Public Housing and Paid Work: Help or Hindrance? by Jennifer Amy Stoloff
A dissertation submitted to the faculty of the University of North Carolina at Chapel Hill in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Sociology. Chapel Hill 2001
Approved by
________________________________________ Advisor: Professor Rachel A. Rosenfeld ________________________________________ Reader: Professor Kathleen Mullan Harris ________________________________________ Reader: Professor James H. Johnson, Jr. ________________________________________ Reader: Professor Ted Mouw ________________________________________ Reader: Professor William M. Rohe
2001 Jennifer Amy Stoloff
ii
ALL RIGHTS RESERVED
ABSTRACT JENNIFER AMY STOLOFF: Public Housing and Paid Work: Help or Hindrance? (Under the direction of Rachel A. Rosenfeld)
The main argument of this dissertation is that living in public housing is not a disadvantage to entering the paid labor force and may be an advantage due to affordable rents. Public housing residents have more time to search for work than non-residents, as well as the option of accepting lower paying jobs without the risk of losing their housing. The scant previous research on this topic finds either a negative and significant relationship between work and public housing or an insignificant relationship. My hypothesis is that if there is a negative effect of public housing on working it should diminish if the proper controls are used in the analysis and if public housing residents are compared to a group of similar people who are not in public housing. In order to explore the effect of public housing on those who are working, I will also look at changes in the number of hours worked per year. In order to define the sample for this study, the technique of “group matching” is used. Because the public housing and non-public housing populations are quite different, it is appropriate to use matching to limit the range of inference. The characteristics on which public housing residents are matched with non-residents include education, work experience, age, family size, race/ethnicity, welfare receipt, and marital status. The Panel Study of Income Dynamics (PSID), a longitudinal data set, is analyzed with event history methods, looking at seven years of data. The main finding is that public housing residence increases the likelihood of leaving work, but has no effect on the likelihood of a transition to work or on whether or not annual hours of work increase or decrease beyond a given threshold.
iii
ACKNOWLEDGMENTS My deepest thanks go to my chair and advisor, Rachel Rosenfeld. Rachel always encouraged me and found positive ways to give me comments and suggestions. It has been a privilege working with her. Jim Johnson has been a mentor, sparking my interest in urban poverty and providing me with many research opportunities. Bill Rohe provided outstanding comments on the proposal and an invaluable perspective. Ted Mouw graciously agreed to join the committee on short notice and I am very pleased to have had the opportunity to work with him. Mike Stegman played an important role in the development of the project. Kathie Harris was vital to the process and has spurred me to do my best work. My parents, Judith and David, have helped me in innumerable ways during my life as a student. They have always supported me emotionally, encouraged me, and never seemed to doubt that I would one day complete my dissertation. There is no way to adequately thank them. I would like to thank all of my fellow graduate students at Chapel Hill for the collegial environment they created. Among my friends (most of whom are no longer students), I am especially grateful to Jennifer Glanville, Pam Paxton, Kathleen Nebeker, Elisa Bienenstock and Marion Hughes for reading drafts, answering questions and listening to my ideas. Finally, I would like to thank my colleagues at the U.S. Department of Housing and Urban Development, without whose tolerance and encouragement I would not have been able to complete the dissertation. I would especially like to acknowledge Kevin Neary, Todd Richardson, Mark Shroder, and Carlos Martín.
TABLE OF CONTENTS Chapter
Page
LIST OF TABLES .......................................................................................................................... VI LIST OF FIGURES........................................................................................................................ VII 1: INTRODUCTION ........................................................................................................................8 1.1: PREVIOUS RESEARCH ON THE RELATIONSHIP BETWEEN PUBLIC HOUSING AND WORK ................................ 11 1.2: GROUP MATCHING ..................................................................................................................... 12 2: PUBLIC HOUSING BACKGROUND ............................................................................................14 2.1: TARGET POPULATION.................................................................................................................... 14 2.2: SITE SELECTION ............................................................................................................................ 16 2.3: FINANCING ................................................................................................................................. 18 2.4: DESIGN ...................................................................................................................................... 19 3: THEORETICAL FRAMEWORK ...................................................................................................21 3.1: PLACE ........................................................................................................................................ 22 3.2: PERSON...................................................................................................................................... 30 3.3: PERSON AND PLACE ..................................................................................................................... 33 3.4: RESEARCH QUESTIONS .................................................................................................................. 34 4: DATA AND METHODOLOGY ....................................................................................................35 4.1: PSID (PANEL STUDY OF INCOME DYNAMICS) ................................................................................... 35 4.2: U.S. CENSUS & COUNTY DATA ...................................................................................................... 37 4.3: SELECTION/PROPENSITY ................................................................................................................ 39 4.4: MULTIVARIATE ANALYSIS ............................................................................................................... 42 5: RESULTS .................................................................................................................................46 5.1: LIFE TABLE ANALYSIS .................................................................................................................... 46 5.2: EVENT HISTORY ANALYSIS ............................................................................................................. 50 5.3: SOCIAL RESOURCES ANALYSIS ........................................................................................................ 79 DISCUSSION ....................................................................................................................................... 82 6: CONCLUSIONS ........................................................................................................................83 APPENDIX A: CATEGORIES OF DATA AVAILABLE IN THE PSID .......................................................89 APPENDIX B: VARIABLES ............................................................................................................91 APPENDIX C: CENSUS ATTRITION ................................................................................................94 C1: TRANSITION FROM WORK TO NON-WORK ......................................................................................... 96 C2: TRANSITION FROM NON-WORK TO WORK ......................................................................................... 99 C3: INCREASE IN ANNUAL HOURS WORKED............................................................................................ 100 C4: DECREASE IN ANNUAL HOURS WORKED........................................................................................... 101 C5: CONCLUSIONS ............................................................................................................................ 103 APPENDIX D: DIAGNOSTIC TESTS .............................................................................................. 103 D1: MULTICOLLINEARITY ................................................................................................................... 103 D2: OUTLIERS .................................................................................................................................. 107 APPENDIX E: PREDICTED PROBABILITIES ................................................................................... 113 REFERENCES ............................................................................................................................ 118
LIST OF TABLES
TABLE 4.1: VARIABLES INCLUDED IN PROPENSITY SCORE MODEL (1986) ........................................... 41 TABLE 5.1: DESCRIPTIVE STATISTICS FOR THOSE WORKING IN 1986 ................................................... 51 TABLE 5.2: DESCRIPTIVE STATISTICS FOR THOSE NOT WORKING IN 1986........................................... 53 TABLE 5.3: TRANSITION FROM WORK TO NON-WORK, BASE MODEL ................................................. 54 TABLE 5.4: TRANSITION FROM WORK TO NON-WORK, PLACE MODELS ............................................. 55 TABLE 5.5: TRANSITION FROM WORK TO NON-WORK, PERSON MODELS .......................................... 57 TABLE 5.6: TRANSITION FROM WORK TO NON-WORK, FULL MODELS ................................................ 59 TABLE 5.7: TRANSITION FROM NON-WORK TO WORK, BASE MODEL ................................................. 61 TABLE 5.8: TRANSITION FROM NON-WORK TO WORK, PLACE MODEL ............................................... 62 TABLE 5.9: TRANSITION FROM NON-WORK TO WORK, PERSON MODEL ............................................ 63 TABLE 5.10: TRANSITION FROM NON-WORK TO WORK, FULL MODELS.............................................. 66 TABLE 5.11: INCREASE IN ANNUAL HOURS WORKED, BASE MODEL ................................................... 68 TABLE 5.12: INCREASE IN ANNUAL HOURS WORKED, PLACE MODEL.................................................. 68 TABLE 5.13:INCREASE IN ANNUAL HOURS WORKED, PERSON MODEL ............................................... 70 TABLE 5.14: INCREASE IN ANNUAL HOURS WORKED, FULL MODELS .................................................. 71 TABLE 5.15: DECREASE IN ANNUAL HOURS WORKED, BASE MODEL................................................... 73 TABLE 5.16: DECREASE IN ANNUAL HOURS WORKED, PLACE MODEL ................................................. 75 TABLE 5.17: DECREASE IN ANNUAL HOURS WORKED, PERSON MODEL.............................................. 76 TABLE 5.18: DECREASE IN ANNUAL HOURS WORKED, FULL MODEL ................................................... 78 TABLE 5.19: JOB SEARCH ...................................................................................................................... 80 TABLE 5.20: CHILD CARE COSTS ............................................................................................................ 82 TABLE C1: DESCRIPTIVE STATISTICS FOR THOSE WORKING IN 1986, CENSUS VARIABLES EXCLUDED 94 TABLE C2: DESCRIPTIVE STATISTICS FOR THOSE NOT WORKING IN 1986, CENSUS VARIABLES EXCLUDED ............................................................................................................................................. 96 TABLE C3: TRANSITION FROM WORK TO NON-WORK, FULL MODELS................................................. 98 TABLE C4: TRANSITION FROM NON-WORK TO WORK, FULL MODELS................................................. 99 TABLE C5: INCREASE IN ANNUAL HOURS WORKED, FULL MODELS ................................................... 100 TABLE C6: DECREASE IN ANNUAL HOURS WORKED, FULL MODELS .................................................. 102 TABLE D2: MULTICOLLINEARITY FOR TRANSITION FROM NON-WORK TO WORK ............................. 104 TABLE D3: MULTICOLLINEARITY FOR INCREASE IN ANNUAL HOURS WORKED ................................. 106 TABLE D4: MULTICOLLINEARITY FOR DECREASE IN ANNUAL HOURS WORKED ................................ 107 TABLE D5: TRANSITION FROM WORK TO NON-WORK, FULL MODEL ................................................ 108 TABLE D6: TRANSITION FROM NON-WORK TO WORK, FULL MODELS .............................................. 109 TABLE D7: INCREASE IN ANNUAL HOURS WORKED, FULL MODELS ................................................... 110 TABLE D8: DECREASE IN ANNUAL HOURS WORKED, FULL MODEL .................................................... 111
vi
LIST OF FIGURES FIGURE 5.1: HAZARD PLOT OF THE TRANSITION FROM WORK TO NON-WORK ...................................................... 47 FIGURE 5.2: HAZARD PLOT OF THE TRANSITION FROM NON-WORK TO WORK ...................................................... 48 FIGURE 5.3: HAZARD PLOT OF AN INCREASE IN ANNUAL HOURS WORKED ........................................................... 49 FIGURE 5.4: HAZARD PLOT OF A DECREASE IN ANNUAL HOURS WORKED ............................................................. 50 FIGURE 5.5: TRANSITION FROM WORK TO NON-WORK .................................................................................... 59 FIGURE 5.6: TRANSITION FROM NON-WORK TO WORK .................................................................................... 65 FIGURE 5.7: DECREASE IN ANNUAL HOURS WORKED ....................................................................................... 76 FIGURE E1: PROBABILITY OF A TRANSITION FROM WORK TO NON-WORK .......................................................... 114 FIGURE E2: PROBABILITY OF A TRANSITION FROM NON-WORK TO WORK .......................................................... 115 FIGURE E3: PROBABILITY OF AN INCREASE IN ANNUAL HOURS WORKED ............................................................ 116 FIGURE E4: PROBABILITY OF A DECREASE IN ANNUAL HOURS WORKED ............................................................. 117
1: Introduction In this research, I will examine the relationship between working for pay and living in public housing. Funding and political support have been granted inconsistently to the conventional public housing program since its initiation in 1937, yet it remains a reliable, if small, source of low-income housing. In 1989, which falls close to the mid-point of the data used in this study, slightly over 4% of all renters lived in public housing (Casey 1992). Unfortunately, the problems faced in the urban areas of the U.S. seem exemplified by the residents of urban public housing. In 1989, 83% of all public housing units were located in urban areas, and 69% of the units were in central cities (Casey 1992). Public housing also disproportionately houses minorities; in 1989, 53% of public housing tenants were Black and 12% were Hispanic, with most of the remainder categorized as White. Even though rural poverty and lack of housing may be as common, urban ‘blight’ and distressed public housing remain popular targets for conservative and reform-minded politicians who believe that public housing causes or, at least, intensifies the problems of its residents. There is a great deal of concern about the negative impact of public housing on labor market participation, in part because only 35% of public housing residents have income from wages or salary (Casey 1992). However, this figure includes the approximately 38% of households headed by an elderly tenant (age 65 or older). The main questions addressed in this research deal with the independent effects of public housing residence on labor market participation. There is evidence to suggest that public housing serves as both an advantage and disadvantage to entering the paid labor force (Edin and Lein 1997). Living in public housing, with its lower rent, may give poor people the opportunity to pursue educational and job training programs, which would lead to increased labor force participation and better wages. Alternatively, public housing projects that are spatially distant from jobs may negatively stigmatize residents, making them less attractive to potential employers (Kirschenman and Neckerman 1991). There is very little direct, empirical evidence on the relationship between public housing residence and employment, though much ethnographic work finds detrimental aspects of public housing residence (see, for example Furstenberg 1993; Popkin, Gwiasda, Olson, Rosenbaum, and Buron 2000; Sullivan 1989). Research that examines this question has decidedly mixed results, though it mostly discusses public housing as a component of some larger phenomenon—e.g., race, poverty or welfare. In this context, it is hard to separate public housing from the many serious problems likely to be faced by residents. The question asked in this research is, after controlling for the many negative correlates of public housing residence, do public housing residents have a lower likelihood of employment than similar non-residents? When considering public housing, large dysfunctional projects such as Cabrini-Greene or Pruitt-Igoe may come to mind. In this study, I will examine whether employment impacts are apparent in a national-level longitudinal sample that compares public housing residents and non-residents. To understand how public housing affects its residents, I compare them to a similar group of nonresidents. 1 It is important to make the correct comparison between actual residents and people
1
The terms “non-public housing resident” and “non-resident” are used interchangeably. When “residents” are mentioned, I am referring to public housing residents.
with a high propensity of being residents. 2 Findings from this research will give a better idea of the labor market consequences for the group of people who are served by public housing, as well as those who are potentially served by public housing. This research is especially relevant because it examines the labor market position of a disadvantaged group of people whose earning power has been deteriorating over the last 25 or 30 years (Wilson 1996). Wilson (1987, 1996) postulates that it is harder and harder for lower skilled workers to find jobs that pay well enough to support their families and, in some cases, it is difficult to find employment at all (Wilson 1987; Wilson 1996). The concentration of a low-income, low-wealth, low-skilled population in public housing developments may make their labor market problems more acute. I will argue, though, that this is only indirectly the fault of public housing as an institution. While the people who are screened into public housing are often those least in demand in the labor market, many income-eligible applicants are not housed (Casey 1992). Public housing, as a social welfare intervention, has received relatively little attention in the literature about poverty and welfare. The labor market problems of the urban poor have been ascribed to many different things, including: lack of skills; distance from jobs or distance from jobs at the appropriate skill level; discrimination and racism; stigma associated with place of residence; welfare dependency; family responsibilities; and lack of social networks and social capital. In this research, I ask if public housing represents a unique barrier to employment, over and above the other disadvantages associated with urban poverty. In other words, I try to see if public housing causes problems for people, or concentrates people with problems. Public housing residents are selected into public housing on characteristics that are inversely related to labor force participation, thus there is the necessity for creating a comparison group of similar non-residents. My starting assumption is that public housing should not harm its residents and, other things being equal, may serve as a benefit to them. This is in contradiction to most of the discussion about public housing, although, as I show in a later section, it is not necessarily contradicted by the empirical research. It is through this comparison, after introducing controls for many of the factors mentioned above and discussed below, that a positive effect of public housing might become evident. I could have chosen several other places as my starting point. It can be argued that the effects of public housing are obviously negative and pernicious; therefore, a negative effect of public housing must necessarily be expected. I will briefly discuss the justification for these perspectives. The hypothesis that residents of public housing are disadvantaged in the labor market, and that this disadvantage is connected directly to their place of residence, is grounded in two schools of thought, one quite conservative the other politically liberal. The conservative perspective views public housing as an exacerbating factor that increases dependency for already disadvantaged people. The residents’ low skills and lack of motivation are emphasized in this perspective. It is thought that people have a better chance of becoming self-sufficient if they have no state supports so they would be forced to be more self-reliant. Since it is hypothesized that welfare and other 2
This type of analysis is similar to recent work by Duncan, Duniform, Doran and Yeung (1998) that compares welfare families to low-SES single-mother working families and finds very few differences in both psychological and skills dimensions.
9
state-supported safety net programs induce dependency (Murray 1984), public housing only increases that dependency further. This perspective assumes that the beneficiaries of public assistance are not self-reliant and are willing to exploit the system rather than work for pay. The more liberal perspective posits that public housing causes or exacerbates structural disadvantages, particularly social isolation and poor social networks. Living in public housing developments that are set apart physically from other parts of a neighborhood causes social and spatial isolation. If public housing tenants are physically separated from their neighbors, it hurts their access to jobs and weakens their social networks. If there are few employed people in the housing project or the neighborhood at large, casual connections to jobs through friends and acquaintances (i.e., social networks) are very constrained. For example, in one neighborhood in Brooklyn, public housing tenants had less access to job information than residents of the same neighborhood who lived in private housing, suggesting a direct negative effect of public housing residence on obtaining employment (Kasinitz and Rosenberg 1996). The stigma of living in projects that are visually and spatially set apart from the community leads to discrimination by some employers and thus lower employment (Kirschenman and Neckerman 1991). Additionally, some very dangerous public housing projects cause parents to choose to stay home to care for children, thus lessening their participation in the labor force (Popkin et al. 2000). Another reason to expect a negative correlation between public housing and work is the potential income disincentive connected to rent structures in public housing. Rents are generally set at 30% of the tenant’s income, which quickly becomes a tax on income as income increases. This high marginal tax rate on residents’ income may be a disincentive to work or, at least, to report total income. In public housing, a working family may pay much more rent than a non-working family living in equivalent public housing, and this may limit the ability of the family to save and potentially move. Tenants are well aware of this situation and readily express their frustration at not being able to save money (Stoloff 1999). This also adds to a feeling of unfairness among residents who are often aware of each other's work status. However, the rent structure can provide a safety net for people who lose their jobs, as well as facilitate job re-entry. While many residents are frustrated with rent increases, others recognize the legitimacy of the rent structure and can manage their work effectively within it (Stoloff 1999). On the positive side, early social reformers thought that by providing better living conditions, many social evils would be held at bay (Riis 1890, Addams 1960 (1910)). A more pragmatic argument in favor of providing decent affordable housing can be made in that if housing is stable, safe and sanitary, it provides a base of security from which work can be comfortably sought. By removing the stress of making high housing payments, or having to move frequently either because of evictions or unsafe or unhealthy living conditions, it is simply easier to focus on other goals. Since housing is affordable, residents have more flexibility in choosing jobs. Public housing could facilitate educational and training opportunities by directly targeting social services to tenants. Public housing could also represent resources for residents, if, for example, neighbors could provide low cost childcare or other kinds of social support (Hasell and Scanzoni 1997). In addition, if social services are made more available to residents because of proximity, public housing may provide a labor market advantage for them (for examples, see Lassen 1995).
10
Research by Edin and Lein (1997) found that for welfare-reliant mothers, public housing residence was something of a financial advantage, compared with similar women who lived in private housing. Public housing residents were somewhat better off and found it easier to pursue work. While AFDC (Aid to Families with Dependent Children, now TANF (Temporary Aid to Needy Families)) was never enough to pay all the bills and unreported side work was almost universal (Edin and Lein 1997), public housing eased the tension by not requiring the tenant to pay an exorbitant proportion of her income for rent. Public housing gave these women the flexibility to take lower paying jobs that made labor market entry easier and also allowed them to enroll in education and training programs. Additionally, these mothers had the option to choose part-time jobs that gave them more flexibility regarding their childcare arrangements. It should be noted that the issues regarding the marginal tax on income are different for public housing residents who received AFDC. When AFDC recipients start working, their grants are reduced one dollar for each dollar of earned income; however, their public housing rent would not increase because it is based on total income. Thus, AFDC recipients living in public housing may be more likely to enter the labor force than non-AFDC recipients because the income disincentives imposed by public housing residence are not as great for them. In fact, one might expect AFDC recipients living in public housing to be more likely to work than AFDC recipients living in private, non-subsidized housing. Another possibility is that public housing has no independent effect on labor market outcomes, positive or negative, but serves as a proxy for the characteristics of a highly disadvantaged group of people with housing needs not adequately served in the private market. If the characteristics of tenants are the reason for their labor market disadvantage, rather than the fact that they live in public housing, then public housing should have no effect on employment. The unmeasured characteristics of tenants, such as poor literacy skills, a criminal record, poor speaking skills or poor presentation skills, would be likely to show up as a negative correlation between employment and public housing residence. Thus, if no effect or a very small negative effect is found, there is still strong support for the hypothesis that public housing, in itself, does not cause harm to tenants. In the following section, I will review the research that relates most directly to my study as part of the explanation for why I chose a positive perspective as my starting point.
1.1: Previous research on the relationship between public housing and work
Most of the previous research on this topic, though scant, finds a negative and insignificant relationship between work and public housing. My previous research, which found a negative relationship between public housing and work, sparked my interest in this area (Glanville, Stoloff, Johnson, and Bienenstock Forthcoming; Stoloff, Glanville, and Bienenstock 1999). The studies I review here all have as their primary focus the relationship between living in subsidized housing and employment. Reingold’s (1997) research asks whether “inner city public housing exacerbates the employment problems of its tenants.” The analysis presented in his article is a logistic regression using the Urban Poverty and Family Life Survey (UPFLS). The UPFLS is a survey of high poverty neighborhoods in Chicago, and is the same study used in research by Wilson (1996) and Kirschenman and Neckerman (1991). Because of the nature of Reingold’s data, there was a de facto group match. All of the respondents in the UPFLS are poor and live in high poverty neighborhoods. In the logit models, all
11
the respondents were Black, female, and received welfare. For each of the dependent variables, “employed,” “self-employed,” “informal work,” and “ever worked,” the effect of public housing residence was insignificant and negative. Reingold argued that, given that Chicago projects are subject to some of the worst stigma effects in the country, this is a lenient test. If no effect of public housing is found in this sample, it is unlikely that effects would be found elsewhere. Newman and Harkness (1999), in a report to the Department of Housing and Urban Development (HUD), examined the “long term effects of housing assistance on self-sufficiency.” They used the PSID-AHD (assisted housing database), a unique merge of the PSID (1968-1993) with several other sources of data about assisted housing. This is not public-use data due to the address matching necessary for its construction. Newman and Harkness (1999) examined the effects of childhood assisted housing receipt on adult outcomes. Using several different estimation techniques (OLS, Ordered Logit, Tobit, and Probit), they looked at four major outcomes: welfare receipt, earnings above poverty, total earnings, and educational attainment. After including controls and using an instrumental variable approach, they found that living in assisted housing between the ages of 10-16 had no negative impact on any of the adult outcomes. In the models estimating total earnings and earnings above poverty, public housing had a positive, though insignificant, effect. They also found no evidence that families living in privately-owned assisted housing, compared with families living in public housing, have better long-term self-sufficiency outcomes. Importantly, Ong (1998) found a positive and significant relationship between receiving a tenantbased housing subsidy (Section 8) and hours worked, though his results for public housing residents were positive and insignificant. He hypothesized that the mobility and residential choice available with a housing voucher, which is person- rather than place-based, improves the employment opportunities for subsidy recipients. Ong’s (1998) research explicitly compared public housing tenants to similar people who were non-residents and his sample was a pool of welfare recipients. A similar effect was seen in the work of Edin and Lein (1997) among residents of subsidized housing, though their sample was not random. With the exception of Newman and Harkness (1999), most of the previous research on this topic used cross-sectional data and local-level studies. This research uses national level longitudinal data and, while it is similar in scope to Newman and Harkness (1999), I use a different method and test different hypotheses. The literature that addresses this particular question has produced little evidence that directly supports or contradicts my hypotheses. The research that does exist is only suggestive regarding the question of whether public housing has a positive or negative effect on paid work. None of the previous studies found a strong negative correlation between public housing and employment. The studies seem to provide support for the hypothesis that public housing has little direct negative impact on paid work.
1.2: Group Matching
Group matching is one of the keys to the study’s value; it provides a vehicle by which to examine the independent effect of public housing on its residents. Group matching allows me to test the hypothesis that, other things being equal, public housing residents may have a labor market advantage compared to similar non-residents because of the benefits of living in affordable housing. It is expected that at baseline, though, public housing residents will be more disadvantaged in the labor market because of the nature of the ways in which they were selected for entry into public 12
housing. However, as discussed above, there are several potential benefits of public housing that could either outweigh or balance the negative characteristics of the residents. Because public housing is inherently selective (specific criteria are used to determine program eligibility, though there is variation across localities), it is not appropriate to compare tenants to all non-tenants. Public housing is a relatively rare event and tenants are quite different than the general population. If we view public housing as a ‘treatment’ and match residents with similar nonresidents, then we will be closer to understanding the treatment effects of public housing on the groups of people who have the potential to be exposed to them. Not all people are equally likely to live in public housing, and because the public housing and non-public housing populations are dissimilar, it is appropriate to use matching to limit the range of inference (Smith 1997). One danger is that long waiting lists and the poor quality of units may select on negative unobservable factors. The intent of this research is to match tenants with respondents from the population at large who have a high likelihood of living in public housing. The goal is to isolate the public housing effect from the other negative observable factors associated with poverty. However, even after matching the public housing group with a similarly disadvantaged group, it will still be necessary to perform multivariate analyses to control for as many of the negatives associated with public housing residence as possible. The characteristics on which public housing residents are matched with non-residents include education, work experience, age, family size, race/ethnicity, welfare receipt, income level, 3 and marital status. Family background characteristics are not included, with the exception of an indicator for parents’ poverty. The purpose is to match to a group of people similar to the group of 1986 public housing respondents. It is appropriate to look directly at the individual and other characteristics of those respondents, rather than using less reliable retrospective measures. Using the Panel Study of Income Dynamics (PSID), 4 a national level longitudinal data source, I explore the independent effects on labor market behavior of living in public housing. Specifically, I examine the effects of public housing residence on the dependent variables “working for pay” (operationalized as a transition from work to non-work and a transition from non-work to work) and changes in annual hours worked. These two types of dependent variables allow me to explore the factors that influence whether one is working at all, as well as the factors that affect changes in labor market participation for those who are already working for pay. By providing historical context and examining various hypotheses about the potential impacts of public housing on labor market participation, empirical models can be tested and the relationship between work and public housing can be more thoroughly understood. I review the history of the public housing program, provide a theoretical framework, propose hypotheses about public housing and work, discuss the data and methods I use to examine them, and present the results of the analysis.
3
Median income is one of the ways potential public housing residents are screened, with low income levels, below 80% or 50% of median income, comprising one of the basic eligibility criteria for entry to public housing.
4
Some of the data used in this analysis are derived from Sensitive Data Files of the Panel Study of Income Dynamics, obtained under special contractual arrangements designed to protect the anonymity of respondents. These data are not available from the author. Persons interested in obtaining PSID Sensitive Data Files should contact PSID staff through the Internet at
[email protected].
13
The next chapter summarizes the history of the public housing program. In Chapter 3, I outline the different theories that form the framework for the models that I test. The discussion is a detailed review of the place (ecological) and person (individual) based elements of the models. Chapter 4 summarizes the data I use for the analysis and presents the methodology for the construction of the propensity score used for the group matching, as well as the multivariate analysis. Chapter 5 presents results for the hazard plots, discrete time logit models, and the social resources analysis. The conclusions developed from the research are found in Chapter 6.
2: Public Housing Background This chapter gives a brief overview of the history of public housing. Public housing is a program introduced at the federal level in 1937 that provides for public financing of low-cost housing in the form of publicly-managed and owned multi-family developments. This chapter is organized, slightly modifying Hays (1995), into six sections that are seen as critical to understanding the history and development of public housing: 1) target population; 2) site selection; 3) financing; and 4) design.
2.1: Target population
Public housing was not originally built to house the ‘poorest of the poor,’ but was intended for select segments of the working class (Atlas and Dreier 1992; Bauman, Hummon, and Muller 1991; Marcuse 1995; United States 1937). Specifically, it was designed to serve the needs of the ‘submerged middle class,’ who were temporarily outside of the labor market during the Depression. After World War II, many working-class people were able to buy their own homes using low-interest mortgages through the VA and FHA (Bratt 1986). These benefits were targeted to Whites and helped move Whites to suburbs and keep Blacks in cities (especially in the North). This made it possible for mostly White working-class people to move out of public housing, and contributed to a downward income shift in the public housing population after the 1940's. The discriminatory nature of these practices has been well documented by Massey and Denton (1993). Public housing has also been thought of as a solution for inner-city poverty and isolation, and as a basic human necessity for less well-off people (Marcuse 1986; Riis 1890; Stegman 1990). The view of many planners, architects and social workers was that good housing was humane and necessary to the well-being of all people and would greatly improve life chances for slum dwellers. They saw public housing as way of fulfilling part of the state's responsibility to ensure that decent, affordable housing was available for all residents of the U.S. Early reformers were appalled by the conditions of the tenements where immigrants lived. They called for the demolition of the tenements, an end to windowless interior rooms, better air circulation and more light. They ascribed many of the undesirable qualities of the poor to their unsafe and unsanitary living conditions. By the turn of the century, housing commissions had been set up in several major cities in order to impose some regulations on landlords (Marcuse 1986). The first national housing legislation was passed in 1937 after a long struggle in the Congress. Beyond providing low-cost housing, the other purpose behind the original 1937 legislation was to improve the lagging economy by providing jobs in the building industry. Indeed, 14
public housing was never conceived of as providing long-term permanent housing. The explicit purpose of the act was “...to alleviate present and recurring unemployment and to remedy the unsafe and insanitary housing conditions and the acute shortage of decent, safe and sanitary dwellings for families of low income...” (United States 1937). The addition of ‘alleviation of unemployment’ as one of the purposes of the act was a way in which the original legislation was modified in order to be accepted by Congress. The act also provided for slum clearance and the provision of replacement “low-rent housing.” This housing was to be consumed by “families of low income,” which had a rather loose definition as, “...families...in the lowest income group who cannot afford to pay enough to cause private enterprise...to build an adequate supply of decent, safe, and sanitary dwellings...” (United States 1937). The only directive for income screening of tenants was that their incomes be no higher than five times the rental cost of the unit (six times in the case of families with three or more children). Some of the earliest advocates of public housing supported tenant screening and they knew that, to have a successful housing development, most residents must be employed (Bauer 1957; Spain 1996). Qualitative tenant screening was the norm when public housing was first built in the late 1930's (Marcuse 1995). These practices were challenged in the 1960’s but there was a paradoxical criticism of the way public housing was managed; on the one hand some managers were criticized for the laxity of rules, while others were maligned for being too strict and moralistic, demonstrating the inconsistent standards by which public housing was judged (Hays 1995). In the 1950’s very strict tenant policies were enforced. Unwed pregnant women could be evicted and large fines for property damage were imposed. Other criteria were that families have two parents, the head of the household hold a job, and the families have some record of good housekeeping skills. In fact, visits were made to future tenants’ previous dwellings to see if they were suitable candidates. It was also common to make spot checks in public housing developments to make sure units were being well cared for. Even into the second and third decades of public housing, spot checks to catch extra tenants (especially men living with unmarried welfare recipients) were not infrequent (Bratt 1986; Marcuse 1995) and still continue in some places. In the 1940’s and 1950’s, income limits had the effect of penalizing residents for upward mobility. Families could be evicted if their income surpassed a strict upper limit. The Housing Act of 1949 introduced subsidized housing programs other than public housing, and included a housing priority for very low-income people, income limits, and maximum rents (rents were required to be 20% less than lowest market rates) (Bratt 1986; United States 1949). This benefited business interests by limiting the program to the very poor and leaving the working class to be housed by private builders. Limiting the program in this way ensured non-competitiveness with the private sector, and was not motivated by a desire to serve the neediest in society (Schill 1991). In the late 1960's, further incentives were introduced to encourage the involvement of private developers and real estate interests in the development of low-cost housing in the form of public financing of private subsidized housing developments (HUD programs such as sections 235, 236, 221d, and 8). These programs “…gave private developers tax breaks, low-cost mortgages, and rent subsidies to house the poor,” (Atlas and Dreier 1992). This marked the beginning of corrupt practices in the administration of some housing subsidy programs that led to the HUD scandals of the 1970s, which were visited again in the 1990s (Atlas and Dreier 1992; Hays 1995). While the public housing program was not directly implicated in the abuses, the problems weakened support for all federal
15
housing programs. Despite problems in the implementation of housing subsidy programs, the direction of housing policy was steadily moving away from supply-based models and towards subsidized private development and demand-based delivery systems, such as housing vouchers (Orlebeke 2000). Ironically, the Civil Rights Act of 1964, while ending legal discrimination by no longer allowing racially segregated projects, contributed to the flight of Whites from projects. Whites moved out of projects when they became racially integrated. Over time, advocates for the poor asked that preferences on waiting lists be given to the most disadvantaged applicants, in particular to the homeless and disabled. This, combined with income limits, ensured that public housing residents were drawn from the least well-off segments of society. Public housing became less attractive to anyone who could afford to live elsewhere. In 1981, rent ceilings were eliminated, which potentially made public housing even less attractive to its higher-income residents. Additionally, the proportion of tenants with incomes over 50% of median was limited. Rents were changed to reflect a payment of 30% of adjusted income, an increase from 25%. Discretion as to how to calculate standard deductions from total income was largely removed from the public housing authorities; in 1983, Congress established standard deductions for minors, elderly heads of households, and for other allowable expenses (Feins, Merrill, Kutty, Heintz, and Locke 1994). Rent ceilings were reintroduced in 1987, but federal preferences for tenants were also implemented (Feins et al. 1994). This continued a trend of less control by the public housing authorities and a shift in the public housing population to a more disadvantaged segment of society. The preferences were for tenants who were involuntarily displaced, living in substandard housing or paying more than 50% of their income for rent. In 1990, a 10% limit on annual rent increases was implemented. In 1992, some flexibility was restored to local public housing authorities and federal housing preferences were largely eliminated. It is impossible to say exactly which standards were used to select tenants, since local housing authorities could impose additional preference criteria when screening for residence. At a minimum, during most of the period of this analysis, federal preferences were in effect though, at any particular housing agency, additional preferences may have been used as well. All of the residents in this study were paying 30% of their adjusted income for rent.
2.2: Site selection
Site selection was, initially, completely under local control. It was not until the court challenges of the early 1960’s that the federal government interfered with many of the discriminatory site selection practices at the local level. Racial segregation in public housing, implemented by site selection strategies, was the norm and reflected larger patterns of residential segregation in the U.S. Projects were often race-specific and more often designated for Whites than Blacks (Bratt 1986; Marcuse 1986; Massey and Denton 1993). The racial segregation of housing projects was often a deliberate decision on the part of the local housing authorities. For example, in New York the Williamsburg Houses project in Brooklyn was built in 1935 for Whites, and Harlem River Houses project in Manhattan was built to house Blacks. Harlem River Houses was seen as a way to prevent
16
demand by African-Americans for access to the housing being provided in all-White communities (Marcuse 1986). The situation in Chicago exemplifies the problems caused by racial segregation. By the l960’s Chicago had some of the most segregated public housing in the U.S. Because city council members had veto power when it came to placing new public housing developments in their wards, almost all units were located in black neighborhoods. The series of court cases now known as Gautreaux successfully challenged this situation, but desegregation is still not a reality in Chicago (Hays 1995). The implementation of Gautreaux faced massive resistance, and public housing development was essentially halted for several years until the laws were changed and the Chicago Housing Authority was allowed to operate without city approval, at least for a short period. The eventual remedy mandated by Gautreaux included issuance of Section 8 certificates that residents could use to move to privately-owned housing in mostly white suburbs. 5 In other cities, similar court cases challenging the pattern of excluding public housing from White, suburban areas were only successful when discriminatory zoning practices could be proved. The result of the successful cases was that HUD (U.S. Department of Housing and Urban Development) issued regulations prioritizing racial deconcentration as a site selection factor. Unfortunately, HUD regulations could not ensure compliance with the goal of desegregation, and some localities turned down federal money rather than follow the new regulations. This raised a dilemma for those who had advocated the changes: was a decline in the number of units being produced preferable to segregated housing? Was it right to cut off federal funding for other programs in order to force approval of new public housing sites (Hays 1995)?
Urban renewal
Slum clearance, while a major focus of the 1937 Act, became even more of an emphasis in the Housing Act of 1949. This served business interests because, by limiting public building to the replacement of demolished slums, publicly provided housing stock created almost no direct competition in the private real estate market. Building lobbying interests promoted many of the changes in the 1949 Act that precipitated the problem of concentrated poverty in public housing (Atlas and Dreier 1992; Bratt 1986). Urban renewal was initiated with Title I of the Housing Act of 1949 and it made large-scale slum clearance possible without the requirement that all cleared housing be replaced (Teaford 2000). Title I did not include a mandate for the construction of low- or moderate-income housing. In some instances, though, public housing was an integral part of a city’s redevelopment plan, such as PruittIgoe in St. Louis (von Hoffman 2000). Early Title I projects in New York, Philadelphia, and Cleveland included low- and moderate-income housing but, by the late 1950’s, this kind of development lost favor. One of the worst examples of urban renewal was the slum clearance project in the West End of Boston, which was undertaken with little support from the neighborhood residents, as documented by Gans (1962). Perversely, the renewal process could be quite lengthy, leaving large
5
The Gautreaux experience was the main motivation for the implementation of a large scale social experiment in housing mobility, Moving to Opportunity (MTO), now being conducted by HUD.
17
barren areas in the center of a city waiting for development to begin. The benefits to the poor of new commercial and retail development, if any, were mostly indirect (Teaford 2000). The 1949 Housing Act mandated 810,000 units of public housing, but by December of 1951, only 84,600 units were under construction. The 1954 Housing Act called for public housing to be built only in areas of slum clearance and urban renewal. Thus, new public housing did not increase the housing supply, but only served to replace demolished housing. Additionally, displacement was a problem for former slum dwellers, as they waited for the promised new housing to be built. As public housing construction declined, investment in urban renewal increased. Between 1957 and 1960 an average of 26,750 public housing units per year were constructed (Biles 2000).
2.3: Financing
Public housing has always been faced with financial difficulties. Congress funded fewer units than were authorized under the first housing act. The 1937 Act funded only capital costs and expected that most operational and maintenance costs would be covered by rental income, though operating subsidies were not explicitly excluded (Schill 1991). Often, though, excess rent was applied to debt payment and maintenance needs were neglected. Congress attributed rising costs in public housing to management problems, although in the 1950’s and 1960’s, high inflation, increasing expenses and aging buildings contributed to higher maintenance costs. Compounding rising inflation, tenant incomes declined from 47.1% to 36.9% of the U.S. median income between 1961 and 1970 (Hays 1995). A small construction boom in public housing between 1969-1970 intensified the existing financial problems. Public housing was attacked, along with other housing programs, by the Nixon administration. Public housing authorities were left with an impossible choice: raise rents, decrease services and maintenance, or both. Rising rents and reduction of services led to widespread tenant discontent, and a series of rent strikes in the 1960’s eventually culminated in the passage of the Brooke Amendment to the 1969 Housing Act (Hays 1995). In 1971 the Brooke Amendment capped public housing rents at 25% of income (30% since 1981) and provided for operating subsidies to housing authorities to pay for shortfalls and deficits (Hays 1995; Bratt 1986). Also, tenants’ incomes were not to exceed 80% of the area's median income in order to qualify for admission. A very strict interpretation of the legislation was applied, and HUD tried to use the operating subsidies to encourage good management. HUD managed to spend only $33 million out of the $75 million 1970 appropriation in this attempt to exert control (Bauman 2000; Hays 1995). Even though there was a great deal of new construction in the 1970’s, older units were crumbling (e.g., Pruitt-Igoe) (Hays 1995). Eventually, the operating subsidies that were designed to fill the gap between rents and expenses were tied to performance. Low-performing housing authorities continued to struggle and a lagging economy forestalled repairs and modernization efforts of troubled projects (Bauman 2000). Critics of the Brooke Amendments argued that deferred maintenance needs were not considered and that the modernization fund that was eventually enacted was never sufficient to fully complete repairs. The new funding did not cover the losses caused by Brooke, after figuring in inflation. During the 1980s, rents covered only 79% of operating costs, down from 97% in the early part of the decade (Feins et al. 1994). Rents were raised in many places, but even with increased rental income, maintenance problems continued and many buildings decayed rapidly (CLPHA 1993).
18
In January of 1973, the Nixon administration imposed a freeze on most federal housing programs. They began a large-scale reorganization and consolidation of programs, with a major emphasis on Section 8 subsidies as a replacement for public housing and several other programs. However, in 1977 public housing was reintroduced to the budget and was maintained until the Carter administration’s last budget in 1981. Public housing was a tried and true program that, despite all of its problems and failures, did succeed in providing low- and moderate-income people with decent, affordable housing. At the time, it was unclear if Section 8 would be implemented successfully and this doubt led to the reintroduction of public housing. Until 1981, annual reservations for between 35,000 and 50,000 units a year were made. An argument in favor of public housing was that it could provide housing to a population that the private sector was unwilling to serve and that public housing was actually more economical, in the long run, than some subsidy programs. Public housing potentially increases the housing stock and remains available to low-income people permanently. There is no risk of resale and market turnover or loss of that housing from the low-income sector (Hays 1995). Since 1981, there has been no large-scale funding of public housing at the federal level. Local governments have built public housing, usually on the scattered site model, and public housing has been used as a vehicle to selectively replace housing. Funding of public housing for Native Americans has continued, but the bulk of the federal housing dollars are used for tenant-based housing vouchers, formerly Section 8, now called “Housing Choice Vouchers.” The recipient pays 30% of her income towards rent and the voucher covers the difference between that and the rental price of the unit. The demand for affordable housing has not diminished, and public housing and Section 8 vouchers have failed to satisfy it. In the late 1990's, HUD introduced a few pilot programs, such as HOPE VI and Moving to Work, that attempt to remedy the problems related to restrictive rent structures and poor design. It will be several years before we know the outcomes of these programs, though HUD is conducting evaluations.
2.4: Design
Poor design of developments has been blamed for many of the problems that public housing residents face (Newman 1972). Modern design was thought to play a key role in improving the environmental conditions for slum dwellers. Social activists argued that children and families could not thrive in the squalid environment of tenements where people often lived in interior rooms with no windows or ventilation. Early ideas of improving tenements emphasized light and air (Franck and Mostoller 1995). In the late 1800’s, reformers such as Jacob Riis decried unsanitary conditions and called for the destruction of tenements (Riis 1890; Riis 1902). By the early 1940’s, many planners felt that high-rises could provide a healthy, unique living environment that would contrast favorably with surrounding slum areas. However, guides to good design for two and three story buildings were still being promulgated (National Housing Agency Federal Public Housing Authority 1946). While high-rise buildings were desirable for their cost effectiveness in construction, they were not necessarily the cheapest forms of housing development. While the per unit cost of high rises is often more expensive than other types of units, when the land cost is extremely high it is efficient to build more units at a higher cost in order to maximize the use of land. It was often the case that land costs for public housing sites were very high, even though the sites were not very desirable (Bacon 1985). In some cases, the original plans
19
of architects were imperfectly executed due to lack of funds, leading to unpleasant housing and poorly designed common areas. Limits on unit amenities were imposed for cost reasons and because it was believed that a lack of amenities would encourage residents to better themselves. In the name of cost control, inexpensive amenities were often sacrificed, and poor quality units were produced, though Congress and the housing authorities often blamed tenants for the poor condition of units. Even in contemporary cases where the buildings are well-constructed, basics such as floor space, closet doors, and reliable elevators might be lacking (Biles 2000). Poorly designed floor plans lead to security problems and the lack of site planning and recreational facilities make the developments stark and unfriendly. The monolithic appearance of many developments has become part of the larger grounds for which public housing has been attacked (Hays 1995). Developments often were designed to separate the new public housing from the existing community. Site plans often placed buildings on diagonals in relation to the existing street pattern. Some large developments imposed ‘superblocks’ for the development, creating blocks that contained two or three normal city blocks. The break in the street grid was considered necessary to distinguish the new housing from the rest of a poor neighborhood (Franck and Mostoller 1995). The institutional look, the uniformity of the buildings, and the peculiar and confusing layout of developments all have made public housing easy to identify visually and to stigmatize and isolate (Franck and Mostoller 1995). This was an unintended and unanticipated consequence. The modern design of many public housing developments was imagined to have positive symbolic value for places, while it usually had the opposite effect once constructed (Bacon 1985). The demolition of Pruitt Igoe in St. Louis in the early 1960’s is an early example of the failure blamed on design that typifies many of the negative stereotypes of high-rise developments (Rainwater 1967). More recent research suggests that the failure of the project was most likely due to the declining economy of St. Louis, though the innovative and modern design did not prove to be very accommodating to residents. One of the reasons that this particular design was chosen was to represent modernity and urban renewal for the struggling city of St. Louis (Bacon 1985). Financing was not available to completely carry out the design and many compromises were made that sacrificed amenities such as children’s playgrounds and green space between buildings. These and other design flaws made the buildings unpleasant to live in, and vacancies were always high even though there was a shortage of affordable housing in St. Louis (Bacon 1985). However, the real failure of the project was more financial in nature than social, though the social problems were much more visible. It became easy to blame the ‘social pathologies’ of residents for the deterioration of the buildings rather than a lack of operational funds to maintain them. Debt retirement was the first priority of all revenues in public housing, and took precedence over maintenance. As living conditions worsened, vacancy rates rose and Pruitt-Igoe became a financial black hole. Because funds were not reinvested in the development, eventually demolition became the most viable, and least costly, solution. By 1968, HUD had prohibited the building of high-rises for families (Biles 2000). The perception is that most public housing is in the form of high-rises and, in fact, they compose a little over a quarter of the public housing stock. The largest proportion of the housing stock was built before 1970 and consists of: 27% high-rises, 32% garden apartments, 16% low-rise walk-ups,
20
and 25% single-family homes or townhouses. Since the 1980s, very little new housing stock has been added (Atlas and Dreier 1992). What has been added for families are mostly scattered-site developments with 50 or fewer units. Approximately 80% of all family units are in medium- or lowrise buildings (CLPHA 1993). Many high-rises originally built for families were eventually converted for elderly use. Despite the many problems of funding and organization that it has faced over the years, public housing remains one of the largest sources of federally funded low-income housing in the United States. There are reasons to think that public housing residence may increase or decrease labor force participation—either overall or to a degree. These effects may be caused, in part, by the physical characteristics or locations of buildings. Well-designed housing projects may help decrease criminal behavior, alleviate fear of crime, and allow residents to keep normal working hours. Poorly designed buildings in unsafe neighborhoods may cause residents to reduce or eliminate labor force participation in order to keep closer watch on their children or due to isolation. In the next Chapter, I will summarize the large body of literature that provides my theoretical framework, and discuss the variety of reasons why public housing residents and similarly positioned non-residents are disadvantaged in the labor market.
3: Theoretical Framework Hypothesis: The zero order effect of public housing on employment is predicted to be negative. The introduction of controls, detailed below, is expected to reduce the magnitude of the effect and potentially reverse it from negative to positive. Public housing is predicted to have a negative relationship with annual hours worked that may be modified by the introduction of controls. Residents of public housing may have characteristics that are labor market disadvantages, but is there a further limitation on labor market participation due solely to public housing residence? There may be some independent negative effects of public housing on labor market outcomes due to stigma, isolation, and work disincentives, but, on balance, public housing residents may have a labor market advantage due to the security provided by having affordable housing. These issues may also affect working residents and cause them to change the number of hours they work annually. Employed public housing residents may decide that they want to decrease their time working, or leave the labor force, in order to care for children or improve their human capital. Individuals who are detached from mainstream institutions are less likely to enter the paid labor force (Browne 1997). Certain classes of people experience a significant disconnect from the labor market and other mainstream institutions. This detachment occurs in structural, cultural, and spatial dimensions (Browne 1997; Wilson 1987; Wilson 1996). Public housing residents are a particularly disadvantaged and isolated group, even when compared with other poor people. Public housing residents are selected on their negative characteristics, particularly low income (Spence 1993). Self-selection may also play a role, adding to negative unobservable characteristics associated with public housing residence. Because of their residential situation, they may be more isolated from some things (e.g., networks rich in job information) than similarly positioned people in the labor market, or even people in similar neighborhoods (Kasinitz and Rosenberg 1996). Individually, public housing residents may be similar to other poor people along the dimensions of skills and education, but the geography of housing projects could potentially have some differential
21
impact on public housing residents by isolating them from work. It is possible, though, that public housing residents may have advantages connected to their place of residence that are not available to other poor people, such as access to social services or on-site child care (Lassen 1995). The theoretical areas with which I deal are personal characteristics and several aspects of place. Place is a mostly structural category and includes social resources that are tied to location and social class. This framework encompasses previous research about the urban underclass, particularly that of Wilson (1987, 1996), as well as research about the effects of living in metropolitan areas and residential segregation on labor force participation, characterized as “industrial restructuring” by Browne (1997). This, in part, reflects the perceived shift to a more service-oriented economy. Human capital, welfare, and family issues have some structural aspects, but also operate on the person level. The issues surrounding welfare receipt and family organization, along with human capital characteristics, form the person part of the theoretical framework. The place factors will be discussed first, followed by a section addressing the person part of the theoretical framework.
3.1: Place Neighborhood & isolation
Is it possible that the physical location of people has a direct effect on their work experiences (specifically, labor force participation and number of hours worked)? What does place mean for people with low socioeconomic status—lower education, fewer job skills and an impoverished background or environment? Neighborhood poverty and isolation can affect the way people approach the labor market. Neighborhood factors may even affect the social resources and social networks to which residents have access, including the social contexts that shape people's interaction with the labor market (Browne 1997; Granovetter 1983). Isolation, social and spatial, could be caused by living in public housing developments that are set apart physically from other parts of a neighborhood, effectively separating residents from their neighbors and jobs. Living in such developments could isolate public housing residents from working neighbors and friends. If there are few employed people in the vicinity, casual connections to jobs through friends and acquaintances may be very constrained for all neighborhood residents. Public housing tenants may have more of a labor market disadvantage than other people living in similar neighborhoods. In one example, employment opportunities seemed to be limited specifically because of place of residence. In their study of a poor, Brooklyn waterfront neighborhood, Kasinitz and Rosenberg (1996) found that public housing residents did not have the same access to job information as their similarly socioeconomically disadvantaged neighbors. The problem was that their social networks did not tie them to nearby industrial jobs, while neighbors in private housing did have such ties. Unfortunately, without social network data, effects at this level cannot be examined. Differences in job search patterns will be examined, though (more details below). In some cases, the public housing developments contrast radically with the surrounding neighborhood. This can impose a structural disadvantage that may constitute an independent negative effect of living in public housing (Briggs 1998; Kasinitz and Rosenberg 1996; MacLeod 1995). While limited qualitative data seem to support this idea (Kasinitz and Rosenberg 1996), it is not clear that such generalizations may be made about all public housing developments. In the 22
‘worst case’ scenarios (Popkin et al. 2000), despite the wretched geography and design of public housing, it is clear that the characteristics of the people living in the housing almost categorically exclude most of them from the labor market—what is striking is the concentration of such a group of people in one relatively small area. Very dangerous public housing, such as much of that found in Chicago, does seem to limit choice in terms of childcare arrangements, which could constrain labor force participation (this will be discussed in more detail below) (Popkin et al. 2000). It is possible that other people living in the same disadvantaged and dangerous neighborhoods would face similar constraints. Unfortunately, the data do not provide detailed information about safety or the housing developments in which respondents reside. Data would be needed for all sample members in order to examine this. Limited neighborhood characteristics, such as tract level poverty and residential segregation, are examined by linking the PSID with Census data. If public housing residents have a labor market advantage, it would be net of these neighborhood factors.
Concentration
Concentration is important to understanding the meaning of place. While measured as the aggregate of individual characteristics that are thought to add up to more than the sum of their parts, ‘concentration effects’ are actually a spatial concept. Concentration effects are hypothesized to occur in ‘underclass,’ ‘ghetto-poor,’ or inner-city neighborhoods, and they can be thought of as the environmental level manifestation of the ‘culture of poverty’ (Lewis 1968). Concentration effects are found when people live in neighborhoods with high levels of poverty (40% or higher), where there are mostly single-parent homes, and where a low percentage of adults are employed. Many of these measures represent more than one concept. For example, concentration effects touch on aspects of isolation, neighborhood characteristics, and discrimination. These characteristics are usually measured at the Census-tract level because better spatial designations of neighborhood boundaries are not widely available. Wilson (1987,1996) posits that concentration effects worsen the prospects of the inner-city poor. Living in an environment (or neighborhood) populated by a preponderance of people with negative characteristics will negatively affect all the residents of the area by engendering, for example, a lower work ethic and high rates of juvenile delinquency. Research along these lines can be traced to Chicago school sociologists who theorized about ‘social disorganization’ (Park, Burgess, and McKenzie 1925; Shaw and McKay 1942). Public housing residents may be particularly prone to disadvantage because public housing brings together large groups of people selected on the exact kinds of traits that are hypothesized to be the basis of harmful concentration effects. This would lead to the hypothesis that they may be less likely to work, or work fewer hours, than non-residents. They may be particularly prone to social disorganization, as exemplified by crime and delinquency. Families living in public housing may not have the organizational strength to fend off aggressive criminals (Popkin et al. 2000). The aggregate measure of concentration used in the analysis is tract-level poverty.
Discrimination
Related to concentration effects are other neighborhood-level effects, such as the evidence that employers discriminate against residents of poor neighborhoods and public housing developments in Chicago (Kirschenman and Neckerman 1991; Wilson 1996). While studying the same Chicago neighborhoods later examined by Wilson (1996), Kirschenman and Neckerman (1991) found that many employers who are familiar with public housing projects hold negative stereotypes about residents and are less likely to hire them. Thus, the stigma of living in projects that are visually and 23
spatially set off from the community may make it difficult for public housing residents to find work. This stigma can effectively be a barrier to employment. Active discrimination extended to residents of poor neighborhoods, where some employers were familiar with the addresses associated with poor neighborhoods. Employers may be less likely to hire people whom they know live in public housing or in a poor area. Their familiarity with certain neighborhoods led them to use the reputation of the area or housing project as a way of screening employees. Chicago has some of the most easily recognized and largest public housing developments in the country. It is possible that address discrimination is a phenomenon that occurs only in similar places, or, in fact, only in Chicago. In Chicago, one’s address promoted discrimination in hiring, and place of residence served to increase labor market disadvantage. It is possible that people living in neighborhoods similar to those of public housing residents would face similar levels of stigmatization, particularly in places where neighborhood characteristics are well known. Living in a poor neighborhood may also be a disadvantage. Employers tend to attribute laziness, tardiness, and poor attitudes to potential inner-city workers. Residents of poor neighborhoods are assumed to not have the qualities for which employers are looking, which can lead employers to exclude whole classes of people from their employee recruitment strategies—for example, by not advertising jobs in the city papers (Wilson 1996). Many employers attributed a better work ethic to Black females than Black males. Women were seen as more likely to have children and be heads of households, thus their commitment to work was taken more seriously. These issues are germane for both public housing residents and non-residents who live in high poverty areas. Such areas can be examined by linking the PSID with Census data. It will be possible to determine whether public housing residence has an effect on work outcomes after controlling for poverty level of the census tract.
Fear
Another place effect is fear for one's family due to neighborhood conditions. Many housing projects have dangerous reputations, and parents may change their behavior accordingly to provide as safe an environment as possible for their children (Furstenberg 1993; Popkin et al. 2000). Even if public housing is not actually subject to higher crime rates than the surrounding areas, residents may share the perception along with outsiders that their development is particularly dangerous (Vale 1995). In Chicago, it does appear that public housing is significantly more dangerous than the surrounding community (Dubrow and Garbarino 1989; Popkin et al. 2000). Parents may be reluctant to leave children unattended in what they perceive as a dangerous environment while away at work, which may lead to a decrease in hours worked. This hypothesized decrease in hours spent working could take effect soon after moving into public housing and level out over the period of residence, or it could take a more sporadic pattern as perceptions of safety vary over time. It seems unlikely (though possible) that one would move into public housing anticipating a decrease in hours worked due to security issues. However, over time, these concerns could exert a strong influence on decisions pertaining to labor market involvement. These choices could take the form of opting for part-time work over full-time work, reducing hours on one’s current job, or even remaining out of the labor market for an indefinite period of time. One might also choose a lower-paying job to gain the benefit of flexibility and convenience.
24
Rosenbaum and Popkin (1991) report the results of research in Chicago where inner-city public housing residents were given housing vouchers (Section 8) that allowed them to move to private housing, either in the suburbs or within the city. The greater likelihood of employment for public housing residents who moved to private housing (using vouchers) in the suburbs of Chicago may have indicated a feeling of safety on the part of working parents (Rosenbaum and Popkin 1991). People who moved within the city expressed no greater feelings of safety. Suburban movers felt safer in the suburbs than in the city and were more comfortable leaving their children alone after school, which increased their ability to work regular hours. It is not known if residents of public housing felt more fear than the inner-city movers. Their neighbors who also live in poor areas with reputations for high crime rates may share the feelings of fear expressed by public housing residents (Vale 1995). While some neighborhood characteristics can be measured, safety concerns are not among them. Ideally, attitudinal data about perceptions of crime and fear levels would be included, but they are not available. Safety concerns are an unobservable factor and would likely have a negative effect on the likelihood of working for public housing residents.
Spatial mismatch
While I do not propose to test the spatial mismatch hypothesis directly, I have included a discussion of it here because it has generated many ideas relevant to this research. In my analysis, I account for many of the factors that would lead to a spatial mismatch, such as race, residential segregation, and urban location. These factors are the most relevant for the question I am asking, which specifically deals with the circumstances of public housing residents. The most germane factor for public housing residents may merely be inner-city versus suburban location; spatial mismatch may reflect the problems caused by residential segregation. The spatial mismatch hypothesis assesses the impact of residential distance from jobs on employment and non-employment as an outcome (Holtzer 1991; Kain 1968; Kain 1992). The basic premise is that non-Whites (Blacks in particular) are spatially distant from jobs at the appropriate skill level and that this puts them at a disadvantage in the labor market. Increasing suburbanization and deindustrialization has led both jobs and people to move away from the center city. Because of historic labor market patterns and residential segregation, Blacks remain disproportionately concentrated in the central city. Racial discrimination and overall disadvantage both make it harder for Blacks to suburbanize than other groups, and they tend to remain spatially homogenized and distant from jobs, even in suburban areas. Even the suburbs to which Blacks tend to move are closer to the city center than the White suburbs. Public housing historically concentrated its residents in single-race dominant groupings, replicating neighborhood segregation patterns and sometimes intensifying them (Popkin et al. 2000). It is open to debate whether or not this practice, to the extent that it continues, further exacerbates the hypothesized spatial mismatch. It may also be that for people living in inner-city areas, particularly Black males (Wilson 1996; Johnson, Farrell, and Stoloff 2000), there are not as many job opportunities. Wilson (1996) asserts that jobs have left the central city, especially those jobs for which most poor residents are qualified, which is similar to Rosenbaum and Popkin’s (1991) findings. This is a relatively recent change, since in the 1950’s the residents of the same Chicago neighborhoods that Wilson studies were poor but mostly employed. Because public housing can concentrate disadvantaged people in disadvantaged neighborhoods, residents may be particularly impacted by a spatial mismatch of jobs.
25
The main motivation of the Gautreaux case, on which Rosenbaum and Popkin (1991) reported, was to give public housing residents the opportunity to move to non-segregated areas. Rosenbaum and Popkin’s (1991) research supports the hypothesis that a low-skilled person living in the city is less likely to be employed than a similar person living in the suburbs. Residents who moved to the suburbs experienced no improvements in wages but were more likely to be employed than residents who moved within the city or those who remained in public housing. The people involved had similar skill levels prior to their moves, but those who moved to the suburbs may have faced lower levels of address discrimination. Additionally, there were more low-skilled job opportunities in the suburbs. The experiment supports the idea that location is important, thus, urban location is included in the model. If the spatial mismatch hypothesis is operating, it is likely to differentially affect public housing residents. Non-residents have a broader range of residential options, though their choices are inevitably still quite constrained. However, newer scattered site public housing developments should at least approximate the distribution of housing available to other low-income people; thus, public housing does not necessarily mean a greater labor market disadvantage. It also may be that the spatial mismatch model is no longer adequate to explain the disadvantage faced by inner-city minorities or project dwellers given the state of the economy in the 1990’s.
Skills mismatch
The skills mismatch hypothesis posits that there are plenty of jobs in the places near poor inner-city residents, but the people living in the inner-city do not have the requisite skills for these jobs. The jobs that are at the appropriate skill level for inner-city residents have moved to the suburbs (Kasarda 1983; Kasarda 1989). Therefore, better public transportation, for example, would help solve the problem. According to the skills mismatch hypothesis there is no deliberate removal of nearby jobs from the vicinity of low-skilled workers. It is considered an historical artifact that innercity residents are distant from jobs of the type appropriate for their skills. There have been many challenges to this idea, and there is some evidence that employers have moved manufacturing plants to the South to avoid minority and union workers (Fernandez 1994). Also, the ‘historical accident’ that concentrates low-skilled workers in the city is, in fact, a form of residential segregation. The trend of moving jobs to suburbs is part of the same phenomenon—a kind of racism or discrimination that is subtler than outright slurs. How do these issues specifically relate to public housing residents? Certainly, many public housing projects are concentrated in the inner-city, which would lead to the likelihood of public housing residents being distant from jobs at the appropriate skill level, making them less likely to be employed. Some public housing residents are more concentrated than their similar neighbors, though this varies by specific locale. Controlling for inner-city location should equalize the effect of housing location on job holding. Living in public housing should have no independent negative effect on people's skills, though there may differences due to selection on unobservable factors. The problem is that public housing, like a poor neighborhood, concentrates people who, on the whole, have fewer job skills than the general population, as Rosenbaum and Popkin (1991) demonstrate. Therefore, when controlling for education and other skill related variables, the negative effect of public housing should decrease and its positive effect should increase. Public
26
housing residents are prime candidates for the interventions proposed by those who agree with the skills mismatch hypothesis.
Residential segregation and racial discrimination
I have grouped these two ideas together because they are intimately related. Residential segregation, while often treated as an independent phenomenon, is caused at least partially by racial discrimination. While it can be argued that residential segregation has pernicious effects independent of racial discrimination and prejudice (Massey and Denton 1993), it is clear that the processes are linked. While it is difficult to measure prejudice, it is clearly a prime mover in the disadvantages that minorities face. The argument that race, not space, leaves poor inner-city Blacks at a disadvantage (Ellwood 1986) is often seen as opposed to the spatial mismatch hypothesis. The idea is that racial prejudice is the cause of labor market disadvantages, not distance from jobs. If this argument were followed, then it implies that it is the racial concentration of Blacks in public housing that would lead to the lower likelihood of employment for public housing residents. Thus, controlling for race in the analysis is necessary to see if it modifies the effect of public housing (as an independent variable) on work status. It may be that such an overwhelming majority of public housing residents are minorities that it is not possible to disentangle the race effect from the public housing effect. However, if there are no differences between residents and non-residents by race then public housing will not be the cause of a ‘race’ effect. It should be noted that due to the structure of the PSID, nearly 100% of the minorities in the study are Black. Is racial discrimination compounded in some way for minority public housing residents? Could the historic racial segregation patterns particular to public housing (race-specific projects) compound the effects of racial discrimination? It is not clear that the distance to jobs is worse for public housing residents than other residents of poor areas. An interaction between public housing residence and race will be included in the multivariate analysis, as well as an interaction between race and poverty. Residential segregation may be the root cause of the hypothesized spatial mismatch, rather than lack of jobs being an independent effect of space. Spatial distance compounds pervasive racial discrimination, and the two cases are part and parcel of the same phenomenon. Public housing is clearly implicated in residential segregation because of the history of building race-specific housing projects (Bickford and Massey 1991). While such practices are now illegal, the overt racial discrimination found in many federal lending programs has led, historically, to widespread residential segregation in the U.S. (Massey and Denton 1993). If there is a spatial mismatch related to public housing, it is almost certainly a result of residential segregation that mostly affects Blacks and, to a lesser extent, Hispanics. It is important to remember, though, that residential segregation in public housing (which certainly exists, see (Goering 1992)) is part of a larger trend of residential segregation in the U.S. Public housing residents, however, live in particularly segregated circumstances. Measures of tract-level racial residential segregation and urban area are included in the analysis.
Social resources
Structurally, one's personal social network can provide information about jobs as well as the social support needed to take and hold a job. However, in order for a social network to serve information needs, one must occupy a position in society where the members of one’s social network are likely to have access to the labor market. The disadvantaged segments of society do not have access to 27
useful social networks or the institutions that would give them the skills needed to enter the labor market in a permanent way. It is possible, though, that public housing residents are even more isolated from job contacts than their neighbors. There may be information flows from which public housing residents are excluded that could connect them to jobs. Exclusion from these social networks would put public housing residents at an especial disadvantage in the labor market (Kasinitz and Rosenberg 1996). Unfortunately, without data on social networks these issues cannot be tested. However, if there are differences in job search patterns, they can be examined. This would reflect, in a limited way, some of the outcomes of having different types of social networks. The importance of job search patterns is discussed in more detail below. Social resources are ‘goods’ (such as emotional support or help in times of trouble) that may be accessed through social networks. Social resources, in the form of support but not job information, may actually be increased for public housing residents if they make contact with their neighbors more easily than in private housing. Public housing developments often have some kind of selfmanagement or tenant council that may facilitate friendships with neighbors and the development of social networks among residents. However, some qualitative evidence suggests that it is more difficult for public housing residents to make connections with their neighbors than for people who live in private housing in disadvantaged neighborhoods (Furstenberg 1993). In fact, the extremely dangerous conditions of some public housing may make residents wary of any kind of contact with neighbors (Dubrow and Garbarino 1989; Popkin et al. 2000). The very structure of poor neighborhoods, especially for resident of high-rise buildings, might discourage the friendliness and trust that would lead to contacts that might provide social support (Furstenberg 1993). In the cases of large, dangerous projects, it is the neighbors, or the family members of neighbors, who may well be the perpetrators of crime and violence. Thus, distrust would be an understandable response (Popkin et al. 2000). However, public housing could present resources for residents, if, for example, neighbors could provide low-cost child care or other kinds of social support (Hasell and Scanzoni 1997). To partially assess this, the child care costs of the residents and non-residents in the matched sample are compared. If public housing residents have lower child care costs, it would suggest that living in public housing improves access to low-cost child care. Access to these resources might make it easier for public housing residents to attain employment or increase hours worked if already holding a job. In addition, if social services were more readily available to public housing residents, because of onsite locations, it may provide a labor market advantage. In some areas, public housing authorities have developed on-site community centers where residents can meet. Often, after-school programs and daycare are also operated in these facilities (Lassen 1995). It is not possible to test the effects of on-site child care or social services due to data limitations, except indirectly through the analysis of child care costs. However, it is possible that these factors represent a benefit of living in public housing, which might positively affect labor market participation.
Job search
The ways in which people search for and find jobs are of interest because the methods used may produce different outcomes. Commonly, job changes are accomplished in a few different ways. Either a person actively looks for a job by using newspaper advertisements, employment agencies, 28
talking to people (friends, relatives, or acquaintances), directly contacting an employer, or the person changes jobs without searching (Campbell and Rosenfeld 1985), that is, someone approaches the person with the job offer, rather than the person actively looking for new employment. This last type of job change is probably not an issue in this analysis since it is most relevant to higher prestige occupations, which are not available to most low-income people or public housing residents. However, the common job search methods of advertisements, direct application, employment agencies, and personal contacts are available to everyone. It might be the case, though, that since public housing residents are less likely to be employed than the general population, they use personal contacts less frequently in their job search process since it is likely that fewer of their acquaintances are employed. Of particular interest is whether public housing residence affects the ways in which people search for jobs. If public housing residence constrains job search options, it would be an important finding. However, it would also be of interest if there were no differences between residents and non-residents. The job search patterns of residents and non-residents will be compared, although they cannot be included as variables in the event history models due to sample size constraints.
Rent structure & welfare reform
Public housing residents may be at a disadvantage to non-residents through the work disincentives that public housing rent structures may impose, a constraint directly tied to place of residence. While one might expect to see a labor market advantage for public housing residents because rents are controlled, it is also true that since rents are set as a percentage of income they impose a tax as income increases. In this respect non-residents may have an advantage, although rent increases in the private rental market are a standard practice and may be less predictable than the rent structure of public housing. Renters in the private market may also be getting less for their money, for example, smaller units, poorer living conditions, or less responsive management. Pragmatically, it is clear that some residents may be disinclined to work since it may leave them financially worse off. The results of this research should give an indication of whether or not public housing is a work disincentive, though it is not an explicit test. The dependent variables that measure changes in annual hours worked will help to test the question of whether working residents are more or less likely than non-residents to change the number of hours they work. If public housing does serve as a work disincentive, then we would expect it to be positively associated with a decrease in annual hours worked. Currently, a variety of reforms and demonstrations are underway, and public housing authorities are experimenting with different kinds of rent structures that may provide incentives to work. In the long run, this may even increase revenue from rents. Some of the ideas are floor and ceiling rents, income disregards, and savings plans for extra income, such as the Family Self-Sufficiency Program. Since the data in this paper were collected prior to these reforms, the results of this research will serve as an indication of the disincentive value, if any, of the prior rent structure by providing a baseline for the likelihood of labor force participation for public housing residents. Some research has shown that labor force participation is higher for welfare recipients with a housing certificate or voucher than those without (Ong 1998). Not only did Ong (1998) find that the effect of public housing residence on work outcomes was not statistically significant, he also concluded that housing certificates or vouchers that were tenant-based actually helped welfare recipients form a stronger attachment to the labor market, despite the hypothesized work disincentive. This research will
29
improve upon Ong’s (1998) study because I use longitudinal data to examine public housing residents and group matching to determine the appropriate comparison group, rather than just focusing on welfare recipients. Housing subsidy receipt is included as an independent variable in the analysis. Since there are very few respondents receiving a housing subsidy in the sample, though, no detailed analysis of their characteristics is undertaken. The group of respondents receiving housing subsidies is mutually exclusive of the public housing group. 6
3.2: Person
Wilson’s (1996) research suggests that many inner-city residents are lacking in the traits they need in order to obtain and retain employment. Much of the work available to low-skilled workers is now service-oriented and requires contact with the public. Many employers feel that inner-city Black job candidates do not have the requisite speaking skills and literacy skills to interact with the public (Kirschenman and Neckerman 1991; Wilson 1996). Because the low-skilled labor market has shifted from manufacturing to services, the supposed lack of these communication skills is a major impediment to labor force participation. People with low socioeconomic status are not only disadvantaged because they are thought to lack certain skills, but also because they lack cultural traits and access to social resources that would make it easier to find and retain employment (MacLeod 1995).
Skills acquisition/human capital
Greater education and past work experience provide an advantage to those seeking paid employment (Blau and Ferber 1992). There may be benefits attached to greater age as well (Harris 1993), though it is mostly a proxy for work experience. These characteristics can be thought of as human capital, which is any investment that increases the productivity of the worker. The worker may increase her human capital by obtaining education, job training or other skills (Becker 1975). The employer may also invest in the worker’s human capital by providing on-the-job training, though employers are not likely to do so unless the skills directly benefit them and are nontransferable. The lack of human capital may be one of the main reasons residents of public housing end up living in subsidized housing. In general, lower levels of education and less desirable job skills are the things that keep certain segments of the population out of the labor force or in lower paying jobs. In addition, for many poor families, welfare payments and housing benefits are much more generous if one is not officially working for pay. However, it is expected that all people make similar cost-benefit analyses and will go to work when it is the financially expedient choice. Thus, those with more human capital will be more likely to work because they will get higher paying jobs, regardless of place of residence. If the person’s main goal is to improve her labor market position, then time and energy will be invested in accruing human capital that will be desirable to employers. However, since economic outcomes are not the only concern of individuals, other factors will affect their decisions about how to invest their time and energy. It may be that, for example, caring for
6
The results from this analysis should not be directly compared to Ong’s work since the PSID definition of “housing subsidy” may include subsidies other than Section 8 certificates and vouchers and is almost certainly an undercount of the total population receiving a housing subsidy.
30
children is a high priority, so human capital investments are delayed until children reach at least school age. The costs and benefits of these investments must be weighed before action is taken. Hypothetically, living in public housing should allow residents to acquire skills that will improve their position in the labor market. Skill acquisition would result from having more time to return to school or engage in other training programs. There may be training opportunities and social services located in or near the projects that would not only facilitate working, but would provide a way to increase skills and, eventually, wages. For example, the Department of Labor is funding several Welfare-to-Work grants that specifically target job training programs to public housing developments (Nightingale and Brennan 1998). HUD, in conjunction with Manpower Demonstration Research Corporation (MDRC), is running a similar demonstration project, JOBS Plus, in seven sites around the country that targets saturation-level job placement, training, and service interventions to all residents of selected housing developments. The time out of labor market needed to make such investments might lead to a decrease in annual hours worked, though, in the long-run, increases to full-time employment would be expected. Currently many local housing authorities offer Family Self-Sufficiency (FSS) programs (Rohe 1995) that allow tenants to save part of the extra rent money they owe as incomes increase in an escrow account, though the number of families involved is small. The program is designed to help tenants save money and achieve self-sufficiency, which may or may not mean moving out of subsidized housing. Typically, with the help of a caseworker, participants are encouraged to spend their savings on education or a home purchase. Thus, for a few public housing families, their living situation will directly aid them in accruing financial capital, and possibly human capital. The presence of social services or job training programs directed at public housing residents may provide a way for them to acquire additional skills that would have a positive impact on employment prospects. For example, if residents were able to increase their education while living in public housing it might improve their employment prospects. These kinds of programs are not available to similarly positioned non-residents. Education (a time-varying measure), work experience, and age are included as controls in the quantitative models.
Welfare reliance
Families with a history of welfare receipt are thought to have a weak labor market connection, which can either be viewed structurally or culturally. A cultural perspective is likely to claim a lack of work ethic as the reason for poor and welfare reliant people's weak attachment to the labor force (Mead 1992; Murray 1984), though there is little empirical evidence that supports these hypotheses (e.g., Harris 1993). The structural perspective views the disconnect of the poor from the labor market as a result of larger societal forces (Wilson 1987; Wilson 1996). In addition, people with small children may prioritize child well-being over work. Due to recent welfare reform (which post-dates the period covered in this analysis), TANF recipients are forced to enter the legitimate labor market. Most people are finding jobs, but these jobs are not lifting them out of poverty. The kind of work that they are qualified to do is only a marginal improvement on a welfare payment. Edin and Lein's (1997) research demonstrates how closely and carefully poor families budget. These families can account for how they spend almost all of their money. They often do not report income because it will cause their welfare subsidy to decrease.
31
Since there is so little money to start with, even minor adjustments can make a big difference in the household budget. The idea of a cycle of welfare dependency can trace its roots back to the historical distinctions between the ‘deserving’ and ‘undeserving’ poor (Katz 1989). These ideas shape the public perceptions of who deserves public assistance and who does not. For example, there is rarely any opposition to providing public housing for the elderly—this group is perceived as deserving. The distinctions between deserving and undeserving led to the development of the idea of a cycle of dependency (Murray 1984). This perspective tends to attribute the problems of the poor to their moral qualities, though it cannot disentangle the starting point—i.e., whether people who lack certain qualities end up in poverty, or whether poverty creates a culture where people develop undesirable qualities (Lewis 1968). Almost all the research done on this topic, though, finds no distinction in terms of moral qualities, at least as demonstrated by willingness to work. Research on welfare reliant women finds that from 50 to 100% of them are, in fact, working (Edin and Lein 1997; Harris 1993). What keeps these women from higher paying work is that they have few desirable job skills and a lower than average education. However, they all know that AFDC does not provide sufficient income to live on and they must provide additional income to support their families, even though they usually do not have the skills necessary to find employment that pays a living wage. They find themselves in a balancing act between working for pay and keeping their welfare payment as a supplement—along with the benefits it provides, such as health care for their children and day care subsidies (Tilly and Albeda 1994). It may be easier to move from welfare to work if one is a public housing resident, because as benefits decrease and income increases, rent will still be calculated as 30% of total income, including welfare. If one's total income does not increase as welfare income is replaced with labor income, one’s rent would not increase either. Their budgets will not be as negatively impacted as AFDC recipients living in private housing. Thus, in practical terms, welfare recipients in public housing may be more likely to work or increase the amount of time they work. This is a contrast to the private rental market where landlords are allowed to increase the rent annually, with no regard to the tenant’s circumstances. This analysis controls for the welfare experience of respondents and includes an interaction between public housing residence and AFDC receipt. This is complicated by the demise of AFDC and the introduction of TANF (Temporary Aid for Needy Families) in 1996, which has different policies regarding earned income. Under the new program, families have a maximum lifetime limit of five years of benefits and work requirements are imposed. The limiting of state welfare benefits may make public housing an even more important resource for poor families, though the restructuring of the program may have different implications for work disincentives associated with housing subsidy receipt. It should be noted that these changes took effect after the data were gathered for the present study.
Family obligations
The household and child-care responsibilities of poor single mothers may be viewed as legitimate reasons not to work. Not only are the jobs these women would be likely to get low paying and without benefits, but they may want to stay home to care for their children. Poor women may simply not have enough financial security to do this, yet the work opportunities available to them may not provide enough income for good quality, paid child care. For poor women, the cost of child
32
care may put them in the net loss conundrum that keeps them reliant on welfare, as discussed above. Public housing residents are not strangers to this dilemma. Many express their frustration that even when benefits are available they do not facilitate working. One example is a current program that offers a subsidized day care voucher after the person is hired. The difficulty is that while the tenant is searching for a job, there is no allowance for day care (Stoloff 1999). Also, many employers would like the new employee to start immediately, but the day care voucher may not be available for up to a week. It should not be surprising that poor women are less likely to work; their jobs are not likely to pay for child care or include medical coverage. Especially when a young child is present, parents may decide to stay home, or work part-time, rather than pursue full-time work. The presence of a young child is included in the models as a time-varying variable. If there are other adults in a home with children, the head of the household may be more likely to work. Certain family configurations may make the other adults more willing to take responsibility for the care of related children, for example, if the other adult were a spouse, parent, or grandparent. If the other adult were non-related, he or she might still be expected to help, either financially or with child care, but the effect might not be as strong (Edin and Lein 1997). Marital status, the presence of a related adult (non-spouse), or the presence of a non-related adult are included as time-varying variables in the models. Another problem that might be unique to public housing is the lack of nearby support networks of relatives who could provide free or low-cost child care. For example, most public housing authorities have waiting lists for housing and if a person's name comes up, she must take the first available unit, often regardless of location. This may mean moving some distance from family and friends and could be quite problematic, especially if most of these families must rely on public transportation or unreliable automobiles to get around. This makes moving a real barrier to maintaining family relationships. In addition, public housing has rather strict occupancy limits, and the inability to have extra people stay in the home to care for children may make it harder to keep steady employment, work a forty-hour week, or have a 9-to-5 shift. On the other hand, as previously mentioned, the organizations that exist in public housing, such as tenant councils, may fill that gap. Public housing developments may also have on-site child care that is free or very low-cost, which would not be available in private housing. Another factor that could make it harder for the residents of poor neighborhoods, as well as public housing residents, to get or keep a job is the possible lack of neighborhood work opportunities that would allow for closer checking on children left at home. Because of the lack of social network data or information about specific child care arrangements, it will not be possible to directly examine this issue, although I will compare child care costs, as discussed earlier.
3.3: Person and Place
Many of the variables used in the place model represent overlapping concepts. The Censuslevel measures, urban area, poverty level, and residential segregation represent the ideas of spatial mismatch, concentration and, possibly, racial discrimination. By including ‘Black’ as a place variable, I hope to tap some of the larger issues facing Blacks in public housing. Even though race is an individual level construct, the implications of being Black in this society are much farther reaching, Blacks are more likely to live in poor neighborhoods, be discriminated against, and be relatively distant from jobs. The place model includes the following variables that operationalize its concepts:
33
housing subsidy; level of poverty in tract; urban area; Black; and two measures of residential segregation. Additionally, the following interactions are included: ‘public housing and Black’ and ‘high poverty tract and Black,’ in order to control for the particular disadvantage associated with race. In the person domain, most of the measures relate to family composition and human capital. Gender and welfare-receipt address issues of disadvantage for women in particular. I include several interactions based on these concepts in the analysis because I want to see if living in public housing interacts in any particular way with family composition or welfare receipt. It is also important to control for disability status, since disability used to be a preference for admittance to public housing. The person model includes the following variables: female; AFDC receipt; young child; presence of other relatives; presence of a non-related adult; married or not; age in years; years of education and work experience; and disabled or not. Additionally, the following interactions are included: ‘public housing and AFDC’; ‘public housing and relative’; ‘public housing and other adult;’ and ‘public housing and young child.’ The interactions between public housing and types of people living in the household are included to explore whether the presence of an additional person, related or unrelated, has an effect on work outcomes specifically for public housing residents.
3.4: Research Questions
The main question for this research is, over a given period, will public housing residence increase the likelihood of employment? If a person is living in public housing at time T, what is the likelihood that s/he will be employed at time T+1? Will living in public housing at time T have a negative effect on the average number of hours worked annually? Using the PSID, longitudinal analysis will be conducted that takes into account changes in work status and housing tenure over time. This will address the problem of the feedback loop that exists between public housing and work. The problem can be summarized simply: those living in public housing are less likely to be working, and lack of work may precipitate a move to public housing. However, after moving into public housing, residents may find that the freedom from worry provided by the low cost of rent makes it possible to more actively pursue paid employment. There are several ways in which the research questions can be conceptualized, including: 1) changes in likelihood of employment over time; 2) changes in hours worked over time; and 3) changes in wages over time. For this analysis, I focus on changes in employment and changes in hours worked. I hypothesize that the effect of public housing on work is non-negative, and possibly positive, once contextual and composition effects are controlled. I also hypothesize that, for those who are working, public housing residence might lead to a decline in annual hours worked because of the opportunity to improve human capital or the need to stay home with children. Some possible negative effects of public housing that account for an initial negative effect are stigma, lack of helpful social networks, lack of relative-provided child care, and fear. The evidence from the literature does not make it clear whether these effects are unique to public housing, or, rather, aspects of the neighborhoods in which public housing is located. Possible positive effects of public housing include the potential for improved education and skills, housing security, better access to social services, and improved living conditions.
34
I am not able to measure all the processes by which public housing may have a positive versus negative effect on work outcomes, but I will look at some of these indirectly through a comparison of child care costs and job search patterns for residents and non-residents, as well as by including interactions between ‘welfare and public housing,’ ‘young children and public housing,’ and ‘race and public housing’ in the multivariate models. ‘Years of education’ is included in the multivariate models as a time-varying variable. The multivariate analysis cannot distinguish among all the different types of contextual effects, but I use neighborhood level measures to control, generally, for these effects. For example, my model does not explicitly test spatial mismatch, but several of the elements that potentially contribute to a mismatch (e.g., race, residential segregation) are measured. In the next chapter, I will describe the data and methodology that is used for the analysis. In Chapter 5, there is a short section about a preliminary analysis using the life table method and a presentation of the results of the event history models. In the final section of Chapter 5, the results of the statistical analysis of job search methods and child care costs may be found. Chapter 6 contains the overall conclusions formed from this research.
4: Data and Methodology The ideal data set for this research would be relatively current and longitudinal, would have measures of labor market outcomes, labor market context, poverty, family context, social networks and neighborhood characteristics, and would have an adequate number of people living in public housing. Ideally, it would also be possible to identify neighborhood characteristics, length of residence, and other types of housing subsidies. There are several longitudinal data sets that might have been useful. I considered using the Survey of Income Program Participation (SIPP) and the National Longitudinal Survey of Youth (NLSY) before deciding that the PSID was the best choice. None of these data sets have social network variables. The SIPP has most of the necessary variables and is longitudinal. It tracks respondents in four-month intervals for approximately 32 months. Every year a new panel is introduced. This presents a level of complexity that makes the SIPP particularly hard to work with. There is no consensus on what represents a longitudinal unit, which makes linking the cross-sectional files difficult. Respondents are dropped when they move outside the scope of the survey. There are also some questions about the accuracy of the responses (Blank and Ruggles 1996). The NLSY contains questions about public housing residence for five years of its administration (1979-1984). The sample includes only people ages 16-24 in 1979, which makes the population too young for an analysis of work outcomes. If the public housing questions were asked for more years, then an older sample would be available for analysis. The data that are available are relatively outdated, especially considering that more current measures are available. Also, its measures of labor market experiences are not as extensive as those in the PSID.
4.1: PSID (Panel Study of Income Dynamics)
The PSID, based at the University of Michigan, is a nationally representative longitudinal study, designed to collect data on economic and demographic characteristics of its respondents. The PSID also includes many sociological, psychological, and health status variables (for more details see the 35
PSID web page: www.isr.umich.edu/src/psid/). Attrition rates for the PSID are relatively low, and, by including new families as they are formed, the sample size has increased from the initial number of 4,800 families to over 7,000 families in more recent waves. Importantly for this study, the PSID contains questions about public housing residence and receipt of housing subsidies for the years 1968-1972 and 1986-1999 (final release, or ‘public release II’, data are currently available through 1993). 7 By using the PSID, a national level analysis of public housing can be conducted. The PSID contains all basic background information necessary for this research, including family composition, education, paid work status, and welfare status. There are questions about job search and extensive information about labor market experiences. With special permission, it is possible to link the PSID to Census and county level data files, thus providing access to some neighborhood level variables. The main drawback to using the PSID is that it has a complicated and difficult to use file structure. The data consist of two types of files, individual level and family level. The individual file has a separate record for each person in every family ever interviewed, over 51,000 individuals through 1993. In each year, some sets of questions are asked about the previous year. Of importance to this study is that annual hours worked is asked about the previous year, which means that in 1986, information was collected about 1985. The family files contain single years of data with a record for each family and individual level variables provided for the head of the household (and in a more limited fashion for the wife or longtime cohabiter). As mentioned earlier, the PSID has increased its sample size since its inception by including ‘splitoff’ families. Splitoffs are created when children move out and form new households, couples separate (due to divorce or otherwise), or when families change for any other reasons. One advantage of the PSID is that by using the individual level file data about other household members may be included if desired. Each individual record may be linked to the family level data. However, much more limited data are available from the individual level files than the family level files (see Appendix A for a list of the categories of variables available from the family and individual files).
Merging the PSID
In order to use more than one year of the PSID family level files, they must be merged with the individual level data. My sample includes all individuals who were the heads of household (head) in 1986. I have only kept people in the sample who remain as the head during the entire period because some of the variables I wish to use are only available for heads. Thus, I would not have complete data on each individual for the entire period unless the respondent were a head for the entire period. The PSID convention is that if there is a male present in the household, he will be named as head. Thus, usually, only households with no adult males present have female heads.
7
When I began this project, the PSID data were divided into two subsets: ‘early release’ (1968-1993) and ‘final release’ (1994 and later). Subsequent to the completion of the bulk of the analysis for this project, the data were renamed, respectively, to ‘public release I’ and ‘public release II.’ The purpose was to indicate the higher quality of the early release data due to new computer assisted data collection techniques that greatly reduce recording errors in data collection and codification. However, since the quality of the ‘early release’ files was inferior to the ‘final release’ files when I began this project, I decided to use only the ‘final release’ files for analysis. At this time, public release II files are available through 1993, though most subsequent years will be available in 2002. For the PSID’s explanation of this situation, please see: http://stat0.isr.umich.edu/psid/data-center/data-center.html.
36
Potentially, this may introduce some bias to the sample. Fewer female-headed households will be included since, if a female head marries, she may be excluded from the sample. Thus, the femaleheaded households included in this sample will mostly be unmarried women. It does appear, though, that some women are marrying and remaining in the sample as heads. In 1986, only 1% of the women in the initial sample were married, compared to 59% of the men. By 1992, 16% of the women were married and 63% of the men were married. 8 48% of the respondents in the initial sample are female and 52% are male. Women head 56% of public housing families, compared to 46% of non-public housing families. The advantage of this, potentially, is that my study focuses equally on men and women in poverty, while many poverty studies focus solely on women. The first step in the data merging process was to obtain the cross-year individual files with data from 1968-1993 (this is one file with data from every year of the survey about each individual in the household). The cross-year individual file contains a limited amount of data about each respondent and is designed for use in conjunction with the family level files. After gathering the individual data, the single year family files were obtained for the survey waves including a measure of public housing residence (1986-1993). Next, a subset of the individual file was made keeping the family id for 1986-1993, status of the person (if she is the head, wife, etc.), and other relevant variables available at the individual level. It is necessary to use both the family data and the individual data when analyzing more than one year of the survey in order to distinguish if the head has changed across years, otherwise it would not be possible to merge the family-level files. After the family and individual files were merged, the data were subsetted accordingly to include only heads and non-elderly respondents (under 56 in 1986). (See Appendix B for a complete list of the variable used for this research.) In order to link the PSID with 1990 U.S. Census data, a special request was submitted to the University of Michigan Survey Research Center. Approval was granted for the use of these ‘sensitive data’ and in December 1999, the staff at the PSID delivered a CD-ROM with the 1968-1997 ids of each respondent, along with the Census geocodes. I attached these codes to the merged PSID file for the years 1986-1993 using the 1986 ids. I used this file to merge the PSID with Census data (CensusCD+Maps, CD-ROM), which I obtained independently. Due to an incomplete address match on the part of the PSID, when the census data is merged with the PSID, the sample is reduced by approximately 20%. I chose to continue the analysis with the smaller sample, but I am including three sets of diagnostic models to demonstrate that the loss of cases did not gravely affect the multivariate models (see Appendix C). 17% of the final sample are public housing residents and 83% are non-residents. I will provide more detail on characteristics of the sample as I discuss methods and models in the next section (see Appendix B).
4.2: U.S. Census & County Data
The 1990 Decennial Census will be used in combination with the PSID to fully explore the hypotheses in this analysis. Since the Census is only conducted once every ten years, the most recent data available are from 1990. In particular, the Census will provide the contextual data necessary to examine residential segregation and neighborhood poverty. Census data can be 8
There is no explanation for the increasing number of married women in the sample. I assume it is simply an anomaly of the data collection process.
37
matched at the tract-level to each of the respondents and each individual observation may be assigned a tract-level score. For example, segregation indexes may be computed at the tract-level and all of the individuals living a particular tract may be assigned the same segregation score. Since there are thousands of tracts represented in the PSID there will be sufficient variation. In addition, county-level data is used to predict the propensity score by accounting for the labor market context of the individuals in the analysis (see Appendix B). None of the Census variables are time-varying. Even if respondents have moved since 1990, or were in a different tract prior to 1990, this information is not included. Without exact address information for each respondent, it cannot be determined whether a move took a respondent out of the 1990 Census tract. Addresses in the PSID are not public access data, thus it is nearly impossible to determine the exact location of a given respondent at a given time, though respondents are asked whether they have moved since their last interview. The utility of the 1990 Census data are as a baseline indicator for the kinds of areas in which respondents have lived, not as a measure of current environment. Two of the measures derived from the Census data merit some detailed attention. Two standard measures of residential segregation, D and P*, are calculated using the Census data and are included in the multivariate analysis. The formulae for these measures are shown below:
Σ ( xi / X ) − ( yi / Y ) D= *100 2
x
Py* =
(∑ (( x / X ) * ( y / t ))) *100 i
i
i
x i =number of Blacks in tract y i =number of non-Blacks in tract X=number of Blacks in MSA Y=number of non-Blacks in MSA t i =total population of tract
D is a measure of dissimilarity and P* is a measure of contact. D, the index of dissimilarity, can have values from 0 to 100. The numerical value of the index indicates the proportion of Blacks (the minority group) that would have to move from one census tract to another in order for Blacks to compose an equal proportion of each of the tracts in the MSA. An index of zero is a perfectly integrated tract, while an index of 100 is a perfectly segregated tract (Farley 1991). P* can be interpreted as the amount of contact between members of different groups within a given census tract. It is an asymmetrical measure and represents the probability that an average Black person in the MSA lives in a tract that includes at least one non-Black person (Massey and Denton 1987).
38
The values for these indices were taken from the 1990 Census and they were calculated for each Census tract that was merged with the PSID. The scores from a given tract were assigned to each respondent living in that tract. This is similar to the way in which the tract-level poverty values were assigned. The percent of poor in each tract is a variable available directly from the Census. I assigned Census tracts to one of three categories: high poverty (above 40%), medium poverty (3920%), and low poverty (19% and lower). For each respondent, a dummy variable indicates to which category of poverty her tract of residence is assigned. For additional details on the coding of the Census variables, and all of the variables used in the analysis, see Appendix B.
4.3: Selection/propensity
Since the main purpose of this project is to gain a better understanding of the effect of living in public housing on work outcomes, it is necessary to construct a sample that is composed of roughly comparable people, while maintaining sufficient variation for a multivariate analysis. The sample was first restricted to the non-elderly population (under 56 in 1986). The sample must include residents of public housing, but cannot be limited to them. It is quite likely that public housing residence will have a negative effect on work outcomes in a sample that includes all households, and indeed this is the case in the PSID, because public housing residents are a select group. 9 It is important to consider non-residents who are economically and socially similar to residents, but who do not live in subsidized housing. The question is how to select the comparison group. By using propensity score matching, a group of people with similar probabilities of living in public housing can be selected using a number of different variables (Freeman and Rohe 1999; Rubin 1997; Smith 1997; Winship and Morgan 1999). The propensity score, which is a predicted probability calculated by running a logit model with public housing residence as the dependent variable, may be used to determine which observations to include in the sample by matching them with a control group. In effect, public housing is seen as the ‘treatment.’ It is ‘given’ to one group and not to the other. These two groups can then be compared in any type of model. The advantage of the propensity score method is that it allows for the maximum use of the available information. As Smith (1997) suggests, I will test a specific outcome, “…the effect of x on y….” (p. 350), that is, the effect of public housing residence or non-residence on work outcomes. Many of the factors that determine the likelihood of admission to public housing can be used to construct the propensity scores. Everyone included in the matched sample used for all analyses has a relatively high propensity of living in public housing, but not all of them will be residents. Since the control group is selected by matching cases to the public housing population, everyone living in public housing in 1986 will remain in the sample. As Smith (1997) recommends, I use a matching algorithm to select cases for the analysis. First, I constructed the propensity score, which is simply a probability based on a linear combination of a group of variables in a logit model. The variables used for this include education, number of children, welfare status, county-level 9
It is not possible to simply select on the actual criteria used by public housing authorities to determine eligibility because these criteria differ across time and place. One universal criterion will be included in the propensity score: whether the family income is below 80% of county median. This is the minimum requirement that most housing agencies impose.
39
median income (whether family is below 80% of this), percent working in county, percent renters in census tract, race/ethnicity, and age. Then, the public housing group will be matched with the rest of the sample, matching one public housing observation with a maximum of fifteen cases (Smith 1997). As the matching process was conducted, I checked to see how close the scores of the matches were to the public housing cases. When the difference between scores of the cases and controls began to exceed one standard deviation of the propensity score, I halted the matching process. When matching the merged PSID-Census data ‘distant’ matches start appearing on the fifth iteration. For this reason, I limited the matching to five cases per treatment. I estimated the propensity score using the 1986 data and one time point. Matched cases are included for the rest of the analysis prospectively. That is, the variables used to construct the propensity score are all drawn from the first year of PSID data that contains a measure of public housing residence, 1986 (with the exception of the Census data from 1990). The matched cases are drawn from a sample that includes all years of the merged PSID data. The propensity score should be considered a time-invariant variable when included in the event history model. Regardless of changes in an individual’s propensity to live in public housing throughout the time period, the person will still be included in the sample if she was selected through the matching procedure because the sample should not vary across years. After the control group was selected it was recombined with the public housing group for the event history analysis (EHA). The propensity score is included as a variable in the multivariate analysis. In this way certain variables of interest, such as welfare status, can be interpreted as net of the propensity to live in public housing. Some of the variables that are used to construct the propensity score will be included separately in the multivariate models. The intercorrelation between these variables and the propensity score were examined to determine how many variables measuring similar concepts might be included in the EHA model. I conducted standard diagnostic tests for multicollinearity (VIF) and did not find any serious problems (for complete results, see Appendix D). Because the propensity score is a vector and does not capture the actual value of any particular variable, it is appropriate to include the variables that were used in constructing the propensity score when there is a theoretical justification to do so. Some variables are not adequately controlled simply by the inclusion of the propensity score in the model. However, any variables that are highly correlated with the propensity score are not represented separately in the EHA model. Some variables used to calculate the propensity score were excluded from the models of labor market participation for substantive reasons. The variables excluded from the multivariate models are: poor parent, never married, Hispanic, income less than 80% of county median, percent working in county, percent renters in tract, and number of children. The fact of having a poor parent might conceivably influence work outcomes as could the percent working in the county, but the relationships were not expected to be strong. An indicator of marital status is included in the multivariate models and would be highly correlated with never being married. Hispanics are a very low proportion of the total sample and are considered part of the reference group for the multivariate analysis. The percent of renters in a tract is not thought to have any direct relationship to the outcome variables. The below 80% of median income variable is endogenous to the work outcome models and must be excluded. The number of children is excluded from the multivariate models and a control for the presence of a young child is used. The presence of a young child is 40
thought to have more of an impact on labor market outcomes and is more substantively relevant to the multivariate analysis. When the propensity score is included in the model, to a large extent, the public housing variable becomes a measure of place, not person, effects. This is the case because the propensity score controls for many of the person effects indirectly, such as number of children. Certain factors that relate more to place and cannot be measured with the available data, such as social services and perceptions of safety, will be represented by the coefficient for public housing residence. Most of the person aspects of residence can be directly controlled, and will be included in the model as covariates. Thus, my hypothesis that the coefficient for public housing residence will be positive is based, at least in part, on the way I measure the effects. These issues are also addressed in the social resources piece of the analysis by comparing job search patterns and child care costs for residents and non-residents. In the event history model, the interaction between public housing residence and a few variables: welfare status, presence of a young child, and race are examined. The change in education level, over time, is also measured in the event history model. For the construction of the propensity score I used a model with the following variables, as measured in 1986: age; years of education; years of work experience; whether or not parents were poor when the respondent was growing up; never married; number of children; whether the respondent was Black or Hispanic; whether the respondent received AFDC; whether the head’s family income was less than 80% of the county median income, adjusted for family size and 1986 dollars; the percent employed in the county; and the percent of renters in the census tract (see Table 4.1). The initial sample size is 1,254 (209 public housing residents plus five matches per case ((209*5) + 209=1,254)). The frequencies and means for these variables can be found in Table 4.1. There are sizable initial differences between public housing residents and the rest of the population. For example, over 80% of public housing residents are Black, while only 32% of non-residents are Black. 12.4% of public housing residents received AFDC and only 3.2% of non-residents did. Public housing residents had a higher percentage of poor parents, indicating a history of childhood poverty. Public housing residents also had a slightly higher number of children and two years less education, on average. Work experience was over five years less on average for public housing residents. It should also be noted that the average age of public housing residents was a little over 31, over three years younger than non-residents. Table 4.1: Variables included in propensity score model (1986) Public housing residents N=209
Non-residents
Non-residents
matched cases
full sample
N=1045
N=2828
N
%
N
%
N
%
AFDC (yes)
26
12.44
83
7.94
89
3.15
Poor parent (yes)
93
44.50
455
43.54
928
32.81
Never married (yes)
97
46.41
388
37.13
627
22.17
Black (yes)
171
81.82
750
71.77
907
32.07
41
Public housing residents N=209
Non-residents
Non-residents
matched cases
full sample
N=1045
N=2828
N
%
N
%
N
%
Hispanic (yes)
5
2.39
37
3.54
119
4.21
Income less than 80% of county median
166
79.43
737
70.53
990
35.01
Mean
SD
Mean
SD
Mean
SD
Propensity score
.23
.17
.13
.11
.06
.09
Percent working in county
.80
.05
.80
.05
.81
.05
Percent renters in tract
.62
.23
.54
.21
.39
.22
Number of children
1.45
1.37
1.27
1.39
1.13
1.24
Years of education
11.40
1.91
11.84
2.14
13.18
2.42
Years of work experience
8.25
8.25
9.75
8.39
13.80
8.98
Age
31.43
8.45
32.08
8.73
34.72
8.81
The tabulations of the matched sample show a much different picture (see Table 4.1). The matched cases are still different than residents, but are more similar than the general non-resident population on the selected characteristics. It can be seen that the matching has accomplished its purpose and that some of the extreme variations have been eliminated. The main differences in the matched sample are found in the variables for AFDC, which 7.9% of matched non-residents receive; work experience, one and a half more years for non-residents; and marital status, 37.1% of matched non-residents are never married, compared to 46.4% of residents.
4.4: Multivariate analysis
Because of the nature of the question and the nature of the data, the best approach to modeling is discrete time event history analysis (EHA or survival analysis) (Allison 1995). The logit model used for the multivariate modeling is as follows (Allison 1995):
P log it = α t + β1 x it +...+ βk x itk 1 − Pit The main outcomes of interest are discrete variables that lend themselves to EHA. EHA is a good way of handling time-varying covariates with several years of data. Because I have seven years of data available and six distinct time points, it behooves me to use a method that takes advantage of the data. Also, by looking at several years of data, a stronger causal case can be made and more of the dynamics of the process may be understood. EHA allows for a more process-oriented approach than a static model. In addition, while causal direction cannot be definitively proven, allowing for several time-varying covariates will lead to a better understanding of the way these factors relate to each other over time. EHA may produce a clearer picture of what is going on—
42
regardless of causality issues. A univariate logit model on the full sample testing the zero order effects of public housing residence in 1986 on work/non-work in 1987 shows a strong negative effect (p=.0001). Thus, the analysis will attempt to diminish this negative effect by controlling for other factors, analyzing several years of data, and by selecting a smaller sample for appropriate comparison through the propensity score matching. One way that time-varying covariates can be included is by coding them as the cumulative number of times an event occurs by a given year (Allison 1995). Alternatively, time varying covariates can be coded as indicator variables in each of the person-years of the observations. The first method allows the effect of additional years of public housing residence on work outcomes to be discussed. If different coding is used, the effect of having lived in public housing in a given year can be assessed. In the current analysis, the former method is used. 10 Most of the time-varying independent variables were coded as indicators in each year. Thus, for example, the measure for ‘young child’ will equal one in any year where there was a young child present and zero in any year where there was no young child present. Age and years of education were measured as the cumulative value of that variable in a given year, similar to the public housing variable. Time itself can be controlled by including a variable for ‘year’ in the model to measure the number of years until the occurrence of an event or by including indicator variables for each year covered by the analysis. EHA gives enough flexibility to adequately address most of the questions raised in this research about the effects of public housing residence on work outcomes.
Outcome variables--EHA
The analysis uses a series of logit models, each tested with four dependent variables: 1) transition from work to non-work; 2) transition from non-work to work; 3) increase in annual hours worked; and 4) decrease in annual hours worked. The time interval for each variable is one year. The sample for each set of models is defined as all respondents who were at risk for the given outcome during the initial observation period. Thus, for the first variable, only sample members who are categorized as ‘working’ in 1986 are included in the models. For the second dependent variable, only those who are ‘not working’ in 1986 are included. For the third and fourth dependent variables, only sample members who have already made a transition to ‘work’ are included (effectively the same group as for the first dependent variable). These models use ‘person-year’ coding to account for the time varying nature of the data. Each year a person is observed as a member of the risk set contributes one observation to the sample. When the person exits the risk set, no more observations are recorded for that respondent. In the case of the transition from work to non-work, observations are recorded for each year in which a person is working. If the person makes the transition to non-work, no more observations are included. If the person is still working in the last period (year 7), it is regarded as a censored observation.
10
Both types of models were tested. Results do not vary significantly. It was decided for substantive reasons to use the time-varying measure ‘cumulative years in public housing.’
43
It should be noted that all of the outcomes being tested in this study are treated as nonrepeated events, though it would be possible, if more complicated, to treat the events as repeated. In this analysis, using repeated or non-repeated events is equally arbitrary since the starting point is not meaningful because these periods of working life do not have any special significance. That is, it is not necessarily the respondent’s first job, last job, or most important job. These are random slices of paid work, examined for people at different stages in the life course. Looking at the first transition and the subsequent transitions would not necessarily be more illustrative, since the nature of the transition could be quite different for different respondents. Also, by only looking at one event, the problem of unobserved heterogeneity introduced by repeated events is reduced substantially (Allison 1995). I intend to pursue this line of research in a subsequent project, but it is outside the scope of this dissertation.
Operationalization of outcome variables
Coding the dependent variables was a somewhat complicated process. The basic idea was to measure the probability of (paid) work vs. non-work. This would test the conceptual question behind the analysis: would living in public housing increase or decrease one’s likelihood of working? Along with this fundamental question, I also examine two further variations: would living in public housing increase the probability of working more hours per year? Would living in public housing increase the odds of a decrease in annual hours worked? I hypothesize that living in public housing might increase labor force participation for those not working (or working very few hours), but that it could also potentially lead to a decrease in total hours worked. There are several variables in the dataset that could be used to operationalize the dependent variables. Eventually, I chose as the basis for each of my dependent variables a raw variable that recorded the number of hours worked in each year. I re-oriented the variable to be contiguous with other variables measuring events in that same year. That is, if the variable measured annual hours worked in 1986 (even though the question was asked in 1987), I treated it as a ‘1986’ variable. Another option was to use the raw variable that asks if a person is working in the current year (that is, it would gather 1986 data in 1986). However, this variable, when compared to the ‘annual hours’ variable, seemed to greatly underestimate the actual work experience of the respondent. Many respondents who showed a value for ‘annual hours’ were recorded as ‘not working.’ While the ‘currently working’ variable might accurately reflect the work status of the respondent at the time of the interview, it does not reflect the work experiences of the respondent over the entire year. After also examining the variable that measured ‘weeks worked’ I decided that ‘annual hours’ was the most reliable variable with which to work. Using ‘annual hours’ to code ‘work/non-work’ presented some additional issues. There is not a standard way to assign a cutoff for the number of hours below which one should be regarded as ‘not working.’ 11 Should a person working more than zero hours per year be considered
11
This may be the case because when continuous data are available they are generally used in their original format. This research presents a case where in order to test the hypothesis it is necessary to transform a continuous variable into a dichotomous variable.
44
‘working’? Should there be a somewhat higher cutoff? Should there be some minimum number of hours worked annually before one is considered to be ‘working’? In order to preserve some variation in the data, it was clear that a cut-off was needed, since very few respondents worked zero hours in a year. After examining the distribution of these variables, and relying on common sense, I decided if anyone were working fewer than 320 12 hours a year, she should be considered ‘not working.’ There were also a few extremely high values on the annual hours worked distribution. After checking with staff at the PSID, it was decided not to exclude these observations from the analysis. PSID staff confirmed, that, indeed, these were accurate values and not due to recording error. It became clear that the exact level of the cutoff did affect the results. However, within a certain range, the changes were not dramatic. I chose 320 hours as the cutoff, after testing cutoffs at 160 and 520. 320 was viewed as the best compromise. For the measurement of increase and decrease in hours worked, the same threshold number of hours is used to measure change. 320 is used as the minimum number of hours needed for an increase or decrease in hours worked per year to be considered an event. Thus, if one were working 1,000 hours in year one and increased to 1,320 hours or more in year two, this would count as an increase. The measurement of an increase or decrease in annual hours worked is only applied to respondents who are regarded as ‘working’ (already working at least 320 hours per year).
Left truncation/censoring
While all of the observations included in any particular model will be in the risk set, there is still a problem of left truncation or left censoring because some respondents have been in the risk set longer than others. It is necessary to include information about the total length of time spent in the risk set, not just the time while the respondent was under observation for this study (Allison 1984; Guo 1993). Because public housing was not measured in the PSID before 1986, 1986 became the starting point for this study. However, there is nothing special about that particular year in the working lives of the respondents. In order to resolve this issue, a variable is included in the model that measures the number of years in the risk set prior to 1986. For example, in the models using transition from work to non-work as the dependent variable, the risk set includes only those people who were working in 1986. The ‘spell’ variable contains the length of the work spell for each member of the risk set prior to 1986. When the risk set contains those who are not working, the spell variable measures the length of the non-work spell prior to 1986. This variable was coded by using data from previous years of the PSID and includes all waves of the study, back to 1968. In the case of the models using a change in hours as the dependent variable, length of prior work spell is still used as a control for the open interval. While an argument could be made for using a measure of hours worked in prior years, to choose a measure of prior hours is not clear-cut, since there are many ways to conceptualize such a measure, for example, as the number of years working the same number of hours as in 1986, or the length of time since the last increase or decrease in annual hours worked. The construction of such a measure is not necessary in this case, because labor market participation, in general, is still the major consideration, not past changes. Because the sample is constrained to those respondents who are already working, it is theoretically important to
12
Approximately six hours a week or eight weeks of 40 hours per week work.
45
control for their prior work history and not as important to control for previous changes in hours worked. The complete work history is accounted for, which is the crucial measure for these models.
Hazard plots
In addition to the logit model for discrete time, which is the main feature of this analysis, I also used lifetables to produce hazard plots. This way of looking at the data is not appropriate when using many covariates. It does not allow for the layers of time that are available in the event history models, but it does provide a useful starting point. I examine plots of hazard estimates for each of the dependent variables for the control group versus the public housing group while holding time constant. The hazard plots provide a useful descriptive look at whether public housing residents differ from non-residents in their employment patterns. In the matched sample, the patterns are very similar. The hazard plots are presented below. They are followed by a discussion of the multivariate event history models and the social resources analysis.
5: Results I have used the more stringent two-tailed test when discussing these results, allowing for a hypothetical negative or positive result for each variable. Given that the literature is not clear on the expected effect of public housing on work, I am giving up power in order to check both directions, even though my hypothesis is directional. All analyses were conducted using the SAS system.
5.1: Life Table Analysis
I will present four hazard plots, one for each of the dependent variables. The models produce estimates of how long a respondent survived in the sample until the event occurred and the plots show the hazard of the event at a given moment in time. The matched sample was used for the analysis. The sample was stratified by whether or not a respondent lived in public housing in 1986. The duration for the analysis includes the entire period for the survey. The maximum duration was 25 years. The work spell and non-work spell variables were used to calculate durations prior to 1986. Six intervals were evaluated at their midpoints. Figure 5.1 (see below) represents the transition from work to non-work for the sub-sample of respondents who were working in the initial period, 1986. However, the “duration” axis represents the total duration of the exposure to work. Thus, if one began working in 1968 (the earliest year for which data are available), and was still working in 1986, and continued to work until 1992, the total duration would be 25 years. The graphs display the hazard in six intervals, graphed at their midpoints (the lifetable algorithm requires that the last interval be included). The hazard of a transition to non-work is usually slightly higher for the 1986 public housing residents, but, in the fourth interval, their hazard rate dips below that for non-residents. Respondents show relatively low hazards of leaving work in the first four intervals observed, but there is large increase in the hazard for the last interval. In the first four intervals, the hazard never rises above 2% for either group. The hazard for the transition to non-work is decreasing in the last period for both groups. When the hazard decreases in the last period, it is often a by-product of attrition; fewer cases remain, and thus the hazard of an event declines. When an increase in the hazard is observed in the last interval, it represents a true increase in the likelihood of an event. Thus, the hazard of a
46
transition to non-work can be said, with some certainty, to increase in the next to last interval, but no conclusions should be made about the last interval. The increase in the fifth interval is most likely an age effect, that is, as one gets older, one is more likely to stop working for a variety of reasons. In this case, those still at risk in the fifth interval are the oldest respondents, who are necessarily at a high risk for an exit from work (mostly likely due to retirement).
Figure 5.1: Hazard plot of the transition from work to non-work In Figure 5.2 (see below), the transition from non-work to work is shown. The hazard for a transition is low for both groups in the first four intervals, but higher for 1986 public housing residents, except in the first interval. The hazard rates cross between the first and second intervals, showing an initially higher hazard of a transition to work for the non-residents. However, for the rest of the period, hazards are always higher for the 1986 public housing residents. In the third interval, both hazards drop off slightly showing that, over time, both groups experience a decline in the hazard of a transition to work. In the fifth interval, there is a sharp rise in the hazard, though the increase is greater for non-residents. The drop in the last interval should not be interpreted. The hazard for a transition to work is consistently higher for public housing residents. Proportionately, in 1986, more public housing residents are not working (30%) than non-residents (19%), thus more public housing residents are eligible to make the transition to work.
47
Figure 5.2: Hazard plot of the transition from non-work to work Figure 5.3 (see below) shows the hazard plot for an increase in annual hours work. The subsample is the group of respondents working in 1986, the same as in Figure 5.1. The pattern of these hazard rates is quite different from the previous graphs. At most points, the 1986 public housing residents show a lower hazard for an increase in hours, though their hazards are higher in the first and second intervals. The hazard for the non-residents shows a steady decline over time, with a steep increase in the fifth interval. The general pattern shows that the likelihood of an increase in hours over time is fairly consistent, though declining until the last interval. The pattern for the public housing residents is strikingly similar to the pattern for non-residents. If public housing residents were very dissimilar from non-residents, a larger difference would be expected. What is observed here, in the matched sample, are almost identical hazard rates for an increase in annual hours worked.
48
Figure 5.3: Hazard plot of an increase in annual hours worked Figure 5.4 (see below) is the plot of the hazard for a decrease in annual hours worked. The sample is the group of respondents working in 1986. The hazards for the public housing residents are almost exactly parallel to the hazards for the non-residents. For both groups, the hazard for a decrease in hours declines over time, until the sharp increase in the final interval. The pattern here is actually quite similar to that in the previous figure for the increase in annual hours. There is a steady, relatively flat hazard for a decrease in hours across most of the period, with a large increase for both groups in the fifth interval. The decline in the hazard, after an initial increase, suggests that the longer one remains in the labor force, the less likely one is to experience a decline in annual hours worked. The hazards are not exactly the same for public housing residents and non-residents, but they are so similar as to be almost equivalent.
49
Figure 5.4: Hazard plot of a decrease in annual hours worked
5.2: Event History Analysis Descriptive statistics
Most of the independent variables included in the multivariate models are coded as timevarying. Housing subsidy is a time-varying dichotomous variable, as is AFDC receipt, presence of related or non-related adults, presence of a young child, marital status, and disabled status. Years of education, age, and public housing residence are cumulative time-varying measures. Length of prior work or non-work spell is a time-constant measure, as are years of work experience, female, Black, urban area, poverty level, and level of residential segregation. The work experience variable is a measure of years of experience, less the number of years spent in the prior work spell, for the models using a sample of currently working respondents. For the models examining transitions from non-work to work, the work experience variable measures total work experience since age 18. See Appendix B for a description of all of the variables included in the analysis. Tables 5.1 and 5.2 present the descriptive statistics for the event history models. Table 5.1 is for the analyses that use the sub-sample of people working in 1986. The total sample size is 987. Public housing residents compose 14.7% of the sample. About 42% of the sample is female and 37% are married. 32% of public housing residents make a transition to non-work and 25% of nonresidents make this transition. The fact that this baseline difference is relatively low is fairly remarkable. Given the extremely negative characteristics thought to be associated with public housing it is striking that they are only 7% less likely to make a transition to work than the matched
50
non-residents. When this comparison is made with the non-matched sample, the baseline percentage point difference is slightly over 19%. 13 A much higher percentage of respondents overall have an increase or decrease in annual hours worked. A slightly higher percentage of public housing residents experience a decrease in annual hours, and slightly fewer public housing residents experience an increase than non-residents. It is slightly perplexing that almost the same number of people experience an increase in annual hours worked as experience a decrease in annual hours worked. However, this is over a seven-year period, which may explain the overlap. Even though this reflects a high level of variability in employment status, that is not too surprising for a low-income population that is likely to experience intermittent employment. Any periods of unemployment are included—that is, in a given year if one were to move from one full-time job to another, but spend eight weeks out of work, that would be shown as a decrease in annual hours for that year. In addition, the same respondents may contribute an event to both the increase and the decrease models. Table 5.1: Descriptive statistics for those working in 1986 N (987) Received housing subsidy 1986 High poverty tract Medium poverty tract Low poverty tract Urban area Black White Hispanic Female Received AFDC 1986 Other relatives present 1986 Non-related adult present 1986 Child age 5 or less in 1986 Married 1986 Disabled head 1986 Event (transition to non-work) Event (increased hours worked/yr) Event (decreased hours worked/yr) Duration times (transition to non-work): 1 2 3
Total %
36 178 374 435 874 711 238 31 412 37 78 42 352 361 6 258 720 759
3.65 18.03 37.89 44.07 88.55 72.04 24.11 3.14 41.74 3.75 7.90 4.26 35.66 36.58 0.61 26.14 72.95 76.90
65 42 35
6.59 4.26 3.55
13
Public housing resident 1986 (145)% 0.0 34.48 31.72 33.79 88.28 77.24 18.62 3.45 45.52 7.59 9.66 4.83 40.00 29.66 0.0 31.72 71.72 78.62 6.90 7.59 2.76
Non-public housing resident 1986 (842)% 4.28 15.20 38.95 45.84 88.60 71.14 25.06 3.09 41.09 3.09 7.60 4.16 34.92 37.77 0.71 25.18 73.16 76.60 6.53 3.68 3.68
I also ran the EHA models described below using the full PSID sample (that is, not using the subsample obtained through the propensity score matching). The results of those models are strikingly similar to what was found using the subsample. One exception is that in all four base models the coefficient for public housing was highly significant. However, in three of the full models, the public housing coefficient was reduced to insignificance. The exception was the case of the model testing the transition to non-work, where a significant public housing coefficient was observed in the final model, just it was observed in the EHA using the matched subsample.
51
N (987) 4 5 6 Duration times (increase in annual hours): 1 2 3 4 5 6 Duration times (decrease in annual hours): 1 2 3 4 5 6
Propensity score Dissimilarity index for county (race) (d) Contact index for county (race) (p) Age of respondent in 1986 Years of education in 1986 Years worked in 1986 (less work spell) Average length of work spell, prior to 1986 (years)
37 35 773
Total %
3.75 3.55 78.32
Public housing resident 1986 (145)% 6.21 4.14 72.42
Non-public housing resident 1986 (842)% 3.33 3.44 79.33
218 172 134 81 66 316
22.09 17.43 13.58 8.21 6.69 32.01
20.69 20.00 12.41 5.52 7.59 33.80
22.33 16.98 13.78 8.67 6.53 31.71
239 163 152 93 58 282
24.21 16.51 15.40 9.42 5.88 28.57 All N (987)
28.28 17.93 15.86 9.66 4.14 24.14 Public housing resident 1986 (145)
mean s.d. mean s.d. mean s.d. mean s.d. mean s.d. mean s.d. mean s.d.
0.13 0.11 1.41 2.56 1.99 5.90 31.45 8.19 12.00 2.03 3.40 6.04 12.09 7.13
0.19 0.14 1.52 2.56 2.56 6.78 31.32 2.04 11.61 1.98 3.67 7.00 10.87 7.55
23.52 16.27 15.32 9.38 6.18 29.34 Non-public housing resident 1986 (842) 0.12 0.10 1.39 2.60 1.89 5.74 31.47 8.23 12.07 2.04 3.35 5.87 12.30 7.04
Table 5.2 (see below) presents descriptive statistics for the sub-sample of respondents who were not working in 1986 (used to model the transitions from non-work to work). The sample size is 267 and 24% of the respondents are public housing residents. 33% are in a high poverty tract and 79% are Black. 70% are female, 19% are disabled, and 13% are married. About 64% of the 1986 public housing residents made a transition to work, compared to 58% of the non-residents. The characteristics of the two sub-samples are quite different and reflect the reasons why some people are not working.
52
Table 5.2: Descriptive statistics for those not working in 1986 N (267) Total %
Received housing subsidy 1986 High poverty tract Medium poverty tract Low poverty tract Urban area Black White Hispanic Female Received AFDC 1986 Other relatives present 1986 Non-related adult present 1986 Child age 5 or less in 1986 Married 1986 Disabled head 1986 Event (transition to work) Duration times (transition to work): 1 2 3 4 5 6 Propensity score Dissimilarity index for county (d) Contact index for county (p) Age of respondent in 1986 Years of education in 1986 Years worked in 1986 Average length of non-work spell, prior to 1986
24 89 111 67 247 210 41 11 186 72 32 11 108 34 51 159
8.99 33.33 41.57 25.09 92.51 78.65 15.36 4.12 69.66 26.97 11.99 4.12 40.45 12.73 19.10 59.55
80 30 17 16 8 116 mean s.d. mean s.d. mean s.d. mean s.d. mean s.d. mean s.d. mean s.d.
29.96 11.24 6.37 5.99 3.00 43.45 0.22 0.15 1.33 2.74 1.38 4.18 33.87 10.10 10.90 2.16 3.92 7.24 6.15 5.50
Public housing resident 1986 (64) % 0.0 43.75 32.81 23.44 90.63 92.19 7.81 0.0 78.13 23.44 12.50 6.25 39.06 6.25 15.63 64.06
0.31 0.19 1.85 3.99 1.44 3.86 31.66 9.49 10.94 1.68 2.42 5.32 6.70 5.98
Non-public housing resident 1986 (203) % 11.82 30.05 44.33 25.62 93.10 74.38 17.73 5.42 67.00 28.08 11.82 3.45 40.89 14.78 20.20 58.13
0.19 0.13 1.17 2.20 1.36 4.29 34.57 10.20 10.89 2.30 4.39 7.70 5.98 5.34
In each of the person and place models for all of the dependent variables, I include categorical variables for year, work or non-work spell, and the propensity score. The variable for year has six categories. Year six is the reference category and does not appear in the tables. The year variables represent the year in which the respondent made a transition or was censored from the sample. All respondents who survived to year six may be regarded as censored cases. Respondents with a value for “year one” made a transition to work between 1986 and 1987. The variables selected for each of the models are designed to control for the factors presented in detail in the theoretical
53
framework. For the most part, only significant coefficients are discussed, with the exception of the public housing variable. In the case of the time-varying variables, they are coded to represent the observations in the periods prior to an event. This should minimize the endogeneity in the models. Only when there is no event (a censored case) do the variables have values for all observation periods and, thus, a full six cases are contributed to the sample for that person. The goodness of fit measures used for discussion are nested likelihood ratio tests. For a different approach to interpreting the results presented here, see Appendix E, which contains a discussion of all significant coefficients using predicted probabilities.
Transitions from work to non-work
The group at risk is those respondents who were working in 1986 (see Table 5.1), a subset of the matched sample. In the base models (see Table 5.3), the public housing variable is always significant. Including an indicator for each year of observation does not diminish public housing’s significance and the significance does not change when the length of the previous work spell is included. Without further controls, public housing is found to increase the likelihood of leaving work, with high levels of statistical significance. All the year variables are insignificant and remain so in every model. The length of time of the work spell before the starting point of the analysis is negatively and significantly related to the outcome, and retains its significance in each model. That is, the longer the total length of the prior work spell, the less likely one is to leave work. A nested likelihood ratio test shows that model fit is significantly improved by the variables added in models 5.3a and 5.3b (see Table 5.3), when compared to Model 5.3.
Table 5.3: Transition from work to non-work, base model Model 5.3 Model 5.3a Parameter Parameter Estimate Estimate (SE) (SE) Intercept -3.1648 -2.9752 (0.0700) (0.1651) Years in public housing
0.1608*** (0.0547)
Model 5.3b Parameter Estimate (SE) -2.1139 (0.191)
0.2420*** (0.0596)
0.2073*** (0.0617)
Year 1
0.2841 (0.2067)
0.1924 (0.2083)
Year 2
-0.1403 (0.2246)
-0.2029 (0.2259)
Year 3
-0.3076 (0.234)
-0.3462 (0.2354)
Year 4
-0.2343 (0.2301)
-0.2603 (0.2315)
Year 5
-0.2702 (0.2332)
-0.2963 (0.2348)
Length of work spell Events
258
-0.0733*** (0.0093) 258
258
54
Table 5.3: Transition from work to non-work, base model Model 5.3 Model 5.3a Parameter Parameter Estimate Estimate (SE) (SE) Person years 5215 5215
Model 5.3b Parameter Estimate (SE) 5215
Log likelihood
-982.9205***
-1021.1263***
-1015.3731***
+
p