US Public Housing Access and Family Structure: 1950 and 1970 Katharine L. Shester Washington & Lee University and

Samuel K. Allen Virginia Military Institute November 2014 Preliminary: Do Not Quote without the Authors’ Permission.

Abstract Between 1950 and 1970 the number of public housing units in the United States grew nearly six fold, expanding the housing opportunities for low income families. During this same period, the U.S. experienced the beginning of a dramatic change in family structure, as the percentage of births to unwed women and the percentage of female-headed households began to increase rapidly. We use available individual-level data to explore and quantify possible connections between this increased access to public housing and family structure decisions made by young women in this era. We use linear probability and probit models to assess whether young women living near higher concentrations of public housing were more or less likely to get married, have children, or have children out of wedlock. Our results suggest a positive and statistically significant association between public housing access and the probability that young non-black women chose to marry. This is consistent with public housing lowering the cost of household formation and marriage. We find no evidence that public housing is correlated with higher rates of single motherhood. Greater access to public housing does not tend to increase rates of marriage or motherhood for young black women.

Acknowledgements The authors thank seminar participants from Washington & Lee University, Virginia Military Institute, The College of William & Mary, and the Virginia Association of Economists for their valuable feedback and suggestions. Carolyn Moehling graciously provided ADC/AFDC data. Charlotte Karp provided excellent research assistance. Remaining errors are our own. 1

Between 1950 and 1970, the percentage of births to unmarried women increased from 4.0 to 10.7 percent. This upward trend in single motherhood continued throughout the late twentieth century, as the percentage of births to unmarried women reached 28.0 percent in 1990 and 33.0 percent in 1999 (U.S. Department of Health and Human Services, 2000). The birth rates of unmarried women followed a similar pattern, increasing from 14.1 births per 1,000 women (ages 15-44) in 1950, to 26.4 births in 1970 and 43.8 births in 1990 (U.S. Department of Health and Human Services, 2000). This sea change in family structure has been a topic of much concern, as female-headed households are more likely to have incomes below the poverty line (Garfinkel and McLanahan 1986) and have been linked with adverse child outcomes, such as lower educational attainment (Dawson 1991, McLanahan and Sandefur 1994), psychological problems (Aseltine 1996, Hetherington and Clingempeel 1992, Zill and Peterson 1986), and illicit activity (Flewelling and Bauman 1990). While these changes in family structure have been well documented, it is unclear what has been driving them. Much of it may be due to changes in cultural norms (Akerloff and Yellen 1996), although these norms have been influenced by policy incentives. There is a large literature that assesses the effects of welfare on fertility and family structure, but results are mixed. Cross-sectional studies have found that states with more generous welfare benefits have more unmarried women (Schultz 1994 and 1998, Moffitt 1994), more single mothers (Moffitt 1994), and higher rates of female household headship (Moffitt 1994, Danziger, Jakubson, Schwartz, and Smolensky 1982), but these relationships are not robust to the inclusion of state fixed effects (Moffitt 1994, Hoynes 1997, Blau, Kahn, and Waldfogel 2004). This suggests that while states with more generous welfare benefits have lower rates of marriage and higher rates of female headship, women do not respond to changes in welfare benefits by altering their marriage 2

and headship decisions. The large increase in single motherhood between 1970 and 1990 was also accompanied by a general decline in welfare benefits, which is inconsistent with the notion that increases in single motherhood were driven by increasingly generous welfare payments. Most of the welfare and family structure literature focuses on 1970-onward, as this was the period during which there was the largest absolute increase in single motherhood. However, Moehling (2007) adds a valuable perspective on this relationship by focusing on the period from 1910 to 1970. This time period is useful as it covers the introduction of Aid to Families with Dependent Children (AFDC) in 1935, as well as the initial rise in single motherhood, which began in the 1940s for blacks and the 1950s for whites. Many states created mothers’ pension programs as early as the 1910s, which were targeted at needy single mothers, and Moehling finds that states that offered generous pensions also offered generous AFDC benefits later on. However, she finds little evidence that more generous welfare benefits influenced family structure throughout the twentieth century. Moehling notes that the variation in benefits existed decades before the rise of single motherhood and that the variation in benefits was not initially correlated with state differences in family structure. This finding casts further doubt on the idea that welfare benefits were responsible for changing family structure norms. In this paper, we assess the relationship between family structure and a related public assistance program, public housing, between 1950 and 1970. We focus on this period as it coincides with both the initial increase in single motherhood in the United States and the rapid expansion of public housing. The first public housing was built in the 1930s, and the program expanded greatly after the Housing Act of 1949, which called for the construction of 810,000 additional units. While construction was slow, the number of public housing units in operation

3

increased by nearly six fold between 1949 and 1970, increasing from 146,000 to 830,000 (Olsen 2012).1 We use individual-level data from the 1950 and 1970 Censuses to assess the relationship between the availability of public housing and the fertility and marriage decisions of young women (age 16-24). More specifically, we use linear probability and probit models to assess whether young women living near higher concentrations of public housing were more or less likely to get married, have children, or have children out of wedlock. It is important to note that the potential effects of public housing on these family structure decisions are ambiguous. In the earlier part of the period, many public housing agencies enforced “morality restrictions,” which allowed them to block the admission of families with “immoral character” into public housing, often times excluding women who had children out of wedlock. However, as these restrictions were loosened, regulations related to admission and rental payment may have influenced the behavior of prospective tenants. Rent was set as a proportion of income, which may have discouraged young women living in public housing from marrying, as this may have increased their household income, and therefore their rent.

Or, if

marrying sufficiently increased household income, marrying may have made the household ineligible for public housing. On the other hand, it is also possible that public housing lowered the cost of marriage, not only for tenants, but for the larger population as well. Assuming that the construction of lowrent public housing did not perfectly crowd out the construction of private housing (Sinai and Waldfogel 2005), the increase in the supply of housing may have lowered the cost of rental

We end our focus in 1970 for a couple of reasons. The first is that while additional public housing units were built after 1970, the relative importance of the public housing program declined in the 1970s with the introduction of the Section 8 Housing Voucher program. The second reason is more practical, which is that our public housing data ends in 1973.

4

housing in the market. To the extent that being able to financially support a household was viewed as a prerequisite for marriage, public housing may have allowed couples to get married earlier by reducing the price of housing. This may have also reduced the propensity of single motherhood if some of the women who would have otherwise had children out of wedlock chose to marry instead. Our results suggest that proximity to public housing increased the probability that young non-black women were ever-married and decreased the probability that non-black women were never-married single mothers. However, there appears to be no measurable effect on marriage or single motherhood for black women. We also find no evidence that public housing influenced the number of women having children, although public housing is correlated with higher fertility rates for women with children.

A BRIEF HISTORY OF PUBLIC HOUSING The United States public housing program began in 1933 under the New Deal, when the National Industrial Recovery Act gave the Public Works Administration (PWA) funding to build the first federally funded public, low-income housing. Construction started in 1934 and by 1937, the PWA constructed approximately 22,000 units in 36 metropolitan areas (National Commission on Urban Problems p. 108 1969, Coulibaly, Green and James 1998). The program was continued in the Housing Act of 1937, which made a few major changes to the way that the public housing system operated. First, while the PWA had the authority to build public housing in any locality, the Housing Act of 1937 transferred the planning, construction and management of public housing, as well as the ownership of projects, to local communities.

5

The Housing Act of 1937 also established the “equivalent elimination requirement”, which stated that for each unit of public housing that was built, a “substantially equal number” of substandard units in the metropolitan area had to be demolished or repaired (National Commission on Urban Problems p. 109 1969). At first glance, this rule seems to suggest that the construction of public housing had no net effect on the supply of housing. If one substandard unit was demolished for every unit of public housing that was constructed, there would be no net effect on the number of housing units, although the quality of units would have increased. However, the housing stock might have increased if the requirement was met by repairing substandard units and the construction of public housing did not perfectly crowd out private construction. Historic accounts suggest that local housing authorities found ways to minimize the demolition and repairs required by the equivalent elimination requirement. According to Elizabeth Wood, a contemporary public housing expert, housing authorities “…counted the units eliminated anywhere in the city due to public works projects, because of the action of local health and building departments, or as a result of the voluntary action of private individuals” and “…credited demolitions to their equivalent elimination accounts for which they had no responsibility and which would have taken place anyway. (Wood, p. 11 1982)” It is therefore unlikely that the equivalent elimination requirement kept housing supply constant. The public housing program was expanded again with the Housing Act of 1949. At the beginning of 1949, there were approximately 170,000 units in operation, and the Housing Act of 1949 approved the construction of an additional 810,000 units (NCUP p. 110 1969). These were initially going to be constructed over a six year period, although construction was much slower than anticipated. By 1970, there were approximately 830,000 units in operation and by 1973, there were over one million (Olsen 2012). The Housing Act of 1949 weakened the equivalent

6

elimination requirement, exempting all projects in rural areas and urban projects that were built on slum sites (Fisher 1959). The Act also required that eligible low-income families that were displaced by government programs, such as slum clearance, public housing, or urban renewal, be given first priority for admission into public housing (NCUP p.111 1969). The characteristics of tenants changed greatly after 1950 as well, which was in large part due to changes in eligibility requirements and rent schedules. The first federally mandated eligibility criteria was passed in 1936 with the George Healey Act which stated that eligible families were required to live in housing of substandard quality, had to be able to afford a rental payment that covered all project operating expenses, and could earn an income no greater than five times the rent of a public housing project (Wood p. 19, 1982). The Housing Act of 1937 contained no additional guidelines, and so local housing authorities were largely on their own to decide exactly how to set income requirements and rents. However, the requirement of rent covering operating expenses did put a lower bound on the incomes of potentially eligible families, excluding the very poor. While local housing authorities could choose their own rental payment schemes, by the mid-1940s most charged their tenants a fixed proportion of their income in rent. While the George Healey Act specified income limits for admission into public housing, it did not directly mention what should happen if public housing tenants experienced an increase in income above the maximum allowed for admission. Some tenants were evicted when their incomes increased above the maximum level for admission, but enforcement was not strict. A manual published by the USHA in 1942 emphasized the importance of tenants feeling like they had a home without fear of being punished, either by being harassed or evicted, if they tried to improve their lives by increasing their income. The manual also suggested that local housing authorities let tenants

7

remain in public housing and pay an additional rent premium if their incomes exceeded the limit and they could not find decent private housing (Wood p.20-21, 1982). In addition to income eligibility criteria, many housing authorities were selective on who was admitted and imposed restrictions based on “moral character”. For example, up until 1968 in New York, women who had children out of wedlock, individuals who had a history of drugs or violent behavior, and individuals with a “…bad previous record as a housekeeper or parent” were usually excluded (US National Commission on Urban Problems p. 116, 1969). By the late 1940s, the federal government insisted that local housing authorities strictly enforce maximum income limits. This was formalized in the Housing Act of 1949, which required local agencies to periodically check the incomes of its tenants and remove any families whose income rose above the maximum limit. This, combined with the requirement that eligible candidates displaced by public action be given first priority, changed the character of public housing tenants. Between 1950 and 1969, the median family income of public housing tenants fell from 63.5 to 42.4 percent of the area median and the percent of public housing tenants composed of single-parent households increased from 19 to 31 percent (Silverman 1971).

POTENTIAL EFFECTS OF PUBLIC HOUSING ON FERTILITY AND MARRIAGE DECISIONS The vast expansion of subsidized housing for low-income households could have created a variety of incentives for women and families. First, the construction of public housing may have lowered the cost of decent, non-dilapidated housing not just for public housing tenants, but for other households as well. By providing housing for some low-income families, public housing may have lowered the demand for private, unsubsidized rental housing. If the

8

construction of public housing did not perfectly crowd out the construction of private units and was not met with an increase in the demand for rental housing, such as through the in-migration of households, then the cost of private rental housing would go down. To the extent that being financially able to form one’s own household was a prerequisite for marriage, lower rental costs, both for public housing tenants as well as others, may have allowed couples to get married and form new households earlier in their lifecycles. For couples waiting until marriage to have children, this may have also influenced their fertility decisions by allowing them to have children at an earlier age. Lower housing costs could also directly reduce the cost of having children if the cost of additional bedrooms was reduced. This reduction would have been magnified for public housing tenants, as rent and unit size were determined independently. The size of a tenant’s public housing unit was a function of family size and may increase with children, while a tenant’s rent was a function of income. Therefore, for public housing tenants, having more children may have resulted in a larger housing unit at no additional housing cost. Economic theory suggests that as the costs of having children go down, the fertility rate should increase. This may be true for all low-income women, and may exist on the extensive (number of women with children) or intensive (children per woman) margins. The “morality restrictions” imposed by some local housing authorities during the earlier part of this period may have also incentivized marriage by offering limited admission to women who had children out of wedlock. Depending on the prevailing local norms, it is conceivable that some young women may have considered marriage as a way to increase their chances of qualifying for subsidized public housing. Or, if single mothers who had children out of wedlock

9

were ineligible for public housing, then the provision of public housing may have, at the very least, created no incentive for women to have children outside of marriage. However, it is also possible that the income eligibility and rental pricing scheme of public housing discouraged marriage. Because rent was linked to household income, tenants received an implicit penalty to earning more income or adding an additional earner to their household. This may have discouraged single women, particularly single mothers, residing in public housing from getting married, if doing so would have caused their rent to increase or caused them to become ineligible for public housing. If single mothers were discouraged from marrying their partners because of reduced housing benefits due to increases in household income, this could increase the percentage of female headed households or unwed mothers.

DATA In this paper, we use public housing data from the Consolidated Development Directory, published by the Department of Housing and Urban Development in 1973. This dataset contains project-level information on the number of units and dates of construction and eligibility for all public housing projects in existence in 1973. We aggregate the number of public housing units available in 1950 and 1970 at the metropolitan statistical area (MSA) level. MSA boundaries change over time, and so we limit our analysis to individuals in 56 MSAs for which we can define consistent boundaries between 1950 and 1970. Depending on the MSA, public housing accounts for between 0 and 30 of every 1000 housing units. The 56 MSAs include all of the largest five cities in 1950, and 49 of the MSAs include a city in the top 100 most populous cities in either 1950 or 1970. Figure 1 plots the MSAs (with circles’ areas approximately proportional to population represented.) The full list of MSAs, and their definitions, is in the appendix.

10

Individual-level data from the 1950 and 1970 Censuses is from IPUMS (Ruggles et al., 2010). We also incorporate MSA characteristic data from the 1950 and 1970 Censuses (Haines 2010) and state AFDC benefit data (Moehling 2007; Moffitt 1994).2 We limit our sample to females, age 16-24, in 1950 and 1970. Our rationale for focusing on young women is two-fold: first, because we are using pooled cross-sectional data, we want to observe the intensity of public housing that women were exposed to while they were making fertility and marriage decisions. Young women are more likely to be making these decisions around the time of observation. Second, public housing may influence the percentage of the population composed of ever-married women, mothers, and never-married mothers either by influencing young women’s fertility and marriage decisions (what we want to measure), or by influencing their migration decisions. For example, it is possible that public housing had no effect on the fertility and marriage decisions of women, but that single mothers chose to move areas with high concentrations of public housing. While we cannot rule out the possibility that public housing influenced these outcomes through migration, it is less likely that younger women had moved from where they lived when they made their marriage and fertility decisions than older women. We do not include individuals from 1960, as the 1960 Census does not identify an individual’s MSA of residence. Summary statistics are reported in Table 1. In 1950 and 1970, approximately 38 percent of our sample of 16-24 year old women were married. The percentage of women with children was 24 percent in 1950, and 23 percent in 1970. While these two statistics remained fairly constant between 1950 and 1970, the percentage of young women composed of never-married moms increased from 0.3 percent to 1.65 percent over the period. The general trends are the 2

The data for 1970 can be downloaded from Robert Moffitt's website: http://www.econ2.jhu.edu/people/moffitt/datasets.html

11

same for whites and blacks; however, the percentage of mothers and single mothers is higher for blacks than for whites in both years. The percentage of black 16-24 year old females with children in 1950 and 1970 was 32-33 percent, compared to 23-24 percent for whites. Between 1950 and 1970 the percentage of young black women who had children and were never married increased from 1.7 to 8.4 percent.

EMPIRICAL STRATEGY AND BASIC RESULTS We use this large, pooled cross-sectional dataset to assess the relationship between family structure and the access to public housing for women aged 16 to 24 in 1950 and 1970. Our variables of interest include whether a woman has ever been married, whether she is a mother, and whether she is a single mother who has never been married. We adopt a linear probability model and probit model framework in which we regress our variables of interest on individual and place characteristics.

Yimt = α + β1 %PublicHousingmt + β2 %PublicHousingmt*Black + Iit'δt + Mmt'γt + Γm

Our primary variable of interest is %PublicHousing, which is the percent of total occupied units within an MSA which is composed of public housing in either 1950 or 1970. We implicitly assume that the impact of public housing access on marriage and fertility decisions is constant across years; however, we do allow black women to have a separate public housing coefficient by interacting our public housing measure with our Black dummy variable. Therefore the estimated “effect” of public housing on non-blacks is 𝛽̂ 1 , and the estimated “effect” of public housing on blacks is 𝛽̂1+ 𝛽̂2.

12

We control for individual-level characteristics, I, which include age, education, and race dummy variables. Age is broken into individual year bins and we interact these bins with a year indicator variable to allow for different effects in 1950 and 1970. Our education bins include high school dropout, in school (high school or below), high school graduate (no college), in college, college dropout, and college graduate. We interact our education bins with a 1970 year indicator variable as well, to allow for different effects over time. Race is defined as black and non-black, and we allow for it to trend differently over time. We include MSA fixed effects, Γm, which will absorb time-invariant differences in marriage and fertility behavior of women across MSAs in 1950 and 1970.3 We also control for time-varying place characteristics, such as the log of average monthly family welfare benefits, MSA employment rate for males age 24-54, and an MSA industrial shift index.4 The effect of welfare benefits is allowed to differ by race. The MSA industrial shift index is meant to capture anticipated MSA trends based on the industrial structure of MSAs in 1950. This index interacts MSA industrial composition in 1950, measured as the percent of the labor force employed in 16 different sectors, with national industrial shifts in employment between 1950 and 1970.5 The presence of MSA fixed effects means that our coefficients of interest, β1 and β2, are being identified by relative changes in MSA public housing intensity over time. In principle, if an MSA has the same percentage of public housing in 1950 and 1970, then its MSA fixed effect would pick up their public housing “effect.” Similarly, if all MSAs experienced the same change

Because we only have data from two time periods, we cannot add MSA-level time trends. Doing so would cause our MSA fixed effects to absorb the effect of public housing in 1950, and the trend would absorb any changes in public housing intensity between 1950 and 1970. 4 Welfare benefits (ADC in 1950, AFDC in 1970) were determined at the state level. For MSAs that cross state lines, all residents in the MSA were given the value of the highest available benefit within the MSA. 5 It has a value of zero in 1950 and a positive value in 1970.

13

in public housing between 1950 and 1970, the 1970 fixed effect would pick this up. Our public housing coefficients could therefore be interpreted as causal effects of public housing IF, controlling for national trends in marriage and fertility decisions by age, education and race, as well as time-invariant differences across MSAs, and differences in MSA levels of employment and industrial composition, there are no unobservable characteristics that are correlated with MSA changes in public housing intensity between 1950 and 1970 that directly influence marriage and fertility decisions. We assess the robustness of these results to further controls later in the paper. Results from linear probability models and probit models are similar, and are reported in panel A of Table 2. The results in Table 2, panel A show robust, positive, and statistically significant effects of public housing on the likelihood that a young, non-black women had ever been married. The coefficient on %PublicHousing in our linear probability model is 0.019 (statistically significant at the one percent level), and suggests that if public housing access rises from 1 to 2 percent of the total housing stock, then the probability of an average young woman in an average MSA having ever married rises from 37.6 to nearly 40 percent. The coefficient on %PublicHousingmt*Black, indicates that the relationship between public housing and motherhood was statistically different for blacks and non-blacks, and the net effect of public housing for blacks is not statistically different from zero. While we find a link between public housing and marriage for non-black women, we find negligible evidence to suggest that public housing access influenced the probability that young women chose to become mothers. The estimated “effect” of public housing on the probability of being a mother is statistically different for blacks and non-blacks, but neither is statistically different from zero.

14

Public housing intensity is, however, weakly correlated with lower fertility among nevermarried women. Our linear probability model estimate suggests that a one percentage point increase in public housing intensity is correlated with a 0.3 percentage point decrease in the likelihood of a young woman being a never-married single mother (statistically significant at the 5 percent level). The magnitude of this coefficient is relatively large, given that the percentage of never-married single mothers in our sample is 0.3 in 1950 and 1.6 in 1970. However, the net effect of public housing on blacks is not statistically significant. The probit results for blacks and non-blacks are smaller and not statistically significant, although they are consistent with public housing lowering the propensity of never-married single mothers for non-blacks. The number of observations used in the probit model is somewhat smaller (91,866 compared to 93,145), as individuals observed in 1950 who were either in college or were college graduates were dropped.6 Linear probability models run with the same limited sample are similar to those reported with the somewhat larger sample. Robustness Checks We test the robustness of our results to a few additional controls. First, we interact our education and education-by-year variables with our black dummy variable to allow for education to have a different effect by race. Next, we include region x 1970 controls to allow for differences in time trends across regions. In our original specification, we included interaction terms of our age and education bins, MSA employment rates, and our race variable with our 1970 dummy variable, in addition to the inclusion of a 1970 fixed effect. However, it is possible that regions experienced different trends over time that were correlated with, but not caused by,

No one in these bins is a never-married single mother and so these bins perfectly identified our lefthand variable.

15

public housing construction. Results for these specifications are reported in panels B and C of Table 3. We find that interacting our education bins with race has a negligible effect on our results. The inclusion of region time trends reduces our ever-married point estimates by approximately 15 percent, but the coefficients remain statistically significant at the five (linear probability model) and ten (probit model) percent levels. The point estimates in our nevermarried single mother regressions are also similar to those in our original specifications.

PUBLIC HOUSING AND FERTILITY, BY MARRIAGE STATUS In the results above we find that exposure to public housing increases the propensity of marriage and decreases the probability that young women were never-married single mothers, although we find no effect on fertility overall. There are a couple of potential mechanical explanations for the negative relationship between public housing and never-married single motherhood. Because women must be both never-married and have children to be never-married single mothers, public housing may influence the likelihood of women belonging to this group simply by increasing the probability that women have ever been married. In this scenario, the probability that a never-married woman has children could remain unchanged, but the probability of any given woman being a never-married single mother would decline simply because there are fewer never-married women. On the other hand, public housing may influence the likelihood of women being never-married single mothers by influencing the propensity of never-married women having children. If public housing discouraged never-married women from having children, the probability of a woman becoming a never-married single mother could decline with no change in the propensity of marriage.

16

Our finding of a positive relationship between public housing and marriage suggests that the increase in the propensity of young women being ever-married is at least partially responsible for the decline in never-married mothers. However, it is unclear whether public housing influenced the fertility decisions of never-married women. To better understand this relationship, we regress our motherhood dummy variable on public housing exposure and our other controls again, this time separating the sample by ever-married status. Results are reported in Table 3. We find no evidence that public housing exposure is correlated with the fertility decisions of ever-married women. However, we find a negative and statistically significant relationship between public housing exposure and the likelihood of never-married women being mothers. A one percentage point increase in public housing intensity is correlated with a -0.5 percentage point decrease in the likelihood of motherhood in our linear probability model (statistically significant at the one percent level). Our probit results suggest that a similar increase in public housing intensity is correlated with a -0.2 percentage point decrease in the likelihood of motherhood, although this is not statistically significant. While our results suggest that public housing access is correlated with a decline in fertility rates among never-married women, this does not necessarily mean that public housing discouraged never-married women from having children. It is still possible that public housing had no effect on anyone’s decisions to have kids, but that public housing increased the likelihood that women with children got married.

17

PUBLIC HOUSING AND QUANTITY OF CHILDREN Results presented in Table 2 suggest that access to public housing had little to no effect on the probability that young women became mothers. However, it is possible that public housing influenced the number of children that women with children had, even if it did not influence the probability that they had children. For public housing tenants, public housing may have encouraged larger families as additional children may have increased the size of a family’s unit while having no effect on rent. This would have lowered the cost of additional children and increased the size of the housing subsidy. It is possible that public housing influenced the family size decisions of non-tenants as well. If public housing lowered the cost of decent rental housing in the private market, then other couples may have found it easier to have more children. This effect could be through public housing’s effect on age at marriage. For women who decided to get married before having kids, public housing may have encouraged them to have children at a younger age by lowering the age at first marriage. This may have caused women to have more children than they would have otherwise, or it may have influenced the age at which women had children without influencing their total fertility. To assess the relationship between public housing intensity and the number of children that young women had, we run similar regressions to those above, using fertility dummy variables as our left-hand variables. These variables indicate whether or not a woman had two or more kids, three or more kids, or four or more kids. We use these variables in separate regressions and our categories are not mutually exclusive, and so a mother with four children would have a value of one for each of the three indicator variables. Results are reported in Table 4. We include all of the young women in our sample in panel A, limit our sample to women with at least one child in panel B, and limit our sample to ever-married mothers in panel C.

18

When limiting our sample to women with children, we find that public housing is positively correlated with the likelihood that mothers have multiple children. While the results are not statistically significant, a one percentage point increase in public housing intensity is correlated with a 1.5 percentage point increase in the likelihood of a mother having two or more children in our linear probability results and a 1.7 percentage point increase in our probit results. The coefficients on our interaction term between %PublicHousing and Black are correlated with an additional 0.5 percentage point increase in the likelihood of multiple children in both models, but are also not statistically significant. This is consistent with public housing increasing the fertility decisions of young women on the intensive margin. The coefficients on public housing are smaller when we look at the relationship between public housing and the probability that women have three or more and four or more children, although they are generally positive. When we look at the relationship between public housing and the likelihood that women had four or more children, however, the coefficients on public housing intensity are statistically significant. The relationship between public housing and fertility changes for black women in these regressions, however, as the coefficient on the %PublicHousing x Black interaction term becomes negative.

CONCLUSION Family structure norms in the United States have changed dramatically since 1950. As young women (aged 16-24 years) in metropolitan areas made choices about whether to marry, have children, or remain single they were likely influenced by their economic opportunities, their educational attainment, the social norms in their area, and the incentives created by state and local government programs, including welfare and public housing. We utilize census data from

19

1950 and 1970 to access the relationship between access to public housing and family structure during this period. We control for observable individual- and MSA-level characteristics, as well as unobservable time-invariant differences across MSAs. Our results emphasize three particularly noteworthy findings. First, there is a positive association between public housing access and the probability that young non-black women chose to marry. This result is robust to a variety of specifications and is both statistically and economically significant. A one percent increase in public housing corresponds to nearly a two percentage point increase in the probability that an average young woman will be married. The second crucial finding is that virtually all connections between public housing and family structure differ substantially for blacks and non-blacks. The coefficients for black women are not always precisely estimated, but in general greater access to public access does not tend to increase rates of marriage or motherhood. Third, while public housing does not appear to influence the probability that young women have children, it does appear to be correlated with the number of children that women have. Public housing is positively correlated the likelihood that young mothers have multiple children. Given the data available, we cannot make strong claims about a causal relationship between public housing and family structure. Nor can we offer a definitive explanation for the diverging implications for public housing access between races. However, the persistence of a robust connection between public housing and marriage propensity for non-blacks suggests that it plays a meaningful part in the calculus of family structure decision-making.

20

REFERENCES

Akerlof, George A., and Janet L. Yellen. "An Analysis of Out-Of-Wedlock Births in the United States." Brookings Policy Brief Series, 1996. Aseltine, R. "Pathways Linking Paternal Divorce with Adolescent Depression." Journal of Health and Social Behavior 37 (1996): 133-48. Blau, Francine D., Lawrence M. Kahn, and Jane Waldfogel. “The Impact of Welfare Benefits on Single Motherhood and Headship of Young Women: Evidence from the Census.” The Journal of Human Resources 39, 2 (2004): 382-404. Coulibaly, Modibo, Rodney D. Green, and David James. Segregation in Federally Subsidized Low-Income Housing in the United States. Westport, CT: Praeger Publishers, 1998. Danziger, Sheldon, George Jakubson, Saul Schwartz, and Eugene Smolensky. “Work and Welfare as Determinants of Female Poverty and Household Headship.” The Quarterly Journal of Economics, 97, 3 (1982): 519-534. Dawson, Deborah A. "Family Structure and Children's Health and Well-Being: Data from the 1988 National Health Survey on Child Health." Journal of Marriage and Family 53.3 (1991): 573-84. JSTOR. Web. 11 July 2014. Fisher, Robert Moore. Twenty Years of Public Housing. New York: Harper, 1959. Flewelling, Robert L., and Karl E. Bauman. "Family Structure as a Predictor of Initial Substance Use and Sexual Intercourse in Early Adolescence." Journal of Marriage and Family 52.1 (1990): 171-81. Garfinkel, Irwin, and Sara McLanahan. Single Mothers and Their Children: A New American Dilemma. Washington, D.C.: Urban Institute, 1986. Haines, Michael R. Historical, Demographic, Economic, and Social Data: The United States, 1790-2000 [computer file]. ICPSR Study 2896. Ann Arbor, MI: Inter-university Consortium for Political and Social Research, 2010. Hetherington, E. M., and W. G. Clingempeel. "Coping with Marital Transitions: A Family Systems Perspective." Monographs of the Society for Research in Child Development 97.2/3 (1992): 1-238. Hoyne, Hilary Williamson. “Does welfare play any role in female headship decisions?” Journal of Public Economics 65 (1997): 89-117. McLanahan, Sara, and Gary D. Sandefur. Growing up with a Single Parent: What Hurts, What 21

Helps. Cambridge, MA: Harvard UP, 1994. Moehling, Carolyn M. "The American Welfare System and Family Structure." The Journal of Human Resources 42.1 (2007): 117-55. Moffitt, Robert. “Welfare Effects on Female Headship with Area Effects.” The Journal of Human Resources 29, 2 (1994): 621-636. National Commission on Urban Problems. Building the American City. New York: Frederick A. Praeger, 1969. Ruggles, Steven J., Trent Alexander, Katie Genadek, Ronald Goeken, Matthew B. Schroder, and Matthew Sobek. Integrated Public Use Microdata Series: Version 5.0 [Machine-readable database]. Minneapolis: University of Minnesota, 2010. Schultz, T. Paul. “Marital Status and Fertility in the United States: Welfare and Labor Market Effects.” The Journal of Human Resources 29, 2 (1994): 637-669. Schultz, T. Paul. “Eroding the economic foundations of marriage and fertility in the United States.” Structural Change and Economic Dynamics 9 (1998): 391-413. Silverman, Abner. “Basic Needs and Social Services.” In Papers Submitted to Subcommittee on Housing Panels on Housing Production, Housing Demand, and Developing a Suitable Living Environment. 579-606. Washington, DC: Committee on Banking and Currency, U.S. Housing of Representatives, 92nd Congress, 1971. Sinai, Todd and Joel Waldfogel. “Do Low-Income Housing Subsidies Increase the Occupied Housing Stock?” Journal of Public Economics 89, no. 11-12 (2005): 2137-64. United States Department of Health and Human Services. “Nonmarital Childbearing in the United States, 1940-1999.” National Vital Statistics Reports 48, 16 (2000). United States Department of Housing and Urban Development. Consolidated Development Directory. Washington, DC: GPO, 1973. Wood, Elizabeth. The Beautiful Beginnings The Failure to Learn Fifty Years of Public Housing in America. Washington, D.C.: National Center for Housing Management, 1982. Zill, Nicholas and James Peterson. "Marital Disruption, Parent-Child Relationships, and Behavior Problems in Children." Journal of Marriage and Family 48.2 (1986): 295-307.

22

Table 1: Summary Statistics Total

Non-black

Black

1950

1970

1950

1970

1950

1970

40,021

53,124

35,809

46,583

4,212

6,541

% ever-married

37.6

37.6

36.3

37.8

48.1

36.5

% mother

23.8

22.8

22.9

21.4

32.1

33.0

0.332

1.64

0.173

0.691

1.69

8.36

N

% mother and never married

23

Table 2 (1) Ever-married Panel A: Basic Specification LPM

(2) Mother

Probit

LPM

(3) Never-married mother

Probit

LPM

Probit

%PublicHousing

0.01943*** (0.00676)

0.01866** (0.00910)

0.00750 (0.00636)

0.00362 (0.00590)

-0.00302** (0.00127)

-0.00152 (0.00102)

%PublicHousing x Black

-0.02110***

-0.02787***

-0.01735**

-0.01094

0.00241

0.00128**

(0.00762)

(0.01077)

(0.00768)

(0.00672)

(0.00446)

(0.00055)

lnADC lnADC x black Black x educ Black x educ x 1970 Region x 1970

Yes Yes No No No

Yes Yes No No No

Yes Yes No No No

Yes Yes No No No

Yes Yes No No No

Yes Yes No No No

Observations

93,145

93,145

93,145

93,145

93,145

91,866

R-squared

0.36703

0.26065

Panel B: Basic Specification + Race x Education x Year Controls LPM Probit LPM

0.05051 Probit

LPM

Probit

%PublicHousing

0.01920*** (0.00675)

0.01795** (0.00912)

0.00730 (0.00636)

-0.00148 (0.00102)

-0.00295** (0.00126)

-0.00148 (0.00102)

%PublicHousing x Black

-0.01886**

-0.02079**

-0.01542**

0.00130**

0.00036

0.00130**

(0.00759)

(0.01016)

(0.00762)

(0.00054)

(0.00445)

(0.00054)

lnADC lnADC x black Black x educ Black x educ x 1970 Region x 1970 Observations R-squared

Yes Yes Yes Yes No

Yes Yes Yes Yes No

Yes Yes Yes Yes No

Yes Yes Yes Yes No

Yes Yes Yes Yes No

Yes Yes Yes Yes No

93,145

93,145

93,145

93,145

93,145

91,866

0.36703

0.26065

Panel C: Basic Specification + Region x Year Controls LPM Probit %PublicHousing %PublicHousing x Black lnADC lnADC x black Black x educ Black x educ x 1970 Region x 1970 Observations

LPM

0.05051 Probit

LPM

Probit

0.01604**

0.01558*

0.00396

0.00140

-0.00278**

-0.00157

(0.00677)

(0.00912)

(0.00641)

(0.00591)

(0.00127)

(0.00101)

-0.01914**

-0.02620**

-0.01515**

-0.00969

0.00227

0.00125**

(0.00762)

(0.01079)

(0.00768)

(0.00676)

(0.00447)

(0.00054)

Yes Yes No No Yes

Yes Yes No No Yes

Yes Yes No No Yes

Yes Yes No No Yes

Yes Yes No No Yes

Yes Yes No No Yes

93,145

93,145

93,145

93,145

93,145

91,866

0.36703 0.26065 0.05051 R-squared Notes: Robust standard errors are reported in parentheses. Marginal effects are reported for probits. All regressions

24

and probits contain controls for individual characteristics (age, education and race dummies and interactions of all dummies with year indicator variable) and MSA characteristics (average welfare benefits (different effects by race), male employment, and industrial shift index), as well as MSA and year fixed effects. Panel B adds race-byeducation and race-by-education-by-year controls. Panel C adds region-by-year controls. See text for more details. *** p