Show Summary Details

Page of

PRINTED FROM OXFORD HANDBOOKS ONLINE ( © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice).

date: 17 November 2018

Homelessness in the United States

Abstract and Keywords

Homelessness has become more common in developed countries, especially the United States. This article reviews definitions of homelessness, the temporal dimensions of how people experience homelessness, the determinants of the volume of homelessness, the characteristics of homeless people, and the policies and interventions that have been successful in reducing homelessness. It discusses two popular views of what homelessness is like and what should be done about it. One view emphasizes that the elimination of personal problems that afflict many homeless people can reduce homelessness. The other view emphasizes that homelessness can be ended by placing everyone who becomes homeless in his or her own apartment (housing first policy). However, understanding and alleviating homelessness requires a more sophisticated view of the interaction among personal problems, market conditions, and incentives. The vast middle ground between policies requires programs that make difficult trade-offs, and good data would be helpful for making these trade-offs.

Keywords: homelessness, homeless people, personal problems, housing first policy, alleviating homelessness

Since the 1980s, homelessness has become more common in developed countries, especially the United States, than it was in the previous several decades. The costs of homelessness, both to those who are homeless and those who are not, are great, and some effective policies and interventions have been discovered. In this chapter, I review what scholars mean by homelessness today, the temporal dimensions of how people experience homelessness, the determinants of the volume of homelessness, the characteristics of people who are homeless, and the policies and interventions that have been successful in reducing homelessness.

I also discuss two popular views of what homelessness is like and what should be done about it. One view emphasizes the personal problems that afflict many homeless people and concludes that little can be done unless those pathologies are eliminated. The other view emphasizes the housing aspect of homelessness and concludes that homelessness can be ended by placing everyone who becomes homeless in his or her own apartment. Both popular views are misguided: neither relieving personal problems in the general population nor improving housing in the general population is likely to make much of a dent in homelessness unless the actions are well targeted, and targeting is hard. Understanding and alleviating homelessness requires a more sophisticated view of the interaction among personal problems, luck, market conditions, dynamics, and incentives.

(p. 278) Definitions

Since the late 1980s, government officials and scholars in the United States have used a definition of homelessness that has remained consistent over time. According to this definition, people are homeless on a particular night if they spent that night in a homeless shelter, or a commercial facility acting as a homeless shelter, or in a place not designed for human accommodation. All parts of this definition are somewhat vague, but enough consensus has emerged around them that they appear to have proved workable. I will use this definition in this chapter.

Two features of this definition are notable. First, it refers to a housing market condition. Homelessness is about where people sleep—not how they dress, what they drink, whether they panhandle, or how they relate to the rest of society. Second, it refers to a condition, not a status or an attribute. There is no such thing as a homeless person; there are only people who are homeless on particular nights. The word homeless is used in many other ways in colloquial discourse; these usages commonly lack one or both of these key features.


Counts of homeless people are rare, and reasonably accurate ones are rarer. But the situation is improving, as the federal government is pushing localities to make counts and compiling the information from local counts. It is strange that the federal government is relying on local service providers to do a job that the Census Bureau or the Bureau of Labor Statistics does in other domains (introduction of the American Community Survey does not affect this arrangement). Quality remains low. Data collection by interested and unsophisticated parties imparts unknown biases.

Counts of homeless people can be designed to answer different questions. The simplest question is: How many people are homeless tonight? This is called the point-in-time (PIT) count and corresponds to the way phenomena like unemployment are measured. More complex questions take a period of time greater than one night and ask how many different people were homeless for at least one night during that period. The most common period used for this purpose is a year, although one-month and five-year counts have also been published.

What question you ask should depend on what you want to know. For questions about housing stock, for instance, PIT counts give the relevant information. PIT counts are also informative about the difficulties homeless people might cause to nonhomeless people.

For questions about the harm that homelessness causes to homeless people themselves, it would help to know how the harm depends on the length of a spell—which we do not know. If the harm of an additional night of homelessness decreases very (p. 279) quickly as a spell lengthens (so that a two-month spell is not much worse than a two-day spell), then the one-year count is more informative than the PIT count. If the harm of a spell is directly proportional to the number of nights in a spell, then the PIT count is exactly the right measure to use. If the harm of an additional night increases as a spell lengthens, then neither measure is informative, although PIT is closer, and a different measure like the PIT count of individuals who have been homeless for a long time is more useful than either of the traditional universal counts.

Table 12.1 Official U.S. Homeless Counts

PIT Counts

One-Year Counts




1990 census



2000 census


2006 AHAR




2007 AHAR




(*) Six-month count.

Sources: 1990: U.S. Bureau of the Census, n.d., American Fact Finder; 2000: Smith and Smith 2001; 2006, 2007: HUD 2008.

Table 12.1 summarizes the official federal counts of homelessness since 1980.

The 1990 census attempted to count both shelter residents and “persons visible in street locations” in late March. The 2000 census counted shelter residents only and made no attempt to publish a separate estimate of street dwellers. Both of these are PIT counts. Starting in 2006, the federal Department of Housing and Urban Development (HUD) has compiled reports from local jurisdictions to produce an Annual Homelessness Assessment Report (AHAR) to Congress. This report includes PIT counts for a night in late January for both shelters and streets and a one-year count for shelter use alone (not streets).

Table 12.1 shows a large increase in shelter population from 2000 to 2006, but this should not be taken as solid evidence of a dramatic rise in homelessness for this period; methods and definitions vary considerably. Probably the strongest conclusions that can be drawn from table 12.1 are that the PIT count of homeless people has six digits, and that it has not fallen significantly since 1990.


Very little is known about spells of street homelessness, but much is known about spells of shelter homelessness. As table 12.1 indicates, most spells of homelessness are short in the context of a lifetime, since about four times as many people experience a stay in a shelter during a year as are in a shelter on an average night.

(p. 280) Generally, families with children have longer shelter stays than adults unaccompanied by children. Some single adults move back and forth between shelters and streets, however, and so their homeless spells may be longer than their shelter spells. Table 12.2 gives information on shelter spells.

Table 12.2 Shelter Spells (Completed)

Number of Days



Single adults:

New York City men (1)


Philadelphia men (1)


Athens, OH, shelter (2)


U.S. men, 2007 (3)


U.S. women, 2007 (3)


NYC men, December 2008 (4)


NYC women, December 2008 (4)



U.S., 2007 (3)


NYC families, December 2008 (5)


(*) Ongoing spells.

Sources: (1) Culhane et al. 1998; (2) Allgood, Moore, and Warren 1997; (3) HUD 2008; (4) New York City Department of Homeless Services 2009a; (5) New York City Department of Homeless Services 2009b.

Transitions to and from homelessness are very hard to predict. Shinn et al. (1998), for instance, use data on New York City welfare families to predict which ones will enter shelters. Their best equations can identify a group of families with an 18 percent probability of entering the shelter system, as opposed to a 1 percent probability in the complementary group. Nothing in their very rich data set allowed them to say what makes that 18 percent different.

Research that tries to predict shelter exits finds similarly weak results (e.g., Poulin 2007). Sometimes characteristics that make people more likely to be homeless make people more likely to leave shelters. Culhane et al. (2007), for instance, found little difference in needs between families with short stays and families with long stays; mental illness predicted shorter shelter spells in their data. The 2007 AHAR similarly finds that longtime shelter residents have few differences in problems for short-time users. Length of stay in a shelter is not very predictive of exit either, unless there are administrative limits on stays. The New York City family shelters have an increasing hazard rate of exit, even though there are no administrative limits. People who have been in shelters a long time are not less likely to exit in the next month.

These stylized facts are consistent with a model of homelessness as bad luck (O’Flaherty 2010). People become homeless because they are surprised by bad shocks—they lose their jobs, for instance, their relatives kick them out, their health (p. 281) worsens. People at risk of becoming homeless cannot predict that they will become homeless, because if they could they would take effective steps to prevent it. Statisticians can expect to do no better than the people they are observing; transitions to homelessness are inherently unpredictable.

Fires are a good analogy. If you knew that a fire would break out in your kitchen tomorrow at 3:15 p.m., you would be present and there would be no fire. Fires in your house are necessarily surprises to you, and they must also be surprises to fire department statisticians.

The most extreme statement of this model can be found in classical consumption theory (Hall 1978). Under strong simplifying assumptions, a household's path of housing consumption over time is a martingale: the expected value of housing consumption tomorrow is housing consumption today. Housing consumption today embodies all the information available today (about everything). Homelessness is a variety of housing consumption. Hence all the information that can be used to predict homelessness tomorrow is already embodied in housing consumption today.

This model has some powerful implications. First, it implies that current housing consumption should be the chief predictor of future transitions to homelessness. This appears to be true. In Shinn et al. (1998), an equation with only housing and demographic variables performed almost as well as equations with many more variables. Households that are doubled up or living in a low-rent neighborhood should be more likely to become homeless, and they are.

Second, the model implies that past homelessness should predict future homelessness, since homelessness is a variety of housing consumption and the martingale property means that the expected value of housing consumption at any future date, not just tomorrow, is housing consumption today. Empirical work has repeatedly shown that past homelessness is a powerful predictor of future homelessness.

Finally, the model implies that exits from homelessness should be unpredictable. Empirical findings are consistent with this implication.

What Determines the PIT Count of Homelessness?

The simple answer is: housing market conditions, weather, and shelter rules. For any night, the history of exits and entries up to that night yields a number of homeless people. Quite a few studies have examined how PIT counts differ across cities, metropolitan areas, states, and time.

Cross-section studies are more numerous, and most are reviewed in O’Flaherty (2004). Most of these studies find that greater PIT homelessness is related to tighter housing market conditions (higher rents for poor people, or lower vacancy rates), and better weather. (Temperature is the usual variable, but some studies also use precipitation.) Demographic variables are almost uniformly insignificant in these (p. 282) studies (and sometimes have the wrong sign): the extent of male, minority, poor, mentally ill, or substance-abusing population rarely matters. The latter two variables are usually measured poorly in these studies, but the former three are not.

Time-series studies are less numerous and focus exclusively on shelter populations. Culhane et al. (2003) is the only such study that uses data outside New York City. These researchers study family and single women's shelters in Philadelphia and find very few significant results. Cragg and O’Flaherty (1999) and O’Flaherty and Wu (2006) look at family shelters in New York City. In contrast to the cross-section papers, they find very little influence of rent on family shelter homelessness, but the time series provides them only with aggregate rent data, and there is little variation during the time they study in the rate of rent increase in New York City. Macroeconomic conditions matter, but the effects are generally not so strong as the effects of administrative rules and procedures. There is a weak tendency for more families to enter in the summer.

Both these papers were motivated by the controversy in New York about whether the prospect of placement into subsidized housing that the family shelters offer draws large numbers of families into the shelters. The researchers find that placements do have an incentive effect (in the 2006 paper, most of the effect is on exits rather than entries—that is, the prospect of subsidized housing slows the rate at which families leave shelters on their own), but that this effect is much smaller than the first-order effect that placements have on population. On net, placements reduce shelter population, and a considerable part of the variation in family shelter population is due to variation in the rate of placements. Other administrative practices matter, too, such as the type of shelters being used. Perhaps the internal aspects of shelter life appear more important than external forces because they are better measured.

The results on shelters for single adults in New York (O’Flaherty and Wu 2008) are similar: external factors like releases of prisoners and inmates make little difference, while internal factors like capacity and placements have significant effects. Macroeconomic conditions seem to have a weak impact on men's shelter population. Single shelter population is highest in the late winter.

For both families and single adults, the authors were unable to reject the hypothesis that shelter population had a unit root. This means that transitory disturbances have permanent effects: history matters as well as current conditions. Cross-section studies should probably include lagged dependent variables or lagged values of explanatory variables.

Who Is Homeless?

The people who are homeless at any point in time are different in many ways from average Americans. Nearly all homeless people are poor, and most are severely poor. (p. 283) A clear majority are male, and almost a majority are African American (44 percent of shelter residents in the 2007 AHAR). Disproportionately few homeless people are old (4 percent of shelter residents in the 2007 AHAR, as opposed to 27 percent of poor people). Disproportionately large numbers of homeless people suffer from severe mental illness (28 percent of PIT shelter residents in the 2007 AHAR) and from substance abuse (39 percent of PIT shelter residents in the 2007 AHAR), although the majority of homeless people do not. Households with one person are disproportionately likely to be homeless (they account for 63 percent of the PIT homeless population but 10 percent of the 2007 U.S. household population). Sheltered homeless people are disproportionately likely to be in central cities (77 percent in the 2007 AHAR, as opposed to 36 percent of poor people and 24 percent of the U.S. population). Veterans also appear disproportionately likely to be homeless (15 percent of the shelter PIT count as opposed to 5 percent of the poverty population and 10 percent of the adult population). Local reports also indicate that foreign-born people are underrepresented in the homeless population, while former prisoners and former foster children are overrepresented.

In national data sets, Early (1999, 2004) has found that individual characteristics are much better predictors of who is homeless than are housing market characteristics. Public awareness of homelessness also often focuses on the characteristics that homeless people have.

But people who are homeless at any time are a minority—usually a tiny minority—of any of the preceding categories. For instance, Early and Olsen (2002) found that about 0.73 percent of the poor people in the average metropolitan area were homeless in March 1990. Frank and Glied (2006, table 7.6), in probably the most complete recent study of where severely mentally ill people lived, estimated that around 3 percent of severely mentally ill people in 1990 and 2 percent in 2000 were homeless on an average night. Based on the 2007 American Community Survey, about 0.8 percent of African Americans and 1.4 percent of one-person households were homeless.

These small percentages are key to understanding why the size of at-risk populations seems to make little difference to the volume of PIT homelessness in either cross-section or time-series studies. The question that these studies essentially answer is: What is the probability that the marginal at-risk person will be homeless on a given night? That is, for instance, by how much would the PIT count decrease if one male or one severely mentally ill person left the general population?

There are many theories about how to answer such a question. One obvious way (though not a good way, it turns out) is to think of the marginal person as the average person. Then the marginal person has the very low probabilities of being homeless that we saw in the previous paragraph. Moreover, what matters is the difference between the probability that an average at-risk person will be homeless and the probability that an average not-at-risk person will be homeless, which is necessarily smaller than the former probability. If this were the case, changes in population characteristics would have small effects on homeless population. (For instance, (p. 284) suppose that always and everywhere 2 percent of mentally ill people are homeless, and 0.1 percent of people who are not mentally ill are homeless. Then a regression of homeless population on mentally ill population would yield a coefficient of .019.) Since the cross-section variation among metropolitan areas in at-risk characteristics is not huge, at-risk characteristics would not explain much of the variation in homelessness.1

If the marginal propensity to be homeless were equal to the average, then the proportion of people who were homeless within each of these at-risk groups would be constant, both in the cross section and in the time series. Abundant evidence indicates that such is not the case. For instance, in Early and Olsen (2002), the standard deviation across metropolitan areas in the proportion of poor people who are homeless is almost as big as the mean. In order to understand the data, more sophisticated models are called for.

For example, consider a very simple cross-section model. Suppose that cities differ in the proportion of the population that is poor (using the federal definition) and in the price of housing. Setting aside household size issues, people are poor when their incomes are below some level P, the same for every city. They are homeless in city c when their incomes are below some level H(c), which differs between cities because the price of housing differs. Then variations in the proportion of the population between P and H(c) will affect poverty population but not homelessness, while variation in housing prices and in the distribution of poverty population below and above H(c) will affect homelessness but not poverty population. Since almost all poverty population is between P and H(c), almost none of the variation in poverty population is associated with variation in homelessness.

Thus in this model on an individual level, being poor would be a strong predictor of homelessness, but in the cross-section variation in the size of the poverty population would explain very little of the variation in the size of the homeless population.

I am not claiming that this rough model is the best possible model of homelessness. What it shows is that it is possible for homelessness to be closely associated with demographic characteristics on an individual level, but for the incidence of those demographic characteristics to explain very little about the volume of homelessness.

(p. 285) How Do Policies Affect the Volume of PIT Homelessness?

Not much is known about how policies affect the volume of PIT homelessness, with two exceptions: shelter policies and rent control. I have already described the results on shelter policies. There is a voluminous literature on how rent control affects homelessness; the consensus is that the impact, if any, is very small. For a review of this literature, see O’Flaherty (2010).

There is also work on housing regulation and on housing subsidies.

Housing Regulation

Raphael (2010) examines the effect of housing regulations on the volume of PIT homelessness. His hypothesis is that housing is more expensive in metropolitan areas where zoning rules and other regulations make building new houses more difficult, and homelessness is greater in places where housing costs more. The first part of this hypothesis has considerable empirical and theoretical support in housing economics. (No studies that I am aware of test for reverse causality, but the relationship has been established a number of indirect ways, not just by simple correlation.)

To test this hypothesis, Raphael uses the Wharton state-level regulatory index (Gyourko, Saiz, and Summers 2006) and the 2007 AHAR counts of homelessness. He uses a two-equation system, with regulation as an instrument for rent-to-income ratios. Regulation works as hypothesized. Raphael's estimates imply that reducing the regulatory burden in the states now above the median to the median level would cut homelessness by 6 to 13 percent.

A much larger literature (e.g., Hoch and Slayton 1989) has tried to connect homelessness to specific regulations on various types of housing for poor people like lodging houses, SROs, and cage hotels (hotels in Chicago where cubicles did not have full walls). While plausible, these claims have never been rigorously tested. The Wharton data set that Raphael uses primarily measures the difficulties of building new housing for middle-class people.

Less plausible are claims that preserving or maintaining specific buildings or uses in particular places will have an effect on the volume of homelessness. In the housing market, what matters are policies that affect large numbers of people like movements of the aggregate rent level, since only in large pools can substantial numbers of people likely to be homeless on any night be found.

Expanding Housing Subsidies as a Policy

Early and Olsen's work (2002), the most straightforward paper on housing subsidies, looks at a cross section of metropolitan areas and regresses PIT homelessness (p. 286) on the availability of housing subsidies, the extent to which subsidies are targeted to extremely poor people, and a variety of the usual explanatory variables. The number of housing subsidies has no effect on homelessness, but better targeting of subsidies reduces homelessness.

Early and Olsen's paper is the only one that tackles the issue of housing subsidies directly, and so needs to be taken very seriously. But it lacks a test for endogeneity: perhaps housing subsidies might be directed, consciously or unconsciously, to places with more homelessness. A way to test this might be to instrument for current housing subsidies with housing subsidies before 1980, when homelessness was not an issue.

Mansur et al. (2002) also address the question of housing subsidies, but its results are based on simulations, not new estimates of the structural parameters. They look at four metropolitan areas in California and simulate several different policies. The most interesting policy is a universal subsidy to poor renters—a subsidy much more widespread but also shallower and more tightly income-targeted than the existing Housing Choice Voucher (HCV) program. This policy substantially reduces homelessness in all four metropolitan areas. However, the reduction in homelessness per 100 households subsidized is not a large number. For every 100 households that receive a subsidy in San Francisco, PIT homelessness goes down by about 2.2 households.

One can make a similar calculation from the Early and Olsen paper, too. What is the reduction in homelessness from an increase of 100 in the number of very poor households receiving subsidies (which in this case comes from a reduction in the number of other households receiving subsidies)? About 7.7.

These somewhat small ratios should not be surprising. Basically, subsidy programs pick some conventionally defined set of housed households and greatly lower the probability of homelessness for that set. Within any conventionally defined set of housed households, the proportion who will be homeless on any given night in the future is quite small, and it is hard to predict which households those will be. Subsidy programs also affect the overall price of housing and so have secondary effects too, but a small number of subsidies is not going to have large effects on the metropolitan price of housing.

How Do Interventions Affect Who Is Homeless?

The literature about whether interventions can keep particular people from remaining homeless or becoming homeless is better developed than the literature about whether policies can reduce the number of homeless people. This is probably because it is easier to observe whether someone is homeless than to count the total (p. 287) number of people who are homeless. Controlled experiments can tell us the effect of treatment on the treated, but they cannot tell us the effect of treatment on the untreated: whether rental subsidies drove up the price of housing and made untreated people homeless, for instance, or whether the prospect of efficacious treatments made people alter their behavior so that they were more likely to need those efficacious treatments.

My distinction between policies and interventions is a rough one. By interventions I mean activities that try to make large changes in the probability of being homeless for designated people, and that are evaluated only by the effects on those designated people. By policies I mean activities that try to make small changes in the probability of being homeless for large numbers of people. A good example of an intervention is a program helping particular long-term homeless people find housing. A good example of a policy is a reduction in housing regulation.

Interventions with People Who Are Not Homeless

Households that receive housing subsidies at least as generous as HCV and public housing very rarely become homeless; receipt of such subsidies drastically reduces the probability that a household will become homeless. The best evidence for this regularity comes from the evaluation of the Welfare to Work Voucher Program (Mills et al. 2006). In this controlled experiment, several thousand randomly selected welfare families received vouchers under HCV, and their progress was compared with that of a control group of equal size. Over a one-year period, voucher receipt reduced the probability that a family would experience at least one episode of homelessness by 9.2 percent of total population; homelessness was virtually eliminated among voucher recipients (Mills et al. 2006, exhibit 5.3). This is a treatment-on-treated (ToT) effect on the probability of being homeless in a year, not the probability of being homeless on a given night. Using ratios for families from the 2007 AHAR, Ellen and O’Flaherty (2010) calculate that this 9.2 percent ToT effect means that for every 100 welfare families receiving vouchers, the PIT homeless count goes down by about 3.5 families. We do not know the effect of “treatment on the untreated” from the voucher expansion, but the ToT estimate is in line with the estimates of Early and Olsen and Mansur et al. that were discussed in the previous section.

Early (1999, 2004) provides an independent estimate that is also in line with this literature. Essentially, he simulates an intervention that expands housing subsidies. First he regresses the probability that a household will be homeless on a set of characteristics. Using this result, he predicts the probability that households receiving subsidies would be homeless if they were not receiving subsidies. He finds that 3.8 to 5.0 percent of subsidized households would be homeless if they were not receiving subsidies. He concludes that an expansion of subsidized housing by 100 units would reduce homelessness by 3.8 to 5.0 households. This estimate is consistent with the other three estimates we have discussed, all arrived at by different methods.

(p. 288) Interventions with People Who Are Homeless

Since people who are homeless today are much more likely to be homeless tomorrow than people who are not homeless today, interventions targeted to people homeless today are the best way of causing large decreases in individual probabilities of being homeless tomorrow. Some of these interventions use housing subsidies alone, some use subsidies in combination with other forms of assistance, and some use only other forms of assistance.

Subsidies alone have been used most frequently with homeless families and are fairly effective in reducing homelessness. Shinn et al. (1998), for instance, found that housing subsidy receipt was the best predictor of whether a family would return to shelter, and that virtually all families who left shelters with subsidies were conventionally housed several years later. Families receiving subsidies, however, were not randomly selected. To my knowledge, no controlled experiments have been conducted.

Combinations of subsidies and services usually are targeted at homeless people with severe mental illness or substance abuse problems. A number of controlled experiments have been conducted on these interventions. Rosenheck (2010) provides a good review.

Supported housing and supportive housing provide participants with subsidized housing in a setting designed to improve their psychiatric functioning. The settings involved vary greatly: sometimes all the participants live in the same building, and sometimes they are spread out in ordinary apartments; sometimes abstinence is required, and sometimes it is not; sometimes case management and therapy are intense, and sometimes they are not.

Controlled experiments often show reductions in homelessness. Most of these reductions appear to come from the subsidies, while the contribution of case management to homelessness reduction appears to be small. These programs also cut use of medical facilities such as emergency rooms.

Housing First is the best known of these interventions. According to Rosenheck (2010), Housing First

puts a high emphasis on client choice and emphasizes rapid placement in housing for severely mentally ill and often dually diagnosed clients, who would otherwise be unlikely to find housing or would find delayed access through multi-stage continuum of care programs. Housing First has among the most robust improvements in housing in comparison to its randomly assigned control group…. There were no benefits in psychiatric or substance abuse outcomes for Housing First clients as opposed to controls, although they experience more choice in their programs. There is clear evidence that Housing First clients experienced significantly less use and lower institutional costs during the first 24 months of treatment, although by 24 months there were no longer significant group differences. (11–12)

Housing First achieves these large reductions in homelessness and institutional costs because its clients are seriously distressed when they enter, and so would have spent a great deal of time homeless or institutionalized in the absence of Housing (p. 289) First. Many enter the program while they are in psychiatric hospital beds. Housing First is very tightly targeted to a proper subset of homeless people.

Finally, a number of case management programs—for instance, Critical Time Intervention—have been shown to reduce psychiatric symptoms, but they have only small impacts on homelessness unless they are accompanied by housing subsidies. Supported work increases employment, but the impact on housing is also small.

To summarize the research on interventions: you get what you pay for.


The fundamental dilemma of homelessness reduction is that the more tightly policies and interventions are targeted, the greater the danger of moral hazard. The best predictor of future homelessness is current homelessness, but programs that reward current homelessness raise the specter of moral hazard. Moral hazard is something that needs to be measured and managed; it is not a reason for doing nothing. Unfortunately, to date little measurement of moral hazard has occurred, despite a pervasive belief among shelter operators and public officials that it is important.

Many strategies have been proposed and attempted to mitigate this dilemma. At one extreme, policies like regulation reduction avoid moral hazard but are very diffusely targeted; they work because the social cost for each person who benefits is small, if not negative. At the other extreme, interventions like Housing First work because although they are very tightly targeted, the characteristics on which eligibility is based are extremely expensive to manipulate. The vast middle ground requires programs that make difficult trade-offs; good data would be helpful for making these trade-offs.


Allgood, Sam, Myra L. Moore, and Ronald S. Warren Jr. 1997. “The Duration of Sheltered Homelessness in a Small City.” Journal of Housing Economics 6:60–80.Find this resource:

Cragg, Michael, and Brendan O’Flaherty. 1999. “Do Homeless Shelter Conditions Determine Shelter Population? The Case of the Dinkins Deluge.” Journal of Urban Economics 46:377–415.Find this resource:

Culhane, Dennis P., and Randall Kuhn. 1998. “Patterns and Determinants of Public Shelter Utilization among Homeless Adults in New York City and Philadelphia.” Journal of Policy Analysis and Management 17:23–43.Find this resource:

Culhane, Dennis P., Stephen Metraux, J. M. Park, Maryanne Schretzman, and J. Valente. 2007. “Testing a Typology of Homelessness Based on Patterns of Public Shelter (p. 290) Utilization in Four U.S. Jurisdictions: Implications for Policy and Program Planning.” Housing Policy Debate 18:1–28.Find this resource:

Culhane, Dennis P., Stephen R. Poulin, Lorlene M. Hoyt, and Stephen Metraux. 2003. “The Impact of Welfare Reform on Public Shelter Utilization in Philadelphia: A Time-Series Analysis.” Cityscape: A Journal of Policy Development and Research 6:173–185.Find this resource:

Early, Dirk. 1999. “A Microeconomic Analysis of Homelessness: An Empirical Investigation Using Choice-Based Sampling.” Journal of Housing Economics 8:312–327.Find this resource:

———. 2004. “The Determinants of Homelessness and the Targeting of Housing Assistance.” Journal of Housing Economics 55:195–214.Find this resource:

Early, Dirk, and Edgar Olsen. 2002. “Subsidized Housing, Emergency Shelters, and Homelessness: An Empirical Investigation Using Data from the 1990 Census.” Advances in Economic Analysis and Policy 2 (1): n.p.Find this resource:

Ellen, Ingrid, and Brendan O’Flaherty. 2010. “Introduction.” In How to House the Homeless, ed. Ingrid Ellen and Brendan O’Flaherty, 1–14. New York: Russell Sage.Find this resource:

Frank, Richard G., and Sherry A. Glied. 2006. Better but Not Well: Mental Health Policy in the United States since 1950. Baltimore: Johns Hopkins University Press.Find this resource:

Gyourko, Joseph, Albert Saiz, and Anita Summers. 2006. “A New Measure of the Local Regulatory Environment for Housing Markets: Wharton Residential Land Use Regulatory Index.” Working paper, University of Pennsylvania.Find this resource:

Hall, Robert. 1978. “Stochastic Implications of the Life Cycle Permanent Income Hypothesis: Theory and Evidence.” Journal of Political Economy 86:971–987.Find this resource:

Hoch, Charles, and Robert Slayton. 1989. New Homeless and Old: Community and Skid Row Hotel. Philadelphia: Temple University Press.Find this resource:

Mansur, Erin, John Quigley, Steven Raphael, and Eugene Smolensky. 2002. “Examining Policies to Reduce Homelessness Using a General Equilibrium Model of the Housing Market.” Journal of Urban Economics 52:316–340.Find this resource:

Mills, Gregory, Abt Associates, Daniel Gubits, Larry Orr, David Long, Judie Feins, Bulbul Kaul, Michelle Woods, Amy Jones and Associates, Cloudburst Consulting, and the QED Group. 2006. Effects of Housing Vouchers on Welfare Families. Washington, DC: HUD Office of Policy Development and Research.Find this resource:

New York City Department of Homeless Services. 2009a. “Critical Activities Report: Adult Services, FY 09.” Accessed May 6, 2009.

———. 2009b. “Critical Activities Report: Family Services, FY 09.” Accessed May 6, 2009.

O’Flaherty, Brendan. 1996. Making Room: The Economics of Homelessness. Cambridge, MA: Harvard University Press.Find this resource:

———. 2004. “Wrong Person and Wrong Place: For Homelessness, the Conjunction Is What Matters.” Journal of Housing Economics 13:1–15.Find this resource:

———. 2010. “Homelessness as Bad Luck: Implications for Research and Policy.” In How to House the Homeless, ed. Ingrid Ellen and Brendan O’Flaherty, 143–182. New York: Russell Sage.Find this resource:

O’Flaherty, Brendan, and Ting Wu. 2006. “Fewer Subsidized Exits and a Recession: How New York City's Family Homeless Shelter Population Became Immense.” Journal of Housing Economics 15:99–125.Find this resource:

———. 2008. “Homeless Shelters for Single Adults: Why Does Their Population Change?” Social Service Review 82:511–550.Find this resource:

Poulin, Stephen. 2007. “On the Way Home: An Analysis of the Influences on Leaving New York City Shelters.” Ph.D. diss., Columbia University School of Social Work.Find this resource:

(p. 291) Raphael, Steven. 2000. “The Deinstitutionalization of the Mentally Ill and the Growth in U.S. Prison Populations: 1971 to 1996.” Working paper, Goldman School of Public Policy, University of California at Berkeley.Find this resource:

———. 2010. “Homelessness and Housing Market Regulation.” In How to House the Homeless, ed. Ingrid Ellen and Brendan O’Flaherty, 110–140. New York: Russell Sage.Find this resource:

Rosenheck, Robert. 2010. “Service Models for Assisting Homeless People with Mental Health Problems: Cost-Effectiveness and Policy Relevance.” In How to House the Homeless, ed. Ingrid Ellen and Brendan O’Flaherty, 17–36. New York: Russell Sage.Find this resource:

Shinn, M., B. C. Weitzman, D. Stojanovic, J. R. Knickman, L. Jiminez, L. Duchon, S. James, and D. H. Krantz. 1998. “Predictors of Homelessness among Families in New York City: From Shelter Request to Housing Stability.” American Journal of Public Health 88:1651–1657.Find this resource:

Smith, Annetta C., and Denise I. Smith. 2001. U.S. Census Bureau, Census Special Reports, Series CENSR/01-2, Emergency and Transitional Shelter Population: 2000. Washington, DC: U.S. Government Printing Office.Find this resource:

U.S. Bureau of the Census. N.d. American Fact Finder, 1990 decennial census, P028, Group Quarters.Find this resource:

U.S. Department of Housing and Urban Development. 2008. The Third Annual Homeless Assessment Report to Congress. Washington, DC: U.S. Government Printing Office.Find this resource:


(1.) What about time-series variation, especially the rise in homelessness after 1980? A popular culprit here is the reduction in the population of state and county psychiatric hospitals. However, much of this population decrease was offset by rises in the populations of mentally people in nursing homes, board and care homes, private and acute care hospitals, jails, and prisons. See Frank and Glied (2006), Raphael (2000), and O’Flaherty (1996, chap. 12). The biggest change in relevant population characteristics in the latter part of the twentieth century was the large decrease in alcohol abuse, which should have reduced homelessness.