Date of Award
5-5-2022
Additional Affiliations
Division of Computational & Data Sciences
Degree Name
Doctor of Philosophy (PhD)
Degree Type
Dissertation
Abstract
Artificial intelligence, machine learning, and algorithmic techniques in general, provide two crucial abilities with the potential to improve decision-making in the context of allocation of scarce societal resources. They can flexibly and accurately model treatment response at the individual level, potentially allowing us to better match available resources to individuals. In addition, they can reason simultaneously about the effects of matching sets of scarce resources to populations of individuals. This thesis leverages these abilities to study algorithmic allocation of scarce societal resources in the context of homelessness. In communities throughout the United States, there is constant demand for an array of homeless services intended to address different levels of need. Allocations of housing services must match households to appropriate services that continuously fluctuate in availability, while inefficiencies in allocation could ``waste'' scarce resources as households will remain in-need and reenter the homeless system, increasing the overall demand for homeless services. This complex allocation problem introduces novel technical and ethical challenges. First, using administrative data from a regional homeless system, we formulate the problem of ``optimal'' allocation of resources given data on households with need for homeless services. The optimization problem aims to allocate available resources such that predicted probabilities of household reentry are minimized. The key element of this work is its use of a counterfactual prediction approach that predicts household probabilities of reentry into homeless services if assigned to each service. To address the inherent fairness considerations present in any context where there are insufficient resources to meet demand, we discuss the efficiency, equity, and fairness issues that arise in our work and consider potential implications for homeless policies. Next, we turn our focus to interpretability and ease of use for homeless caseworkers by using counterfactual predictions to develop decision rules for allocation of resources to homeless households. We use calculated treatment effects of homeless services to develop simple allocations that reduce rate of reentry into the homeless system. We compare these to the original allocation on reentry rate, estimated financial cost, and potential biases in group fairness. Finally, we examine justice in data-aided decisions in the context of a scarce social resource allocation problem. We empirically elicit decision-maker preferences for whether to prioritize more vulnerable households versus households who would best take advantage of more intensive interventions. We find that, for a subset of about one-third of decision-makers, these preferences change from vulnerability-oriented to outcome-oriented when they are previously exposed to outcome predictions in a different context. In a separate task, when decision-makers assign homeless households to scarce services based on a random presentation of household descriptions with or without algorithmically derived risk predictions, we find that risk predictions reinforce decision-maker preferences. Among those who prioritize the most vulnerable, presenting the risk predictions in addition to the household descriptions leads to a significant increase in allocations to the more vulnerable household, whereas among those who prioritize households who could best take advantage of more intensive interventions, presenting the risk predictions leads to a significant decrease in allocations to the more vulnerable household. These findings emphasize the importance of explicitly aligning data-driven decision aids with the allocation goals – an essential element of social policies that frequently determine whom to serve with scarce resources.
Language
English (en)
Chair and Committee
Patrick Fowler, Chien-Ju Ho
Committee Members
Sanmay Das, Douglas Luke, William Yeoh
Recommended Citation
Kube, Amanda Rose, "Data-Driven Decision-Making: Using Counterfactual Predictions to Allocate Scarce Homeless Services Fairly and Efficiently" (2022). Arts & Sciences Electronic Theses and Dissertations. 3262.
https://openscholarship.wustl.edu/art_sci_etds/3262