background image
The Problems of Locality and Time
Any model for targeting those at risk of homelessness will be based on local
data that may not apply elsewhere. For example, both the percentage of homeless
families who have been evicted and the percentage of families who are evicted but
never become homeless vary by location. Weitzman, Knickman, and Shinn (1990)
found that 22% of first-time shelter users in New York city had been evicted, com-
pared to 6% of the public assistance caseload. Bassuk et al. (1997) found that 26%
of homeless families in Worcester--and 17% of housed poor families--had been
evicted or locked out, suggesting that a prevention program based on evictions
would have even more false alarms for every case of homelessness prevented than
such a program in New York. In other cities, reported percentages of homeless
families who had been evicted ranged from 14% to 57%, with the high figures
sometimes including other housing problems (Bueno, Parton, Ramirez, & Vieder-
man, 1989, pp. 89).
Even where local contingencies can be taken into account, data must be con-
tinuously renewed because, unlike the cases of PKU or polio, the correlates of
homelessness shift over time. The phenomenon itself changes (homelessness
today is not like the mass dispossession of the Great Depression or the more ambig-
uous homelessness of postwar skid rows). Routes to shelter also change (very few
arrivals on skid row came from psychiatric hospitals), thus reconfiguring the popu-
lations found there (Hopper & Baumohl, 1994, 1996). Any predictive model, then,
is in jeopardy of becoming rapidly outdated and progressively inefficient. Most of
what we know about correlates of homelessness today comes from studies con-
ducted a decade ago, when economic conditions, for instance, were very different.
Today's knowledge may not apply tomorrow when, for example, a smaller fraction
of the poor is eligible for welfare support. Exit predictors, too, need to be
contextualized. Even if one can specify "heavy users" of shelters, for example,
using "months in shelter" as an outcome variable is problematic because of local
choices that channel scarce resources and bias likelihood of exit in favor of certain
groups. In Philadelphia, people with severe mental disorders were found to exit
more quickly from shelters, probably because those with serious disorders were
eligible for specialized services (Culhane & Kuhn, 1998). As noted above, for
families in New York city, months in shelter predicted subsequent stability in
housing, because a long shelter stay signified movement to the top of the queue for
subsidized housing (Shinn et al., 1998). These and other apparent anomalies rein-
force a more general point: Homelessness is a dynamic phenomenon, chased but
never really captured by research.
A Review of Prevention Programs
Most actual programs to prevent homelessness are indicated programs using
simple targeting mechanisms. Because of the inefficiencies of such programs and
Shinn, Baumohl, and Hopper