当前位置:首页 >> 英语学习 >>

论文 The long term Effects of pubic housing on sufficiency

Association for Public Policy Analysis and Management

The Long-Term Effects of Public Housing on Self-Sufficiency Author(s): Sandra J. Newman and Joseph M. Harkness Source: Journal of Policy Analysis and Management, Vol. 21, No. 1 (Winter, 2002), pp. 21-43 Published by: John Wiley & Sons on behalf of Association for Public Policy Analysis and Management Stable URL: http://www.jstor.org/stable/3325946 Accessed: 16/10/2010 06:03
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=jwiley. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

John Wiley & Sons and Association for Public Policy Analysis and Management are collaborating with JSTOR to digitize, preserve and extend access to Journal of Policy Analysis and Management.


The Long-Term Effects of Public Housing on

Sadra J. Newman

Joseph M. Harkness


Recentyears have witnessed an intensification of the debateabout the fundamental purpose of public assistance to the poor and the effects of these programs on children. This study uses enriched data from the Panel Study of Income Dynamics to examine the effects of living in public housing as a child at some point between 1968 and 1982 on four young adult outcomes: welfarereceipt;individual earnings; household earnings relative to the federalpoverty line; and employment.Living in public housing during childhood increased employment, raised earnings, and reduced welfare use, but had no effect on household earnings relativeto the poverty line. Thebeneficialeffectscould have arisen becausepublic housing improvedphysical living conditions, reducedresidentialmobility,or enabledfamilies to spend more of their income on items that benefit children'sdevelopment.Whetherthese effects apply to contemporarypublic housing is unknown. ? 2002 by the Association for Public Policy Analysis and Management. INTRODUCTION Over the last ten years, debate about the fundamental purpose of housing assistance to the poor has intensified. The core question is whether the traditional goal of decent, affordable housing should continue to be viewed as an end in itself, or also-or instead-as a means to economic independence. Two major factors precipitated this re-evaluation of housing assistance policy. First, after years of consciously distinguishing itself from the nation'ssocial welfare system by servingworking families and the lower-middle class along with the very poor, housing assistance programs were transformed in the 1980s to assistance largely for the very poor alone.1 These changes moved housing assistance squarely into the safety net. A second factor was the apparent shift in public opinion regarding the principles that ought to govern assistance programsfor the poor.The focus is increasinglyon eliminating dependency, not simply alleviating poverty. This is particularly the case for programs as large as federal housing assistance, which spend more than $28 billion a year and serve about 5 million households-more on both counts than AFDCat its 1994 height (U.S. House of Representatives, 1998). The change in societal views is most clearly demonstrated by the debate surrounding the passage of both the 1988 Family Support Act and the 1996 Personal Responsibility and Work Opportunity Reconciliation Act.
Because housing assistance is not an entitlement, the more liberal income eligibility rules in effect in the 1960s and 1970s were not synonymous with larger numbers of households being served. Journal of Policy Analysis and Management,Vol. 21, No. 1, 21-43 (2002) ? 2002 by the Association for Public Policy Analysis and Management Published by John Wiley & Sons, Inc.

22 / Effects of Housing Assistance

Proponents of this reorientation of housing assistance from a focus on bricks and mortar to a broadened concern about self-sufficiency effects argue that performance measures emphasizing housing outcomes, such as the number of dwellings meeting housing codes or that are affordable to low-income households, be supplemented by such outcomes as labor force participation, earnings, and lack of dependence on welfare. Like the broader welfare policy debate, a major focus has been children. But because the debate over this reorientation is relatively recent, little research has been done on this broader array of effects of housing programs.2 This paper begins to fill this gap by investigating the effect of housing assistance on four long-term outcomes of young adults who lived in public housing when they were between the ages of 10 and 16: welfare receipt between ages 20 and 27; earnings between ages 25 and 27; family earnings between ages 25 and 27 relative to the federal poverty line; and labor force participation between ages 25 and 27. Nationwide, about 470,000 households with children live in public housing, 22 percent of all households with children in federally subsidized housing (HUD, 2000). Excluded from this study because of data limitations are the two other major forms and vouchers ("tenant-based assistance"), of federal housing assistance-certificates which tenants use to pay a portion of their rent in the private market, and privately owned developments built or rehabilitated with federal subsidies.3 Public housing is the federal housing assistance program most commonly associated with dire conditions for children's development, a perception confirmed by the evidence (Newman and Schnare, 1997). Thus, if anything, the effects of public housing on children's development should be worse, not better, than other forms of housing assistance.4 HOUSING EFFECTS PUBLIC OF POTENTIAL Why is it plausible to expect public housing to have long-term effects on children? Five possibilities emerge. First, children may be positively affected by the superior physical quality of public housing relative to the dwellings of similar families in marketrate housing. Recent empirical evidence confirms the expectation that public housing should improve the physical adequacy of a family's dwelling unit (Currie and Yelowitz, 2000; Newman and Schnare, 1993). However, limited research links housing quality to children's outcomes. Recent work by Mayer (1997) and Klebanov and colleagues (1997) suggests that there is some effect. Other studies indicate that overcrowding, one measure of dwelling adequacy, is related to children's poor health (Coggon et al., 1993; Galpin, Walker, and Dubiel, 1992; Mann, Wadsworth, and Colley, 1992). Public housing may thus be more physically adequate than other low-income rental

2Some exceptionsare Currieand Yelowitz(2000), Meyerset al. (1995), and Newman and Harkness(1999, 2000). 3We did not include tenant-basedhousing assistance for two reasons. First, neither HUD nor local jurisdictions retain data on the addresses of certificatesor voucher recipients, which thereforecould not be used in an address match over the period of study.Second, tenant-basedassistance was introducedonly toward the end of the period in which youth residence in assisted housing (1968-1982) was observed;it became a largeprogramonly after 1982. Privatelyowned developmentsare excludedbecause most of this housing was built in the 1970s and the address-matchover the period 1968-1982 did not yield a sufficient numberof observationsfor reliableestimation. 4 Early results from the Movingto Opportunityexperiment,in which one experimentalgroup of public housing residentsis given a Section 8 certificate,supportthe expectationthat tenant-basedassistance has better implications for children than public housing (Ludwig, Duncan, and Hirschfield,2001; Ludwig, Ladd,and Duncan, forthcoming).

Effects of Housing Assistance

1 23

properties, but the extent to which this difference matters for children's outcomes is not well understood. Second, public housing might also result in more stable housing either because the subsidy might make it easier for the family to pay its rent, or because provisions in administrative law make it more difficult to evict families living in public housing. Children who move often are also likely to change schools more frequently, putting them at greater risk of grade repetition and poor academic performance (GAO, 1994). In addition Astone and McLanahan (1994), Haveman, Wolfe, and Spaulding (1991), and Jordan, Lara, and McPartland (1996) find that the number of residential moves adversely affects the likelihood of a child's graduating from high school. Unfortunately, there is no empirical work on the residential stability of children in public housing. Therefore, while positive effects are consistent with hypothesized greater housing stability, it is impossible to predict how large such effects might be. Third, by reducing the rent burden, public housing may increase the amount of income a family can spend for items that benefit a child's development, such as nutritious food, books, games, or educational aids.5 Families without housing assistance often devote a significant proportion of their household income to rent. In 1995, more than a third of very low-income households spent more than 50 percent of their income for rent (HUD, 1998). Families with housing assistance, on the other hand, pay roughly 30 percent of their income for rent, with government subsidies making up the balance.6 However, evidence is minimal that more generous cash or in-kind benefits contribute to better outcomes for children. While Meyers and colleagues (1995) find that residence in public housing was associated with greater nutritional adequacy in young children, the literature on the effects of Aid to Families with Dependent Children (AFDC) and non-cash benefit programs on children is inconclusive (Currie, 1995). Mayer (1997) argues that it is not income, but the parental characteristics associated with stable employment, that leads to better outcomes for children. Finally, enhanced income is likely to be most effective during early childhood because these are the critical developmental years (Duncan and Brooks-Gunn, 1997). Therefore, income effects may not be observed in this study, which focuses on older children between 10 and 16 years old. Fourth, children may be negatively affected by the degree of concentrated poverty in their residential environment. A growing body of research is attempting to identify the effects of concentrated poverty, crime, joblessness and other indicators of neighborhood distress on children (e.g., Brooks-Gunn, Duncan, and Aber, 1997; Jencks and Mayer, 1990; Moffitt, 2001). If such "neighborhood effects" exist, then to the extent that public housing increases a child's contact with highly disadvantaged neighbors (or reduces contact with better-off neighbors), his or her prospects will be diminished. Because nearly 90 percent of public housing is located in large developments with 50 or more units, neighborhood effects could occur at two levels: the housing development, or its surrounding neighborhood (Newman and Harkness, 2000). The sparse research in this area suggests that it may be the former that matters most for children's outcomes. Shlay and Holupka (1991) report that for children living in large public housing developments, the orbit of activity is confined to the development itself and venturing out into the neighborhood is rare. Newman and

5 Of course, there is no guarantee that increased income will actually be spent on goods that enhance child development. If additional income is used to sustain a parent'saddiction to drugs or alcohol, for example, children could be adverselyaffected. 6 Before 1982, which covers the period of youth observation in this study,assisted renters paid 25 percent of their income for rent.

24 / Effects of Housing Assistance

Harkness (1998) find that for children living in public housing, the development itselfnot the surrounding neighborhood-matters more for educational attainment. The household characteristics of public housing developments and their neighborhoods are closely correlated. Newman and Schnare (1997) find that 43 percent of units in family public housing developments, compared with 12 percent of welfare households, are located in census tracts with a poverty rate of 40 percent or more.7 Public housing developments also exhibit severe concentrations of households characterized by low income, welfare dependency, and low educational achievement (Newman and Schnare, 1993).8 If the concentration of disadvantaged neighbors hurts life chances, as posited by collective socialization and epidemic models of neighborhood effects (Jencks and Mayer, 1990), both neighborhood aspects of public housing suggest worse outcomes for children who grow up there.9 Finally, children may be indirectly affected by public housing through its effect on the behavior of their parents or guardians. There are three major pathways for these effects. First, public housing may discourage employment either because of work disincentives inherent in the subsidy formula, or because of locational disadvantages and non-working parents may be poor role models for children. Families in public housing generally pay 30 percent of their income in rent, retaining 70 cents out of each additional dollar of increased earnings. This "implicit tax" on earnings may dampen a parent's enthusiasm for work. A "spatial mismatch" between the location of public housing developments and job opportunities could also play a role in reducing work behavior among adults living in public housing (Allard and Danziger, 1999). However, unlike the body of empirical studies on the labor market effects of other transfer programs, such as AFDC, Medicaid, and food stamps (Fraker and Moffitt, 1988; Moffitt, 1992; Moffitt and Wolfe, 1992), there are fewer studies of the influences of public housing on adult work behavior and results are inconsistant. For example, Murray (1980) estimated that living in public housing reduced labor supply by just 4 percent, Reingold (1997) found no effect, Ong (1998) detected a slight increase in work hours for welfare recipients living in public housing, and Houser and DickertConklin (1998) reported that a 10 percent increase in expected public housing benefits reduced labor force participation by 1 percent for single parents and had no effect on primary earners. If these estimates are correct, it is unclear how children 10-16 years old would be affected by such small changes in parental work behavior. A second possibility is that public housing relieves financial pressure on parents, thereby reducing stress, depression, and other symptoms of psychological distress, with potentially beneficial effects on their children. On the other hand, the distressed neighborhoods where public housing is located could impose psychological burdens on adults, resulting in behaviors that could, in turn, adversely affect children. DATA This research relies on a unique database, the Panel Study of Income Dynamics (PSID)Assisted Housing Database (AHD), developed by matching all sample addresses for
7Thesedata are drawnfroma period significantlylaterthan the periodof youth observationin the present study.But the contemporaneousevidence stronglysuggests that assisted housing developments,and public housing in particular, tended to be located in the worst when were built. 8 As with tractcharacteristics,these data are drawnfrom aneighborhoods the they of period later than period youth observation in the present study. The profile of public housing residents was significantly less disadvantaged during the time of the present study than it later became. 9 An obvious problem here is that there is no evidence that public housing actually worsens a youth's neighborhoodcontext over what it would have been in the absence of subsidized housing. Preliminary analysis suggests that public housing leads to worse neighborhoodconditions (Newman and Harkness,

Effects of Housing Assistance

/ 25

28 years of the PSID to the addresses of assisted housing units across the nation.10 The PSID is an ongoing longitudinal survey of U. S. households begun in 1968 by the Survey Research Center of the University of Michigan. Low-income families were initially over-sampled,but statistical weights have been developed to adjust for both the differential initial sampling probabilities and differential nonresponse that has arisen since the beginning of the study. By following all members of its sample over time, including children as they leave their parents' home, the PSID maintains a representative sample of the nonimmigrant U. S. population and of major subgroups in the population. The database of assisted housing addresses constitutes the closest approximation to a national census of assisted housing that the authors are aware of. It is based on eight sources including several of HUD's (U. S. Department of Housing and Urban Development) administrative and program databases, and surveys conducted with housing agencies across the nation including state departments of housing and community development, Housing Finance Agencies, and Farmer's Home district offices. Programscovered by the database include public housing, other HUD projectbased developments,Farmer's Home Section 515, the Low-IncomeHousing TaxCredit, and state rental assistance programs (Newman and Schnare, 1997). Because the PSID-AHD identifies whether the sample member receives public housing through address matches to public housing units and not respondent selfreports, it overcomes the problem of reportingerrorsof respondents answering survey questions about whether they live in assisted housing. Recent evidence suggests that such self-reports are highly inaccurate (Shroder and Martin, 1996). Two criteria frame the analysis sample: A consistent period of youth observation for every sample member; and observation of outcomes as far into adulthood as possible. Balancing these requirements with the objective of obtaining as large a sample as possible, the analysis sample is drawn from nine PSID cohorts born between 1957 and 1967. It includes both those who lived in public housing at some time between the ages of 10 and 16 and those whose family was income-eligible for public housing but was unassisted between the ages of 10 and 16.11The period of observation when these individuals were 10-16 years old covers the 14 years from 1968 through 1982. Adult outcomes from ages 20 to 27 cover the period from 1978 through 1993.
METHODS Dependent Variables and Estimation Techniques

Figure 1 summarizes the effects studied, how each was measured in the data, and the statistical approach used to analyze each. The policy question is whether these outcomes were substantially affected by residence in public housing between the ages of 10 and 16. Exposure to public housing is measured by years of residence in public housing. Most of the children in the analysis sample who ever lived in public housing lived there for most of the period between ages 10 and 16. The mean exposure is 5.3 years, and the median is 6 years.12
1 Children identified as living in privatelyowned assisted housing developments were eliminated from the sample. 12 We also tested models using "everin public housing" as the policy variable. The main difference between those results and the ones reported is that coefficients were 4-5 times larger,except for the "ever employed" outcome, where the coefficients were 1.8-2.4 times larger. Since mean exposure is about 5 years, coefficients 4-5 largerusing "everin public housing"as the policy variableare consistent with those reported.
10Details on the construction of the PSID-AHDcan be found in Newman and Harkness (1999).

26 /

Effects of Housing Assistance

Impact Welfare receipt' Earnings Worked

Measures Years off welfare Average annual earnings of individual Whether individual had any earnings

Age when observed 20-27 25-27 25-27 25-27

Estimation techniques Tobit Tobit Probit Tobit

Average annual earnings of individual plus spouse divided by federal poverty line for family size 'AFDC, food stamps, "other welfare," SSI excluded. Earnings-toneeds

Figure 1. Impacts, measures, and statistical estimation techniques. In the analysis of welfare receipt, the dependent variable is the number of years a person lived in a household that received no welfare from age 20 to age 27. "Welfare" The includesAFDC, welfare(e.g., generalassistance).13 authors food stamps,or "other" relied on household welfare receipt rather than individual welfare receipt because the data set reportedon individualwelfare receipt for household heads only,and only some of the sample members were always household heads from ages 20 to 27, while others never were. Because this variable was restricted to values between 0 and 8, with observations tending to cluster on both poles, this model was estimated using two-sided tobit. In the analysis of the effect of public housing on earnings, the dependentvariableis the person'saverageannual earnings from age 25 to 27. Tobitis used to estimate this model, because average annual earnings for 14 percent of individuals are zero.14 In the third analysis, the dependent variable is defined as the average earnings-toneeds ratio, or the average ratio of the combined earnings of an individual or, if married, his or her spouse, to the federal poverty level between ages 25 and 27. For example,the federalpovertylevel for a family of three in 1990 was $10,530. Therefore, for a single mother of two children with annual earnings of $7,000, the earings-toneeds ratio would be 66.5 percent (7,000 - 10,530). Because the federal povertylevel varies with family size, the focus here shifts from the individual to the individual's family unit. The calculation of family size counts the individual, his or her spouse, and their children.Again, tobit is used to estimate this model, because earnings in 12 percent of cases were zero. Finally,to examine the effect of public housing receipt on labor force participation, the dependent variable is whether an individual had any earnings between ages 25 and 27. This model is estimated with probit. Because 84 percentof individualswere siblings from the same family and, therefore, are not independent observations, Huber-White standard errors should be used. Unfortunately,it is difficult to implement these standard errors with the nonlinear estimation techniques applied here. However,tests with linear estimation techniques (ordinary least squares and two-stage least squares) showed that Huber-White
1 The results are not sensitive to the inclusion of food stamps in this definition. The correlation between years off welfare with, and without, food stamps is nearly 0.9. 14The distribution of earnings was highly skewed because of many zero or near-zero wage earners. Although the usual technique for coping with a skewed distribution of the dependent variable is to transform it by taking the log or performing a Box-Cox transform, neither of these techniques can cope with zeros in the dependent variable.

Effects of Housing Assistance

/ 27

standard errors exceed conventional standard errors by a maximum of 20 percent, and usually by much less. The authors report results using conventional standard errors and describe how statistical significance is affected when they are inflated by 20 percent in the discussion of results. Unobserved Variable Bias A major difficulty in using non-experimental data to gauge the effects of public housing on early adult outcomes is that individuals who have lived in public housing may be systematically different from those who have not, and measures of all of these differences are unlikely to be available. To the extent that unmeasured characteristics affect the self-sufficiency outcomes being examined here, the failure to control for them could bias results. Public housing is not unique in this respect. The participants in any social program always represent a self-selected sample. But because public housing is not an entitlement, the process through which families obtain it is probably a good deal more complex than for other social programs. Administrative rules on the allowable mix of tenant income have varied from requiring a broader mix of income groups to more tightly targeting most assistance on the very poor. Within the bounds set by federal regulation, however, public housing authorities (PHAs) have considerable discretion in setting admission standards, thereby introducing some selection on the part of program administrators. Consequently, any family's expected time on the waiting list for public housing is uncertain. These considerations imply that explicit modeling of the selection process would be extremely difficult. While this process is certainly not random, some research suggests that the receipt of public housing is unrelated to its expected benefits to the household (Keane and Moffitt, 1998). Under these conditions, use of instrumental variables (IV) is an appropriate analytical strategy (Heckman, 1997). Because all of the outcome measures are treated as limited-dependent variables, an Amemiya generalized least squares (AGLS) estimator (Amemiya, 1978,1979,1983; Newey, 1987) is used as the IV estimation technique. The instrument for public housing capitalizes on the marked spatial variation in the supply of public housing per income-eligible household (Kingsley and Tatian, 1999).15This instrument is appealing because households are more likely to be assisted if they live in places where public housing is more readily available. However, the supply of public housing could be endogenous with the sorts of self-sufficiency outcomes being studied. To purge this endogeneity, a preliminary regression of the number of public housing units per income eligible family in the area was performed on a vector of seven area characteristics, and the residuals were used as instruments in the first-stage models.16
15Instrumentsbased on the changes in income eligibility rules of housing programs that occurred during the study period were also considered;this allows definition of two distinct housing policy "regimes":one for 1968-1974, and the second for 1975-1984. As a first approximation,housing programswere governed by two differentsets of rules concerning such key features as the definition of income and income eligibility in each of these time periods (24CFR860;24CFR1272.102;R. Leonard,personal communication, April 16, 1999). These "policyregime"instruments had little predictivepower, however,and were highly correlated with the time trend variable in our models. 16 The characteristicsof the county or metropolitan area in this regression were: the population (logged), percentageof families whose income was below poverty,percentagewhite population, percentageof population aged 65 or older, percentage of adults with a college degree, the ratio of median rent to median income, and the percentage of households with a female head. All of these characteristics were derived from decennial census data, interpolating for inter-census years. See Newman and Harkness (1999) for details.

28 / Effects of Housing Assistance

Two alternative area definitions were explored, one based on county, the other on metropolitan areas. Although the latter corresponds more closely to the housing market areas used by HUD in specifying income-eligibility thresholds and fair market rents, the requisite data do not exist for non-metropolitan areas. Therefore, using metropolitan areas reduces the size and generalizability of the sample.17 Because the results produced by the two approaches were very similar, the discussion focuses on those produced using the county-based instruments, while the results produced using instruments based on metropolitan areas are briefly noted. Even after controlling for local characteristics, the instruments could still be endogenous if public housing were concentrated in states or regions where public programs, educational institutions, or labor market conditions affect youth outcomes. For historical reasons, for example, more public housing is found in northeast and north-central regions that witnessed significant industrial decline through the 1970s and 1980s, and these conditions might affect outcomes. Therefore, in addition to county-level characteristics, two ways of controlling for geography in the preliminary regressions were tested: states and census divisions. For each sample member, there were seven residuals from the preliminary regression (one for each year between ages 10 and 16). While all of these could theoretically be included in the first-stage equation, they were highly correlated. Therefore, models were tested using the mean, the maximum, and both the mean and the maximum of these residuals. Models using the maximum alone proved strongest.18 With controls for state, the instrument is statistically significant at the p = 0.01 level. But with controls for census division, the coefficient on the instrument is twice as large, and it is significant at less than thep = 0.0001 level. Thus, although the instrument performs satisfactorily in predicting years of public housing in both first-stage models, it is clear that controls for states substantially reduce the predictive power of the instrument. Results are reported for both state and census division fixed effect models. In the state fixed effect models, tests for statistically significant differences between states within census divisions generally detected none. Dummy variables were included for the few states that were different from others in their census division in the models with census divisions controls.19 Attrition Recent studies of attrition bias in the PSID conclude that while attrition is substantial, it does not bias model estimates (Fitzgerald, Gottschalk, and Moffitt, 1998a,b; Zabel, 1998). Similar tests of attrition bias conducted for the present analysis produced the same finding of no attrition bias (Newman and Harkness, 1999).

17 One strategyfor includingnon-metropolitanareaswould have been to treat the non-metropolitanareas of each state as a single area, as HUDdoes in definingincome-eligibilitythresholds.However,a source for some of the requireddata (e.g., median incomes in 1970) could not be identified for such areas. Another approachwould be to treat all non-metropolitanareas in the countryas a single area. But this approach seemed too crude to merit testing. 18 AppendixTableA.1 furnishesthe definitions and univariatestatistics of the variablesused in the analysis, and AppendixTableA.2 shows the results of the first-stageregressions. 19In the mid-Atlanticcensus division, earnings and earnings-to-needswere significantlyhigher in New Yorkand New Jersey than they were in Pennsylvania.Illinois was similarly statistically different from other states in the East-NorthCentralcensus division, as was Floridacomparedwith other South Atlantic states.

Effects of Housing Assistance

/ 29

AnalysisSample The analysis sample includes two groups: those who lived in public housing at some point between the ages of 10 and 16; and a comparison group of those who were for "eligible" federally assisted housing but were unassisted. The goal was to compare the self-sufficiency outcomes of youth exposed to public housing with the same outcomes of similar youth who were unassisted.20 Corresponding to the two alternatives tested in defining instruments, were two income-eligibility thresholds, one based on counties and the other based on metropolitan areas. The earlier discussion of the merits of using counties versus metropolitan areas for the instruments applies here as well. Metropolitan areas more closely approximate HUD's income limit areas, but using them restricts the sample because they do not cover the whole country. Also as mentioned, the two approaches yield similar results. Again, to simplify the exposition, the discussion mainly describes only the county-based models, briefly mentioning the results based on metropolitan areas. The authors used 80 percent of area median family income, using HUD'sadjustments for family size, as the sample cutoff because it was the legal eligibility threshold for more of the 14 years between 1968 and 1982 than any other definition.21To be part of the analysis sample, individuals had to have a family income below this threshold for at least 2 of the 7 years when the child was between 10 and 16 years old. Also included were siblings of those who fell into either the public housing or unassisted samples because the family in which these children grew up was a member of the target population for HUDprograms,accordingto the definition used here, for some period.22 Independent Variables The policy variable in each model measures the number of years an individual lived in public housing between the ages of 10 and 16. Control variables also measured at ages 10-16 are: * demographics(whetherblack,whether female, year born, family size, and mother's age at birth); * welfare dependence (whether ever relied on welfare, averagecash value of transfer income); * earnings and employment of household head (average annual labor income specified as a piecewise linear spline function with a knot at $10,000), whether very few work hours (less than 200 hours annually); * other characteristicsof household head (educational attainment, number of years disabled); * family structure (number of years with single parent, whether a marital change occurred); * housing tenure (number of years as homeowner); and

20 Tests of several alternative sample definitions, including an unrestricted sample, produced estimates similar to those presented. 21 The base income (80 percent of median) is the income limit for a family of four: 10 percent is subtracted from this base for each person fewer than four, 8 percent is added to the base for each additional person. 22 This group constituted about 6 percent of the samples. These siblings were not unique in their failure to meet the 80 percent income eligibility cutoff. More than 10 percent of public housing residents were also under this definition. Inclusion of these individuals did not materially affect results. "ineligible"

30 / Effects of Housing Assistance

* location (indicators for number of years in a big city (population more than 500,000) and in a small city (population 100,000-500,000), indicators of the primary state or census division of residence). For the most part, these variables were selected as general measures of characteristics of the individual or family disadvantage.23 By contrast with other studies of similar outcomes, housing tenure and location were also controlled for. Housing tenure was included both as an indicator of household disadvantage and because some evidence indicates that homeownership may exert a positive effect on child development (e.g., Green and White, 1997).24 City size was controlled for because Page and Solon (1999) demonstrated "the importance of being urban" on adult earnings. Two alternative indicators of location were tested to control for geographic variations in public programs and labor market conditions that could potentially affect outcomes: primary state and census division of residence between ages 10 and 16. But not included were indicators of neighborhood quality, such as the tract poverty rate, or the number of times a child moved, because these variables are likely to be influenced by whether a child lived in public housing and, therefore, are endogenous. The public housing variables can be interpreted as capturing "the whole package" of living in public housing: the public housing subsidy, the public housing unit, its surrounding neighborhood, and possible effects on residential mobility. SAMPLE PROFILE This analysis pertains to the effects of public housing that existed-and the rules under which public housing operated-between 1968 and 1982. During this period, income eligibility rules were more liberal than they became in the 1980s. The very poor were also not given priority for public housing as they ultimately were with the institution of "preference rules" in 1988. As a result, in sharp contrast to the largely disadvantaged tenant profile today, the profile of public housing residents included a mix of working poor, the lower-middle class, as well as the more disadvantaged. The dramatic changes in the tenant profile of public housing between 1970 and subsequent decades are shown in Table 1. During the 1970s, there were sharp declines in the fraction of married household heads and in earnings, and a sharp increase in dependence on public assistance. Thus, for children living in public housing during the earlier part of the observation period of this study, the profile of tenants was dramatically different than it is today. (Table 1). The time trend variable in the regression models should control for the effects of this change on outcomes. Another difference between public housing programs in the 1970s and those in the 1990s was the 1969 Brooke amendment to the U.S. Housing Act, which limited public housing rents to 25 percent of tenant income. This ratio was not raised to 30 percent until 1981, nearly the end of the observation period. It is possible that lower rent burdens may have an effect on outcomes.
23Not listed are measuresthat were contemporaneouswith outcomes from ages 20 to 27 because they are likely to be endogenous (e.g., whether the individual experienced a teen birth, was married, or was a household head). 24 It may be argued that the decision to own a home may be simultaneous with the decision to obtain public housing, and that homeownershipshould thereforebe excluded from the first stage probits in the instrumentalvariablemodels. Such simultaneityseems unlikely,however.Moreplausibly,the rent-or-buy decision precedesthe decision to seek rentalpublic housing, which is a credibleoption only for those who chose to remainin, or cannot escape, the rentalhousing market.Practically, issue is moot, because the the inclusion of years of homeownershiphas virtuallyno effect on the key results.

Effects of Housing Assistance

/ 31

Table 1. The changing profile of public housing. 1970 Percentage black Percentage high school graduates Percentage married Percentage receiving welfare Labor earnings mean SD 61.9 28.0 50.4 46.5 $10,087 $9,771 1980 80.8 33.5 24.4 67.1 $4,471 $6,735 1990 80.5 44.2 30.3 72.3 $5,921 $8,184

Source:PSID-Assisted Housing Database. Notes: aSamplelimited to families with children;b1990dollars used for monetary values; cPSIDweights includes AFDC,food stamps, and used to correct for over-samplingthe poor and for attrition; dWelfare "otherwelfare,"SSI is excluded.

Table 2 illustrates several key characteristics of the sample. (Appendix Table A.1 reports the full set of sample univariates.) The sample includes 178 children who lived in public housing during the observation period. Residents of public housing have more disadvantagedbackgroundsand worse outcomes on virtuallyall measures. More than half lived in public housing for at least 6 of the 7-year period between ages 10 and 16. Virtually all of these families received some welfare, and most received welfare for 6 of the 7 years. Families in public housing earned less than eligible unassisted families and lived in neighborhoods with higher poverty and high school dropout rates. The differences in family earnings, welfare receipt, and single parenthood for children who lived in public housing and eligible unassisted children who did not are all statistically significant. The adult outcomes of children who lived in public housing between ages 10 and 16 were also poor. More than 80 percent spent some time on welfare when they were between ages 20 and 27, and more than half spent at least 4 of these 8 years on welfare as young adults. Their adult earnings were 14 percent below those in the eligible but unassisted group, and their adult household earnings relativeto the federal poverty line were 35 percent lower. Differences in years of welfare receipt, individual earnings, and household earnings-to-needs for the two housing groups are all statistically significant. By contrast, there is no difference in the labor force participation rates of children who lived in public housing and those who did not. In both groups, about 90 percent worked for at least some time between ages 25 and 27.

Table 3 presents the regression results for the effects of public housing from both the models that include state dummies and those that include census division dummies. (Complete instrumental variables results from the models using census division dummies are shown in AppendixTableA.3 and are representativeof the other models.) The uninstrumentedresults,which control for observedcharacteristicsonly,indicate that the worse outcomes of children who lived in public housing are entirely attributableto their more disadvantagedfamily background,not public housing itself. All of the statistically significant adverse outcomes associated with public housing observed in the raw differences (Table 2) disappear when individual and family backgroundcharacteristics are controlled for.In particular,being black, being female, renting rather than owning, having parents whose educational attainment is low, and growing up in a welfare-dependent household are the dominant predictors of poor outcomes on the four self-sufficiency indicators tested here. The uninstrumented

32 / Effects of Housing Assistance

Table 2. Illustrative characteristics of analysis sample (weighted). Public housing (N= 178) Unassisted (N = 1005)

Characteristics of children ages 10-16
Number of years in public housing mean median Percentage black Percentage female Annual earnings of head mean median Years in single-parent family mean median Percentage high school graduates (head) Percentage receiving any welfare (household) Number of years receiving welfare mean median Mean neighborhood poverty (tract) Mean percentage high school drop outs (tract)

5.3 6 72 53

0.0 0 28 53

$12,122 $11,635 3.5 4 42 89

$17,328 $15,670 2.8 2 45 54

4.6 6 28 22

2.1 1 18 17

Adult outcomes 20-27
Percentage receiving any welfare 20-27 # yrs. without welfare 20-27 mean median Average earnings 25-27 mean 82 45

4.4 4

6.3 8

Any earnings 25-27 Average household earnings-to-needs 25-27 Percentage high school graduates



88 1.5 67

90 2.3 73

Source:PSID-AHD. Notes: al990 dollars used for monetary values; b"Unassisted" defined as eligible using 80 percent of county median familyincome adjustedfor family size; cPSIDage 27 individualweights used to correctfor includes AFDC,food stamps, and "otherwelfare,"SSI is oversamplingthe poor and for attrition;dWelfare excluded.

Effects of Housing Assistance

/ 33

Table 3. Regression results for years in public housing between ages 10-16. Uninstrumented Coef. With Controls for States Years off welfare 20-27 (tobit) Individual earnings 25-27 (tobit) Household earnings to needs ratio 25-27 (tobit) Ever employed 25-27 (probit) With Controls for Census Divisions Years off welfare 20-27 (tobit) Individual earnings 25-27 (tobit) Household earnings to needs ratio 25-27 (tobit) Ever employed 25-27 (probit) -0.02 49 -0.01 0.00 (0.07) (164) (0.02) (0.01) 0.71 1861 -0.04 0.07 (0.38)* (952)* (0.13) (0.03)** -0.10 -64 -0.01 0.00 (0.07)(170) (0.03) (0.01) 1.77 1616 -0.15 0.11 (1.28)(2325) (0.34) (0.09) SE Instrumented Coef. SE

- p < 0.20, *p < 0.10, **p < 0.05 Source:PSID-AssistedHousing Database. Notes: aWelfare includes AFDC,food stamps and "otherwelfare."SSI is excluded. bl990 dollars used for monetary values.
cN = 1183.

dForprobit estimates, the marginal effect of a 1-yearchange in years in public housing is shown, with all control variables set to their means.

model results that control for states are virtually identical to those that control for census divisions. The suspicion that failure to control for states could lead to biased estimates is therefore unsupported by these models. The instrumented results, which control for both unobserved and observed family background characteristics, show slight positive effects of public housing residence on most outcomes with controls for state, and stronger positive effects with controls for census division. In the most statistically significant result, with controls for census division, every year of public housing residence between ages 10 and 16 is estimated to increase the probability of working between ages 25 and 27 by 7 percentage points (p = 0.01, 90 percent confidence interval = 3-12 percentage points). Less significant, but still notable, every year of public housing residence is also estimated to reduce years of welfare dependence between ages 20 and 27 by 0.71 of a year (90 percent confidence interval = 0.08-1.3 years) and to increase annual earnings between ages 25 and 27 by $1,861 (90 percent confidence interval = $295-$3,427). Both of these results are statistically significant at the 6 percent level, and they remain significant at the 12 percent level when their standard errors are inflated by a liberal 20 percent to account for the possible non-independence of sibling observations.25 Despite the positive effect of public housing on earnings, its effect on household earnings-to-needs is negligible. This result could only occur if children who lived in public housing had larger families when they grew up, or if their spouse had lower earnings.26 Regressions (not shown) on family size, number of children, whether
25 In models using the sample based on metropolitan area income limits and instruments, the estimated effect of a year in public housing on years off welfare was much stronger (b = 2.05, p = 0.02), the effect on employment was weaker (b = 0.10, p = 0.09), and the effect on earnings was larger but marginally less statistically significant (b = 3317, p = 0.07). The estimated effect on household earnings-to-needs was much more positive, but still not statistically significant (b = 0.32, p = 0.16). 26 Recall that this variable is based on earnings of an individual plus his or her spouse relative to the poverty line for their family size.

34 / Effects of Housing Assistance

married, and spouse's earnings if married indicate that living in public housing has no effect on the numberof children an individualhas at age 25. But it raises a female's likelihood of being married at age 25 and, if married, having a spouse with low For males, public housing residence also has a positive, but much smaller,effect on the likelihood of marriage,and it tends to increase spouse'searnings if married.Thus, even though public housing helps children'sindividual long-term outcomes, there is also evidence that public housing may indirectlyhurt long-termoutcomes for females by leading them to make poor choices in the "marriagemarket."This is consistent with the concentration of public housing in high poverty neighborhoods (Newman and Schnare, 1997). Why the same pattern does not apply to males remains unclear. Although the estimates using state controls are consistent with those substituting census division controls, only the latter are statistically significant. Models with state controls were tested because of suspicion that the instrument for public housing might be correlatedwith state-level features that could affect outcomes, potentially biasing results. This worry appearsto be unwarrantedbecause the gains in statistical significance with census division controls are largely achieved by shrinkingstandard errors, not increasing the magnitude of coefficients. In addition, tests identified few cases of states that had significantly different effects within census divisions, and these were controlled for in the models using census divisions. These considerations suggest that the full set of state controlsis unnecessary,and that their inclusion reduces the power of the instruments by absorbing interstate variation.
DISCUSSION earnings.

The results suggest that, during the period 1968-1982, public housing enhanced children's long-term outcomes. The analysis demonstrates that the poor adult outcomes of children who grew up in public housing are entirely attributableto their more disadvantaged family background characteristics, both observed and unobserved, not public housing itself. Every year of public housing residence between ages 10 and 16 is estimated to increase a youth's probability of working between ages 25 and 27 by 7 percentage points, raise annual earnings by $1,860, and reduce welfare use between ages 20 and 27 by 0.70 of a year. These positive effects could arise because public housing improves physical living conditions, reduces residential mobility,or enables families to spend more of their income on items that benefit children's development. But despite its positive effects, public housing is not associated with an above poverty income for females who grew up in public housing. Perhaps this occurs because girls are exposed to a more disadvantaged set of peers than they would otherwise encounter, narrowing their marital prospects. This detrimental effect is a kind of neighborhood effect. The results of this analysis complement those of Currie and Yelowitz (2000), the only other study of the effects of public housing on children'soutcomes. Theyestimate that children who live in public housing are 11 percentagepoints less likely to be held back in school than other children. Our results suggest that these short-termeffects may be translated into long-term gains. It is unclear whether these findings, which pertain to children who lived in public housing during 1968-1982, extend to currentpublic housing conditions. The period coveredby this study predatesthe very heavy concentrationof disadvantagedfamilies in public housing, as well as the concentration of neighborhood poverty (Jargowsky, 1997). While these trends suggest a more negative effect for children living in public

Effects of Housing Assistance

/ 35

housing in the 1980s and 1990s, Currie and Yelowitz (2000) find positive effects on one educational outcome using 1990 census data. Although this analysis focuses on public housing only, it is likely that living in either privately owned assisted housing developments or tenant-based housing assistance would be associated with even more positive long-term outcomes for children than public housing, since families pay the same portion of income (30 percent) for rent, but live in substantially better neighborhoods (Newman and Schnare, 1997). It is important to examine whether public housing affects younger children in the same way it affects the older children studied in this research. The child development literature strongly suggests that early childhood is the critical formative period in a child's life (Duncan and Brooks-Gunn, 1997). If housing quality, residential stability, reduced parental stress, better home learning environments, or more adequate nutrition in early childhood contribute to development, housing assistance during early childhood may have particularly beneficial long-term effects. This research was funded by the U.S. Department of Housing and Urban Development (COPC-05929), the Ford Foundation, and the Rockefeller Foundation. We are grateful for the helpful insights of Robert Moffitt, MarkShroder,Rob Leonard, CarlaPedone, Ophelia Basgal, Linda Campbell, Ellen Merry,and two anonymous reviewers. We also thank Sally Katz and LauraVernon-Russellfor their expert production assistance.

Table A.I. Univariate statistics of variables used in the models. Mean Dependent Variables Years off welfare 20-27 Average annual earnings 25-27 Had any earnings 25-27 Average earnings-to-needs ratio 25-27 Instruments (see text for description) With state controls in preliminary regression With census division controls in preliminary regression Policy Variable Years in public housing ages 10-16 (0, 1) Demographics Cohort (year born + 10) Black (0, 1) Female (0, 1) Mother's age when born Family background Head a high-school graduate (0, 1) Years in one-parent family ages 10-16 Parents changed marital status ages 10-16 (0, 1) Years with disabled family head ages 10-16 Number of children in family Family economic characteristics Whether any time on welfare ages 10-16 (0, 1) Mean annual cash value of welfare ages 10-16 ($ 1,000s) Mean annual labor income ages 10-16 ($1,000s) Mean annual work hours of head < 200 ages 10-16 (0, 1) Geographic and housing characteristics Years in a big city (> 500,000) ages 10-16 Years in a small city (100,000-500,000) ages 10-16 Years in home-owning family ages 10-16 Source: PSID-AHD. a 1990 dollars used for monetary values. b Welfare includes AFDC,food stamps and "other welfare," SSI is excluded. c PSID age 27 individual weights used for weighted statistics. 5.38 10,077 0.86 1.66 2.30 3.54 0.77 1972 0.65 0.55 27.97 0.32 3.35 0.29 2.11 3.90 0.73 3.50 13.39 0.16 2.13 1.28 3.38 Unweighted Median 7 8,504 1.00 1.35 -0.08 1.57 0.00 1972 1 1 28 0.00 3.00 0.00 1.00 4.00 1.00 1.20 11.61 0.00 0.00 0.00 3.00 SD 2.87 9,899 0.35 1.58 10.45 10.99 2.02 2.50 0.48 0.50 6.18 0.47 3.14 0.46 2.54 2.02 0.45 4.61 11.39 0.37 3.11 2.57 3.10

Table A.2. First-stage regression models predicting years in public housing between ages 10-16. Controlling for states Variable Instrument (exogenous supply of public housing)


Coef. SE 0.017 (0.007)
0.415 (0.149)


C 0


Female 0.081 (0.101) Cohort (year born +10) 0.035 (0.021) * Mother's age at birth 0.009 (0.009) Head a high-school grad 0.147 (0.124) Whether family received any welfare 0.095 (0.149) Mean annual welfare receipt ($1,000s) 0.031 (0.020) 0.014 (0.027) Average annual earnings ($1,000s) -0.011 (0.030) Average earnings > $10,000 (spline) Less than 200 hours worked annually 0.355 (0.222) Years in one-parent family 0.013 (0.022) -0.270 (0.116) ** Ever experienced a marital change Number of children in family 0.010 (0.031) Number of years head disabled -0.045 (0.025) * -0.143 (0.020) *** Number of years homeowner -0.003 (0.026) Years in city with > 500,000 pop ulation Years in city with 100,000-500,000 population 0.053 (0.025) ** -70 Constant (42) (State and census division dummies not shown, both are jointly significant at p < 0.001.)
Source: PSID-AHD.

0 0 0 0 0 0 -0 0 0 -0 -0 0 -0 -0 0 0 -56



-p <0.15, *p <0.10, **p <0.05, ***p < 0.01. See Appendix Table A.1 for variable definitions and univariate statistics.

Table A.3. Complete model results, instrumented, with census division controls. Years off welfare 20-27 Coefficient (SE) Years in public housing 10-16 Black Female Cohort (year born + 10) Mother's age at birth Head a high-school grad Whether family received any welfare Mean annual welfare receipt ($1,000s) Average annual earnings Average earnings > $10,000 (spline) 0.71* (0.38)

Earnings 25-27 Coefficient (SE)

Household ear 25-27 Coefficient (SE

(0.38) -1.57*** (0.28) -0.03 (0.06) -0.06** (0.02) 1.34*** (0.35)

(0.41) -0.16*** (0.05) 0.10 (0.07) -0.06 (0.08)

(952) -3215*** (845) -5806*** (619) 3 (125) -88 (55) 2036*** (773) -946 (880) -365*** (120) 247 (163) -230 (180)

-0.04 (0.1 -0.86 (0.1 -0.11 (0.0 0.0 (0.0 -0.01 (0.0 0.5 (0.1 -0.06 (0.1 -0.04 (0.0 0.0 (0.0 0.0 (0.0

Table A.3. continued. Years off welfare 20-27 Coefficient (SE) Less than 200 hours worked annually Years in one-parent family Ever experienced a marital change Number of children in family Number of years head disabled Number of years homeowner Years in city with population > 500,000 Years in city with population 100,000-500,000 Census division = New England Census division = Mid-Atlantic Census division = W. N. Central Census division = Mountain Census division = Pacific Census division = E. N. Central Census division = W. S. Central -0.58 (0.59) 0.15*** (0.06) -0.02 (0.32) -0.08 (0.08) -0.01 (0.07) 0.41*** (0.09) -0.08 (0.06) -0.09 (0.08) 1.88 (1.41) -1.25** (0.54) -0.95 (0.73) -0.73 (2.35) 0.10 (0.65) -1.80*** (0.60) -0.24 (0.60) Earnings 25-27 Coefficient (SE) 453 (1402) 210 (129) -108 (724) -129 (189) 148 (171) 630*** (209) -256* (134) -338* (188) 5751** (2693) -2032 (1524) -867 (1725) -35 (4365) 3855** (1516) -719 (1554) 946 (1469)

Household earn 25-27 Coefficient (SE

0.24 (0.20) 0.04** (0.02) -0.11 (0.10) -0.01 (0.03) 0.00 (0.02) 0.05*


-0.03 (0.02) 0.00 (0.02) 0.26 (0.38) -0.48** (0.22) -0.55** (0.23) -0.78 (0.62) 0.04 (0.20) -0.68** (0.20) -0.48** (0.19)

Table A.3. conitnuzed.
Years off welfare Coefficient (SE) 20-27 Earnings 25-27 Coefficient (SE)

Household earn 25-27 Coefficient (SE



= E. S. Central Jersey

-0.35 (0.57)

New York/New Illinois Florida Constant

60.42 (109.80)

299 (1404) 5535*** (1966) 4534** (1815) 7709*** (2164) 5526 (245739)

-0.30* (0.18) 0.96* (0.28)

0.75* (0.26)

3.21 (35.33)

<0.05, ***p <0.01. *p<0.10,**p Source: PSID-AHD a Welfare includes AFDC, food stamps, and "other welfare." SSI is excluded. b 1990 dollars used for monetary values. NN= 1183. d For probit estimates, the marginal effect of a one-year change in years in public housing is shown, with all control variables set to their means. eExcluded census division is South Atlantic. ' 25 cases with New England and Mountain census divisions dropped from "any earnings" model because all sample members in these areas had earnings. g Controls for individual states are included only in models where the state had significantly different effects from others in the same census division.

Effects of Housing Assistance

/ 41

SANDRA J. NEWMAN is Professor of Policy Studies, and Director of the Institute for Policy Studies, Johns Hopkins University, Baltimore. JOSEPH M. HARKNESS is a Research Statistician at the Institute for Policy Studies, Johns Hopkins University, Baltimore. REFERENCES Allard,S.W.,& Danziger,S. (1999). Proximity and opportunity:How residence and race affect welfare recipients. Paper presented at the Association for Public Policy and Management Research Conference, November 4-6, Washington, DC. Amemiya, T. (1978). The estimation of a simultaneous equation generalized probit model. Econometrica 46, 1193-1205. Amemiya, T. (1979). The estimation of a simultaneous equation tobit model. International Economic Review 20, 169-181. G2SLS in Amemiya, T. (1983). A comparison of the Amemiya GLS and the Lee-Maddala-Trost a simultaneous equations tobit model. Journal of Econometrics 23, 295-300. Astone, N., & McLanahan,S. (1994). Family structure, residential mobility, and school dropout: A research note. Demography 34, 575-584. Brooks-Gunn,J., Duncan, G., & Aber,J. (Eds.). (1997). Neighborhood poverty,vol. 1: Context and consequences for children. New York: Russell Sage Foundation. Coggon, D., Barker,D. J., Inskip, I., & Wield, G. (1993). Housing in early life and later mortality. Journal of Epidemiology and Community Health 47, 345-348. Currie,J. (1995). Welfareand the well-being of children. Chur,Switzerland: Harwood. Currie,J., & Yelowitz, A. (2000). Are public housing projects good for kids? Journal of Public Economics 75, 99-124. Duncan, G., & Brooks-Gunn, J. (1997). Income effects across the life span: Integration and interpretation.In J. Brooks-Gunn,G. Duncan, & J. Aber(Eds.), Consequences of growing up poor (pp. 596-610). New York: Russell Sage Foundation. Fitzgerald, J., Gottschalk, P., & Moffitt, R. (1998a). An analysis of sample attrition in panel data: The Michigan panel study of income dynamics. Journal of Human Resources 33,251299. Fitzgerald,J., Gottschalk,P., & Moffitt, R. (1998b). An analysis of the impact of attrition on the second generation of respondents in the Michigan panel study of income dynamics. Journal of Human Resources 33, 300-344. Fraker,T., & Moffitt,R. (1988). The effect of food stamps on labor supply: A bivariate selection model. Journal of Public Economics 35, 25-56. Galpin, O. P.,Walker,C. J., & Dubiel, A. J. (1992). Helicobacter pylori infection and overcrowding in childhood. Lancet 339, 619. GAO [U. S. General Accounting Office]. (1994). Elementary school children: Many change schools frequently, harming their education. HEHS-94-45. Washington, DC: Government Printing Office. Green, R., & White, M. (1997). Measuring the benefits of homeowning: Effects on children. Journal of Urban Economics 41, 441-461. Haveman, R. B., Wolfe, B., & Spaulding, J. (1991). Childhood events and circumstances influencing high school completion. Demography 28, 133-157. Heckman, J. (1997). Instrumental variables: A study of implicit assumptions used in making program evaluations. Journal of Human Resources 32, 441-462. Houser,S., & Dickert-Conklin,S. (1998). The effects of after-taxwages, transferpayments, and child care expenses on labor market and transfer program participation. Photocopy.

42 / Effects of Housing Assistance

HUD [U. S. Department of Housing and Urban Development]. (1998). Rental housing assistance: The crisis continues. Washington, DC: Government Printing Office. HUD [U. S. Department of Housing and Urban Development]. (2000). Recent Research Results. Washington, DC: Office of Policy Development and Research. Jargowsky, P. (1997). Poverty and place. New York: Basic Books. Jencks, C., & Mayer, S. (1990). The social consequences of growing up in a poor neighborhood. In L. E. Lynn & M. G. H. McGeary (Eds.), Inner-City Poverty in the United States (pp. 111186). Washington, DC: National Academy Press. Jordan, W., Lara, J., & McPartland, J. (1996). Exploring the causes of early dropout among race-ethnic and gender groups. Youth and Society 28, 62-94. Keane, M., & Moffitt, R. (1998). A structural model of multiple welfare program participation and labor supply. International Economic Review 39, 553-589. Kingsley, T., & Tatian, P. (1999). Housing and welfare reform: Geography matters. In S. Newman (Ed.), The home front: Implications of welfare reform for housing policy (pp. 81-122). Washington, DC: Urban Institute Press. Klebanov, P., Brooks-Gunn, J., Chase-Lansdale, P.L., & Gordon, R. (1997). Are neighborhood effects on young children mediated by features of the home environment? In J. BrooksGunn, G. Duncan, & J. Aber (Eds.), Neighborhood poverty, vol. 1 (pp. 119-145). New York: Russell Sage Foundation. Ludwig, J., Duncan, G., & Hirschfield, P. (2001). Urban poverty and juvenile crime: Evidence from a randomized housing-mobility experiment. Quarterly Journal of Economics 116,655680. Ludwig, J., Ladd, H., & Duncan, G. (forthcoming). The effects of urban poverty on educational outcomes, in W. Gale & J. Rothenberg Pack, Brookings-Wharton papers on urban affairs. Washington, DC: Brookings Institution. Mann, S.L., Wadsworth, M.E., & Colley, J.R. (1992). Accumulation of factors influencing respiratory illness in members of a national birth cohort and their offspring. Journal of Epidemiology and Community Health 46, 286-292. Mayer, S. (1997). What money can't buy: Family income and children's life chances. Cambridge, MA: Harvard University Press. Meyers, A. D., Frank, D., Roos, N., Peterson, K., Casey, V., Cupples, A., and Levenson, S. (1995). Housing subsidies and pediatric undernutrition. Archives of Pediatric Adolescent Medicine 149, 1079-1084. Moffitt, R. (1992). Incentive effects of the U.S. welfare system: A review. Journal of Economic Literature 30, 1-61. Moffitt, R. (2001). Policy interventions, low-level equilibria and social interactions. In S. Durlauf & P. Young (Eds.), Social dynamics (pp. 45-82). Cambridge, MA: MIT Press. Moffitt, R., & Wolfe, B. (1992). The effect of the Medicaid program on welfare participation and labor supply. Review of Economics and Statistics 76, 615-626. Murray, M. (1980). A reinterpretation of the traditional income-leisure model, with applications to in-kind subsidy programs. Journal of Public Economics 14, 69-82. Newey, W. (1987). Simultaneous estimation of limited dependent variable models with endogenous explanatory variables. Journal of Econometrics 36, 231-250. Newman, S., & Harkness, J. (1998). Neighborhood poverty, assisted housing, and the educational attainment of children. Photocopy, Institute for Policy Studies, Johns Hopkins University, Baltimore, MD. Newman, S., & Harkness, J. (1999). The long-term effects of housing assistance on self-sufficiency: Final report. Washington, DC: U. S. Department of Housing and Urban Development Research Report. Newman, S., & Harkness, J. (2000). Assisted housing and the educational attainment of children. Journal of Housing Economics 4, 40-63.

Effects of Housing Assistance

/ 43

Newman, S., & Schnare, A. (1993). Last in line: Housing assistance for households with children. Housing Policy Debate 4, 417-455. Newman, S., & Schnare, A. (1997). '...and a suitable living environment': The failure of housing programs to deliver on neighborhood quality.Housing Policy Debate 8, 703-742. Ong, P. (1998). Subsidized housing and work among welfare recipients. Housing Policy Debate 9, 775-794. Page, M. E., & Solon, G. (1999). Correlationsbetween brothers and neighboring boys in their adult earnings: The importance of being urban. Presented at the Joint Center for Poverty Research Conference, Neighborhood Effects on Low-Income Families, September 10. Chicago. Reingold, D. A. (1997). Does inner city public housing exacerbate the employment problems of its tenants? Journal of Urban Affairs 19, 469-486. Shlay,A., & Holupka, S. (1991). Steps toward independence: The early effects of the Lafayette Courts Family Development Center.Photocopy, Institute for Policy Studies, Johns Hopkins University,Baltimore. Shroder,M., & Martin, M. (1996). New results from administrativedata: Housing the poor, or, what they don'tknow might hurt somebody.Washington, DC: U. S. Department of Housing and Urban Development, Office of Policy Development and Research. U. S. House of Representatives Committee on Ways and Means. (1998). 1998 Green Book. Washington, DC: Government Printing Office. Zabel, J. E. (1998). An analysis of attrition in the panel study of income dynamics and the survey of income and program participation with an application to a model of labor market behavior.Journal of Human Resources 33, 479-506.