Struggling to stay out of high-poverty neighborhoods: housing choice and locations in moving to opportunity's first decade

Abstract

Improving locational outcomes emerged as a major policy hope for the nation's largest low-income housing program over the past two decades, but a host of supply and demand-side barriers confront rental voucher users, leading to heated debate over the importance of choice versus constraint. In this context, we examine the Moving to Opportunity experiment's first decade, using a mixed-method approach.

MTO families faced major barriers in tightening markets, yet diverse housing trajectories emerged, reflecting variation in: (a) willingness to trade location – in particular, safety and avoidance of “ghetto” behavior – to get larger, better housing units after initial relocation; (b) the distribution of neighborhood types in different metro areas; and (c) circumstances that produced many involuntary moves. Access to social networks or services “left behind” in poorer neighborhoods seldom drove moving decisions. Numerous moves were brokered by rental agents who provided shortcuts to willing landlords but thereby steered participants to particular neighborhoods.

Keywords:

• low-income housing
• vouchers
• neighborhood
• markets

Introduction: the “locational turn” in low-income housing policy

Should policymakers seek to improve individual decision-making by re-arranging “choice architecture” – to present healthy choices as easier and more attractive than unhealthy ones – as Richard Thaler and Cass Sunstein argue in their aptly titled book, Nudge (2008)? Or do those arguments, and much of the behavioral economics research on which the arguments are based, largely assume that choosers have access to a relatively good set of choices, which is seldom the case for the very poor and disadvantaged (Briggs, Popkin, and Goering 2010)? This distinction belongs at the center of debates over choice-driven policy in education, healthcare, and – less visibly for the body politic – housing. This is particularly true in light of the enormous economic and fiscal challenges we face as a country.

America's largest rental housing assistance program for low-income people – the means-tested Housing Choice Voucher program that currently serves nearly two million households1 – was created in 1974 primarily to reduce rent burden by subsidizing units of acceptable quality. But thanks to influential research and policy debate on the severity of concentrated minority poverty in central cities (e.g. Massey and Denton 1994; Wilson 1987), the past two decades have expanded interest in another policy objective: that of improving the “locational outcomes” of assisted households, meaning: the quality of the neighborhoods in which they are able to live while using federal housing assistance. Both aims pose important dilemmas about choice – what it means and how to protect it.2

The foreclosure crisis and its many spillover effects on markets, neighborhoods, and families have only increased the attention paid to the effects of low-income housing assistance, including the types of neighborhoods that Housing Choice Voucher holders live in and whether low-income families with vouchers contribute to decline when they concentrate in particular neighborhoods. In 2008, The Atlantic magazine and New York Times, among other major media, sounded the alarm, suggesting that changing market dynamics and decisions to demolish public housing might be opening once-stable neighborhoods – in cities and suburbs alike – to a wave of crime arriving with poor people of color using housing vouchers.3

Stepping back from these immediate controversies, since 1992, the hope of improving locational outcomes through low-income housing policy – which has also been linked to the controversial transformation of public housing since the early 1990s (Popkin et al. 2004; Popkin and Cunningham 2005; Vale 2003) – has been pursued through the voucher program in four ways: a broad budgetary shift away from supply-side project subsidies to vouchers; reforms to the voucher program that make it a more flexible tool for deconcentrating poverty and/or promoting racial desegregation, for example, through higher rent ceilings and “portability” across local housing agency jurisdictions (Priemus, Kemp, and Varady 2005; Sard 2001); judicial consent decrees in which the federal government agreed, as part of racial desegregation efforts, to promote a wider array of neighborhood opportunities in particular jurisdictions (Briggs 2003; Polikoff 2006; Popkin et al. 2003); and MTO, a voucher-based experiment launched by the US Department of Housing and Urban Development (HUD) in five metro areas in 1994. MTO aims to examine the effects of voluntary relocation from public or assisted housing in high-poverty neighborhoods to privately owned apartments in low-poverty neighborhoods.

Though HUD under President Bush was criticized for undermining the focus on locational outcomes (e.g. Priemus, Kemp, and Varady 2005), that focus, nonetheless, represents a major shift – a “locational turn” – in the nation's low-income housing policies since the 1980s. True, relocation or “dispersal” programs have been discussed and implemented, often on a small scale and without formal evaluation, since the urban unrest of the 1960s. But MTO's immediate antecedent is a court-ordered racial desegregation effort, the landmark Gautreaux program ordered in metro Chicago in 1976 and examined by social researchers in the decades since (cf. Polikoff 2006; Rubinowitz and Rosenbaum 2000). MTO represents a shift, however, toward economic integration and away from explicit racial integration policy, given the political resistance to race-based preferences. Though they rarely draw much attention in policy debates, dispersal strategies found their way to the headlines in the wake of Hurricane Katrina, which forced an unprecedented relocation of hundreds of thousands of families from New Orleans, many of them black and poor.4

But how much does – or can – demand-side housing assistance actually help? Research has generated mixed evidence that the housing voucher program significantly improves neighborhood quality for users over time. There are glass-is-half-full and half-empty assessments, depending on the reference point: Housing Choice Vouchers do much better, on average, than public housing at avoiding high-poverty neighborhoods, for example, but a relatively small share of voucher users, particularly if they are racial minorities, live in low-poverty or racially integrated areas.5 Vis-à-vis the reformer's benchmarks, then, and national policy statements about neighborhood quality from the Housing Act of 1949 to the Millennial Housing Commission report a half century later,6 the nation's largest housing assistance program for low-income people falls short.

To explain this, previous research, as well as the informally reported insights of program staff at all levels, has highlighted a range of supply-side barriers, such as discrimination and a scarcity of affordable and otherwise appropriate rental housing units for voucher holders, as well as varied demand-side (client-side) barriers, such as: debilitating physical and mental health problems; limited time, money, transportation, information, and other resources vital for effective housing search; a fear of losing vital social support and institutional resources; and ambivalence about moving itself (Goetz 2003; Pashup et al. 2005; Pendall 2000; Varady and Walker 2007). Based on a review of these studies, only programs that emphasize relocation to low-poverty neighborhoods appear to achieve such outcomes to any significant degree. Moreover, the evidence that positive effects of special supports – that is, “assisted” mobility – on locational outcomes persist over the long run is thus far limited to administrative data on the Gautreaux program, which indicate sustained racial and economic integration over more than a decade (DeLuca and Rosenbaum 2003).

These patterns have led some observers to wonder whether deconcentrating poverty is more a reformer's ideal than a priority for the families served by housing programs and to question both the feasibility and the wisdom of intervening in the complexities of housing choice for low-income people (Clark 2005). Yet to date, researchers relying on structured surveys or location mapping have generated limited answers for these fundamental debates about voucher assistance, which we tackle through two research questions. First, beyond short-run success or failure at finding units in particular kinds of neighborhoods, what are the neighborhood trajectories over time for families served by assisted housing mobility? Second, how do housing supply- and demand-side factors interact over time to shape those trajectories, and as part of that question, why do some families – but not others – end up moving back to much poorer neighborhoods after leaving them? The first question is about where families move, the second about why, how, and with what prospects.

MTO has produced a range of locational outcomes over time, and correspondingly varied interpretations by the policy community, not a simple success-or-failure story. As such, we employed a mixed-method approach at three of the five sites: new analyses of the MTO interim survey data, combined with census and administrative data on changing neighborhoods and metro areas, plus in-depth qualitative interviews and intensive ethnographic fieldwork with MTO families (another research team employed a somewhat different approach in the remaining two sites). We examine housing choices in the context of families' broader strategies for their lives, as well as demographic and other changes in metro areas that shifted the distribution of quality, affordability, and other traits among housing locations. We detail specific ways in which choice decisions did matter over time – but almost always in the context of having very limited choices to make. This is a major challenge, albeit one that is often muddled or avoided in policy debates, as we rethink options for tackling persistent inequality in America.

Background

The study of MTO lies at the intersection of two research literatures generally kept apart: one on residential mobility and the other on neighborhood effects. The former is about why families move and where they move. The latter is about whether, how, and how much where they live (neighborhood context) affects child and family well-being. Well-being is a broad construct that includes physical and mental health, economic self-sufficiency, and more (reviews in Ellen and Turner 2003; Leventhal and Brooks-Gunn 2000; Sampson, Morenoff, and Gannon-Rowley 2002). Research on well-being has largely focused on what conditions might be sufficient to produce neighborhood effects. In this paper, we focus instead on a key necessary condition, especially for many low-income and minority families in the housing market: Moving to and staying in comparatively advantaged neighborhoods. We begin with a brief review of the foundational literature on the residential mobility of unassisted households before focusing on the distinctive patterns for assisted ones. We include a brief discussion of policy design and implementation dilemmas in assisted housing mobility as well, tied to our research questions.

Unassisted households: locational choices and outcomes

A large research literature examines residential choice and locational outcomes, with a focus on a majority of households that do not receive the housing subsidies targeted to low-income households. First, centered on the residential satisfaction model, research on mobility decisions emphasizes the importance of life-cycle factors, such as age and family status, and the salience of both housing unit traits and traits of the surrounding neighborhood in triggering moves (Clark and Dieleman 1996; Newman and Duncan 1979; Rossi 1955; Speare 1974; Speare, Goldstein, and Frey 1975). In addition, this literature on why families move reminds us of the importance of what the Census Bureau terms “involuntary” factors, such as job loss, death, divorce, eviction, fire, unaffordable mortgage or rent, or non-renewal of lease (for example, due to property sale), as triggers for moves. Notably, residential mobility has declined for most demographic groups in America in recent decades. But it has increased for low-skill, low-income households. They are much more likely than higher-skill counterparts to be renters, who move four to five times as often as owners, and to make involuntary moves (Fischer 2002; Schacter 2004). Conversely, such households are much less likely to make nonlocal moves toward economic opportunity, for example, to take a job in another region (Fischer 2002). Involuntary moves and the long-run loss of housing affordable to the lowest income households may explain why children move much more often in the United States than other wealthy nations (Long 1992). This gap reminds us that some forms of residential mobility, especially frequent moving in search of a secure and affordable setting, can be a big negative for families.7

But residential satisfaction and mobility rate studies do little to explain where families move to, whether at points in time or in trajectories of moves over time. On the latter front, a second literature has focused on the where of housing preferences and outcomes. The economic model of locational choice emphasizes that households try to optimize a multi-attribute bundle of housing traits, which includes safety, access, and other traits of locations (Galster 2003). But research across disciplines highlights, among other factors, the importance of racial attitudes, discrimination, and patterns of neighborhood change over time. First, most households prefer some racial or cultural “comfort zone” – a factor that interracial class differences alone do not explain well (Charles 2005). Yet there is frequently a mismatch between such neighborhood make-up preferences and the neighborhoods actually available (Bruch and Mare 2006; Schelling 1971). Minority households, for example, consistently express a desire to live in more integrated areas but find a limited supply of available, affordable neighborhoods that fit their preferred range; some rely on referral networks that lack information on such places (see review in Charles 2005). Whites in America report a growing tolerance of, if not always an appetite for, greater neighborhood integration but tend to define their comfort zone in ways that lead to avoidance of areas with substantial black presence (Charles 2005; Ellen 2000).

Second, while racial discrimination in rental and ownership housing markets has declined overall in recent decades, it continues to affect minority as well as white housing choices, adding an informal “tax” (higher lease-up or other fees) to the transaction costs of moving and/or steering households – both whites and minorities – toward particular neighborhoods in ways that reproduce segregation (Turner and Ross 2002; Yinger 1995).

Third and finally, most demographic research on housing patterns describes aggregate patterns for groups over time, not the trajectories of individual households, obscuring important features of housing choice and also of supply. Since the 1960s, geographers, sociologists, and other analysts of “vacancy chains” have modeled churning, filtering, and other systems-level dynamics, whether with simulations or empirical data (Chase 1991; Clark 1965; Emmi and Magnusson 1994; Persky and Felsenstein 2008; White 1971). But a newer body of research, focused on MTO-type prospects of households, finds that as minority poverty concentration soared in the 1970s and 1980s, blacks were about as likely as whites to “exit” poor neighborhoods (South and Crowder 1997). Most exited by moving, not because neighborhood change led to a much lower-poverty rate over time (Quillian 2003). But blacks were far more likely than whites to move from one poor neighborhood to another and also to re-enter a poor neighborhood fairly quickly after residing outside of one. The latter factor – “recurrence” – helps explain blacks' much longer exposure than whites to neighborhood poverty over time (Quillian 2003; Timberlake 2007), a gap that is not explained by racial differences in income or household structure. That gap persisted into the 1990s, even as extreme poverty concentration declined, and appears to be dominated by black renters; it is not accounted for by blacks moving to transitional neighborhoods that became poor after they moved (Briggs and Keys 2009). Until recently, data limitations made it impossible to compare transitions and exposure over time for unassisted versus assisted households, whose fortunes we turn to next.8

Assisted households: locational choices and outcomes

By some measures, voucher holders appear to be quite clustered in particular neighborhoods or types of neighborhoods. According to a 2003 HUD report that examined the nation's 50 largest housing markets, the spatial clustering of vouchers is far greater than the dispersion of housing units at affordable rents alone would predict: 25% of black recipients and 28% of Hispanic recipients live in high-poverty neighborhoods, compared to only 8% of white recipients, and yet the voucher program utilizes only about 6% of all units with rents below the HUD-designated Fair Market Rents (US Department of Housing and Urban Development 2003). This study could not determine the units actually available to interested voucher users, of course. If landlords are unwilling to rent to them, for example, rent levels do not matter much.

Voucher holders typically cluster in moderate-to-high-poverty neighborhoods of housing markets (Feins and Patterson 2005; Newman and Schnare 1997), sometimes in distinct corridors or “hot spots” where affordable rental housing tends to be more abundant and minority concentration high (Hartung and Henig 1997; McClure 2001; Wang and Varady 2005). At least some of these areas are poorer neighborhoods or racial ghettos that are relatively vulnerable to decline (Galster, Tatian, and Smith 1999; Varady and Walker 2007). According to HUD (2000), as of Census 2000, voucher recipients in the five MTO cities – that is, the overall program populations, beyond the relatively small population of MTO participants at each site – lived in a census tract that ranged from 71% minority, on average, in Boston to 91% minority in Chicago. We return to the issue of voucher concentration and submarkets in the “Results” section.

But what are voucher users' neighborhood trajectories over time, that is, considering those who remain on housing assistance? Analysis of the nearly 630,000 households that entered the Housing Choice Voucher program between 1995 and 2002 indicates that most lease-ups were in neighborhoods with poverty rates of about 20% (about the same as the pre-program neighborhoods for those who leased up in new units) and that, over subsequent moves, voucher holders tended to make only modest improvements in neighborhood traits such as poverty rate and percent minority (Feins and Patterson 2005). The analysis also found that the voucher households most likely to move repeatedly are black, lower income, with younger children, and households living in moderately poor (20–39% poor) neighborhoods, not low-poverty or high-poverty neighborhoods. Yet if they moved, black households experienced larger mean neighborhood improvement than whites or Hispanics.9 Similarly, using the geocoded PSID data linked to data on housing assistance receipt, Newman and Harkness (2000) found that MTO-like moves, from high- to low-poverty areas, are extremely rare for those on housing assistance (about 6% of all moves); nearly two-thirds (62%) of moves are from one extremely poor neighborhood to another.

What explains these patterns? Some researchers, typically using structured surveys of clients, have focused on voucher users' preferences and resources, as well as the supply-side barriers they face in the marketplace. As for preferences, research has largely been confined to identifying priorities: Safety and proximity to relatives and friends rank particularly high for assisted households, and there is some evidence that these are threshold concerns – more important, on average, for clients than good schools or proximity to job locations (Basolo and Nguyen 2006; Johnson 2005; Priemus, Kemp, and Varady 2005). Voucher holders – very low-income family, senior, and disabled households – also tend to identify proximity to public transportation as a priority; housing counselors similarly tell researchers that “accessibility” or “getting around” are top concerns for their clients, especially those who live in relatively transit-rich central cities and are asked to consider moving farther out (Varady and Walker 2000, 2007). Understanding locational priorities is important for obvious reasons, but so is understanding a willingness to make trade-offs among those priorities, and prior research has had little to say about that.

Research has also emphasized important demand-side barriers, such as: (a) the debilitating mental and physical health problems found disproportionately in the housing-assisted population – including the so-called “hard to house” – where public or assisted housing has been demolished and “vouchered out” in favor of mixed-income redevelopment (Popkin, Cunningham, and Burt 2005; Popkin and Cove 2007; Snell and Duncan 2006; Varady and Walker 2000, 2007) and in recent desegregation programs (Pashup et al. 2005); and (b) limited time, money, transportation, information, and other resources important to effective housing search (Basolo and Nguyen 2006; Pashup et al. 2005). Where information for search is concerned, Varady and Walker's (2007) multi-city study of vouchering out found that voucher holders leaving public housing were more likely to find out about available apartments from friends and relatives, newspaper ads, or real estate listings than from housing counselors.10

Beyond preferences and demand-side barriers, researchers consistently identify a range of supply-side barriers as well. First, there is reported discrimination based on race, family status (e.g. presence of young children), or source of income, that is, the use of the voucher itself. In most states and localities, landlords are not required to accept vouchers, and requirements elsewhere are loosely enforced. There is some evidence that source-of-income discrimination is more prevalent than racial discrimination, at least in some local housing markets (Varady and Walker 2007).

Next, there is the scarcity of affordable and otherwise voucher-appropriate housing units in many communities, especially in tight housing markets where much job growth is happening in America (Basolo and Nguyen 2006; McClure 2006; Pendall 2000). This scarcity reflects the dwindling supply of rental housing affordable to those at the lowest incomes, including the working poor. Some 1.2 million low-rent units (units costing $400 or less per month, including utilities) were lost between 1993 and 2003 (Joint Center for Housing Studies 2006), and the number of households experiencing “worst-case housing needs,” defined by HUD as falling below 50% of area median income and either paying more than half of household income for housing or living in substandard housing (or both), surged by 16%, or some 817,000 households, between 2003 and 2005 alone (US HUD 2007). The tendency of many low-poverty jurisdictions to exclude such housing is at issue here as well. That is, the scarcity problem reflects the geographic concentration of accessible supply, not just the limited volume of that supply (Briggs 2005; Pendall 2000).

Finally, market conditions also shape outcomes: As of 2001, voucher recipients in very tight markets were about 20% less likely than those in loose ones (61% vs. 80%) to lease up anywhere (Finkel and Buron 2001), and market tightness, as we will show, was especially important for MTO families who wanted to stay in low-poverty areas but had to move repeatedly. Yet program capacity also matters: Research on certain less studied markets – such as Alameda County, California, where local housing agencies are well-managed – shows relatively high lease-up rates over the long run even among families who used vouchers in unfamiliar suburbs, where the rental market was often tight (Varady and Walker 2007).

With the exception of Gautreaux, where long-run administrative data are available, and of Varady and Walker's survey evidence on “returnees” in the Alameda sample (those who initially relocated to suburbs but later returned to the city), research on these issues has focused on initial relocation outcomes. There is no empirical research we know of on the important question of how supply- and demand-side factors interact over time to shape locational outcomes for particular voucher users – an interaction that is best understood, we argue, when housing choices are viewed in the context of families' larger life strategies and challenges, that is, more holistically than conventional survey studies of household priorities and locational outcomes can do. Before we outline our approach to these research gaps, we briefly review the distinctive policy dilemmas facing assisted mobility interventions such as MTO.

Policy design and implementation dilemmas

Assisted mobility programs confront important dilemmas about which clients to target, how to operationalize “choice,” which locations to target, and how to implement effectively. First, it is not clear that the most disadvantaged populations – those that are not only income poor but face barriers to life functioning in the form of chronic physical or illness, substance use, or other problems – are well suited to assisted relocation. At least, such hard-to-house populations may not be suited to relocation strategies right away and not without intensive social services or other post-relocation supports (Briggs and Turner 2006; Popkin 2006). To date, most attention has focused on the rigors of involuntary relocation by these extremely disadvantaged households, such as when public housing projects are demolished, but significant barriers to functioning are also evident in MTO, wherein families volunteered for the chance to escape public and assisted housing in high-poverty neighborhoods. These major barriers were highlighted in the early assessment of counseling challenges in MTO (Feins, McInnis, and Popkin 1997) but largely ignored in research on MTO thereafter.

Second, given the range of constraints faced by assisted households, simple “choice” may never be enough to dramatically change locational outcomes – and some not-so-simple alternatives pose legal and political dilemmas of their own. Most local housing programs appear to lack the will and/or the way (capacity) to inform voucher holders' choices about the range of neighborhoods that have affordable, eligible units in them; the focus is on leasing up affordable units quickly (Johnson 2005; Katz and Turner 2001; Varady and Walker 2007). McClure (2006) argues that in the tight markets where voucher holders struggle most, it may be that “intensive housing placement” – à la Gautreaux, wherein placement counselors “lined up” the units in racially integrated communities – and not simply helping families search, is the key to lowering poverty concentration and racial segregation in the voucher program.11 Also, families' unwillingness to make particular kinds of moves might be a major determinant of MTO housing trajectories and locational outcomes over the long run. This was an issue for initial lease-ups in both Gautreaux and MTO (Feins, McInnis, and Popkin 1997; Rubinowitz and Rosenbaum 2000), as well as the major desegregation consent decrees of the 1990s (Briggs 2003; Popkin et al. 2003).

Third, as for which locations to target, voucher users and policy analysts and advocates may not be on the proverbial same page in terms of what neighborhood “quality” means. Poverty rates, racial composition, local school performance, and other measures are obviously proxies for the value of a particular residential location, and as noted above, safety and proximity to loved ones may be the dominant considerations for most assisted households. In some instances, these factors conflict with the aim of improving locational outcomes, for example, when families feel torn between seeking safer neighborhoods (which may be further away) and staying close to relatives (which often means moving nearby).

Fourth and finally, effective implementation is a challenge. Because the success of voucher-based assisted housing mobility programs, like that of the voucher program generally, hinge on a chain of cooperative action by landlords, tenants, housing agencies, and sometimes organized interest groups, Briggs and Turner (2006, 59) conclude, “This element of the nation's opportunity agenda is particularly vulnerable to the strong-idea-weakly-implemented problem.” Given the risk of NIMBY-ism and other sources of resistance, as well as a history of limited cooperation among local housing agencies in each metropolitan housing market, implementing effectively at scale becomes a particularly challenging prospect (Goering 2003; Polikoff 2006).

Methodological background: why launch a social experiment? What is MTO testing?

Mobility patterns and locational outcomes were important but intermediate concerns – means toward an end – when MTO was launched. The experiment's intended end outcomes were improvements in the well-being and economic prospects of participating children and families, that is, improvements in education, economic self-sufficiency, health and mental health, youth risky behavior, and other domains. In this section, we highlight selected features of the MTO design, which is well documented elsewhere (e.g. Orr et al. 2003), and how the experiment has evolved as a window on housing choice and locational outcomes.

In 1994, MTO's local program managers invited very low-income residents of public housing and project-based assisted housing to participate. All were in high-poverty neighborhoods of Baltimore, Boston, Chicago, Los Angeles, and New York. The mean baseline location was a striking 56% poor in 1990 – much higher than the 40% threshold that analysts have used to define extreme or “ghetto” poverty concentration (Jargowsky 1997). Over 5300 families applied, and just over 4600, 93% of whom were black or Hispanic, met payment record and other basic eligibility requirements. Those families were randomly assigned to one of the three treatment groups: a control group (families retained their public housing unit but received no new assistance), a Section 8 comparison group (families received the standard counseling, centered on how to lease up but not emphasizing the benefits of choosing low-poverty areas, and standard voucher subsidy, for use in the private market), or an experimental group. The experimental-group families received relocation counseling (focused on opportunities to live in low-poverty areas) and search assistance (often in the form of accompanied visits and transportation to vacant units); the supplemental services provided and the specific roles played by public versus nonprofit housing agencies varied considerably across the sites (Feins, McInnis, and Popkin 1997). The experimental group also received a voucher useable only in a low-poverty neighborhood (less than 10% poor as of the 1990 census), with the requirement that the family live there for at least a year. After the initial placement, no families received additional relocation counseling or special assistance from the program, nor did any face program-imposed locational restrictions after the first year. So there was no feature of the demonstration to specifically encourage families to choose another low-poverty neighborhood if and when they moved on.

Of the 1820 families assigned to the experimental group, just under half (47% or 860) found a suitable apartment and moved successfully (leased up) in the time allotted, becoming the program's “compliers” – a 20% improvement over the Gautreaux program. Those who did not successfully lease-up are noncompliers (still members of the experiment group: MTOX in our plots). The experimental-group families most likely to lease up had fewer children, access to a car, more confidence about finding an apartment, greater dissatisfaction with their origin neighborhood, and no church ties to the origin neighborhood; a looser rental market and more intensive counseling services were also significant predictors of success (Shroder 2003). The program's early impacts included dramatic improvements in neighborhood poverty rates and participants' reports of safety and security – but not rates of racial integration (Feins 2003).

Housing tenure, assistance receipt, and the frequency of moves, which are both cause and effect of broader housing choices and opportunities, did not change significantly for any of the treatment groups. By the interim mark, that is, the evaluation point four to seven years after random assignment, about 70% of MTO households continued to receive some form of housing assistance, about 90% (in all three groups) were still renters, and the low-income renters who dominate the MTO patterns showed the comparatively high rates of residential mobility that characterize poor renters nationwide; there were no substantial differences among treatment groups in length of time residing in their current housing unit (Orr et al. 2003, 61–66, C-16, D-1). Members of the experimental group were still much more likely to report feeling safe, to report less social disorder in their neighborhoods, and to report feeling satisfied with their neighborhoods (Orr et al. 2003, 42).

But what, in fact, is the treatment, and what is MTO testing? Like other social experiments, MTO has evolved in the real world and not under controlled laboratory conditions. First, 67% of the experimental complier group had moved at least once more by the interim mark, and, according to Orr et al. (2003), that group was only half as likely (18% vs. 38%) as compliers who stayed put to be living in a neighborhood less than 10% poor. The most common reasons for compliers' moving on were involuntary: problems with the lease (22%), which may include failed unit inspections, rent increases, and decisions to sell the unit or for other reasons not to renew the voucher holder's lease; and conflicts with the landlord (20%). But almost as many families (18%) reported wanting a bigger or better apartment.

As we noted above, MTO's program content helped families get to particular kinds of neighborhoods, not stay in them or move to similar neighborhoods over time. Still, a strong desire to stay in similar neighborhoods was evident by the interim mark, when two-thirds of those who had moved since initial relocation reported searching for housing in the same neighborhoods. The larger point is that additional moves introduce additional family-level selection effects on locational outcomes over time, making it difficult to attribute particular outcomes to the intervention as one moves beyond treatment-group differences to analyze outcomes for distinct subgroups within the treatment groups (as we do). For this reason in particular, most of our results are descriptive in nature, not presented as unbiased estimates of treatment effects or “program causality” (see Kling, Liebman, and Katz 2007).

Second, many of the low-poverty areas that served as initial destinations for the experimental group have changed over time, through no choice of the participants. Census data show that while most were relatively stable in terms of income levels, almost half (45%) were becoming poorer in the 1990s (Orr et al. 2003), even as many inner-city neighborhoods were becoming less poor. Third, about 70% of the control group had also moved by the interim mark – most to other poor neighborhoods but with a mean reduction in neighborhood poverty rate from 51% to 34% (when compared to control-group members who did not move). One reason for these moves was public housing demolition and revitalization programs, HOPE VI most importantly, which received a major boost from federal policy and local mayors and advocates just as MTO was starting up. But the key point is that many members of the MTO control group are movers too, some with vouchers, about one-quarter living in neighborhoods below 20% poverty by the interim point, rather than members of a fixed-in-place comparison category.

Still, at the interim point that preceded our fieldwork by about two years, families in the MTO experimental group were about 13% more likely than the control group, and experimental compliers 27% more likely, to be living in very low-poverty areas; the experimental group had also lived in such areas for longer periods of time (Orr et al. 2003, 42). The experimental group has had an exceptional experience vis-à-vis the dominant pattern for low-income housing assistance nationwide. MTO is thus a test of at least two important things for families who used to live in high-poverty public housing and project-based assisted housing: (a) the experience and effects of living in much lower-poverty neighborhoods over some period of time; and (b) the experience and effects of relocating, after initial counseling and search assistance, to low-poverty neighborhoods, and, in some cases, relocating again to a range of neighborhood types, while raising children and handling other life challenges.

Data and method

The Three-City Study of Moving to Opportunity was designed to examine key puzzles that emerged in the MTO interim impacts evaluation. Based on HUD authorization of our team and another research team, we conducted our study in three of the five MTO sites – greater Boston, Los Angeles, and New York – but for comparison, conducted certain analyses using data on greater Baltimore and Chicago as well. We focused on “how” and “why” questions: To better understand what statistical analyses of close-ended surveys have been unable to explain, we employed a mixed-method strategy that included extensive qualitative fieldwork. Qualitative approaches are particularly important for (a) understanding why participants in social programs make the choices they do and (b) understanding significant variation in outcomes within and not just between treatment groups. But these aims are distinct from (c) making causal claims about the effects of the treatment.

The study featured three main components: statistical analyses of changing metro areas and neighborhoods, using the 1990 and 2000 censuses plus administrative data (for types used in this paper, see sources in table notes), and mapping for key themes; a large random sample of qualitative interviews; and a random subsample selected for follow-on ethnographic fieldwork. This paper includes a fourth component as well: statistical analysis of geocoded MTO interim survey data, obtained under special agreement with HUD and Abt Associates, to examine housing trajectories and locations over time.

Our family-level qualitative data, from the interview and ethnography components, were collected in 2004 and 2005 – about six to 10 years after families' initial placement through the MTO program and about two years after the interim survey data were collected. First, we interviewed 122 randomly selected families, conducting a total of 276 semi-structured, in-depth qualitative interviews with parents, adolescents, and young adults in all three treatment groups, including compliers and noncompliers in the experimental and Section 8 comparison groups (sampling randomly within the stratum of families who had an adolescent child resident in the home at the time of the interview). Overall, we conducted 81 interviews in Boston, 120 in Los Angeles, and 75 in New York, offering monetary incentives (for details on coding, interview length and content, and more, see Briggs et al. 2008). The combined cooperation rate (consents as a share of eligible households contacted) for the interviews was 79%, and the response rate was 70%.12 The sample covers the full range of outcomes for all three MTO treatment groups and both complier statuses, a key to generating representative results.

To enhance validity and extend our data on priority themes, the ethnographic fieldwork added direct observation to what participants reported about their attitudes, choices, and outcomes. We did “family-focused” ethnography (Burton 1997; Weisner 1996), visiting a subset of the interviewed families (n = 39) every few weeks, over a period of six to eight months, for about two hours at a time (for a total of 430 visits). To select these 39, we sampled randomly within the 122 interview households, oversampling for “locational success” among experimental compliers (where success was defined as residing in a low-poverty neighborhood at the time of the interview). The response rate for this subsample was 70% (households that consented initially and with whom we completed data collection on each “core construct,” see below). We provided monetary incentives (cash and/or gift cards) for both the formal interviews and ethnographic field visits.

Unlike more established traditions in ethnography – for example, community, in-school, or peer-group studies – family-focused ethnography centers on developing rich, valid accounts of family-level decisions and outcomes, including efforts to support and advance children, parents, elders, or other family members (Burton 1997). The fieldwork focused on the core constructs of families' lives, such as a daily routines to “get life accomplished” (Burton 1997), important social relations, and the details of engagement (or lack of same) in their neighborhood of residence and other neighborhoods (such as those where relatives or close friends lived). The fieldwork was a blend of naturalistic (unstructured) interviewing, semi-structured interviewing, and direct observation of family life inside and outside the home.

Chi-square analyses confirm that both samples are quite representative of the much larger population of MTO families surveyed at the interim mark in terms of background traits, employment status, and a range of other social outcomes ( Table 1). In terms of sample outcomes, we modestly under-represent Hispanics and over-represent families on welfare in the ethnographic component (p < 0.05). Based on refusal reports, nonworking parents were somewhat more available for repeat visiting, but they may also have been more enticed by the monetary incentives we offered.

Table 1. Descriptive Statistics: Three-City Interview and Ethnographic Samples Compared to Interim Impacts Evaluation Sample.

	Interim Evaluation Sample	Qualitative Interview Sample	Ethnographic Subsample
	BOS-LA-NY only
	(n = 2,720 households)	(n = 122 households)	(n = 39 households)
Demographic traits
Adult female	98%	98%	95%
Children female	53%	53%	49%
Household head is black non-Hispanic	43%	54%	46%
Household head is Hispanic	46%	35%	39%
Family size: 2–4 children under 18	64%	68%	62%
Locational and social outcomes
2000 Neighborhood poverty rate (2002 locations)
Living in a neighborhood less than 10% poor	8%	10%	14%
Living in a neighborhood more than 30% poor	58%	50%	35%
Residential mobility: Moved 1 to 3 times	58%	61%	69%
Adult ever completed high school	37%	41%	36%
Adult employed	52%	55%	44%
Total household income (mean)	$16,703	$18,514	$16,278
Respondent or child receiving TANF	31%	40%	53%
Adult Body Mass Index (BMI, mean)	30	30	30
Adult psychological distress index (mean)	0.33	0.32	0.35
Child psychological distress index (mean)	0.27	0.21	0.22
Child ever arrested	12%	16%	28%
Youth risky behavior index (mean)	0.40	0.36	0.38
Note: Column headers indicate total sizes for each sample; cell sizes vary minimally due to missing data. Locations are as of the 2002 Interim Impacts Survey (Orr et al. 2003). “TANF” is Temporary Assistance to Needy Families. BMI of 30 or higher indicates obesity. When our qualitative sample and subsample are compared to the survey sample (which is most representative of the program population), statistically significant differences are found only for race/ethnicity and welfare receipt.

Consistent with a mixed-method, and not just multi-method, research strategy, we triangulated our data analyses both within and between key components of the study, for example: between the ethnographic fieldnote analysis and statistical survey analysis to understand trajectory types, and within the ethnographic analysis to understand important social relations as reported by MTO participants and also observed directly by our fieldworkers. The integration of distinct types of data to answer central research questions is crucial for generating richer, more valid results and actionable specifics to guide decision-makers. Mixed-method approaches are also crucial for building better theory, over time, from a base of complex and mixed results.

But we caution the reader about the need to appropriately interpret the different types of data. For example, the ethnographic field data, while drawn from a random sample that generated wide range in the phenomena under study, follows a case study logic rather than a sampling logic. The case-study approach allows us to understand family circumstances as integrated constructs – families as cases that are revealing for the conditions that covary within them – without indicating how common those constructs are across the program population as a whole (Ragin 1987; Small 2008; Yin 1994). Survey results, if the measures are valid, tell us what we can reliably conclude about a large population but with little insight into the underlying social processes of interest. On the other hand, ethnographic and other qualitative methods provide the depth and texture that illuminate such processes – housing choices as the subjects themselves perceive and make them in a social context, for example – but typically without precise population inferences. The results are not less “true” where we cannot indicate, with precision, what share of the larger population particular cases represent. Put differently, small-N results can be “big” (in importance), but this does not settle the issue of how prevalent they are.13

Results

Two main research questions guided this study. First, beyond short-run success or failure at finding units in particular kinds of neighborhoods, what are the neighborhood trajectories over time for families served by assisted housing mobility? Second, how do housing supply and demand-side factors interact over time to shape those trajectories?

We begin by placing MTO relocation patterns in their metropolitan contexts, including: (a) the large-scale population settlement shifts that MTO relocations paralleled to a significant degree; (b) the spatial patterns for the full population of voucher holders (not just MTO participants) in each housing market, that is, as local benchmarks for the MTO treatment groups; and (c) the significant differences, between MTO metro areas, in the distribution of neighborhood types (for example, the number and share of neighborhoods in a given metro that featured concentrated minority poverty versus the prospect of racial or economic integration). Then we examine the housing trajectories of MTO families and the supply- and demand-side factors that shaped their locational choices and outcomes.

Relocations in metropolitan context

Earlier MTO research has focused on mean locational outcomes in terms of census traits (Feins 2003; Orr et al. 2003). But tracts lie within larger zones – such as transitional inner-ring suburbs – that often change in distinctive ways as metro areas change. We begin by describing these subareas as MTO destinations, including voucher-clustering patterns. We grouped relocation outcomes into rings according to distance from the central business district (CBD) of each metro, measuring the distance from the CBD core to the centroid of each tract in the metro, then grouping the distance measures into rings corresponding to the first, second, and third thirds within the central city, as well as thirds within the suburbs. Table 2 reports on compliers only, and due to limitations on the administrative data available, it presents 2004 locations for “all voucher holders” alongside 2002 locations for MTO participants.

Table 2. Neighborhoods of MTO Compliers: Urbanicity, Distance from CBD, and Clustering Rates (five sites).

	Experimental-Group Compliers			S8 Comparison Compliers	Metro area
	Origin neighborhood	First relocation	Interim (2002) Location	2002 Location	All vouchers 2004	All households 2000
Number of households	–	815	695	1,027	301,245	11,848,989
Number of tracts	117	380	505	617	7,771	7,771
% households by city/suburban location
Central city	100	62	75	86	56	42
Suburbs	–	38	25	14	44	58
% by location in central city
City CBD + Ring 1	52	9	28	30	24	26
City, Ring 2	31	26	32	33	38	30
City, Ring 3	16	65	40	37	39	44
City total	100	100	100	100	100	100
% by location in suburbs
Suburbs, Ring 1	–	35	40	52	36	27
Suburbs, Ring 2	–	53	37	38	31	33
Suburbs, Ring 3	–	12	24	10	33	40
Suburbs total	–	100	100	100	100	100
% households by Section 8 voucher holders as % households in tract
< 2%	71	56	50	21	38	76
2 to 4.9%	14	31	22	33	30	16
5 to 9.9%	12	11	17	28	22	6
10% or more	3	2	11	18	10	2
Total	100	100	100	100	100	100
Source: Tabulations of Geocoded MTO Interim Impacts Evaluation Survey and Tracking data (provided by Abt Associates); US HUD Housing Choice Voucher Program data.
Note: Locaton rings represent concentric thirds, grouping census tracts by the distance from the CBD to tract centroids.

Relocation zones

Whereas the Gautreaux program's desegregative lease-ups in suburban communities placed families in middle-class, mostly white areas 15–20 miles from their mostly black origin neighborhoods in Chicago, in all five MTO metros, the first and only assisted relocation made by MTO experimental compliers was typically to a low-poverty, majority–minority neighborhood in the outer ring of the central city (about two-thirds of all compliers) or in an economically diverse inner suburb (by which we mean an area with a wide range of income levels) proximate to the central city (about one-third), not to more distant, affluent, or racially integrated communities.14

By the interim mark, the distribution of MTO voucher holders across rings, in both treatment groups, matched that of all voucher holders in those cities at roughly the same time. But experimental-group compliers remained more dispersed: Half were in census tracts where fewer than 2% of the households held vouchers, compared to just one-fifth of the Section 8 comparison-group compliers and 38% of all voucher holders in these metro areas.

What area-wide changes affected the zones MTO families lived in? A doubling of the suburban poor population in the 1990s, for example, was concentrated in older, inner-ring suburbs, where minority suburbanization also tends to be concentrated (Orfield 2002). By 2005, the suburban poor in the nation's largest 100 metro areas outnumbered the poor in their central cities by more than one million persons (Berube and Kneebone 2006). And a resurgence of housing demand in central cities pushed rents and sales prices upward much faster than incomes in many urban neighborhoods.

We will focus briefly on the market-specific character of these changes at our three study sites alongside the MTO-triggered “starting points” (initial relocations) for the experimental compliers, since policymakers' hopes were highest for them. In Gautreaux, housing counselors acted as placement agents, lining up units, in a relatively loose market, that were offered on a take-it-or-leave-it basis to those on the waiting list. In MTO, where clients would choose their units, supply-side constraints quickly led to a variety of local compromises: Early assessment suggested that while all five sites tried to expand the pool of participating landlords, limited staff capacity and limited payoff curtailed such efforts (Feins, McInnis, and Popkin 1997). Counselors found their pre-existing landlord lists most “productive” as sources of vacancies. Also, vacancies for certain types of rental housing were not advertised and thus were difficult to learn about through mailers and other conventional outreach. Finally, rental brokers provided shortcuts to landlords who were willing to accept vouchers, at least at some sites. Boston program staff estimated, for example, that 20% to 25% of their placements were secured through brokers.

Relocation vis-à-vis metropolitan restructuring

New York City compliers were concentrated initially in small rental properties in the Northeast Bronx – where the nonprofit placement agent's landlord contacts were concentrated and where vacancies were numerous – with a handful moved to Staten Island, having relocated primarily from public housing in Central Harlem and the South Bronx. This program-induced mobility tracked a larger movement of people of color, including middle-income black and Hispanic homebuyers, to the city's outer core and inner suburbs; Harlem and Brooklyn, the historic centers of black settlement, gentrified and became home to more whites, while the South Bronx became ever-more Hispanic thanks largely to immigration (Furman Center for Real Estate and Urban Policy 2005). But deep pockets of poverty and black and Hispanic concentration remained in those three areas, where many of the city's most affordable rentals are concentrated; according to HUD, inflation-adjusted gross rents jumped 23% in New York City from 1990 to 2005.15 Finally, a large-scale black migration out of the New York region, mostly toward the Southern US, where the cost of living is much lower, accelerated in the 1990s and 2000s (Frey 2006). A handful of MTO families tracked that migration as well.

Boston MTO families relocated from the inner-city neighborhoods of Dorchester, Roxbury, and South Boston mainly to small rental properties in the city's economically diverse outer core neighborhoods or to transitional inner-suburb communities on the North and South Shore, such as Brockton, Quincy, Revere, and Randolph. Those suburbs became poorer and more racially diverse in the 1990s as the job-rich western suburbs along the Route 128 high-tech corridor, where school districts are strongest, remained overwhelmingly white and middle- to upper-income (McArdle 2003). Figure 1 shows the 2002 locations of MTO voucher households and the growing poverty in these outer-core and inner-ring suburban areas on the region's north-south axis (the “stacking” of data points in central Boston masks the drop in poverty in inner-city neighborhoods). One experimental complier we interviewed in suburban Boston conveyed her perception of growing poverty concentration quite bluntly: “I left the ghetto, but the ghetto followed me.” As of the 2000 Census, three-quarters of metro Boston's poor lived in the suburbs. And Stuart (2000) found that half of the home purchases made by black and Hispanic homebuyers outside the central city between 1993 and 1998 were made in just seven of the metro region's 126 municipalities; those seven were relatively affordable towns, where poverty and fiscal distress grew in the 1990s and where school district performance is much poorer than the affluent suburbs. Meanwhile, many central-city neighborhoods gentrified dramatically; gross rents jumped 15% in real terms between 1990 and 2005, reports HUD, and some neighborhoods saw much bigger increases.

Figure 1. 2002 Metro Boston MTO housing locations by tract poverty rate change 1990–2000.

In Boston and New York, MTO experimental compliers left behind high-poverty and high crime but transit-rich areas that were close to the CBD for more car-reliant areas with dispersed services and job locations. In sprawling Los Angeles, their counterparts left inner-city neighborhoods in South and East LA for transitional southern suburbs nearby (e.g. Compton and Lynwood), as well as communities in the sprawling San Fernando Valley (about 15 to 30 miles from origin, to the north), Long Beach to the southwest, and rapidly expanding and increasingly diverse eastern suburbs and satellite cities 40 to 60 miles from origin, mainly in adjacent Riverside and San Bernardino Counties – the once-agricultural “Inland Empire” where many low- and moderate-income Angelenos have moved in response to the city's desperate shortage of affordable housing. For the MTO families that lacked reliable access to a car, transit options in most of these destination communities were poor.

Metro Los Angeles saw a large outmigration of non-Hispanic whites, together with rapid immigration from Asia and Latin America, throughout the 1990s and into the new decade (Frey 2006), as well as growth in extreme poverty concentration, counter to the national trend (Jargowsky 2003). Closer to the streets, those aggregate changes were reflected in dramatic patterns of ethnic succession and competition as well (Zhou and Myers 2006). For example, many long-black neighborhoods in South LA – known as “South Central” until the 1992 riots inspired a name change – became mixed areas of poor black and Hispanic, or even majority-Hispanic and Spanish-language-dominant settlement.

Housing market trends

MTO has operated in some of the nation's costliest housing markets. Four of the five MTO sites had tight housing markets before the demonstration began, and they remained significantly tighter than the national average over the course of the demonstration (Figure 2).16 In 1990, only Chicago's vacancy rate essentially matched the national rate, with Baltimore and Boston close behind. LA and New York were, even at that recessionary point, at or below the 4–6% vacancy rates estimated for the typical rental market's “natural” vacancy rate, that is, equilibrium corresponding to no downward or upward pressure on real rents (Gabriel and Nothaft 1988, 2001). By the end of the decade, as rental markets grew tighter in many metros nationwide, vacancy rates plummeted in all five MTO metros and most of all in those places that began the decade as tighter markets. Greater Boston, LA, and New York – our study sites – became extremely tight (rate < 3–4%), and in the LA case, the trend toward an ever-greater scarcity of vacant rentals persisted into the new decade, right through the latest recession.

Figure 2. Vacancy rate for MTO housing markets, 1990–2004.

As Figure 3 shows, the HUD-defined fair market rent (FMR) trends mirror image those vacancy rate trends: While the increases were not monotonic, our three study sites began and remained the most expensive of the five MTO metropolitan housing markets, clustered in the $1100 to $1300 per month range by 2006. LA saw gross rents jump 13% between 2000 and 2004 alone, compared to the national increase of 6%, while Boston and New York saw sharp increases (9%) as well (data not shown). By 2004, when we visited MTO families, the rental markets in Los Angeles and New York were imposing very widespread hardships. Here, we must rely on county-level data from the 2004 American Community Survey: More than half of all Los Angeles and New York City renters (54% and 51%, respectively), and nearly half in Baltimore, Boston, and Chicago (48%), paid more than 30% of their income for housing, compared to the national average of 44%. Each of the five MTO markets saw an 8% jump in that hardship rate between 2000 and 2004 alone – twice the national increase.

Figure 3. Fair Market Rents (FMRs) for MTO housing markets, 1994–2006.

Neighborhood traits by MTO metro area

Here, we consider differences in the distribution of neighborhood types – an important arbiter of locational opportunity – between the metro areas. Though some trends, such as sharp increases in real rents and the suburbanization of poverty, affected all five MTO metros, several notable differences remained among them in the typical neighborhoods experienced by participants in the demonstration by the interim mark and by all other households in the metro area, as Table 3 shows. For example, the share of experimental-group compliers living in neighborhoods with poverty rates below 20% ranged from a high of 71% in metro Boston to just 45% in metro Los Angeles; even members of the control group in Boston were three times more likely to be living in such neighborhoods than were their counterparts in Los Angeles and four times more likely than counterparts in New York. While these outcome differences may owe in part to site differences in program effects – for example, effects of differential approaches to counseling or assistance early on – these locational outcome differences in 2002 track the sharp differences in mean exposure to poverty for all households in those metro areas. That is, there are fewer housing opportunities in low-poverty areas overall in metro Los Angeles and New York than in the other three MTO metros.

Table 3. MTO Neighborhood Locations, 2002, by Treatment Group and Site (metro area).

	All sites	Baltimore	Boston	Chicago	Los Angeles	New York
Percent in tracts with poverty rate <20%
Experimental Group (all)	35	39	44	31	31	26
Experimental compliers only	59	57	71	64	45	57
Section 8 Comparison compliers	26	29	33	21	22	12
Control Group	17	26	23	19	7	5
All metro households (2000)	78	86	90	85	67	63
Percent in tracts with share minority <60%
Experimental Group (all)	18	21	42	7	12	8
Experimental compliers	27	29	61	13	18	12
Section 8 Comparison compliers	16	19	38	5	5	7
Control Group	13	13	34	3	4	6
All metro households (2000)	69	80	93	74	44	53
Source: See Table 2.
Note: Control group N = 1133; Row N's here are column N's in Table 2. For the experimental group, “all” includes both compliers and noncompliers, as explained in text.

A parallel pattern emerges for racial mixing: Boston is the standout in terms of MTO locational outcomes at the interim point (with more families in comparatively mixed neighborhoods), and the striking gap between the two metros with the lowest rates (LA, NY) tracks the big gap in racial exposure for all households when those two markets are compared to the other three. In metro Boston, 93% of all households live in neighborhoods with moderate or lower minority concentration, while fewer than half of all households do so in metro New York and Los Angeles.

Summary

Initially and for four to seven years after, the locational quality secured by MTO's experimental-group compliers was far better in multiple dimensions than that in origin neighborhoods but – for reasons that are still uncertain – less and less distinct from that of the unassisted, voucher-holding complier group in the demonstration and also less distinct from the universe of all voucher locations in these metro areas. In additional analyses we have not presented here, the trends reshaping that larger set of locations – a decline in rental housing affordability, shifting patterns of racial and economic segregation, a spike in subprime lending targeting poor and minority neighborhoods, sharp declines in city crime rates, the shift to majority–minority make-up in the central city in some cases (Boston) and the metro in others (LA, NY), and more – indicate how dramatically the geography of housing opportunity was changing around the MTO families over time. But again, that geography – particularly with respect to the number of racially mixed, low-poverty neighborhood options – remained very different across the three MTO metro areas on which our study focused.

Over time: what household trajectories led to the locational outcomes observed?

This section addresses our first main research question: what were the neighborhood trajectories over time? We cannot interpret the outcomes of MTO in a meaningful way without understanding the trajectories that led to those outcomes.

Against a changing distribution of locational quality, MTO families have been mobile – but not necessarily more so than low-income renters as a whole. On average, experimental and Section 8 families moved the same number of times (2.6 moves), while control-group families moved somewhat less often on average (2.1 moves; Orr et al. 2003, C-16).

How did the experimental compliers fare and why? Figure 4 shows locational trajectories for the select group of experimental compliers who leased up in areas that were very low-poverty in 2000 (10% or less poor). Most initial relocation happened in the latter half of the decade, though the program employed the available 1990 tract poverty data. We employed Census 2000 rates for tract poverty and thus: (a) isolate transitions across neighborhood poverty levels that reflect residential mobility from any that owe to substantial change in neighborhood poverty levels; and (b) focus on those compliers who were positioned, through their initial relocation, to receive the treatment planners intended (prolonged exposure to a neighborhood that retained a very low-poverty rate). This is revealing for descriptive purposes but clearly does not allow us to attribute treatment effects, given the selectivity. We further limit this reporting to cases for which valid address data were available at multiple observation points: at initial relocation, in the year 2000, and again at the interim evaluation point in 2002.17 To describe representative patterns for this group, we limit neighborhood types to three levels of poverty: very low (<10%), low (10–20%), and moderate to high poverty (more than 20%). This consolidation does obscure exposure to the extremely poor areas (over 40% poor) that most experimental compliers left behind at first relocation but captures most of the variation in trajectories over time.18

Figure 4. Locational trajectories for MTO experimental complier households that initially relocated to neighborhoods that were very low poverty in 2000 (N = 193).

Four trajectories are evident in Figure 4. Type 1 households are those that managed to stay in very low-poverty tracts at all three observation points, whether they moved or did not move, while type 2 households remained in tracts that were low poverty (or better). Types 1 and 2 are distinct, then, only as a matter of degree. Type 3 households “bounced” from the initial, very low-poverty neighborhood to a moderate-to-high-poverty neighborhood by the second observation and then back to a very low-poverty tract by 2002. Type 4 households, who reverted quickly, were already in a moderate-to-high-poverty tract by 2000 and remained in one (though not necessarily the same one) when observed again in 2002. The bifurcated pattern is striking: Most households either stayed in a very low to low-poverty tract (40% of these particularly successful compliers, combining types 1 and 2) or moved on within a few years to a moderate-to-high-poverty tract and then remained in that type of tract (56%). Only a small fraction (4%) followed the type 3 trajectory, re-attaining a very low-poverty location after some time in a much poorer one.

A second pattern – the enormous variation by site – is not shown in this aggregate plot. For example trajectory type 1 ranged from 13% for the greater LA site and 16% in New York to 43% in Boston. If types 1 and 2 are combined, those two trajectories represent a slight majority (56%) in Boston, compared to just 34% for New York and 24% for LA. Conversely, type 4 is a large majority in New York (66%) and even more so LA (74%), compared to just over one-third (36%) for Boston. Type 3 is a very small share at each site, from 0% for New York to 8% for Boston. A third and final pattern, also not shown, reflects those who leased up in areas that were very low poverty in 1990 and became low-poverty (10–20% poor) areas in the 1990s – and who remained in those areas, stably, over the observation points. This includes 100 cases total, or one-third of all who leased up successfully in the target census tracts at the three sites.

The first pattern adds to Clampet-Lundquist and Massey's (2006) evidence that initially relocating to a racially integrated tract, rather than simply to a very low-poverty one, was highly predictive of living in both racially integrated and very low-poverty tracts at the interim mark some four to seven years later. Again, removing neighborhood change from the patterns and treating the shares summarized above as odds, relocating initially to a tract with a stable, very low-poverty rate gave MTO households at our three study sites a roughly 50–50 chance, on average, of being in a very low to low-poverty tract at the interim point – but much lower odds in New York and LA.

As for the second pattern, we cannot rule out the possibility that program effects, including initial counseling and assistance differences across sites, help explain site differences in the distribution of trajectories. But per the discussion above, the significant differences by site in the pool of neighborhoods available – in the availability of very low and low-poverty tracts, as well as racially integrated tracts – appears to be a significant contributor as well. The third pattern includes some of the most stably housed families, including those who did not move at all, whose locational outcomes by 2000 reflected the growth in poverty around them.

Now we turn briefly to the comparison group of voucher holders in the demonstration who did not face initial locational restrictions or receive special counseling or search assistance (Figure 5). For this group, typing is more complicated, since starting-point neighborhood poverty levels were much more variable. But one may describe all cases with the three starting points indicated with the frequencies on the left side of the starting points in the graph (e.g. 62% relocated initially to a neighborhood that was more than 20% poor according to the 2000 census). And almost 90% of the cases can be described with the four trajectory types indicated, based on starting point: type 5 households (5% of the total) initially relocated to areas that were very low poverty as of the 2000 census and remained in very low or low-poverty areas (for simplicity, the graph shows the end outcome as “low”); type 6 households (11%) started in those places but reverted to moderate or higher-poverty areas; type 7 (17%) started in low-poverty areas and remained in them throughout the four to seven-year observation window; and the largest group, type 8 (56%) started in moderate or higher-poverty neighborhoods and remained in such areas throughout the window. The remaining 10% of cases reflect more mixed patterns, with small cell frequencies. The path frequencies that appear in the middle of the graph, then, can be read as a measure of how predictive the starting points are for the households' locational trajectories and outcomes over time. In simple terms, just under a quarter (22%) of the Section 8 comparison group households had trajectories similar to types 1 and 2 for the experimental group compliers. But the vast majority either stayed in moderately poor to extremely poor neighborhoods over time or moved back to such areas after initially moving to less poor areas when they enrolled in MTO and exited public housing.

Figure 5. Locational trajectories for all Section 8 comparison group complier households (N = 281).

The MTO interim evaluation reported on the rarity of Section 8 comparison group compliers leasing up in the very low-poverty areas targeted for the experimental group, but this trajectory analysis adds a striking view of stability and instability over time. The only type that is not stable is the small share of cases that relocated initially to very low-poverty areas. By comparison, not a single case that began in a low-poverty area reverted to a higher-poverty one, that is, like the origin tracts, over time (though some moved to very low-poverty areas). Yet such reversion was the dominant pattern for those who relocated to the least poor areas and specifically to areas that were very low-poverty at the end of the decade. Two-thirds of the 47 Section 8 compliers who made that initial relocation had reverted. And the other types reflect stable residence in the kinds of neighborhoods that most voucher users in these metros live in: moderately to extremely poor areas, with Boston showing consistently lower mean poverty rates than LA or New York.

Drawing on our most in-depth source of data on MTO families, the ethnographic fieldwork, Table 4 outlines the social context for observed trajectory types 1 through 4. These distinctions are based on the ethnographic sample of 28 experimental compliers, which included an oversample of those who were living in suburban areas, and so we seek here to shed light on the underlying social mechanisms, not to indicate their prevalence in the program population.

Table 4. Experimental Compliers: Trajectory Types by Circumstance and Choice factors.

Types	%	Description	Residential Mobility rate	Context (circumstance and choice factors)
1 & 2	40	Remained in very low to low poverty neighborhoods	None to low	Strong preference for low poverty areas evident. Either retained original rental unit or managed to find an additional rental(s) in similar areas. Daily routines more centered on neighborhood of residence; kin attachments to inner city limited/bounded. High average neighborhood satisfaction.
4	56	Reverted to moderate to high poverty areas and remained in such areas.	Low to high	Variable neighborhood preferences. Stronger kin attachments and/or obligations in inner-city neighborhoods. Involuntary moves common (apartment sold, unit poor, landlord conflict, etc.), but some moves motivated by preference for better unit, access to loved ones, etc. Lower average neighborhood satisfaction.
3	4	Reverted to moderate to high poverty areas and then again to low poverty area.	High	Variable patterns: life shocks, changes in employment status, and family obligations shifted housing demands. Preference for low poverty areas evident, but opportunities insecure.
(N)	(193)
Source: Three-City Study interview and ethnographic field data. See types plotted in Figure 4.

Not only were the types 1 and 2 families in our ethnographic sample luckier (on average) in the marketplace, but they tended to express particularly strong preferences for “better” areas (defined as safer and more economically diverse than the inner city) and more limited kin attachments and obligations in inner-city areas. It is not surprising, in that context, that their social lives had moved with them – even, in some cases, over multiple moves across a wide geography. In simple terms, these families were both satisfied and well adapted. For example, Roxanne, who lost her apartment in one LA suburb when her landlord opted to sell the property, found out about another “good” neighborhood through a friend. While the new neighborhood was roughly 15 miles away, Roxanne and her family once again centered their lives on the new place. Ditto Sabrina in suburban Boston, who complained about ghetto neighbors moving in from the inner city but focused her children on the safety, recreational programs, and shopping in her suburban neighborhood, not the inner-city neighborhoods where most of her relatives continued to live.

In contrast, type 4 families (“move-backs”) were generally drawn back to living in the inner city through an involuntary move but sometimes through social obligation and preference. Sick or otherwise needy kin loomed large for the most constrained families, whose social lives revolved around relatives and close friends back in the inner city even when the (subject) family resided in a low-poverty area elsewhere in the metro. Though our sample sizes are small, parents in this group also appear less likely than those in type 1 or 2 to have access to cars. This was especially serious for the LA move-back cases who relied on welfare or had unstable jobs. But it applied to a transit-reliant family living in a poor section of Staten Island, too, whose kin support networks were concentrated in the South Bronx.

Some type 4 cases endured not one but a series of bad breaks in the rental market. In New York, Lanelle loved living in a low-poverty area in the Northeast Bronx and chose her neighborhood based on a teacher's recommendation of a strong elementary school there. But when the heat did not work for two weeks in the winter, Lanelle got sick, the housing authority refused to pay the landlord, and Lanelle and her children were evicted. After a brief spell living with her grown son, they found a new place with a great landlord in a moderately poor area. But Lanelle's health problems made the fourth floor walk-up apartment untenable. Through her stepfather, Lanelle learned about a good building near Yankee Stadium in the South Bronx. During our fieldwork, several relatives moved into the building. Though the area is poorer and noisier than the Northeast Bronx, and though no one in the family will walk alone there at night, services, shopping, schools, friends and family, and the subway are all nearby. Marlena, Lanelle's youngest daughter, can walk herself to school and play outside.

Type 3, the rarest trajectory type in the MTO interim survey sample, showed variable patterns along these same dimensions. The few cases in our sample that fit this type struggled to align life goals – which included a better neighborhood for the children – with insecure or inadequate housing and employment opportunities as well as hard-to-reach social support, such as vital childcare provided by a parent or sibling living at a distance. Anique is the single mother of Clara, age 11. She has moved five times since the initial relocation. She initially relocated from the housing projects in South Los Angeles to an apartment in the nearby southern suburb of Gardena. But Anique soon moved back to South LA because she wanted more space and because she worried about Clara living too close to a swimming pool. The new home was larger, but the neighborhood turned out to be too dangerous in the evening. So Anique and Clara soon moved again to a home in Compton, also an inner suburb to the south. Anique's failed attempt to buy this house caused her to lose her housing voucher. But she landed in a job in Riverside County, more than 70 miles to the east, where an aunt and uncle lived and were willing to provide childcare. Anique and her daughter moved there. But before long, their relatives left California, and lacking alternative sources of safe, affordable childcare, Anique moved with her daughter to Long Beach to live with Anique's mother and sister. Then Anique was laid off from her job, needed financial help, and so stayed with her mother. Then she got a new job in Riverside County, so her commute was nearly 80 miles each way, and she left the house at 4 AM each day. By the time of our final fieldwork visits, Anique had scraped together enough money to rent a small one-bedroom apartment across the street from her job. Anique is a revelatory case (Yin 1994): an extremely persistent single mother in the experimental complier group whose job, housing, and support locations remained unstable for a long period of time, challenging her to bring them into alignment.

In the final section of results, we look more closely at the demand- and supply-side factors that shaped locational outcomes, including the nature of preferences and trade-offs, for all three MTO treatment groups.

Demand-side factors and choices

Having examined neighborhood change and its metropolitan context above – that is, factors outside the control of MTO families – we focus here on the two types of choices that families did make to influence their locational outcome: whether to move and where to move. The choices made and those available address both of our main research questions, as we will show. In some instances, we present data on several treatment groups (to allow comparison), while in others, as noted below, we are focused descriptively on the experience of the compliers. The latter is revealing of key social processes and decisionmaking factors in spite of the fact that it cannot indicate treatment effects per se.

Reasons for moving on after initial relocation

Choices about where to move would not be so important if the overall rate of mobility were lower in the demonstration. Corroborating and extending the interim survey results, we find, drawing on the large and representative sample of qualitative interviews, that this moving was for a variety of voluntary and involuntary reasons, the most important of which may be categorized as: dissatisfaction with their housing quality, landlords, or neighborhoods; leasing problems (such as a unit being sold, rented above the voucher program price ceiling, or removed from the voucher program); or life changes such as birth, death, job getting and job loss, divorce, or domestic dispute. Based on our large sample of qualitative interviews, these drove second and additional moves and continued to be the major reasons for moving years after the interim evaluation.19 But we find that only rarely did being closer to loved ones act as a reason for moving for experimental compliers; more importantly, it factored into the assessment of neighborhood options and helped shape important daily routines – around the accessibility of childcare provided by a relative, for example, or where socializing took place.

Section 8 comparison group movers were particularly likely to cite dissatisfaction with their neighborhood – a lack of safety and sometimes noise or more generally “the wrong environment for my children” – as a main reason for moving. This is consistent with the comparison group's greater exposure to high-poverty areas and with their lower neighborhood satisfaction scores over the course of the demonstration.

More than one-third of the reasons cited by experimental complier families, on the other hand, reflected dissatisfaction with their landlord or housing unit. Substandard physical conditions were a major culprit. Either the family chose to move because they were not satisfied with landlord maintenance and repair, or the unit failed to meet inspection standards set by the housing voucher program. Some families reported health problems related to toxic home environments, such as carbon monoxide poisoning and mold, and we observed serious problems firsthand in some units: kitchens overrun with cockroaches as we sat to conduct interviews, heat that barely functioned in the cold winter, and more. But there was dissatisfaction with landlords as well: some families found the lack of privacy too restrictive (e.g. where the landlord lived in the same building), especially when it prevented family and friends from visiting; and some parents could not handle landlords' expectations about keeping their children quiet. We are not arguing right and wrong here, particularly since we do not have data from the landlords, merely noting the factors that MTO heads of household cited as important to them.

There were a few notable site differences, regardless of treatment group. For example, New York families talked about landlord or unit-based “push” factors twice as often as Boston or Los Angeles families. This was especially the case for members of the NY experimental group, many of whom leased up initially with small, live-in landlords in multi-family homes. Conversely, about one-third of LA MTO families who moved on complained more (in both treatment groups) about their neighborhoods being undesirable.

MTO control-group families who had moved from public housing projects either highlighted “pull” factors (such as receiving a voucher outside the MTO demonstration), the decision to leave subsidized housing altogether, the desire to get better units and/or neighborhoods, or push factors beyond their control: most importantly, being vouchered out through HOPE VI or other redevelopment programs that required resettlement.

Understanding locational preferences: avoidance behaviors, hopes, and priority setting

Beyond main reasons for moving, our qualitative interviews and ethnographic fieldwork help shed light on why MTO households made the specific housing unit and neighborhood choices they did, showing how needs changed (for example, as the household grew) and also how a willingness to make tradeoffs among desired outcomes – for example, the willingness to stay or not stay in a lower quality housing unit in order to stay in a safer neighborhood – varied significantly among families.

Prior research on MTO has emphasized neighborhood safety – in particular, the chance to get away from the drug dealing and violence in high-risk neighborhoods – as the primary motivator for participating families' initial relocation (Orr et al. 2003). Years afterward, parents in all three treatment groups continued to emphasize this avoidance factor (as distinct from the attractions of a resource-rich neighborhood), sharing stories of the neighborhoods left behind. But distancing children from what subjects perceived as undesirable “ghetto” behavior was a factor as well, parents recalled. We stress that these are perceptions, years after the fact, of behavior in a challenging social and economic context. Distinct class behaviors invariably lead to perceived social boundaries and often to sharp judgments as well, as ethnographic studies of “street” versus “decent” cultures at play in urban neighborhoods have found for more than a generation (e.g. Anderson 1991; Hannerz 1969; Small 2004; Pattillo 2007). What is important for our purposes is that the perceptions remained distinct and strongly negative, constituting a second avoidance factor in neighborhood choice over time.

Some complier parents in our ethnographic sample recalled loud and frequent partying, hanging out on the streetcorner, gang banging, being confrontational and quick to fight, young girls acting “fast” (promiscuous), and dressing or carrying oneself inappropriately (“ghetto style”). For example, April, a mother in the Boston experimental-complier group who is originally from Haiti, has lived in the same neighborhood since her first relocation. It is an economically diverse area in which, says April, people “are nice, go to work, dress nice.” She contrasted her “suburb” neighborhood (her label) with the “ghetto place” the family moved away from, and she emphasized how much more important these locational qualities were to her than housing unit features:

You know, over there [in the old neighborhood], people are “Blah! Blah!” Loud! The music is high, there's ghetto people. You even hear 8-year old kids F-talking! … You know those kids are trouble … I don't care if people give me $5,000 and I get a big apartment, with three bedrooms or more in [my old neighborhood]. I never want to live in the ghetto. (Fieldnote)

These judgments carried over to compliers who noticed in-migration of poorer families in their low-poverty areas – the “ghetto followed me” dynamic we mentioned earlier. Sabrina, an experimental complier in Boston, explained her intent to move again, though she and her children were well integrated into the routines and institutions of their neighborhood:

Sabrina said, “I'm looking for another apartment, better suited for my needs, something more, on a better street. Not so close to the people around here.” Fieldworker: I asked who she didn't want to be around and she replied, “The ghetto people in the building down the street. The whole building is low-income and everyone is coming from Dorchester [in inner-city Boston]. They bring with them their Dorchester behavior, and I don't like that.” (Fieldnote)

But making these negative judgments about some neighbors did not stop less wary MTO parents from relying on those neighbors for support. For example, Roxanne, an experimental complier in Los Angeles, thinks her neighbors are mostly “ghetto” because they like loud music and “let people hang out.” But she also thinks she could turn to any of them for help in an emergency, in part, she reasons, because they are parents in their thirties and forties, and they care about children. Other experimental complier parents expressed the same sentiment, distinguishing their immediate neighbors' watchfulness and helpfulness from the issue of the neighborhood's character, which they hoped would not turn all poor and all minority.

Avoidance aside, and regardless of treatment group, there were common attractions to particular kinds of neighborhoods and neighbors: living near people who are working and/or “middle-class,” as well as “respectful” or “peaceful,” and – while the emphasis on privacy versus social engagement varied – wanting to live in places where “everybody minds their own business,” which is “a nice family environment.” Some MTO parents specifically emphasized homeownership, neighbors' investment in place, and maturity. As one parent in New York told us, “There's nice places in the Bronx. It's the people. Got to find people who care about the community. I should do my research and find a place with less kids, older people” (Interview).

Social supports

Some researchers have suggested that voucher holders' locational outcomes reflect unmeasured preferences for access to family and friends, as well as familiar local institutions, and not just market or program-based barriers to wider housing choice (e.g. Varady and Walker 2000). But our qualitative data point to a range of types: parents who prioritize proximity to loved ones or cherished institutions, such as a church; those who factor in such proximity but do not make it a priority when deciding where to live; and those who use vouchers to distance themselves from needy or risk-bearing relatives and friends. Cross cutting this variation in priorities was the geography of families' ties: Members of the control group often reported extended family networks in and around their public housing developments, whereas compliers in both treatment groups tended to report strong ties residing at greater distance (outside their neighborhoods). This is consistent with Shroder's (2003) finding that experimental-group members with fewer social ties to the old neighborhood were more likely to successfully lease up in low-poverty areas.

Decades of research on social networks confirm that extensive kin reliance is particularly common among the chronically poor, also that their supportive ties tend to be more localized, limited in number, and strained than those of higher income people (Briggs 1998; Fischer 1977, 1982; Stack 1974). Most MTO families fit this bill, organizing their social worlds around relatives and a few close friends rather than new social contacts garnered in workplaces or – even more rarely – in new neighborhoods. But some chose new residential locations that kept relatives particularly close at hand. For example, Larissa, a Section 8 complier in New York, originally had trouble finding an adequate apartment with her voucher. She was relieved to find one near her mother:

It is down the block from my mother, because I try to stay next to my mother, because I have two brothers, but they don't help her. I do, so I try to stay. And it was convenient. She helps me. We all help each other. So it was good. I took it. I was like, okay, it is two bedrooms and better than what I had before. (Interview)

That priority led a small share of experimental compliers (<10% of our interview sample) to move back to inner-city neighborhoods, though not necessarily the ones left behind at first relocation, in order to be close to loved ones and sometimes the church. Almost all of these cases were in Los Angeles, where compliers moved much farther on average and where public transportation is famously inadequate. In the shorthand of social network analysis, the attractions of a move back were both instrumental (social support) and expressive (socializing). Overwhelmingly, for example, compliers in our ethnographic sample reported that socializing with loved ones meant driving or taking transit back into the inner city; their close ties rarely with them in their new neighborhoods. Patricia, an experimental complier in Los Angeles who had no car access and felt particularly isolated, explains why she moved from a low-poverty area in the San Fernando Valley back to South LA (though she later called the former safer and “nicer”):

The reason why I moved by here, because, uh, I wanted to come closer to my family down here because I was the only one in the Valley and everybody stayed over here, or over there, and nobody would come visit me or my kids because they was like you stay too far, you stay too far, you know. And I was like, you know, but still, can't you all come get us … We used to be down here like every weekend catching the Metro all the way from the Valley all the way here. I found a church home down here in L.A., and I liked it and I wanted to be closer to my church home, so I moved down here with my mother, and my sister, and my family and stuff. I liked it out there [in the Valley], but I wanted to move closer to my loved ones … My kids, they was like, “Mama, don't nobody come visit us.” (Interview)

Other families adapted better, whether because they had more resources under their own roofs initially, economic opportunities worked out, or for other reasons.

Finally, some MTO families used voucher-based relocation to distance themselves from the neediness or perceived negative influence of particular relatives, including those with a criminal past, no housing, no steady work, an addiction problem, or all of these. Wary social relations, including self isolation from the strains of kinship and other strong ties, has been a theme of qualitative work on ghetto poverty, including public housing environments, at least since classic studies such as Rainwater's Behind Ghetto Walls (1970) and Stack's All Our Kin (1974). But the debates over public housing transformation in recent years have generally de-emphasized this feature of social life among the chronically poor, highlighting the opposite pattern of cohesiveness and kin reliance (e.g. Greenbaum 2006, Venkatesh 2000, 2006).

The distancing strategy appeared in all three treatment groups, including control-group cases such as Jeanine, who left public housing in Los Angeles after an escalation of gun violence. When we visited her, she and her children were still in a risky neighborhood, with prostitution taking place right outside their front door. But Jeanine said she felt much safer there and added,

I don't want my family to know where I stay. I have three aunties and two other uncles who don't know I'm here. “Fieldworker: What would happen if they knew?” They would come visit, and they would become a problem, wanting to borrow, coming to stay. I got a cousin who has been in jail over 13 years, looking for a place to stay. I was like, ‘Oh no you're not. You been in jail. I woudn't be comfortable in the same house with you.’ My kids say, ‘Why you keep me away from my family?’ I'm like, ‘Protecting you from the bullshit!’ (Fieldnote)

Institutional resources and amenities

Notably, very few MTO mover families identified neighborhood institutional resources or amenities as main reasons to live in particular neighborhoods, and almost no families in the complier group who stayed in low or very low-poverty areas were engaged in the associational life of those neighborhoods, whereby they might have formed useful ties with neighbors. Some movers – notably the families in our ethnographic sample who had remained in low-poverty areas for more than five years – did most of their shopping and some of their socializing in those “new” neighborhoods, while other mover families preferred to attend church and to shop in another neighborhood. Many did comment on the greater convenience and affordability of shopping in their old neighborhoods, where more retailers targeted low-income consumers.

Trade-offs

Complier families faced more complex choices after the initial relocation, including unwelcome trade-offs between the things they valued: a decent housing unit and a decent neighborhood. Above, April makes clear that she would not trade “the right place” (as she defines it) for a much better housing unit – or for all the proverbial gold in Fort Knox. But most parents in our ethnographic sample of experimental compliers maintained close ties with kin or a small circle of close friends in high-poverty neighborhoods left behind and/or emphasized unit features more than April did. These parents weighed the location-unit trade-off very differently, especially if they were not as lucky as April had been to find a decent unit. In moving on, some had more bad luck in the housing market, landing in a poorly maintained unit and needing to move on quickly again or ending up on a street that was more dangerous than it seemed during the search. Some faced changing housing needs, too, as the make-up of their households changed, with sick or homeless relatives moving in or a newborn requiring an additional bedroom – trading away a decent neighborhood to get a bigger or better place was not about preferences in the abstract but problem-solving under tight constraints.

Shifting preferences and life-stage factors

Without consistent baseline measures and a larger sample of in-depth ethnographic cases to allow multivariate analysis, we cannot confidently assert that living in particular kinds of neighborhoods changed preferences in significant ways. But living outside the inner city clearly provided new information on the trade-offs that places present: the projects presented a variety of risks, for example, but also some conveniences and a kind of social acceptance that some parents and children valued more consciously after relocating to a very different environment.

In addition, our fieldwork highlights how the gaps between adult and child preferences (especially if the children are adolescents) sharpen under relocation to much lower-poverty areas. Neighborhoods that are “peaceful” to parents are often “boring” for teenagers, especially males, offering the latter little to do (according to them) and, in some cases, posing strains of acculturating to a different class culture, with its largely unwritten rules of appropriate speech, dress, and conduct. Adolescents with strong kin ties to inner-city neighborhoods were more likely to visit those neighborhoods often, maintain peer relationships there (often with cousins), and seek out some of the very risks their parents feared. MTO parents were often sensitive to these desires while trying to buffer their children from the most serious risks.

Search process and supply-side factors

Beyond preferences and trade-offs, the qualitative data indicate how opportunities and constraints – most reflecting the structure of the voucher program and its weaknesses in expensive housing markets, some reflecting the challenges of low-income single parenting – contributed to housing choices over time by shaping families' information sets, housing search behavior, and more. We avoid any estimates of prevalence here and employ a case-study approach, because we were able to collect valid data on multiple housing searches only for a subset of families (n = 19).

First, while complier families relied on a mix of search strategies for the initial relocation – about one-third relied on program counselors, and the rest were evenly divided among public housing agency lists, private real estate agents who broker rentals, and newspapers – the majority who moved on again relied heavily on their own devices, including rental agents they paid directly, classified ads, and word of mouth (referral networks) to find adequate housing units and landlords willing to accept the housing voucher. One reason, as affordable housing became scarcer, is that housing agency lists were routinely out of date and useless, said interviewees. The qualifier for this is that the MTO program counseling clearly got some families to consider areas they had never heard of and were not (until then) exploring – areas where we found them more than five years later. For those families fortunate enough to find housing units and locations that worked and to avoid (or manage) the chance events that trigger many involuntary moves for low-income households, counseling effects on locational outcomes endured for years. This underscores the pivotal role of housing stability.

Second, when additional moves proved necessary or desirable for whatever reason, landlord refusal to accept housing vouchers, especially in “better” neighborhoods, and the prevalence of units that did not meet program quality standards constrained the search significantly. For these reasons and because of the time and other constraints on wider search, voucher users did focus on what parents perceived to be their best prospects, which generally meant units in poorer and more dangerous neighborhoods – a pattern hypothesized in the MTO interim evaluation report (Orr et al. 2003,31). MTO parents expressed surprise and relief when areas they perceived to be much better offered them a housing opportunity.

Consider Tameka, an experimental complier in Los Angeles who initially relocated to a low-poverty neighborhood in the suburbs that she liked very much. When the landlord stopped accepting housing vouchers, she was forced to find a new apartment, and this drove her back to the central city, though she looked hard in the suburbs too. With a great deal of effort, she found a house in LA to rent that she liked, but she had to help the landlord prepare the unit to pass the required inspection. She recalls:

Well … it's really a tricky thing when you're moving, because … especially when you're a working, single parent … I know I, I have a certain amount of time to be out this apartment, to be relocated. So I went to this place where people find a place for you. You pay them this fee, and they look for like apartments and houses for you, for people that work and stuff. And you pay ‘em a fee, like $150 or something … And all the, they give you like a listing every week. And every time they give you a listing of maybe, I'd say roughly 100 places, maybe 80 of the places is already rented, been rented out already. … But it took me about 45 days and, um, at work, even on my break, I was in newspapers and doing anything I had to do to find a place. Actually, I found this house … in La Opinion [an Hispanic newspaper], and I had my coworker to like read it for me. … And they had a open house and I came, and I was like, oh, I know a lotta people gonna be for this. Don't think like that, don't think negative. And when I came for open house, I was the only person here. I looked at a lotta places in [an inner suburb to the south of Los Angeles] … a lot of places that were vacant and available, they did not accept Section 8. So you run into that … . and if you're not determined and a focus person, you will really give up. (Interview)

These were not static markets, as we showed with vacancy and FMR data earlier. Martina, a Latino Section 8 complier in Los Angeles, explains,

It's been very difficult] to find an apartment with Section 8. … [It took] (a)bout three years to find the one in Larga. You do find them, but not in good areas. I have children and I do it for them, not for myself. There are less expensive areas, but you don't wake up alive, or they rob or kill you. You can find them, but very far away from here. This area is expensive. In Larga, we paid $800, and the man is currently renting it for $1,500. That is why he asked us to move out, so he could raise the rent. [I lived there] five years. He wanted to raise the rent. Section 8 does not allow that. If $800 was being paid, he was not going to be allowed. That is why he told us to leave. It was hard to find an apartment. [Mine is] very expensive … I'm going to see what I can do to pay for my rent because Section 8 will give me $500. I will pay the rest. This apartment costs $1,100. (Interview)

Amber, a Section 8 complier in Boston, describes the change in apartment hunting in the Boston area and why relying on a broker was important as the market tightened and she tried to find a new place in “nicer” neighborhoods:

Looking for a new apartment, just doing the newspaper, don't help. So I end up, um, going to a real estate, the real estate agent had to help me find a apartment. I found this apartment, which I had to pay them a fee. And, um, the rent was going up to like $1,600, $1,800. So it was very hard calling [by] myself. And when I went to Section 8, I tried to do the same list thing. That wasn't working out. They either wanted me to move way far, further than Mattapan, meaning at least 45 [minutes] to an hour away from my family. I didn't want to do that. So my best [bet] was to go to a real estate, and that's what kind of helped me to get this apartment in Hyde Park [a low-poverty neighborhood in the outer core of Boston]. I didn't want to go to Mattapan. I didn't want to go to Roxbury [also inner city]. I wanted to stay, live in Hyde Park or West Roxbury [another low poverty, outer-core area in the city], which I think both the neighborhoods are a little bit nicer. The schools are a little better. (Interview)

Real estate agents made the search for an adequate, voucher-accepting unit more efficient, in part by steering MTO participants toward particular locations. With time, information, and other search resources at a premium, agents probably expanded the locational alternatives for some families who would have otherwise settled on the easiest-to-search locations, those in the weakest rental submarkets – high-risk neighborhoods.

Summary

While the broad contours of changing housing markets are evident in many of these family experiences – those who moved most often, in particular – the range of experiences is striking. For some families, getting a housing voucher is “like [winning] an Oscar” (as one told us), inspectors and landlords cooperate in textbook fashion, and the unit remains affordably priced for years. For others, the dearth of minimally acceptable units, the insecure opportunity to live in a safer neighborhood when one does gain a foothold, and each arduous new search are all reminders of what it means to rent housing on the bottom of the income ladder in extremely costly and tight markets – and with a government housing subsidy that is often stigmatized and rejected in “better” neighborhoods.

Discussion

Prior housing research has had little to report about locational outcomes over time for participants in “assisted mobility” programs – even though exposure to particular kinds of neighborhood environments over time is a necessary condition for certain positive “neighborhood effects” on children and families. Nor has research prior to MTO revealed the processes that put families who receive low-income housing assistance on one housing trajectory versus another; even MTO research to date has been dominated by results for mean treatment effects, which imply an “average” experience that does not, for practical purposes, exist for all of the outcomes we care about.

In lieu of more evidence on these, our two research questions, the debate often revolves around stylized versions of the supply- versus demand-side explanations of segregation in the nation's biggest low-income housing program. In the strong – that is, unqualified – form of the supply-side narrative, at least for tight housing markets, poor families who win the “lottery” of housing assistance are desperate to live in more racially and economically integrated areas. But market discrimination and scarcity thwart their dream of a better life in a better place. In this telling, even voucher holders who receive information, transportation, or other supports have little meaningful choice. In the strong demand-side narrative, families only integrate when they are obliged to do so by government planners. Assisted housing mobility, in this telling, reflects the integrator's ideal and not the preferences of families served. Yes, all parents may want the safest possible places for their children, say the demand-side purists, but the inner-city poor, most of whom are racial minorities, also want the comfort of familiarity and social acceptance, as well as support from loved ones – even if that means enduring more dangerous and resource-poor areas.

Based on the decade-plus experience of families in the MTO experiment, we find that the supply-siders are right about constraints (though our fieldwork was not set up to detect discrimination as a contributor) while the demand-siders largely misconstrue the role of preferences, at least in the tight housing markets where much economic growth and inequality are concentrated in America. Yes, intense market pressure in greater Boston, Los Angeles, and New York over MTO's first decade, as well as the limits and flaws in the housing voucher program were huge constraints for many families. The less stably housed the family, the more this was true – because each new move forced the family to navigate anew, with little room to maneuver in the choice of best-possible neighborhoods – and this appears to have contributed to many trajectories that led experimental compliers (the focus of hopes in the program) to poorer neighborhoods of residence over time. This helps explain why locational outcomes converged over time for the treatment groups even though two-thirds of experimental compliers who had to move on or who chose to do so reported looking for a new apartment in the same neighborhood. In lieu of better locations, affordable units with landlords willing to rent to voucher holders, families take what they can get, making the most of proximity to loved ones, managing in substandard or crowded units for the sake of their children, and otherwise settling.

The first major policy and research implication of this study is clear: In tight markets, relocation-only interventions, even the best assisted, are unlikely to produce enduring improvements in locational outcomes without focused attention on the geography of housing supply that will remain affordable and available. This calls for expanding and accelerating the focus on supply-side strategies with an inclusionary approach in many markets. In addition, it means searching on behalf of families in order to generate wider options, as Gautreaux placement agents did in the program's first wave, and then working with private landlords to ensure that decent, leased units will remain affordable and in program compliance as long as possible. This need not deny families the opportunity to lease up elsewhere, but it would put the onus of the arduous search task in the most competitive markets on the agencies offering the housing assistance, which too often fails to live up to national policy declarations about the importance of a suitable living environment for all families.

In the shorthand of optimization, we have, in the Housing Choice Voucher program, a low-income housing assistance policy engineered to minimize cost to the taxpayer; subject to an inconsistently enforced minimum standard of unit quality. The program lacks a robust rule or incentive to ensure the best-possible locational quality or stability in good locations, especially in the tight markets where those mechanisms are needed most. Stability is a pre-condition, frequently over-looked in policy debates that rely on point-in-time data on housing locations, for more productive engagement by low-income families in schools and community life, especially in less poor, less racially isolated, and also less familiar places: Without stability, no community and fewer positive effects of place.

The second policy implication has to do with improving places rather than helping people relocate away from them. Since safety looms so large in the calculus of low-income families on housing assistance – just as it does for most households with children, regardless of income – policymakers should redouble efforts to making the neighborhoods where very low-income households are concentrated much safer, particularly from the most unpredictable acts of gun violence that so troubled families in the MTO experiment, both on the streets and in schools.

Yet the demand-siders are right that choice (individual agency) also matters, not just in principle but in the significant choices parents make for themselves and their children over time. Rarely, however, did this take the form of an unconstrained preference for neighborhood A over neighborhood B. We have underscored, based on families' in-depth accounts of their choices and circumstances as well as our direct observations of those circumstances, the importance of trade-offs. Where they had a meaningful choice to make, some MTO parents were willing to trade away attractive unit features (including size and quality) in order to stay in a better neighborhood. Others, particularly if they had had to endure the worst of the dilapidated and poorly maintained housing stock in the voucher program, would not make the same choice. They preferred a better apartment in a risky environment, and they were willing to manage the risks.

Only rarely did the location of relatives, friends, or other loved ones trigger a move or determine where families moved. But pre-established ties, most of all the networks that MTO participants did not choose – the kin networks into which they were born – remained the center of most participants' social worlds and so factored into life routines and assessments of neighborhoods. Yes, some families who moved out later moved back and valued the access they regained to loved ones; this was especially true, in our small ethnographic sample, for families without reliable access to a car. But it is also the case that those ties proved burdensome and draining sometimes and that some parents moved in part to distance themselves from perennially needy relatives or relatives who posed special risks, such as addicts and ex-offenders that MTO parents perceived to be bad influences on their children. Similarly, some parents had to deal with dissatisfied, adolescent children who found safer neighborhoods boring.

Some reforms could “change the default,” in the language of choice architecture (Thaler and Sunstein 2008): from renting up where it is most familiar as well as easiest to use the voucher to considering recommended options first – and opting for an alternative, even in a riskier neighborhood, if one prefers that for any reason. But choice architecture can only help so much when the available choice set is very limited. Like the finding about search and constraint, our finding about the role of choice implies that policymakers should re-assess the issues that define available supply for housing voucher holder, in particular the enforcement of quality standards and the pivotal issue of landlord acceptance. It is vital that assisted relocation not be thought of as simply a matter of counseling, more generous payment levels, or locational restrictions on vouchers – the latter would be very unpopular, as well as vulnerable to litigation. Wider landlord participation demands responsive housing agencies, and while we do not think our data offer definitive evidence on the question of regional versus municipal management of the voucher program, the integral role of housing quality assurance, appropriate payment standards and FMR geographies, and wider landlord participation should not be overlooked in the effort to focus on the demand side, say with better information and other search supports for families. Having said that, policymakers should also assess the role that private rental agents can and should play in brokering the voucher market. And they should rethink the rules and incentives facing voucher management agencies (not just tenants) as “choosers” (Sard 2008). Voucher reform legislation submitted in 2007 addresses several of these concerns,20 but as of this writing, the prospects for such legislation remain unclear, and the increasing cost of the voucher program is likely to remain a top policy issue for HUD and the Congress, at least in the near term.

A final implication of this trade-off finding is that car vouchers and other tools could mitigate the trade-off between living in a safer neighborhood and having the desired level of access to one's social supports and cherished institutions, such as “church homes.” In related analyses, we have found that the employment challenges for work-ready MTO participants were not merely a reflection of their limited skill levels but of the difficulties of lining up three-way jobs-housing-social support matches. Difficult commutes and transportation constraints figure into that triangle in predictable ways – and not just for those families who use housing assistance to leave unsafe but transit-rich neighborhoods and then lack access to a car.

Limitations and future directions for research

There are key limitations of these data and the analyses we have been able to conduct. First, by design MTO served only voluntary movers, and some factors we have analyzed, including the preference to stay close to social supports that are concentrated in inner-city neighborhoods, may be more important for involuntary movers. Second, as we underlined above, our discussion is based on family experiences and outcomes in some of the nation's tightest markets; we are not making more general claims about all rental markets or all metros, and longitudinal evidence on these dynamics in less tight marketplaces would be useful. Third and finally, a number of our insights reflect a small-n, case-study logic, which considers multiple dimensions of each family's life. We have tried to show the variation in participants' experiences but often without knowing the prevalence of particular experiences. Inferences should therefore be made carefully, but in general, we have underscored how misleading the notion of an “average” program experience can be.

Modeling choice

Where does this study leave us in terms of a model of housing choice by very low-income families in America? Based on our findings as well as prior research, and adapting the classic critique of rational planning by Lindblom (1959), we propose that a model of limited, segmented comparisons captures the observed variation in outcomes well. Choices are limited first by the structure of the metropolitan housing market and the choices that real estate agents, landlords, and others make and second by the limited time, information, money, and other resources that the poorest families possess to make the best choices. The result is limited comparison shopping by many very low-income families. And choices are segmented on two dimensions: in terms of preferred trade-off, there are families more willing to cope in risky neighborhoods than deal with lousy housing units in better neighborhoods, while other families have the inverse priority; and in terms of key social relations, there are families whose social worlds are organized around kin contact and those who act to distance and buffer themselves from such contact. Individually, each of these distinctions has some relevance across the income spectrum, of course – almost all housing consumers face some important constraints, and we differ in terms of preferred contact with relatives – but as a construct, this model of limited, segmented comparisons helps explain how choice and constraint act together to shape locational outcomes for the poorest renters, many of them racial minorities in a still highly segregated nation.

Given this, should MTO be the model for the Housing Choice Voucher program, the nation's largest effort to address the housing needs of very low-income people? In terms of its guiding principles, we think the answer is “yes.” Many residents of distressed, high-crime neighborhoods want the opportunity to escape those areas, and we not only know how to help them do so – how to assist relocation more wisely than we did when MTO was launched – but we also know how to mitigate the tough trade-offs some families face when relocating. In terms of design and implementation, however, MTO was not enough – in that it relied on poverty rate alone to target “opportunity” areas, failed to properly screen hard-to-house families with severe health and other problems, and failed to enable most participants to stay away from, not just get out of, the high-risk areas they wanted to avoid. In sum, the experiment was founded on a strong and politically legitimate idea – the meaningful choice of a better living environment for one's family – that could be much more effectively implemented.

Acknowledgments

Our study was made possible by generous support from the US Department of Housing and Urban Development and a consortium of private, philanthropic foundations, including the Annie E. Casey, Fannie Mae, Rockefeller, Smith Richardson, and William T. Grant Foundations. For helpful feedback, we are grateful to John Goering and Susan Popkin (Co-Principal Investigators on the study), as well as Phillip Clay, Langley Keyes, Barbara Sard, Margery Austin Turner, and anonymous HPD reviewers. We also benefited from the discussions of the Social Science Research Council's Mixed-Income Housing Research Group, of which the first author was a member, which received support from the John D. and Catherine T. MacArthur Foundation.

Notes

¹See US HUD (2007). The number assisted in a given month varies according to administrative action and utilization rates, distinct from the number of households “authorized” through annual Congressional appropriations.

²We use “minority” to refer to any group other than white, non-Hispanics.

³See Hanna Rosin, “American Murder Mystery,” The Atlantic (June 2008); Xavier de Souza Briggs, Peter Dreier et al., “Memphis Murder Mystery? No, Just Mistaken Identity,” Posted on Shelterforce Online (July 2008); and Solomon Moore, “As Program Moves Poor to Suburbs, Tensions Follow,” New York Times (August 8, 2008).

⁴See Leslie Kaufman, “An Uprooted Underclass, Under the Microscope,” New York Times (25 September 2005); “A Voucher for Your Thoughts: Katrina and Public Housing,” The Economist (24 September 2005); Xavier de Souza Briggs and Margery Austin Turner, “Fairness in new New Orleans,” The Boston Globe (5 October 2005); and Briggs (2006). The Katrina relocation also created a “natural experiment,” with moves from segregated, high-poverty, and often high-crime areas in pre-storm New Orleans to a range of different neighborhood contexts in a variety of metro areas.

⁵There is a large literature. See, in particular, Hartung and Henig (1997), Khadduri (2006), Newman and Schnare (1997), McClure (2006), and Turner and Williams (1998).

⁶See Newman and Schnare (1997) and Meeting Our Nation's Housing Needs: Report of the Bipartisan Millennial Housing Commission Appointed by the Congress of the United States (Washington, DC 2002).

⁷Clearly, some types of moves have long been associated with social mobility as well as escape from undesirable places. But as every parent knows, moving can be harmful as well. Recent research on child and adolescent development has underscored the deleterious effects of frequent moving on children and adolescents, net of other factors, including poorer emotional health, weaker academic outcomes, strained family relationships, smaller and less stable peer networks, and even a greater risk of gravitating toward deviant or delinquent peers after arriving in new schools and communities (Barlett 1997; Haynie and South 2005; Haynie, South, and Bose 2006; Pribesh and Downey 1999). Drawing on fieldwork among low-income African-Americans, researchers and family therapists have emphasized the importance of securing “the homeplace” – comprising “individual and family processes that are anchored in a defined physical place and that elicit feelings of empowerment, rootedness, ownership, safety, and renewal” (Burton et al. 2004, 397) – and the difficulty many families face in securing such a homeplace.

⁸Latinos appear to occupy an intermediate position, with more favorable locational trajectories than blacks but less favorable ones than whites (South, Crowder, and Chavez 2005), and also to show substantial variation among nationality groups (e.g., Cuban, Mexican, Puerto Rican); data limitations have made it impossible to study longitudinal patterns among Asians.

⁹As the researchers note, the finding that those in the most disadvantaged places are less likely to move may reflect negative selection: The fact that households that lease up in those areas face additional, unobserved challenges. In a controlled experiment, Abt Associates et al. (2006) found that voucher-holding welfare families enjoyed some improvement in locational outcomes over time, reflecting both economic and racial integration, when compared to welfare families that did not receive housing vouchers. Families who entered the demonstration while living in “stressful arrangements,” including high-poverty public housing, were particularly likely to experience locational improvements.

¹⁰Most voucher users relocated to predominantly black or racially changing areas of the local market. In addition, more than half of the voucher users in their study reported wanting to move again, and even many who were satisfied with their current housing voiced that wish – citing pressure to move out of public housing quickly and feeling that they had “settled” for a satisfactory unit rather than one that was “just right” for their family (p.153).

¹¹Such placement is the defining feature of the relatively uncommon, small-scale unit-based (as opposed to voucher-based) approaches to housing mobility for low-income families, such as in scattered-site housing programs (Briggs 1997; Hogan 1996; Turner and Williams 1998). It also defines supply-side strategies such as inclusionary zoning and area “fair share” requirements—at least when they include very low income households—and efforts to preserve affordable supply in “better” neighborhoods, such as in the federal Mark to Market reforms for project-based Section 8 housing.

¹²The cooperation rate was calculated by excluding those we could not contact, due to death or invalid address (after all options for updating the address for each hard-to-find household had been exhausted).

¹³All personal names below are pseudonyms, and sublocal places are disguised.

¹⁴Note how inadequate it can be to denote neighborhoods or localities according to poverty rate alone. Some low-poverty areas are home to households with a wide range of incomes and socio-economic status (including the poor, up to some threshold level), while other low-poverty areas, such as affluent suburbs, have almost no diversity on that dimension.

¹⁵Rents were measured in 2005 dollars. See US HUD State of the Cities Data System (SOCDS) at http://socds.huduser.org/index.html [accessed August 2, 2007].

¹⁶We used HUD-designated and provided fair market rent (FMR) geographies to track price trends. These areas are not determined solely by housing market analysis by HUD but by the laws and regulations governing the HCV program.

¹⁷Households not included in this analysis were somewhat more likely to be Hispanic, in the labor force, with a high school/GED, and living in a lower poverty neighborhoods at the time of the interim survey, though differences were generally modest (p < .05, data available).

¹⁸As a final caveat, because the Los Angeles site completed MTO lease up several years after Boston and New York, the intervals between observation points are shorter for Los Angeles. Future research on a longer observation window will mitigate this and other site differences, but all households in our study were observed at least 4 years post random assignment.

¹⁹Families reported reasons for moving for each of their address spells, from original lease-up to the qualitative interview in 2004.

²⁰See the Section Eight Voucher Reform Act (SEVRA), S2684 and HR1851.