Three Essays on Homelessness Policy, Federal Housing Assistance, and Rental Market Dynamics in the United States

Abstract

This dissertation examines three dimensions of homelessness and housing policy in the United States: barriers to housing access for voucher holders, the effectiveness of federal homeless-assistance grants, and the measurement of homelessness during crisis-driven data disruptions. The first essay studies the effects of source-of-income protection laws on Housing Choice Voucher household mobility and rental market dynamics. Using staggered adoption across cities between 2013 and 2018 with a staggered difference-in-differences design, I find no detectable improvement in neighborhood quality or spatial dispersion for voucher households, but document a 4.9 percent increase in twenty-fifth percentile rents and a 1.8 percentage point decline in affordable rental stock, with effects concentrated where enforcement is stronger and vacancy is low. The second essay estimates the causal effect of federal Continuum of Care and Emergency Solutions Grant funding on homelessness and shelter capacity across 370 Continuums of Care in 2019. Instrumenting per-capita funding with the pre-1940 housing share from the Community Development Block Grant formula, I find that additional funding increases sheltered counts and expands emergency and transitional bed capacity but produces no detectable short-run reduction in unsheltered homelessness, with impacts varying sharply by geography, household type, and demographic subgroup. The third essay develops a two-phase machine learning framework to impute the 61.6 percent of unsheltered Point-in-Time counts missing in 2021 due to the COVID-19 pandemic. The first phase predicts baseline counts from pre-pandemic relationships; the second models COVID-specific deviations using the subset of communities that did conduct counts, with propensity score weighting to address selection bias. The resulting national estimate of 195,191 unsheltered people restores continuity to the national homelessness dataset and reveals that emergency relief funding dominated the pandemic adjustment. Together, these essays demonstrate that expanding legal protections and scaling federal funding face implementation frictions that limit their reach, and that the administrative data systems underlying policy evaluation remain vulnerable to disruption precisely when accurate measurement matters most.