Association for Public Policy Analysis & Management

Panel: Algorithmic Solutions to Social Services in Cities? Evaluating How Data-Driven Policies Best Enhance Effectiveness and Equity

(Methods and Tools of Analysis)

Much attention has been paid in recent years to algorithmic approaches to policy and practice, including predictive analytics and other systems that automate operations and decisions previously made by humans. There has been much excitement about the potential for such tools to improve the efficiency of services and equity through the objectivity derived from the use of quantitative data.  But, there have also been high profile failures, revealing the vulnerability to simplistic assumptions, bias, and lack of nuance. Of particular concern, such weaknesses can create or exacerbate inequities across populations in unanticipated ways. Thus, a data-driven policy will only be as useful as the quality of the algorithms models that support it.

 

This panel will feature four efforts to construct or evaluate algorithmic systems intended to underlie particular policy programs. These include: an Opportunity Index built by the Boston Area Research Initiative with Boston Public Schools (BPS) that quantifies place-based inequities that can impact academic achievement and is now being used to inform the distribution of funding for extra-curricular programming; an evaluation of the consequences of BPS’ recently-implemented “home-based” school access and assignment plan for equity, and how it has and has not been able to overcome the geographies of racial and economic disparity; a predictive analytic system developed by Allegheny County’s Human and Health Services that uses individual-level characteristics to identify calls to child protective services that are likely to portend more serious incidents; and an effort by Chapin Hall at the University of Chicago that uses historical data to strategically place new apprenticeship programs in Chicago that fill existing gaps in services.

 

A main theme of the panel is the need for attention to both local context and theory when developing algorithms to inform or drive a given policy, and each of the talks will highlight major decision points that had the potential to both increase equity or unintentionally reduce equity. In some cases, these critical junctures were drawn from lessons learned by previous work—for example, how the correlation between race, poverty, and outcomes can accidentally lead to biased predictions—and in others from newly discovered interactions between the policy and existing behavioral or institutional dynamics. In sum, the panel presents the technical opportunities for leveraging data to innovate on and improve policy and practice, while also offering lessons about the emerging best-practices when developing algorithms for such purposes.