Ending homelessness: More housing or fewer shelters?

Over the past decade, a major effort to "end homelessness" has lead to a marked expansion in permanent housing for the homeless relative to shelters. In this paper, I use community-level data over the period 2007–2014 in the United States to estimate short and long run associations between homeless populations and inventories of homeless assistance beds — in shelters and permanent supportive housing. I find that in the first year, one additional permanent housing bed is associated with 0.12 fewer homeless people on the streets and in shelters; however, this negative association is fully muted after the first year. The muting effect is driven entirely by the homeless subpopulation with relatively shorter spells of homelessness. Shelters are not associated with long-run reductions in the unsheltered homeless population, and are thus strongly and positively associated with the sheltered homeless population. Ultimately, sustained reductions in homelessness are strongly associated with eliminating shelters but not with housing the homeless. Aside from providing new evidence regarding homelessness policy, this paper is also the first to use national panel data to explain how within-community variation in homelessness relates to non-policy factors. The homelessness rate is significantly associated with median rent but weakly associated with unemployment and weather.