Course of Homelessness

Paul Koegel. Homelessness Handbook. Editor: David Levinson & Marcy Ross. Berkshire Publishing, 2007.

Many of the early “first-generation” studies on homelessness—cross-sectional studies seeking a better descriptive sense of who the contemporary homeless are, what their needs are, and whether those needs are being met— collected data on history of homelessness but did so only with regard to current episodes of homelessness. Homeless respondents, in other words, were simply asked how long they had been homeless. The implicit assumption seemed to be that homelessness is something into which individuals fall and remain.

As researchers learned to ask more-pointed questions about homelessness history, however, they realized that homelessness is not a chronic condition for the majority of those referred to as “homeless.” Rather, it is a dynamic state that individuals enter, exit, and then often reenter repeatedly over time. This finding prompted the realization that in addition to understanding why people become homeless in the first place, we must understand what is now commonly referred to as the “course” of homelessness—the process by which homeless people exit and reenter homelessness over time.

In response, a “second-generation” set of prospective, or longitudinal, studies took on the considerable challenge of tracking samples of homeless individuals in order to better understand movement out of, and back into, homelessness. Some of these studies focused on community-based probability samples of homeless adults. Others relied on comprehensive administrative databases that documented stays in municipal shelter systems. Still others addressed this issue in the process of evaluating interventions designed to end homelessness.

As a result, a growing literature exists on the extent to which people move into and out of homelessness, the factors that determine who eventually exits from homelessness, and the factors that distinguish successful from unsuccessful exits.

Cross-Sectional Studies: Early Evidence

Most of the studies of homelessness that began emerging in the mid-1980s were cross-sectional in nature. These studies drew samples of homeless individuals at one point in time and interviewed them to determine their demographic and diagnostic characteristics and their life circumstances. With regard to homelessness itself, questions tended to focus on the amount of time individuals reported being homeless, usually in simplistic ways that obscured as much as they revealed. In some cases questions were posed only in terms of an individual’s current episode of homelessness, ignoring the possibility that there may have been past episodes. In other cases the question of duration was posed without reference to timeframe, leaving it unclear whether the answer pertained to lifetime homelessness or current homelessness. More often than not, the question of how long the individual had been homeless was asked without precisely defining for the individual what an entry into homelessness, or an exit from homelessness, actually was. Given this confusion, people often had difficulty interpreting the results.

With time more precisely worded questions —and more precise data—emerged. These data suggested that recurrent homelessness is in fact a common experience among cross-sectional samples of homeless individuals. For example, in the Los Angeles Skid Row Study, conducted by Paul Koegel, Audrey Burnam, and Rodger Farr between 1984 and 1986, an extensive set of questions on homelessness history was asked. It yielded information not only on the current episode of homelessness but also on the extent to which people had moved into and out of homelessness since their first entry into homelessness. In this study respondents were asked when they first became homeless, “homelessness” being defined as not regularly sleeping in a room, apartment, or house of their own but instead sleeping in shelters, the streets, or other places not meant for sleeping. They were then asked whether, since that time, they had ever lived in their own room, apartment, or house for a month or more and whether, since their first episode of homelessness, they had ever lived with family or friends for a month or more. Those individuals who responded affirmatively were then asked when they had last done so.

The answers to these questions made it possible to determine not only the precisely defined length of a respondent’s current episode of homelessness but also how many times a respondent had experienced homelessness since first becoming homeless. The results were surprising. The vast majority—two-thirds—had experienced multiple episodes of homelessness during the course of their adult lives. Indeed, slightly more than one-quarter (26 percent) had experienced six or more such episodes. Clearly, homelessness was not a one-time phenomenon for most of these people.

By crossing the number of times respondents had been homeless by the length of respondents’ current episode, it was also possible to use these data to begin understanding patterns of homelessness. Doing so made it clear that “newly” homeless individuals—people who were experiencing a relatively short-lived first episode of homelessness—were a distinct minority: Only 13 percent of the sample were in a first episode of homelessness lasting less than six months, a figure that rose to only 18 percent if the definition was stretched to include those in a first episode lasting less than a year. By the same token, those who had fallen into homelessness once and remained there for long periods of time were a small minority as well: Only 10 percent were experiencing first episodes of homelessness lasting more than two years, a figure that climbed to 15 percent if the definition was stretched to first episode of homelessness lasting more than one year. Again, multiple episodes were the rule. Homelessness, it was becoming clear, is not so much an event as a process.

The Los Angeles Skid Row Study, of course, focused on a restricted geographic area in which almost the entire local homeless population consisted of unattached homeless men, raising questions about generalizability. The patterns revealed by this study, however, were confirmed by findings from other locales as well, although their intensity proved to be more closely associated with the single adults who dominated the Los Angeles Skid Row sample than adults with children in their care. The 1996 National Survey of Homeless Assistance Providers and Clients (NSHAPC), conducted by Martha Burt and colleagues of the Urban Institute, provides the clearest evidence of this. The NSHAPC study drew a nationally representative sample of homeless people from a broad range of homeless assistance programs, including not only shelters, vouchers, and housing programs, but also meal, mental health, substance abuse, medical, outreach, and drop-in programs. In this sample single men were still the modal (largest) group (61 percent), but single women (15 percent), women with their own children in their care (12 percent), men with children in their care (2 percent), and a remaining group of men and women attached to others in a myriad of ways (10 percent) were also represented.

In findings that were remarkably similar to those for Los Angeles, only 13 percent of the single men and 15 percent of the single women in this national sample were in a first episode of homelessness that had lasted six months or less. The experience of people who were homeless with their children was quite different, however. Fully 50 percent of the adult men in homeless families and 30 percent of the women in homeless families were in a first episode of homelessness lasting six months or less, providing a vivid reminder of the fact that the experience of homeless families is often different. Even so, approximately half of the singles and adults in homeless families alike had experienced multiple homeless episodes during the course of their lives, suggesting again that homelessness is a dynamic phenomenon and that understanding transitions between being homeless and being housed is critically important.

Because the studies from which these understandings were emerging relied on point-in-time samples, they likely overrepresented people with long-term and multiple episodes of homelessness relative to those who experience brief episodes of homelessness and never become homeless again. We know that those who are homeless at one point in time are only a fraction of those who experience homelessness during longer periods of time. Bruce Link and colleagues, for instance, determined that 3.1 percent of a random sample of households across the nation with telephones—a total of 5.7 million people—had experienced homelessness (living in a shelter or in public spaces) during the previous five years, a number that is significantly higher than the (one-week) national estimate of 508,300 reported by Martha Burt and Barbara Cohen in 1987. Similarly, Dennis Culhane and colleagues found that although the estimated size of the homeless populations in New York and Philadelphia on a given night was between 0.2 and 0.3 percent of the city’s population, the number of different shelter users during a single year was approximately 1 percent of each city’s population, whereas the three-to-five-year rates increased to 2.8-3.2 percent. This high turnover in the homeless population over time suggested that a significant number of people exit homelessness and do not return. Even so, the recurrent nature of homelessness among a significant segment of the homeless population remained a startling fact that warranted further attention.

Cross-sectional studies, then, pointed to an issue that needed exploring. They were also the basis for the first attempts to systematically address the factors that were affecting course of homelessness. These attempts involved comparing currently homeless individuals with similar groups who were not currently homeless on relevant characteristics. For instance, Michael Sosin and his colleagues utilized a cross-sectional design to identify predictors of homeless transitions by retrospectively examining the experience of 535 randomly sampled persons in Chicago who were using free meal services. Approximately one-third (34 percent) had never been homeless, 34 percent were currently homeless, and 32 percent had previously been homeless but were currently housed. Domiciled, never-homeless individuals were compared to the first-time homeless to consider determinants of first entry into homelessness, and domiciled, previously homeless individuals were compared to repeatedly homeless individuals to understand factors associated with returns to homelessness. Comparisons between these groups suggested that variables related to social institutional factors—that is, the “safety-net” resources that potentially protect people from homelessness, such as living with others or receiving public assistance—were more closely associated with homeless transitions than were disabilities or occupational deficiencies.

Although illuminating, studies of this nature were limited in their explanatory power because cross-sectional studies, by definition, can establish only associations between variables; they cannot disentangle predictors from consequences or cause from effect. To achieve these kinds of understandings, longitudinal designs were necessary. Longitudinal studies, needless to say, are far more challenging than cross-sectional studies. The challenges faced by homelessness researchers interested in exploring homelessness over time were even greater than is usually the case given the unique circumstances in which homeless people lead their lives.

Challenges of Conducting Longitudinal Research

Of the many challenges confronted by those researchers interested in course of homelessness, three stand out as particularly noteworthy: (1) arriving at the optimal design; (2) appropriately defining the phenomena at hand; and (3) successfully tracking homeless people over time.

The optimal design for a study of course of homelessness would begin with a sample of individuals who are “at risk” for but haven’t ever experienced homelessness and follow the individuals for a long time to determine who becomes homeless and what happens to them after they do. The idea would be to follow people long enough that the universe of patterns of homelessness had time to emerge. Such a design would make it possible to establish rates of and predictors of first entry into homelessness, suggesting critical points of early intervention. Even further, assuming the baseline sample was large enough and people were followed long enough, it would ultimately produce unbiased rates and predictors of the different ways in which homelessness unfolds over time.

Unfortunately, such a design was not a viable option for researchers who, in the late 1980s, were contemplating how to study course of homelessness longitudinally—and probably isn’t today. To begin with, too little was known about the prehomeless backgrounds and communities of homeless individuals to design an efficient yet representative sample of those “at risk” for homelessness. Moreover, even if such information were available, homelessness is probably a rare enough event that baseline samples would have to be extremely large and follow-up periods extremely long for researchers to have enough first entries into homelessness to support an analysis of determinants. The sample would have to be even larger and followed for even longer periods of time to analyze transitions out of and back into homelessness. As a result, studies that began with a baseline sample of at-risk individuals, even if possible to conduct, would be both prohibitively expensive and slow to produce findings capable of addressing what was an immediate policy need.

Recognizing these problems, researchers interested in course of homelessness instead chose as their starting point samples of individuals who were already experiencing homelessness. They knew that such a design would not allow examination of first entry into homelessness but reasoned that following samples of homeless individuals prospectively would at least allow researchers to identify predictors of exits from and reentries into homelessness and to understand what distinguishes successful from unsuccessful exits. In some cases (for example, in studies of homelessness transitions in Minneapolis, Minnesota, and Alameda County, California), researchers attempted to focus exclusively on samples of newly homeless individuals in order to eliminate the bias introduced by what is referred to as “left-hand censoring.” This refers to the fact that when you draw a sample of homeless adults, you capture the adults while they are in the midst of homeless episodes of varying durations, so that they are not all beginning at an equivalent “starting gate.” Such studies ultimately discovered that identifying newly homeless individuals within a short enough time period was difficult, and researchers ended up supplementing their “recent arrivals” with a cross-section of homeless adults. Other studies, such as the Course of Homelessness Study in Los Angeles, intentionally focused not only on newly homeless individuals but also on individuals at later stages of their homeless careers in an attempt to estimate what might be found if newly homeless individuals were followed for longer periods of time.

Ultimately, then, community-based studies that were explicitly designed to examine course of homelessness relied on designs that involved following baseline samples of currently homeless adults over time. Admittedly, this alternative sacrificed information on first entry into homelessness and introduced biases associated with using cross-sectional comparisons of “slices” of the homeless population (e.g., the new versus the experienced homeless) to inform individual changes over time. Nevertheless, the longitudinal nature of these “second-generation” designs represented an important advance over the “first-generation” cross-sectional studies that were the norm during the first half of the 1980s. This was equally true of the longitudinal designs, typically associated with Dennis Culhane, that relied exclusively on the analysis of large municipal shelter system administrative databases to understand the movement of homeless people into and out of shelter and of the service demonstrations and evaluations that followed homeless people over time to understand the impact of innovative interventions.

A second challenge faced by researchers interested in course of homelessness prospectively in community-based samples was how to define exits from homelessness. Defining homelessness itself was a straightforward matter. Virtually all researchers agreed that a person is homeless if he or she spends at least one night in a temporary dwelling designated for homeless individuals or in any number of places not meant for sleeping (although there was often disagreement on how to handle doubling up with family and friends and long-term but temporary stays in housing programs). Defining exits was more complicated. For instance, should an individual be counted as exiting homelessness if he or she spends a week in a hotel? Does it matter whether he pays for the hotel or whether it is paid for with a voucher received by a community-based organization? What about stays in board-and-care facilities? What about jails or hospitals? What about a long stay in the home of family or friends? Does it matter whether the individual is contributing to the rent?

With time it became clear that researchers had to consider two dimensions in defining homelessness exits. The first of these dimensions pertains to the type of place to which a person is exiting. Some places unequivocally connote an exit from homelessness, such as when someone who is homeless obtains a room, apartment, or house that he or she owns or pays rent for. These exits have been referred to as “independent” exits. Other exits clearly involve changes in homelessness status that are less closely associated with what we think of as a “home,” such as doubling up with family or friends, living in voucher-paid housing, living permanently in lodging facilities for which pay is not expected, or staying in institutions such as hospitals or jails. These exits have been referred to as “dependent” exits.

The second relevant dimension is duration—the length of time someone has to remain in an exit category before researchers can say that an exit has taken place. The first longitudinal study of homelessness, conducted in Minneapolis by Michael Sosin, Irving Piliavin, and Herb Westerfelt, set a minimum threshold of fourteen consecutive days. This was rejected by most researchers as too lenient, especially given that during a given month, many homeless people—especially those who receive some kind of public assistance—find housing for as long as two or three weeks at a time but regularly find themselves homeless when their monthly income runs out before the months does. Researchers eventually arrived at thirty consecutive days in housing as the most common threshold, although at least one study, seeking to be sensitive to the idea that even thirty-day exits may be too shallow to be real, also looked at a ninety-day threshold.

Crossing these two dimensions—type and duration—yields a wide variety of homelessness exit definitions. For instance, one can define an exit from homelessness as thirty consecutive days in independent exit settings, or thirty consecutive days in dependent exit settings, or thirty consecutive days in either of the preceding. Similarly, one can look at the same categories but defined on the basis of sixty or ninety, rather than thirty, consecutive days. Each of these definitions is defensible depending on what the researcher is trying to explain. In practice, most community-based studies of course of homelessness have selected thirty-day independent exits as their definition of exit from homelessness, although substantial variation exists and must be attended to in interpreting results. Similarly, longitudinal studies relying on municipal shelter system databases have used departure from the system for thirty days as their primary exit criterion. Alternatively, evaluations of innovative services aimed at ending homelessness have often relied on a continuous measure of days homeless after the intervention as a measure of program success.

Researchers who follow people over time traditionally rely on the anchors that tie people to a particular place—their addresses, phone numbers, and workplaces—and, even so, struggle mightily to track people successfully. Homeless people, almost by definition, lack these anchors and thus present a challenge that goes well beyond the traditional one. Indeed, the extent of this challenge was reflected in the substantial rates of attrition that characterized early attempts to follow homeless people. More than 40 percent of the baseline sample was lost to follow-up in the Minneapolis study during a six-month period, for instance. Similar proportions were lost to follow-up in early longitudinal studies of homeless people in St. Louis and Baltimore as well. Such levels of attrition are generally viewed as unacceptable because of the potential bias they introduce. This bias stems from the possibility that those lost to follow-up may differ from those who have been retained in ways that haven’t been measured and accounted for.

More recent longitudinal studies of homeless samples have achieved significantly higher retention rates. For instance, in the Alameda County and Course of Homelessness studies, 85 percent and 87 percent of those interviewed at baseline were successfully recontacted at least once. Likewise, in a longitudinal study of homeless persons with alcohol and other drug problems conducted in New Orleans, at least one follow-up interview was completed with 93 percent of the sample during a twelve-month period.

These improvements are the outcome of a series of innovative techniques and strategies that was developed simultaneously and independently by a number of research teams. Basically, researchers learned that to successfully recontact homeless individuals, researchers had to (1) collect the right information from respondents at baseline; (2) make sure that respondents are given meaningful incentives to remain in contact and that doing so is as easy for them as possible; and (3) be ready to use a myriad of tracking strategies to find people if they fail to keep in touch. Each of these issues is worth talking about in greater detail.

The first steps toward achieving successful follow-up occur at the end of the baseline interview. Successful researchers have learned to collect detailed recontact information. This means learning as much as possible about the daily, weekly, and monthly cycles of individuals in their samples so that researchers can determine the people, places, and institutions with which respondents periodically interact that might know about their whereabouts. This also means recording as much contact information on these collaterals as possible—and confirming that this information is accurate. This also means learning about aliases and nicknames, the multiple Social Security numbers under which individuals might be known, and their distinguishing physical characteristics. Successful researchers have taken photographs of subjects (with their permission) and have made sure that subjects sign consent-to-be-tracked forms that can be used later to reassure collaterals who are asked for help in finding the subjects.

Researchers also learned that they must make it as easy as possible for subjects in longitudinal samples themselves to initiate contact at the appropriate time so that tracking resources can be channeled toward those not able or willing to take on this responsibility. Along these lines, recontact or calendar cards are provided to research subjects to remind them when they are supposed to be back in touch. Cash and other incentives, which are generally available at each follow-up interview, are increased for those subjects who initiate contact at the appropriate time so that subject initiative is rewarded. Incentives are similarly available for those who touch base between interviews to update contact information. Researchers set up field offices and/or toll-free, twenty-four-hour phone lines so that subjects can do this as easily as possible. Research participants are recontacted immediately after baseline to solidify the relationship and to enhance the rapport between interviewer and participants. Indeed, rapport is often nurtured as another incentive that motivates participants to continue staying in touch with researchers.

Despite these efforts, a substantial number of participants will not make contact at the time of their follow-up interview and will have to be more aggressively tracked. Successful studies have used a combination of phone tracking, mail follow-up, agency and systems-level tracking, and field-level tracking, usually pursuing each of these strategies simultaneously. Systems-level tracking, for instance, may involve searching shelter records, calling jail and prison information numbers, obtaining state-level Criminal Information and Investigation Reports, having the Social Security Administration forward mail, checking Department of Motor Vehicles Records, linking into the Veterans’ Administration record system, and using other agency-related strategies. Field-level tracking may mean making frequent visits to local places where a missing person is known to hang out, sitting in the local welfare or assistance office and listening to names being announced, staking out local check-cashing businesses at the beginning of the month, or, as reported by Wright and his colleagues, posting flyers “which are printed on obnoxiously colorful paper so as to be impossible to miss” (Wright, Allen, and Devine 1995, 273). Effective organization of interview staff can be critical as well. As part of the Course of Homelessness Study, for instance, interviewers were regularly given time to peruse a “lost list” and the study’s “Family Album,” which included photographs of all participants. Interviewers were organized into teams who were given collective responsibility for missing individuals. A “Most Wanted” incentive program was initiated to provide rewards to interviewers who were successful in finding hard-to-reach people. Those interviewers who exhibited a flair for successful sleuthing were designated “Recontact Experts” and assigned to follow-up tracking full-time.

Studies that have followed homeless people over time tend to endorse what Wright and colleagues refer to as the “90-10 rule”: It takes 10 percent of the time to accomplish 90 percent of the work, and the remaining 90 percent of the time to accomplish the remaining 10 percent of the work. Although that rule may often be more of an 80-20 rule, or even a 70-30 rule, the point is an important one. Easily implemented strategies will account for the vast majority of recontact success. However, the difference between marginally acceptable retention rates and exemplary retention rates will likely be related to a study’s willingness to engage in any number of additional labor-intensive efforts. Each of these may yield only a smattering of successful recontacts. Experience has proven, however, that collectively they make a critical difference.

What We Know from Longitudinal Studies

As indicated earlier, three types of longitudinal studies inform issues related to course of homelessness. These are (1) community-based panel studies that have followed and reinterviewed baseline samples of homeless people over time; (2) studies that rely entirely on large municipal shelter system administrative databases to trace the movement of people into and out of shelters; and (3) experimental evaluations of innovative service programs for homeless adults.

Community-based panel studies are probably our best source of information on course of homelessness because their reliance upon probability samples of community-based homeless adults ensures broad representation and because they have focused explicitly on collecting data that can inform homeless/domicile transitions. These panel studies generally tell a similar story—a story that carries both good and bad news. The good news is that among samples of homeless adults followed over time, the vast majority of them get out of homelessness. The actual percentage of those who exit, of course, is in part a function of how exit is defined. For instance, in the Course of Homelessness Study, 72 percent exited homelessness during a fourteen-month period using the least conservative definition of exit—a thirty-day exit in either a place the person paid for (independent exit) or other arrangements, including doubling up, staying in institutions, and so forth (dependent exit). That figure falls to 52 percent if the definition is thirty days in an independent exit setting only, to 51 percent if the definition is ninety consecutive days in either independent or dependent exit settings, and 30 percent if the definition is ninety days in an independent exit setting only. Exit rates also differ by subpopulations. For instance, in the Alameda County study, 94 percent of women with children exited within a year (defined as at least thirty days, excluding institutional stays), whereas 82 percent of women without children in their care and 65 percent of men exited. Still, the general pattern is the same: Substantial numbers of homeless people exit from homelessness within relatively short periods of time.

The bad news is that most of these individuals fall back into homelessness within those same time periods. For example, in the Course of Homelessness sample, more than three-quarters of the 72 percent who experienced a thirty-day exit to either independent or dependent exit settings experienced homelessness again; two-thirds of the 30 percent who experienced at least ninety consecutive days of housing for which they paid fell back into homelessness again. (Clearly, type of exit affects likelihood of return, a fact that emerges from the Alameda County study as well.) Moreover, multiple cycles of exiting and reentering were not unusual: More than half of those persons with thirty-day exits and approximately one-quarter of those with ninety-day exits experienced two or more exits during the fourteen-month follow-up period. Again, differences were quite apparent with regard to subgroups. More than half of the women without children in their care and more than two-thirds of the men who exited homelessness in the Alameda County study transitioned back into homelessness during the time they were followed, although only one-third of the women with children in their care did so. Still, the general trend is the same. Particularly among single homeless adults, exits are shallow for most people, making the typical pattern of homelessness, as Sosin and colleagues have pointed out, “one of residential instability, rather than constant homelessness over a long period” (Sosin, Piliavin, and Westerfelt 1990, 171). Moreover, what appears to drive that residential instability, based on multivariate (considering multiple variables simultaneously) analyses of predictors of exit and reentry, is not the individual deficits that people may have (such as psychiatric disorders or substance abuse problems) but rather the institutional and economic resources available to them (such as formal or informal income support or subsidized housing).

The Alameda County data show that homeless families appear to differ significantly from single people in their movement into and out of homelessness. This fact is even more apparent in a study of homeless families in New York City that was conducted by Beth Shinn, Beth Weitzman, and colleagues. This study interviewed a sample of families as they were requesting shelter for the first time and then reinterviewed them approximately five years later. Among the 256 families who actually entered a shelter, 79 percent were housed in their own apartments five years later, and another 17 percent were living with family or friends. Only 4 percent were in a shelter. Receipt of subsidized housing proved to be the best—and virtually only—predictor of who was in their own housing. Controlling for all other factors, those who received subsidized housing at some point between time 1 and time 2 were twenty-one times more likely to be housed in their own places than were those who hadn’t. (Receiving subsidized housing was not confounded with individual characteristics, eliminating the hypothesis that factors affecting selection into subsidized housing are responsible for this finding.) Closer analysis of these data suggested that 96 percent of families who left a shelter with subsidized housing were housed in their own apartments at time 2; only 15 percent had returned to a shelter at any point, usually because of serious building problems or safety issues related to their housing. Alternatively, 71 percent of the families who went into unsubsidized housing situations were in their own place at time 2; 43 percent had returned to a shelter at some point during the interim between time 1 and time 2. These findings point to the critical contribution of subsidized housing, foreshadowing the findings from service demonstrations, discussed later.

Several studies have taken advantage of the comprehensive administrative databases that document stays in New York’s and Philadelphia’s large, centralized municipal shelter systems to try to understand residential transitions among homeless adults. Dennis Culhane and Randall Kuhn, for instance, used these databases to examine shelter utilization first among homeless single adults over time and then, along with Yin-Ling Irene Wong, among family shelter users. Tracking people through administrative data confers many advantages. These advantages include the large numbers of people captured, the ability to focus on new entries and to deal with both left (see above) and right (ending data collection before an episode has played itself out) censoring, the actual documentation of stays (thus avoiding the need to rely retrospectively on self-report and all the bias associated with this), and the ability to avoid underrepresenting short-term stays. The disadvantages, of course, are that information on periods of homelessness outside of the public shelter system are not captured and that specific information on potential predictors of interest may not be available. Overall, such studies are best suited to understanding the dynamics of shelter utilization per se, rather than homelessness more generally, and speak more specifically to policy issues related to shelter administration.

In general, the findings of these studies parallel those of panel studies. All of the index cases (those tracked over time) in each of these studies exited the shelter system within a two-year period, but substantial numbers returned. Again, recidivism (relapse) was much higher among single males (50 percent) and single females (33 percent) and lowest among adults with children in their care (22 percent). Interestingly, among the single adults, short-term use of the shelter system was the rule. The majority of single men (55 percent) and single women (65 percent) had only one episode of shelter use during a two-year period. Moreover, even when all shelter stays were combined, half of these adults spent fewer than forty-five days in a shelter. For many, then, shelter stays represented a short-term transition. On the other hand, a smaller group of adults made significant use of the shelter system—18 percent of users accounted for 53 percent of the shelter usage during that two-year period—suggesting a critical target group for intervention. A host of demographic and personal deficit variables was associated with probability of exit, but no resource utilization variables were available for inclusion in the multivariate analyses of predictors. Findings from the family shelter study were somewhat more complex but paralleled the findings from panel studies in that exiting to subsidized housing was clearly linked with a lower rate of readmission to the family shelter system.

Panel and shelter studies provide a window into what happens to homeless people in naturally occurring situations—people who are receiving what might be referred to as “usual care.” A growing literature has also documented the extremely positive impact of subsidized and supportive housing interventions on subsequent course of homelessness. Debra Rog and Scott Holupka recently reviewed this literature, focusing not only on single-site studies but also on large-scale, multisite supportive housing initiatives such as the Robert Wood Johnson (RWJ)/HUD Homeless Families Program and the McKinney-supported studies of housing and services for seriously mentally ill homeless adults. Several lessons emerge from these studies. First, the vast majority of homeless people placed in housing stay housed. In the McKinney demonstration projects, for instance, 78 percent of those people placed in housing were stably housed in the community twelve to eighteen months later. In the RWJ/HUD program, more than 85 percent of the homeless families who received Section 8 housing certificates (subsidies that make apartments affordable) were still in permanent housing eighteen months later. Housing by itself, however, may not always be enough. Subsidizing the cost of housing increases stability, especially where subsidies allow people to live in safer, more decent housing. Supportive services may be needed to help establish, maintain, and enhance stability. However, providing secure housing first is the most effective way of stabilizing homeless adults and families.

Where We Go from Here

Clearly we need to learn more about who becomes homeless, patterns of homelessness over time, and the factors that predict movement into and out of homelessness. We do understand, however, that a large number of people experience homelessness and then never experience homelessness again, that homelessness is a chronic condition for a much smaller number of people, and that in between a substantial number of people experience ongoing residential instability, moving into and out of homelessness repeatedly over time. Similarly, we understand that access to institutional resources, including the safety net meant to support economically marginal households, plays a critical role in determining which group people find themselves in. Finally, we know that proven models for successfully intervening with the full range of homeless sub-populations exist. What remains is the will to translate these understandings into broad policy initiatives that are capable of both preventing homelessness and ameliorating it when it occurs.