, infrastructure conditions, and social ties to develop a more comprehensive understanding of the complex interactions between race, wealth, and disaster outcomes. These findings suggest the need for policies and interventions to address these inequalities, promoting equity and resilience in disaster preparedness and recovery.

The study's limitations include the focus on a single three-month period in the Houston MSA, potentially missing larger displacements or long-term impacts. Sampling bias related to smart device penetration rates remains a challenge, although efforts were made to mitigate it. The analysis didn't fully control for other factors like home damage and social ties that may influence evacuation behavior, although a regression model incorporating data on flooded roads was tested and found not to be significantly predictive. Future studies incorporating more comprehensive data on these factors could further enhance understanding of the complex interplay between race, wealth, and disaster outcomes.