Aggregate-Level Inferences from Individual-Level Data: The Case of Permanent Supportive Housing and Housing First

Brendan Andrew O'Flaherty
I estimate the "simple mechanical effect" of permanent supportive housing and Housing First as studied in the At Home/Chez Soi and HUD-VASH experiments on point-in-time counts of homelessness (HUD definition). The simple mechanical effect is the effect that would occur in the absence of any behavioral responses aside from those in the experiments. The estimates of the simple mechanical effects overlap the confidence intervals in Corinth's (2017) regression study of the total effect. This
more » ... suggests that the net effect of behavioral responses outside the experiments is small. The essay illustrates how useful inferences about aggregate-level phenomena can be derived from individual-level data. I am grateful to Emmy Tiderington and Yi-Ping Tseng for helpful comments and information. Strategies that work on an individual level do not always work the same way on an aggregate level. Running your air conditioner on high all summer may be a good way to keep yourself cool in the era of climate change, but it will not keep the planet cool (given today's electricity generating technology). Stockpiling groceries and toilet paper at the start of a pandemic may be a wise precaution for a household to take, but if everyone runs to the supermarket to stockpile, the pandemic may spread faster and the rise in demand may create a shortage that would not otherwise have existed; stockpiling may create and exacerbate the problems it was trying to avoid. Similarly, interventions that produce promising results on an individual basis do not translate automatically into policies that reproduce the same results on an aggregate level. Translation is difficult and uncertain. This essay is about an example of such translation. It shows how translation can be done, what its uncertainties are, and how it can be valuable. The effects of permanent supportive housing in general and (Pathways) Housing First 1 in particular are areas where the implications of individual level patterns on aggregate level outcomes have not always been clear, and have sometimes been stated incorrectly. Several high-quality randomized controlled trials (RCTs) have shown that permanent supportive housing and Housing First cause improvements in many housing outcomes for their participants (see reviews in, for insance, Kertesz and Johnson 2017 and Miler et al. 2021), but these are individual-level results. Aggregate-level results would require RCTs that randomly assigned permanent supportive housing (PSH) and Housing First (HF) beds to different cities, or natural experiments where such beds were 1 In this essay when I say "Housing First" or "HF," I mean Pathways Housing First.
doi:10.7916/658n-r008 fatcat:tzdnzo6aa5etvpze45wbowo5ui