Who Pays the RENT? Implications of Spatial Inequality for Prediction-Based Allocation Policies

Targeting high-risk individuals can be more effective than neighborhood-wide approaches even in segregated cities: The study found that individually targeted policies reached significantly more high-risk households compared to neighborhood-based approaches, even in highly segregated metropolitan areas where eviction risk is concentrated. This challenges some prior research suggesting targeting is ineffective in unequal settings.

Where inequality is concentrated matters more than how much inequality exists: The paper introduces the RENT metric to measure how spatial distribution of risk affects policy effectiveness. The key insight is that the pattern of where high-risk households are located—not just the overall level of inequality—determines whether targeted versus broad-based approaches work better for resource allocation.

Previous conflicting research results can be explained by two overlooked factors: The apparent disagreement in prior studies about targeting effectiveness can be reconciled by considering (1) who bears the costs of deployment and (2) whether researchers are measuring actual observed risk concentration versus modeled predictions. This suggests the field needs more careful attention to these methodological details.

Real-world eviction data shows targeted outreach policies have practical promise: Using actual eviction court records from a U.S. city, the researchers demonstrated that AI-powered targeting could meaningfully improve which households receive assistance, suggesting these approaches may have genuine value for social service delivery when properly designed.

One-size-fits-all AI deployment recommendations are insufficient: The findings indicate that responsible AI allocation policies must be tailored to specific applications and local geographic contexts rather than applying universal rules about whether targeting or broad approaches work better.