Making Policies, Making Lives: A Trajectory of Homelessness

Making Policies, Making Lives: A Trajectory of Homelessness

Policies directed at reducing homelessness must aim, in the end, at changing the course of people's housing histories. This involves permanently housing those who have been persistently homeless, generating stable housing conditions for those sporadically homeless, and eliminating the risky situations of those often teetering on being unhoused.

Surprisingly, however, we don't know much about these housing histories. Evidence suggests that three percent of the U.S. population becomes homeless over a five year period, that three-quarters of those becoming homeless have up to three homeless spells, and that about half of these spells last twelve months or less. Such evidence hints at the temporal structure of homelessness, but abbreviates and collapses time in ways that do not let us see homelessness as people live it-day to day, month to month, year to year. Eliding time in these ways may obstruct our ability to sufficiently understand what causes homelessness and to gauge the impact of policies and programs (or lack thereof).

A group of researchers-Mary Clare Lennon, Li Kuang, and Daniel Herman at the Mailman School of Public Health and Jennifer Hill and I on the Morningside Campus-are trying to unpack the temporal character of homelessness. What does this mean exactly? The basic idea is pretty straightforward: people have different histories of homelessness, different "biographies" when it comes to housing, and our bet is that it would be useful to know what these biographies are. We all have some idea as to what biography usually is-a story of a person's life as it unfolds over time. Our research tries to do something similar, but with two major differences. Biographies commonly purport to tell the entire life story. We don't make such a claim. We tell that part of the biography that-s theoretically useful. And biographies are commonly about a single thing: a person, an idea, an organization, or some other solitary entity. But we want to say something about lots of people. We want, that is, to find groups of people who have common biographies relative to homelessness, who have pretty much the same sequencing, duration, and timing of housing circumstances over some period of time. Sequencing: people who are homeless and then hospitalized have a different history from those who are hospitalized and then become homeless. Timing: people who are homeless at the beginning of a time period have a different history from people homeless at the end of that period. Duration: people homeless for several months have a different history from those homeless for several years. The trick is to group people who are similar on all three characteristics.

How can this been done? The answer to that question is at the heart of a study funded by the National Institute of Mental Health. We are looking at two possible answers: optimal matching analysis and group-based trajectory model analysis. The idea of optimal matching is to find similarly patterned biographies by first comparing each pair of housing histories among homeless people to see how different they are from each other based on differences in the timing, sequencing, and duration of housing circumstances and then grouping together biographies that are most alike. These groups thus comprise a measure of the temporal character of homelessness based on similarities in housing histories.

This result is also the goal of group-based trajectory modeling. But instead of the data-mining approach of optimal matching, group-based modeling assumes an underlying statistical model generated the observed data. This model estimates the form of the relationship linking where people actually lived to a series of time points. Groups result by estimating the shape of these housing trajectories and the probability of individuals belonging to a particular trajectory group. Because these estimates are allowed to vary across groups, it is possible to find more than one group in a population. Thus, with either method, if we know, say, whether people are homeless or housed each night for a period of time, we can group together those who are homeless at the same time, for the same duration, and in the same sequence over that time period. And these groups can be used to evaluate theories or policies.

To suggest the analytic and policy potency of this way of thinking, we present some results from an analysis of a random assignment study of a program designed to help long-term homeless men stay housed as they leave a New York City shelter. For this analysis, we used the group-based trajectory modeling approach. Trajectories in the following figures express the probabilities groups of individuals had of being homeless at least one day each month over the eighteen-month observation period of the study. The first figure shows results for those not in the program and the second for those in the program.

Viewed biographically, the results suggest that a life of perpetual homelessness can be eliminated by the program (the absence in the program figure of subgroup 4 in the controls figure) and that the program made possible a life almost completely free of homelessness for twenty percent more of the sample (the percentage change in subgroup 1 in both figures). We might also speculate more subtly. An important question, for instance, is, When should policies and programs intervene in people's lives? These results suggest a subgroup of people (# 2 in the program figure) that stays housed at the beginning of the time period only to seriously falter-perhaps they need an additional intervention, some kind of "booster shot." And another subgroup (# 3 in the program figure) is arcing towards permanent homelessness but then, somehow, moves toward permanently being housed-perhaps intervention was unnecessary for these people, as for those in subgroup 1. More subtle still: subgroups 2 and 3 for those not in the program have similar trajectories as subgroups 2 and 3 respectively for those in the program. Why? One explanation is that each subgroup is comprised of the same kinds of individuals (e.g., similar ages, ethnicities, prior housing or other trajectories, and so forth). But perhaps individuals in these subgroups are responding to the structure in which they found themselves: both sets of subgroups seem to change at important points in the design and functioning of the program as it operated over the first nine months of the observation period. This suggests, more generally, that using these approaches might enable us not only to better understand policy and intervention effects on individual biographies, but to learn how those biographies express (and construct) the social structures in which our lives unfold.

William McAllister is a Senior Research Fellow at ISERP, where he also directs its Graduate Fellows Program and teaches in its Quantitative Methods in the Social Sciences Program. His research interests include political elites; methods for analyzing biographical information; homelessness and homeless policymaking; and the importance of time and change in explaining political phenomena. For more information, contact him at wm134@columbia.edu.

See Also

• Research Grant: Temporal Dynamics, Intervention Analysis and Policymaking: Latent Trajectory Group Analyses of Service Interventions for Homeless Mentally Ill
• Working Paper: Temporality and Intervention Effects: Latent Trajectory Analysis of a Homeless Mental Health Program
• Working Paper: Policymaking and Caseload Dynamics: Homeless Shelters