Policy History

The federal public housing program began in the 1930s after decades of concern regarding the condition of the housing stock inhabited by low-income families. Proponents of public housing argued that slums led to high rates of disease, crime, fire, and delinquency, and that the market could not profitably provide better housing for the poor (Radford Reference Radford1996). According to this logic and in the presence of assumed externalities, the government was called upon to intervene. The Great Depression brought a significant expansion of federal activity, which allowed public housing to become a significant and entrenched federal policy. The first federally funded public housing was built under the New Deal when, between 1933 and 1937, the Public Works Administration (PWA) built 21,640 units in 36 metropolitan areas across the country (Coulibaly, Green, and James Reference Coulibaly, Green and James1998).Footnote ² This was followed by the Housing Acts of 1937 and 1949, among others, which expanded the program's geographic coverage and intensity.

The Housing Act of 1937 replaced the Housing Division of the PWA with the United States Housing Authority (USHA). Its goals were broad: “To provide financial assistance … for the elimination of unsafe and insanitary housing conditions, for the eradication of slums, for the provision of decent, safe and sanitary dwellings for families of low income, and for the reduction of unemployment and the stimulation of business activity…” (U.S. Housing and Home Finance Agency 1964). The act delegated the clearance and construction of projects to Local Housing Authorities (LHAs), and the USHA assisted LHAs by providing loans to cover up to 90 percent of the costs of constructing the public housing projects (Coulibaly, Green, and James Reference Coulibaly, Green and James1998).Footnote ³ The act also introduced the “equivalent elimination” requirement, which required that for every unit of public housing built, an unsafe or insanitary unit must be demolished or repaired (Fisher Reference Fisher1959). Between 1937 and 1949 a total of 160,000 units were built.Footnote ⁴

This act was followed by the Housing Act of 1949, which was the most far-reaching piece of legislation regarding public housing. Although the public housing framework remained virtually unaltered from the Housing Act of 1937 (the laws on loans remained the same), an additional 810,000 units of public housing were approved (Bingham Reference Bingham1975). The Housing Act of 1949 also weakened the equivalent elimination requirement in the original 1937 Act by requiring equivalent elimination only for urban projects that were not built on slum sites. Rural projects and urban projects built on slum sites were exempt (Fisher Reference Fisher1959).Footnote ⁵

By the late 1950s, some public housing projects, particularly those in large cities, had received intense criticism from both the program's original supporters and long-standing skeptics. Catherine Bauer, an early and active supporter of public housing in the 1930s, wrote an article in Architectural Forum entitled, “The Dreary Deadlock of Public Housing” in 1957, in which she stated that the poor architectural design of public housing projects made it obvious that each housed “the lowest income group.”Footnote ⁶ Additionally, the income limits caused a “…trend toward problem families as the permanent core of occupants (Bauer Reference Bauer1957).” In 1958 The New York Times writer, Harrison E. Salisbury, wrote about the failure of the New York City public housing system in The Shook-Up Generation, in which he accused public housing of institutionalizing slums. Salisbury described “…the broken windows, the missing light bulbs, the plaster cracking from the walls, the pilfered hardware, the cold, drafty corridors, (and) the doors on sagging hinges…” in the Fort Green project and claimed that public housing “…create(d) human cesspools worse than those of yesterday (p. 75).” Public housing received a great deal of criticism in other large cities as well. In 1965 Chicago Daily News published a series of articles that referred to the Robert R. Taylor homes as the “$70 Million Ghetto” (Friedman Reference Friedman1966). In St. Louis, the Pruitt-Igoe public housing projects became so dilapidated and crime-ridden that all eleven buildings were demolished between 1972 and 1976 (Meehan 1979).

The physical deterioration of public housing projects was widespread and due in part to the design of the public housing program. Price ceilings on the cost per room established in the Housing Acts of 1937 and 1949 put downward pressure on construction quality, which was exacerbated by a lack of maintenance.Footnote ⁷ The way that the fiscal arrangements were set up, the federal government paid for the capital costs of the public housing projects, but local housing authorities (LHAs) were responsible for all other expenses. LHAs received rental income from tenants and used this income to pay for utilities, maintenance, and repairs. In years in which a LHA had money left over, it could place the remaining funds in cash reserves. However, the total amount that a LHA could hold was limited to no more than 50 percent of one year's rent. Once it held its maximum in reserves, any additional revenue was required to go towards paying off the capital costs. This essentially prohibited LHAs from building up the funds necessary for major repairs, such as roof maintenance (Meehan 1979).Footnote ⁸

There was a decline in the “quality” of tenants during this period as well, due partially to changes in eligibility requirements. Under the Housing Act of 1937, potential tenants could make an income no more than four times the local fair market rent to become eligible for public housing and, once living in public housing, were evicted if their income increased to more than 25 percent over the eligible income limit (Meehan 1979). Rent was set to approximate operating costs, which implicitly put a lower bound on tenant income as tenants had to be able to cover operating costs (Weicher Reference Weicher1980). In the late 1940s and 1950s the federal housing agency forced local housing agencies to enforce these income limits, removing some of the better-off tenants (Von Hoffman Reference von Hoffman2000).Footnote ⁹ The Housing Act of 1949 also influenced the pool of potential public housing tenants by displacing very poor families through Urban Renewal and highway construction, and relocating them into public housing (Jones, Kaminsky, and Roanhouse Reference Jones, Kaminsky and Roanhouse1979).Footnote ¹⁰ In 1959 Congress gave LHAs the right to set their own income limits and rents. The majority of housing authorities set rent as a proportion of income for most tenants, but required that all tenants' rent cover operation costs. While this maintained the lower bound on tenant income that existed earlier (i.e., they must afford the operating costs), the escalations in rent with income blunted work incentives. High inflation in the 1960s caused operation costs to increase faster than tenant income and LHAs responded by increasing rents in an attempt to cover their costs. Congress reacted to the growing rents in 1969 by passing the Brooke Amendment, which limited rent to 25 percent of a tenant's income for tenants with earnings less than a project-specific threshold (Weicher Reference Weicher1980). The combination of these changes led to notable changes in the predominant character of the tenant population, from the temporarily unemployed and working class, to households on welfare, the otherwise homeless, and the disabled (Epp Reference Epp1996). Between 1950 and 1969 the median family income of public housing residents fell from 63.5 to 42.4 percent of the national median, the percent of nonwhite residents increased from 38 to 52 percent, and the number of single-parent families increased from 19 to 31 percent (Silverman Reference Silverman1971).Footnote ¹¹

As Bauer (Reference Bauer1957) suggested, there are several characteristics of the public housing program which may well have doomed its success from the very start. The poor location, the architectural design and quality of construction of the projects, the lack of available maintenance funds, and the decline of tenant standards made it difficult for the program to live up to the expectations of early supporters. On January 8, 1973, President Nixon announced that all housing programs would be suspended pending a thorough policy review (Orlebeke Reference Orlebeke2000). Congress subsequently passed the 1974 Housing and Community Development Act, Section 8 of which gave low-income families subsidies to pay the difference between the “fair market rent” (FMR) on a standard quality unit in their particular locality and 25 percent of their income (Weicher Reference Weicher1980). This movement from a unit- to a tenant-based subsidy marked a sea change in public housing policy. The public housing program was reactivated by Congress in 1976, however the program's relative importance in the provision of low-income housing began to decline.Footnote ¹²

Empirical Strategy and Basic Results

To assess the impact of public housing on communities, I start by running regressions of a variety of county-level economic outcomes (Y) on public housing intensity (PH), an extensive set of pre-program community characteristics (X), and state-fixed effects (Λ).

(1)

My main variables of interest, Y _{county, 1970}, are the log of median owner-occupied property values, the log of median family income, the percent of families with low income, and the log of population density.Footnote ¹⁸ The concentration of public housing, PH_{county ,1970}, is measured as the percentage of occupied dwelling units comprised of public housing (public housing units in 1970 / total occupied units in 1970 * 100). The pre-program control variables, X_{county ,1940}, are extensive and include housing stock characteristics (percent owner occupied, median persons per room in rental units, percent of units in good condition, percent of units with electricity, percent of units with water, log median value), population characteristics (percent urban, percent of high school graduates, percent of college graduates, log population density, percent black and percent black squared), economic characteristics (percent of the labor force employed in manufacturing, percent of the labor force employed in agriculture, the per capita value of World War II contracts between 1940 and 1945, the per capita value of war facilities projects between 1940 and 1945), and some social and political characteristics that could underpin differences in support for public housing programs (the percentage of votes for Roosevelt in 1940 and percentages of Baptists and Catholics in 1950).Footnote ¹⁹ State-fixed effects account for unobservable differences at the state level and standard errors are clustered by state.

The coefficient of interest, β, is identified from within-state variation in PH, adjusting for observable characteristics in 1940. Therefore, the estimated coefficient could be interpreted as a causal effect of public housing if, within-state, there are no omitted variables that are correlated with public housing intensity and that influence the outcome variables of interest. Public housing, of course, was not randomly distributed across counties, and so one should hesitate to assign a causal interpretation to the coefficient. Nonetheless, the rich set of pre-program control variables and the existence of idiosyncratic variation in local public housing intensity within states (due to local politics surrounding the issue) may allow some scope for uncovering public housing effects. Further investigation into the robustness of the results to omitted variable bias is discussed below and suggests that the relationship between public housing and 1970 community-level characteristics might well be causal.

FIGURE 1

THE DIFFUSION OF PUBLIC HOUSING, 1940–1970

Notes: Percent of Public Housing Units is defined as public housing units / total occupied units * 100 for each year.Sources: Public housing data are from the Consolidated Development Directory (HUD 1973). Total occupied units are from the Census of Housing (Haines 2004).

Estimation results of equation 1 are reported in Table 1, panel A, where each coefficient is the estimated \beta macr from a separate regression. Controlling for X_{county , 1940} and state-fixed effects, counties with relatively high levels of public housing in 1970 also had relatively high concentrations of low-income families, as well as lower median family income, median property values, and population density. A one percentage point increase in public housing concentration is associated with a 0.42 percentage point increase in families with less than $3,000 income (compared to an average level of 16.7 percent) and is statistically significant at the 1 percent level.Footnote ²⁰ The same increase in public housing intensity is also correlated with a 1.6 percent decrease in median property values, a 1.6 percent decrease in median family income, and a 3.6 percent decrease in population density (all at a 1 percent level of statistical significance). Results are similar when the sample is restricted to counties with a positive amount of public housing in 1970 (also reported in Table 1, panel A) and are robust to the exclusion of counties with the highest intensities (above the 95th percentile) of public housing.Footnote ²¹

Observables, Unobservables, and Causal Interpretation

The robustness of the results to SEA-fixed effects and to pre-program population trends suggest that public housing may indeed have had negative effects on outcomes in 1970. However, omitted variables might still confound the estimated correlation between public housing and outcomes. For example, a county declining in ways that are unobservable to an econometrician might experience a fall in income and an increase in demand for (and supply of) public housing. In this scenario, the cross-place variation of PH is not driven by quasi-random, idiosyncratic local factors, but rather by unobserved negative trends.

To assess how strong such unobserved factors would have to be for the true causal link between PH and Y to be zero, I use a procedure formulated by Joseph G. Altonji, Todd E. Elder, and Christopher R. Taber (2005). John Bellows and Edward Miguel (2009) adapt this technique for a linear framework, and I follow their structure below.Footnote ²⁵

TABLE 1

PUBLIC HOUSING AND COUNTY-LEVEL ECONOMIC OUTCOMES IN 1970

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Each coefficient is from a separate OLS regression of an economic outcome on public housing intensity and control variables, where public housing intensity is defined as (public housing units in 1970 / total occupied units in 1970 * 100). Pre-program controls included in panels A–C are: 1940 percent owner occupied housing, median persons per room in rental units, percent of units in good condition, percent of units with electricity, percent of units with water, log median value, percent urban, percent high school graduates, percent college graduates, log population density, percent black and percent black squared, percent of the labor force employed in manufacturing, and percent of the labor force employed in agriculture; the value of 1940–1945 World War II contracts and war facilities projects per capita; the percentage of votes for Roosevelt in 1940; and percentages of Baptists and Catholics in 1950. Panel C includes additional controls for the percent change in total population, black population, and number of dwelling units between 1900–1910, 1910–1920, 1920–1930, and 1930–1940. State fixed effects are included in panels A, C, and D. State economic area fixed effects are included in panel B. In all panels, standard errors are reported in parentheses and are clustered by state. Ratios reported in panel D are calculated as β_C / (β_NC − β_C), where β_C is the estimated coefficient of the percent of public housing units in 1970 in a regression with controls (panel A) and β_NC is the estimated coefficient of the percent of public housing units in 1970 in a regression with no controls (panel D).

Sources:

County population and housing data are from Haines (Reference Haines2004). Data on 1940 presidential election results are from Leip (Reference Leip2009). Public housing data are from the Consolidated Development Directory (HUD 1973).

The Altonji, Elder, and Taber approach centers on a comparison of coefficient estimates from regressions with and without controls for observables. In theory, if the variable of interest were essentially randomly distributed (i.e., there is no selection based on observable or unobservable characteristics), then the coefficient estimated with and without control variables for observables would be the same. In practice, therefore, one might be less concerned about unobservables if adding extensive controls for observables does not diminish the coefficient on the variable of interest. On the other hand, if the coefficient of interest is substantially diminished in magnitude when controls for observable characteristics are added, then one might be particularly concerned that the coefficient estimate would move even closer to zero if one could actually control for additional, unobservable characteristics.

To be more precise, assume that an outcome variable is a function of public housing intensity and an index of community characteristics that may be correlated with public housing. Also assume that the index of community characteristics is a linear function of observable characteristics (X) and unobservable characteristics. The probability limit of an estimated \beta macr is the sum of the true value of β, denoted β₀, and the omitted variable bias. In a regression without controls, (where \beta macrNC stands for “no controls”) is equal to β ₀ plus the total omitted variable bias (from both observable and unobservable characteristics). Similarly, in a regression with controls for observable characteristics X, the probability limit of \beta macr is equal to β ₀ plus the portion of the total omitted variable bias that is not controlled for by X. The difference between and \beta macr_NC and \beta macr_c is therefore equal to the portion of the total omitted variable bias that can be explained by X.

If the true effect of public housing intensity on outcomes is zero (β ₀ = 0), then the value of \beta macr_NC is simply equal to the total omitted variable bias, and the value of \beta macr_C is the portion of the omitted variable bias attributable to unobservables. Therefore, if the true value of β is zero, then the ratio of the bias explained by unobservables and the bias explained by X (later referred to as the “sensitivity ratio”) can be calculated as \beta_nc(\beta_nc - \beta_c) If the addition of controls diminishes the point estimate by less than half, the sensitivity ratio is greater one. Altonji, Elder, and Taber (Reference Altonji, Elder and Taber2005) argue that “…the ratio of selection on unobservables relative to selection on observables is likely to be less than one…(pp. 176–77),” implying that it is unlikely that the inclusion of unobservables would reduce the point estimate by a greater amount than the inclusion of observables. Therefore, if the addition of controls diminishes the point estimate by less than half (i.e., the sensitivity ratio is greater than one), then the true value of β is not likely to be zero. Footnote ²⁶

I use this technique to assess the direction and strength of selection on unobservables that would be necessary for the true effect of public housing to be equal to zero. Of course, this cannot rule out omitted variable bias in the point estimate, but it allows some appraisal of the plausibility that unobservables drive the basic results. I run regressions without controls for X (but with state-fixed effects), and estimate sensitivity ratios, comparing the estimates from regressions with (Table 1, panel A) and without controls. “No controls” regression results and estimated sensitivity ratios are reported in Table 1, panel D. Without controls for X, public housing intensity is positively correlated with 1970 log median property values, log median family income, and log population density and negatively correlated with the percent of families with low income within states. In fact, it is the addition of my extensive set of controls that causes the public housing coefficient to become negative (or positive in the case of the percent of low-income families) and statistically significant. This causes the sensitivity ratios to be negative. Results are similar using SEA-fixed effects, and suggest that, if anything, county-level selection into public housing on observables was apparently positive.Footnote ²⁷ If the addition of unobservable characteristics pushed the point estimate in the same direction, then the point estimates reported in panel A of Table 1 would be underestimates of the true effect.

However, it is possible that there is positive selection into the program based on observables (for example, urban areas are more likely to adopt public housing), but, conditional on these observables, selection on unobservables is negative. In this scenario, the panel A results would be overestimates. Even if this is the case, the sensitivity ratios indicate that the correlation between public housing and the index of unobservable characteristics would need to be 0.70–0.94 times the magnitude of the correlation between public housing and the index of observable characteristics and be of the opposite sign in order for the true effect of public housing on median property values, median income, and the percent of low-income families to be zero. While one cannot rule out the possibility of negative selection on unobservables, the extensive nature of my controls makes it unlikely that the addition of unobservables could completely reverse the public housing estimate.

As a further check, I run these regressions without pre-program controls (but with state-fixed effects) again, limiting my sample to counties with positive amounts of public housing in 1970. If, conditional on observables, there is negative selection on the extensive margin, then excluding counties that did not participate in the program should address this issue. Results are reported in Table 1, panel D. With the limited sample, public housing is positively correlated with the percent of low-income families and negatively correlated with median property values and median income in regressions with no controls. The addition of 1940 controls diminishes the public housing point estimates slightly, but the sensitivity ratios are large (3.1, 2.9, and 1.9 for log median property values, log median income, and the percent of low-income families, respectively). This suggests that, after controlling for selection on the extensive margin, the correlation between public housing and the index of unobservable characteristics would have to be two to three times the magnitude of the correlation between public housing and the index of observables in order for omitted variable bias to explain the entire effect.Footnote ²⁸

URBAN AND NONURBAN COUNTIES

Given that most of the criticism of public housing predominantly focuses on large cities, it is possible that the experiences of urban areas were different from those of nonurban areas. To assess these differences, I estimate equation 1 again, separating my sample by urban status. I define “urban” counties as counties with more than 25 percent urban population in 1940. Because my urban definition is broad, I also run the regressions with a more restricted sample of counties in a Standard Metropolitan Area (SMA) in 1950. Results are reported in Table 2, and suggest that public housing had negative effects in urban and nonurban counties. The point estimates for the log median property value and log median income regressions are similar for urban and nonurban counties (–0.014 and −0.017 for urban counties, compared to −0.015 and −0.017 for nonurban counties, respectively) when using the whole sample. However, the public housing effects appear to be larger for urban areas when the sample is limited to counties with public housing in 1970. For example, using the limited sample, the point estimates for the log median property value and log median income regressions are −0.024 and −0.025 for urban counties, compared to −0.0098 and −0.016 for nonurban counties. The effects of public housing are larger for counties located in SMAs in 1950 for all outcomes. For example, the point estimate of public housing in the log population density regression is −0.083 for counties in SMAs, compared to −0.024 for counties with more than 25 percent urban population. For log median income, the public housing estimate is −0.026 for counties in SMAs, compared to −0.017 for urban counties.

THE EFFECTS OF PUBLIC HOUSING OVER TIME

Effects of Public Housing in 1950 and 1960

Public housing appears to have had negative effects on adopting communities in 1970, but it is unclear whether these effects existed in earlier periods. It is possible that public housing, when relatively new and under the original rules of tenant selection, had no negative effects(or potentially positive effects) on local communities. However, given the disappointment many supporters of the program felt with public housing as early as the late 1950s, these negative effects may have been present earlier. To assess whether the negative relationship between public housing and community outcomes was in place prior to 1970, I estimate equation 1 for similar outcome measures in 1960 and 1950, using public housing intensity measures that are specific to those years. The set of 1940 control variables are the same as for the earlier analysis and state-fixed effects are included.

TABLE 2

PUBLIC HOUSING AND COUNTY-LEVEL ECONOMIC OUTCOMES IN 1970, BY URBAN STATUS

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Each coefficient is from a separate OLS regression of an economic outcome on public housing intensity and control variables, where public housing intensity is defined as (public housing units in 1970 / total occupied units in 1970 * 100). See Table 1 or the text for a list of independent variables. State fixed effects are included in every regression. “Urban” is defined as having more than 25 percent urban population in 1940. Standard errors are reported in parentheses and are clustered by state.

Sources:

County population and housing data are from Haines (Reference Haines2004). Data on 1940 presidential election results are from Leip (Reference Leip2009). Public housing data are from the Consolidated Development Directory (HUD 1973).

TABLE 3

PUBLIC HOUSING AND COUNTY-LEVEL ECONOMIC OUTCOMES IN 1950 AND 1960

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Each coefficient is from a separate OLS regression of an economic outcome on public housing intensity and control variables, where public housing intensity is defined as (public housing units in 1970(1960, 1950) / total occupied units in 1970(1960, 1950) * 100). See Table 1 or the text for a list of independent variables. State-fixed effects are included in every regression. Standard errors are reported in parentheses and are clustered by state.

Sources:

County population and housing data are from Haines (Reference Haines2004). Data on 1940 presidential election results are from Leip (Reference Leip2009). Public housing data are from the Consolidated Development Directory (HUD 1973).

Results are reported in Table 3, and suggest that public housing did not have negative effects on adopting communities before 1970. Public housing was not statistically significantly correlated with log median property values, log median income, the percent of families with less than $3,000 income, or log population density in 1960. Public housing was also not statistically significantly correlated with log median property values, the percent of families with low income, or log median income in 1950. The coefficient on public housing is statistically significant in the log population density regression for 1950, however the coefficient is positive.Footnote ²⁹

Several hypotheses might account for this pattern of results. One possibility is that there were changes in the types of counties adopting or expanding public housing over time. Counties with negative unobservable characteristics could have rapidly adopted or expanded public housing in the 1960s, causing public housing to be correlated with poor community outcomes. Of course, the variety of robustness checks presented above (using SEA-fixed effects, controlling for pre-program trends, and assessing the necessary relative size of the omitted variable bias) suggest that negative selection does not explain the negative public housing effects in 1970.Footnote ³⁰

Another possibility is that long-run changes in the characteristics of public housing are driving the results. For example, the type (high rise, low rise, scattered site) or quality of public housing built could have varied by decade of construction. If higher density or lower quality public housing was built in the 1960s, then public housing built during this decade may have led to larger negative effects than older, higher-quality, less densely populated public housing. On the other hand, the negative effects may have nothing to do with public housing built in the 1960s specifically, but could be due to the long-run deterioration of projects like Salisbury (Reference Salisbury1958) observed in New York City. Even though the process of decay started decades earlier in some places, it may have taken until the 1960s for local economies to adjust and for these effects to be detectible at the county level. Or, the effects of public housing may not be related to the physical characteristics of the units (such as age or style), but to issues related to poor management by local housing authorities and changing tenant characteristics (Weicher Reference Weicher1980).

Public Housing and Local Human Capital, 1950–1970

It is difficult to decipher the channels through which public housing negatively affected outcomes in 1970. However, one possibility is that public housing worsened economic outcomes by negatively affecting the educational composition of the population. To better understand the relationship between public housing and human capital, I first assess whether public housing is correlated with the education level of county residents between 1950 and 1970. I regress the percent of the population composed of high school graduates in each decade on public housing intensity in that decade, 1940 pre-program characteristics, and state-fixed effects (equation 1). I also run the 1960 and 1970 regressions again, adding a control for the percent of high school graduates in the previous decade.Footnote ³¹ Results are reported in Table 5.

I find that public housing is positively correlated with the percent of high school graduates in 1950 and 1960. A one percentage point increase in public housing intensity is correlated with a 0.25–0.32 percentage point increase in the percent of high school graduates in each of these two decades. By 1970, however, the relationship between public housing and the percent of high school graduates reversed. By this point, a one percentage point increase in public housing is correlated with a 0.21–0.31 percentage point decline in the percent of high school graduates.

Next, I assess whether public housing's negative correlation with economic outcomes in 1970 is accounted for by the negative correlation between public housing and the local population's educational attainment. I estimate equation 1 again for log median property values, log median income, the percent of low income families, and log population density, this time adding a control for the percent of high school graduates in 1970. The percent of high school graduates in 1970 is likely endogenous to public housing, and so the coefficient on PH should not be interpreted as a causal effect of public housing intensity when the 1970 control is included. However, observing how the public housing coefficient changes with the addition the endogenous control will shed light on whether the change in the educational composition of the population is driving the base results. Results are reported in Table 6.

TABLE 4

PUBLIC HOUSING AND COUNTY-LEVEL ECONOMIC OUTCOMES IN 1970, BY DECADE OF CONSTRUCTION

Notes:

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Public housing intensity built in each decade is defined as (# public housing units built in the 1930s (or 40s, 50s, 60s) / total occupied units in 1970 * 100). See Table 1 or the text for a list of independent variables. State-fixed effects are included in every regression. Standard errors are reported in parentheses and are clustered by state.

Sources:

County population and housing data are from Haines (Reference Haines2004). Data on 1940 presidential election results are from Leip (Reference Leip2009). Public housing data are from the Consolidated Development Directory (HUD 1973).

In these regressions, controlling for 1970 educational composition reduces the coefficient on public housing by between 24 and 40 percent. For log median property values, controlling for the educational composition of the 1970 population causes the coefficient on public housing to fall from −0.016 (Table 1, panel A) to −0.0098 (statistically significant at the 5 percent level), while the coefficient in the log median family income regression falls from −0.016 to −0.012 (statistically significant at the one percent level). The coefficient falls from 0.42 to 0.30 for the percent of families with low income, and falls from −0.036 to −0.027 for log population density (both statistically significant at the 1 percent level). This pattern suggests that part, but not all, of the estimated effects of public housing on the outcome variables of interest might work through effects on the educational attainment of the local population.

Public Housing and Selective Migration

There are several ways in which public housing may have negatively affected educational attainment. As discussed earlier, it is possible that public housing lowered human capital by creating perverse education and labor market incentives for local residents. Or, public housing may have influenced human capital through selective migration, either by encouraging low education residents to move into the locality, or by encouraging high education residents to move out. Depending on the mechanism, public housing may have decreased the percent of high school graduates either by increasing the population of high school dropouts, or by decreasing the population of high school graduates. If public housing created negative educational incentives for locals and/or encouraged the in-migration of low-human capital residents, then public housing would be correlated with an increase in the population of high school dropouts. On the other hand, if public housing encouraged the outmigration of high school graduates, it is possible for public housing to be negatively correlated with the percent of high school graduates, even if it is not correlated with an increase in the number of high school dropouts.

TABLE 5

PUBLIC HOUSING AND COUNTY-LEVEL EDUCATIONAL ATTAINMENT, 1950–1970

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Each coefficient is from a separate OLS regression of an economic outcome on public housing intensity and control variables, where public housing intensity is defined as (public housing units in 1970 / total occupied units in 1970 * 100). See Table 1 or the text for a list of independent variables. State-fixed effects are included in every regression. Standard errors are reported in parentheses and are clustered by state.

Sources:

County population and housing data are from Haines (Reference Haines2004). Data on 1940 presidential election results are from Leip (Reference Leip2009). Public housing data are from the Consolidated Development Directory (HUD 1973).

TABLE 6

PUBLIC HOUSING AND COUNTY-LEVEL ECONOMIC OUTCOMES IN 1970, CONTROLLING FOR THE PERCENT OF HIGH SCHOOL GRADUATES IN 1970

Notes:

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Each coefficient is from a separate OLS regression of an economic outcome on public housing intensity and control variables, where public housing intensity is defined as (public housing units in 1970 / total occupied units in 1970 * 100). See Table 1 or the text for a list of independent variables. State-fixed effects are included in every regression. Standard errors are reported in parentheses and are clustered by state.

Sources:

County population and housing data are from Haines (Reference Haines2004). Data on 1940 presidential election results are from Leip (Reference Leip2009). Public housing data are from the Consolidated Development Directory (HUD 1973).

To better understand this relationship between public housing and human capital, I assess whether public housing is correlated with a larger population of high school dropouts or a smaller population of high school graduates in 1970. I do this by running equation 1 again, using the log number of high school graduates and dropouts as my variables of interest. Results are reported in Table 7. The point estimates on public housing intensity in the regressions of the log number of high school graduates and the log number of high school dropouts in 1970 are −0.033 and −0.022, respectively (statistically significant at the 1 and 5 percent levels). This suggests that a one percentage point increase in public housing intensity is correlated with a 3.3 percent decline in the population of high school graduates, but only a 2.2 percent decline in the number of high school dropouts. When I run the regressions again, adding controls for the log number of high school graduates (dropouts) in 1960, the point estimates fall to −0.020 and −0.0082. However, both remain statistically significant and the coefficient for the number of high school graduates is more than twice the magnitude of the coefficient for the number of high school dropouts. This is consistent with public housing encouraging the out-migration of both high- and low-education residents, however the low-education residents appear to leave more slowly than those with a high school degree.

Public Housing and Rates of In-Migration

The results above indicate that places with high intensities of public housing experienced a net loss in the population of high school graduates and dropouts. However, it is still possible that public housing encouraged the selective in-migration of high school dropouts who wanted to take advantage of public housing, while simultaneously encouraging a larger outflow of others without a high school degree. To further assess the relationship between public housing and selective migration, I incorporate data on migration rates for 1960 and 1970. Data from the 1960 and 1970 Population Censuses (U.S. Bureau of the Census Reference Grigsby1963, 1973) and the 1970 micro-Census (Ruggles et al. Reference Ruggles, Alexander, Genadek, Goeken, Schroeder and Sobek2010) report whether individuals lived in the same county five years previously. I use these data to assess whether areas with high densities of public housing had higher or lower percentages of their populations composed of recent in-migrants in 1960 and 1970, and to assess whether these migration trends differ by educational attainment in 1970.

TABLE 7

PUBLIC HOUSING AND THE POPULATION OF HIGH SCHOOL GRADUATES AND DROPOUTS IN 1970

Notes:

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Each coefficient is from a separate OLS regression of an economic outcome on public housing intensity and control variables, where public housing intensity is defined as (public housing units in 1970 / total occupied units in 1970 * 100). See Table 1 or the text for a list of independent variables. State-fixed effects are included in every regression. Standard errors are reported in parentheses and are clustered by state.

Sources:

County population and housing data are from Haines (Reference Haines2004). Data on 1940 presidential election results are from Leip (Reference Leip2009). Public housing data are from the Consolidated Development Directory (HUD 1973).

I begin by assessing whether areas with high densities of public housing in 1960 or 1970 had higher or lower percentages of recent in-migrants in those years.Footnote ³² I regress the percent of recent in-migrants (the percent of the 1970 (1960) population who migrated into a county between 1965–1970 (1955–1960)) on 1940 pre-program characteristics and state-fixed effects (equation 1). I then run the regressions again, adding controls for log population density and the percent of high school graduates in the previous decade. Finally, I run the regressions a third time, controlling only for log population density and the percent of high school graduates in the previous decade, for comparison with future results. Results are reported in Table 8.

In 1960 public housing intensity is negatively correlated with the percent of recent migrants in all specifications, although none of the point estimates are statistically significant. In 1970 this negative relationship increases in magnitude and becomes statistically significant at the 1 percent level in all three specifications. Depending on the specification, a one percentage point increase in public housing intensity is correlated with a 0.34–0.62 percentage point decline in the percent of the population composed of recent migrants. This is not consistent with an in-migration story.

TABLE 8

PUBLIC HOUSING AND THE PERCENT OF RECENT MIGRANTS IN 1960 AND 1970, BY COUNTY

Notes:

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Each coefficient is from a separate OLS regression of an economic outcome on public housing intensity and control variables, where public housing intensity is defined as (public housing units in 1970 / total occupied units in 1970 * 100). See Table 1 or the text for a list of independent variables. State fixed effects are included in every regression. Standard errors are reported in parentheses and are clustered by state.

Sources:

County population and housing data are from Haines (Reference Haines2004). Data on 1940 presidential election results are from Leip (Reference Leip2009). Public housing data are from the Consolidated Development Directory (HUD 1973). Migration rates come from the Population Census of 1960 and 1970 (U.S. Bureau of the Census, Reference Grigsby1963 and 1973).

Public Housing and Rates of In-Migration, by Educational Attainment

Finally, I assess whether public housing had different migration “effects” on high and low human capital residents by incorporating data from the 1970 micro-Census (Ruggles et al. Reference Ruggles, Alexander, Genadek, Goeken, Schroeder and Sobek2010). The 1970 individual-level census data does not report an individual's county of residence, but it does report the “county-group” in which an individual resides.Footnote ³³ The census also reports whether an individual lived in a different county in 1965. Using these data, I divide each county-group's 1970 adult population (age 20+) into four groups based on educational attainment (high school graduate or dropout) and 1965 residence (same county (stayer) or different county (migrant)).

For comparison with the county sample, I begin by regressing the percent of a county-group's 1970 population (both high school graduates and dropouts) that lived in a different county in 1965 on 1970 county-group public housing intensity, 1960 county-group log population density and percent of high school graduates, and census division fixed effects.Footnote ³⁴

(2)

Next, I estimate equation 2 again, using the percent of the “stayer” population (those who lived in the same county in 1965 and 1970) composed of high school graduates and the percent of the migrant population composed of high school graduates in 1970 as my variables of interest. If public housing is negatively correlated with the educational attainment of stayers, then this would be consistent with public housing encouraging the out-migration of high human capital residents, or discouraging the educational attainment of local residents. If public housing is negatively correlated with the educational attainment of recent migrants, then this would be consistent with public housing encouraging the in-migration of low-human capital residents. Results are reported in Table 9.

I find that public housing intensity is negatively correlated with the percent of the 1970 population composed of recent migrants, which is consistent with the earlier county-level results. The estimated coefficient on public housing intensity is −0.34, although it is not statistically significant. I also find that public housing intensity is negatively correlated with the percent of stayers who are high school graduates in 1970. A one percentage point increase in public housing intensity is correlated with a 0.32 percentage point decrease in the percent of stayers who are high school graduates (statistically significant at the 1 percent level). However, public housing is not correlated with a less educated stock of recent migrants. In fact, public housing is correlated with a 1.77 percentage point increase in the percent of recent migrants with a high school degree. This is consistent with public housing encouraging the out-migration of high school graduates or discouraging educational attainment among locals, but not consistent with public housing encouraging the in-migration of low human capital residents.

TABLE 9

PUBLIC HOUSING AND THE PERCENT OF RECENT MIGRANTS BY EDUCATIONAL ATTAINMENT 1970, BY COUNTY-GROUP

Notes:

* = Significant at the 10 percent level.

** = Significant at the 5 percent level.

*** = Significant at the 1 percent level.

Notes:

Each coefficient is from a separate OLS regression of an economic outcome on public housing intensity and control variables, where public housing intensity is defined as (public housing units in 1970 / total occupied units in 1970 * 100). Census division fixed effects are included in each regression. Standard errors are reported in parentheses and are clustered by census division.

Sources:

County-group 1970 public housing intensity, 1960 log population density, and 1960 percent of high school graduates are calculated by aggregating county-level data (HUD 1973; Haines Reference Haines2004). The percent of the population who migrated between 1965 and 1970, the percent of stayers who are high school graduates, and the percent of migrants who are high school graduates are aggregated from individual-level 1970 Census data (Ruggles et al. Reference Ruggles, Alexander, Genadek, Goeken, Schroeder and Sobek2010).