background image
factors associated with longer stays and repeat use of shelter (age, mental health,
substance abuse, and sometimes medical problems) but did not discuss how effi-
ciently these high consumers can be identified, which is crucial to the practical
application of such data.
The Problem of Effectiveness
After selecting people at risk for homelessness, based on a more or less sophis-
ticated model, one must then determine what interventions will most readily pre-
vent homelessness and at what cost. The best design for evaluating a prevention
program is to randomly assign some proportion of people who meet some risk cri-
teria to receive the specialized program. People who did not receive specialized
services would remain free to use other services. Both groups would need to be fol-
lowed for some reasonably long period of time (years rather than months) to deter-
mine meaningfully the number of cases or months of homelessness prevented.
Remarkably few studies of prevention programs have used anything approximat-
ing this design. Many programs have no comparison group, much less one that is
randomly assigned, and authors make implausible assumptions about the numbers
of people who would have become homeless in the absence of intervention (typi-
cally 100%). Studies frequently have little or no follow-up to determine whether
homelessness was prevented, merely postponed, or not affected at all, and often
presume success rates of 100% for those who received services. Cost-benefit anal-
yses derived from such studies present an illusion of specificity. Different and
more plausible assumptions lead to conclusions markedly at odds with those
The Problem of Queue Jumping
Some observers have likened homelessness to a game of musical chairs in
which the players are poor people and the chairs are the housing units they can
afford (McChesney, 1990; Sclar, 1990), or in a slightly more sophisticated anal-
ogy, the chairs represent the housing poor people can purchase or otherwise
occupy by drawing on their personal networks (Koegel, Burnam, & Baumohl,
1996). Where there are more poor people than affordable housing units and where
personal networks are attenuated or materially impoverished, some will be left
homeless when the music stops. Although individual characteristics may deter-
mine who becomes homeless, it is resources relative to needs that determine over-
all prevalence rates (Koegel et al., 1996; S. Schwartz & Carpenter, 1999; Wright &
Rubin, 1991). Thus, although homelessness can be prevented by creating resources
or reallocating them from those who are not at risk to those who are, reallocation
among groups at similar levels of risk is unlikely to affect overall prevalence rates.
Shinn, Baumohl, and Hopper