White Suburbanization and African-American Homeownership 1940-1980 Pt 2

Chapter 2 C. Predicted highway rays as an instrument for white departures OLS estimates will be biased if white location decisions are directly influenced by black homeownership or are correlated with unobserved characteristics of the city that also predict black homeownership. A correlation between white suburbanization and black homeownership could arise if, for example, the demand for black homeownership increases and the housing stock in black neighborhoods is not conducive to owner-occupancy, leading prospective black homeowners to move into white neighborhoods. White households concerned about living near black neighbors may then move to the suburbs, generating a positive relationship between black homeownership and white suburbanization and biasing the OLS estimates upward (that is, too large in absolute value). Alternatively, the coefficient may be biased towards zero by measurement error. To address endogeneity bias, we look for an instrumental variable that is correlated with the share of whites living in the suburban ring but is otherwise uncorrelated with the black homeownership rate. We instrument for the white suburban share using the predicted number of interstate highway rays built within one mile of the central city between 1950 and 1980 (PREDICTED_RAYS). The original plan for the Interstate Highway System was drafted in 1947 with the dual goals of serving national defense and inter-city commerce. Baum-Snow (2007) determined the total number of rays that were assigned to each central city in the 1947 plan. After the plan was established, local politicians could lobby the federal government to build extra highway miles through their city and were more likely to do so if there was a high demand for suburbanization in their area. Therefore, we predict the number of completed rays in each city i at time t by interacting the number of assigned rays in the 1947 withfinance the national share in certain neighborhoods began to break down. That being said, we barriers in plan mortgage and in residency find no relationship between white departures and black homeownership from 1970 to 1980. of highway construction completed by date t. Our assumption is that the national rate of highway construction is not influenced by the demand conditions in any one city.23 Our first stage regression relates the number of white households in the central city to the predicted number of highway rays passing through the city, controlling for metropolitan area and Census year fixed effects and the full set of population controls in equation 2: NUM_WHITE it = ρ(PREDICTED_RAYS)it + ΦXijt + αi + δt + εit (4) Baum-Snow (2007) demonstrates that this instrument is correlated with overall population loss from central cities; thus, not surprisingly, we find a strong first stage relationship between predicted highway rays and white departures. The fourth column of Table 4 (last row) presents our estimate of ρ from equation 4. The coefficient is negative and large; each new planned highway that we predict to be built through the central city leads to the departure of 8,000 white households from the central city (on a base of around 100,000). The F-statistic on predicted highway rays is 16.94, well above the conventional value of 10 used to judge an instrument’s strength. The fourth column of Table 4 (first row) presents the IV coefficient of white departures using predicted highway rays to instrument for the number of white households in a city. The IV coefficient is slightly larger (in absolute value) than its OLS counterpart, but the two estimates cannot be statistically distinguished from each other. According to the IV estimate, every 1,000 white household departures from the central city generates 108 new black homeowners. The typical city lost 25,700 white households from 1940 to 1980 and gained 10,400 black 23 Because construction of the interstate highways began in 1954, we set the number of highway rays in every city at zero in 1940. homeowners. By our IV estimate, white departures would have generated 2,700 new black homeowners (= -108 · -25.7), which can explain 26 percent of the growth in black homeowners (= 2,700/10,400); if instead we use the OLS estimate, white departures can explain 21 percent of the growth in black homeownership (= 2,200/10,400).24 The identifying assumption for the instrumental variables procedure is that highway construction is only related to black homeownership through its effect on white departures. Some scholars argue that interstate highways tended to be built through black neighborhoods, thereby reducing the stock of housing available to black households (Sevilla, 1971; Frieden and Sagalyn, 1989; Lewis, 1997). The evidence cited in these earlier studies, however, is largely anecdotal. In a recent paper, Collins and Shester (2013) show that urban renewal projects, similarly accused of clearing black neighborhoods, had no effect on a city’s black population share or its degree of racial residential segregation. In addition, if highways did reduce opportunities for black homeownership, we would expect the estimated effect of white departures in the IV specification to be smaller in absolute value than its OLS counterpart, yet we find the opposite. D. Household-level data on black homeownership, 1940 and 1980 The individual characteristics of black households that settle in cities that are losing white population may be systematically different from those of black households in growing areas. We address this concern in Table 5 by using the 1940 and 1980 micro data to control for householdlevel characteristics. In particular, we estimate the relationship between the probability of black 24 Over this period, around 50 percent of white households in central cities were homeowners, suggesting that the typical city lost roughly 13,000 white owner-occupiers (= 25,700 * 0.50). If every new black homeowner moved into a housing unit recently vacated by a white owner, then 80 percent of vacated units would be filled by new black homeowners (=10,400/13,000), leaving only 20 percent vacant or converted into a rental unit. If, instead, some portion of new black homeowners moved into newly constructed units in growing central cities, then perhaps up to 40 percent of the units vacated by whites remained vacant or were converted to rental occupancy. These patterns are consistent with the high transaction cost of converting a unit conducive to owner-occupancy into a rental unit, due both to structure type and to the reinforcing factors of lot size and zoning regulations. homeownership (equivalent to the black homeownership rate at the aggregate level) and the number of white households living in the central city from equation 3, controlling for other aspects of the city and metropolitan population. The first column of Table 5 estimates the main relationship in the micro data without adding household-level controls. In this case, we find that 1,000 white departures increase the black homeownership rate by 0.05 percentage points. For comparison with the aggregate data, column 2 presents a weighted regression, ensuring that each city, rather than each individual, is counted equally. The strength of the estimated relationship increases to 0.08 percentage points. Column 3 then includes our extensive set of household characteristics. Including these controls reduces the estimate by around 15 percent to 0.07 percentage points but we still find a large and statistically significant relationship between white departures and the black homeownership rate.25 The coefficients in the individual-level regressions are nearly identical to those derived from a regression of the aggregate black homeownership rate on the number of white households living in the central city (coeff. = -0.00079, s.e. = 0.00033).26 Finally, in column 4, we add the city-wide ratio of median home values to median rents. Adding this measure of housing price pressure cuts the responsiveness of the black homeownership rate to white departures in half, suggesting that a portion of the effect of suburbanization on black homeownership can be explained by the price mechanism outlined in the conceptual framework.27 25 Results are qualitatively similar if we instead use probit estimation (coeff. = -0.00086, s.e. = 0.00026). The implied response to white departures is smaller in regressions that use the homeownership rate as a dependent variable than in the main specification, which uses homeownership counts. The typical city in our sample had 42,000 black households in 1980. According to coefficients from the homeownership rate regressions, the loss of 1,000 white households from the central city would have increased the black homeownership rate by only 34 households (= 42,000 * 0.0008), which is only half as large as the point estimates in Table 4. This discrepancy can be explained by the fact that white departures had a greater effect on black homeownership in cities with a large black population (see Table 6),a pattern that generates larger point estimates in the ownership count specification. The two estimates would be reconciled in a city with around 100,000 black households (87 = 109,000 * 0.0008). 27 Prospective black homeowners may be more responsive to the value of below-median homes. Information on the distribution of housing values for each of our cities can be computed from the 1940 and 1980 IPUMS; in particular, 26 E. Heterogeneous effects of white departures by city type We find a sizeable relationship between black central city homeownership and white suburbanization nationwide over the period 1940 to 1980. This section explores variation in the strength of this relationship across metropolitan areas. Table 6 examines how the core relationship between the number of white households and the number of black homeowners in a city varies across different subsamples. In the first column, we interact the change in the number of white households with two features of the housing market: the share of units that are owner-occupied and the black share of all households (as a proxy for the black share of prospective homebuyers). We measure these characteristics in 1940, before the rise of either white suburbanization or black homeownership, to capture initial differences in housing markets across cities. We expect the relationship between white departures and black homeownership to be stronger in cities with a large initial stock of units conducive to owner-occupancy. If the housing stock is instead primarily made up of rental units, the effect of white departures will be attenuated. We find that cities with a larger initial stock of owner-occupied housing experience a sharper increase in black homeownership for every white departure. In 1940, the urban owneroccupied share ranged between 15 and 55 percent. This 40 percentage point difference in initial owner-occupancy is associated with an additional 147 black owners for every 1,000 white departures (= -367 · 0.4). The relationship between white departures and black homeownership may also depend on the racial composition of the pool of prospective homebuyers. If whites constitute the majority of we calculated the ratio of home values at the 25th percentile to median rents across cities in the 1940 and 1980. We find a very similar effect of each ratio on black homeownership in these years. In particular, the coefficient on the median-median ratio is -0.012 (s.e. = 0.005) and the coefficient on the 25th percentile-median ratio is -0.014 (s.e. = 0.006). These figures imply that a one standard deviation increase in the price-rent ratio would reduce black homeownership by 7.2 percentage points. prospective homebuyers, white sellers can easily find white buyers without needing to compromise on price. However, if many prospective homebuyers are black, sellers who chose to limit their market to white buyers would incur a large price penalty for doing so and, thus, sellers would more likely opt to sell to black buyers. Indeed, we find that white departures have a stronger effect on black homeownership in cities with a large initial black household share. In 1940, the black urban household share ranged between zero and 50 percent. This 50 percentage point difference in initial racial composition is associated with an additional 144 black owners for every 1,000 white departures (= -289 · 0.5). The final heterogeneous effect that we explore concerns the role of overall white population growth or decline.28 The arguments outlined in section 2 identify white suburbanization with literal departures of whites from the central city. This is an accurate description for many of the industrial cities of the Northeast and Midwest that attracted black inmigrants during the first (World War One) and second (World War Two) waves of the “Great Migration” from the rural South. However, in growing metropolitan areas in the South and the West, the white population – both city and suburban -- expanded through in-migration. The treatment effect we are interested in would arguably be strongest in cities that were losing white residents, rather than in cities that were otherwise expanding and adding new housing units. The third and fourth columns of Table 6 examine this hypothesis explicitly by splitting the sample into metropolitan areas that experienced net gains or losses in the white central city population from 1940 to 1980. Consistent with a simple model of housing supply with durable 28 Following up on footnote 14, we also explored the impact of variation across cities in the proportion of black households with income higher than the level at the 25th percentile among white owner-occupiers. In particular, we included an interaction term between this variable and the number of white households leaving the central city using the specification reported in Table 6. The coefficient on the interaction term is positive, meaning that white departures have a smaller impact on black home purchases in areas where black income is high; this finding is consistent with the idea that blacks in some areas faced credit constraints. However, the standard error is quite large and so we hesitate to interpret this finding (coeff. -= 44.16; st. err. = 148.56). housing, we find essentially no relationship between the number of white households in a city and black homeownership in areas that were gaining white population. For the typical growing city, the estimates suggest that 1,000 white departures were associated with 6 fewer black homeowners, a tiny and statistically insignificant relationship.29 In other words, white arrivals do not compete with existing or new black residents for owner-occupied housing, presumably because, in these growing cities, new units are being constructed to house the expanding population. In contrast, there is strong relationship between the number of white households and black homeownership in cities that were losing white population over our period. For the typical city with net white population loss, 1,000 white departures were associated with 222 additional black homeowners.30 As we have argued, as white households left these cities and the housing stock remained, a portion of the vacated units were sold to prospective black buyers. It is important to recognize that a sizeable majority of urban black households over our period – 75 percent in 1980 – lived in such cities and, thus, potentially, was “at risk” of experiencing the particular pathway into homeownership emphasized in this paper. . Among the 53 cities that lost white population from 1940 to 1980, the strength of the relationship between white departures and black homeownership varied according to initial housing market characteristics. Figure 1 uses the coefficients from the third column of Table 6 to predict the number of new black homeowners for every 1,000 white departures by city. 1,000 white departures would lead to only 50 new black homeowners in New York City, in which 15 29 The typical growing city had an initial housing stock that was 37 percent owner-occupied and was 10 percent black in 1940. 30 Similarly, the typical declining city had an initial housing stock that was 34 percent owner-occupied share was 10 percent black in 1940. percent of the housing stock was made up of owner-occupied units in 1940.31 On the other extreme, 1,000 white departures generated 450 new black homeowners in Birmingham, AL, a city with a high black population share (41 percent) and a larger share of owner-occupied units (29 percent) in 1940. Figure 2 combines the predicted strength of white suburbanization by city from the prior figure with the actual number of white departures over the 1960s. According to our estimates, 3,000 black households in the typical city became homeowners due to white departures during the 1960s. However, in a few cities over 10,000 new black homeowners were created by white suburbanization over this decade. One such city happens to be Chicago, the subject of Berry’s (1976) original study; Detroit and Oakland (CA) are other examples. These three cities were characterized by sizeable central city black populations during the period, a large number of white departures from the urban core, and a city housing stock initially conducive to owneroccupancy – prime examples of the effect documented in this paper. IV. Concluding Remarks The share of metropolitan white households living in the suburbs increased from approximately one-third in 1940 to two-thirds in 1980. Over the same period, the rate of homeownership among black households living in central cities more than doubled from 15 percent to 42 percent. This paper argues that there was a causal link between white suburbanization and rising black homeownership. As white households left central cities, the price of housing units conducive to owner-occupancy fell and many black households on the 31 Central cities exhibited tremendous variation in the share of housing units that were owner-occupied in 1940. The mean owner-occupied share in that year was 36 percent with a standard deviation of 8 percent. At the low end were old cities in the Northeast, including New York City, Boston and Hartford, CT. At the high end were small cities in the Midwest (Fort Wayne and South Bend, IN) and the West (Tacoma, WA). margin of renting and ownership chose to become homeowners. Although city neighborhoods were residentially segregated by race, black and white neighborhoods were interconnected through the urban housing market. We estimate that, on average, around 100 black households became homeowners for every 1,000 white households leaving the central city between 1940 and 1980. The effect of white suburbanization on black owner-occupancy in central cities was largely confined to the mid-twentieth century for three historical reasons. First, the “Great Migration” of African-Americans from the rural South to industrial cities in the North and West began during World War One and peaked in the decade of World War Two (Gregory, 2005). By mid-century, metropolitan areas like Detroit, Cleveland, Pittsburgh, and Chicago had substantial black populations living in their central cores. Secondly, the particular timing of changes in transportation technology and infrastructure facilitated mass white suburbanization only after World War Two. Had technical advances in the internal combustion engine and the development of America’s highway system occurred a half century earlier, the prospective home buyers interested in these centrally-located homes would likely have been immigrants from Europe. Thirdly, black households eventually gained access to the suburbs. Beginning in the 1970s, increasing numbers of middle-class urban black families in search of owner-occupied housing could (and did) access the suburbs directly, as Table 1 shows. To such families, a home in the suburbs was arguably a more desirable as a place to live and as an investment than remaining in the central city, particularly in cities that had experienced net outflows of whites between 1940 and 1980, where our estimates show that “treatment effect” of white suburbanization on black homeownership were largest historically. This paper has stressed the direct effects of white suburbanization on black homeownership. However, at the neighborhood level, higher rates of homeownership can have positive effects on social capital and may forestall urban decay (Dietz and Haurin, 2003; Rosenthal 2008). A useful extension therefore would be to investigate if the formation of socalled “bad ghettos” after World War Two (see Collins and Margo, 2000) was slower in central cities in which relatively larger numbers of black households became homeowners due to white suburbanization. Table 1: Race, Residential Location, and Owner-Occupancy, 1940-1980 1940 Metro population, Share in suburbs White Black 0.355 0.194 Black, Share owner-occupier Total City Suburbs 0.192 0.150 0.349 White, Share owner-occupier Total City Suburbs 0.414 0.347 0.522 1960 0.515 0.190 0.374 0.330 0.468 0.661 0.543 0.767 1980 0.682 0.276 0.462 0.421 0.572 0.702 0.561 0.770 2000 0.738 0.382 0.479 0.423 0.570 0.742 0.589 0.796 2010 0.742 0.466 0.457 0.384 0.541 0.720 0.563 0.774 Notes: Authors’ computations from IPUMS. Samples include all metropolitan households whose place of residence (city or suburb) was reported. Table 2: Structure type and homeownership, Central cities in 1960 Housing Type Detached Single Family White, Percentage in 43.6% White, Owner rate 82.9% Black, Percentage in 31.5% Black, Owner rate 63.6% Attached Single Family 8.6 55.1 15.1 35.4 Two Family 13.5 37.3 13.5 25.3 3+ housing units per building 34.3 8.0 39.9 5.6 Total 48.9% 31.1% Number HH in sample 110,236 19,417 Notes: Authors’ computations from 1960 IPUMS (www.ipums.umn.edu). Samples include all metropolitan households whose place of residence (city or suburb) was reported. Households in mobile homes, trailers, boats, etc. are excluded. Table 3: White departures and the relative price of owner-occupied housing, 1940-1980 White departures from city = Change in # of white households < 0 Dependent variables: Value, owner-occupied Value/rent ratio 1. # white HH in city (1000s) 74.007* 0.025* (28.351) (0.006) 2. # white own in city (1000s) 118.439* (55.802) 0.089* (0.012) # white rent in city (1000s) 40.431 (46.077) -0.024* (0.009) Mean of dependent variable 68,188 13.939 488 488 N Notes: Regressions follow the format of equation 2 in text. “Value” and “Rent” are median values for the central city. All regressions contain vector of population controls (the number of black households in the central city, the number of black households in the metropolitan area, and the number of white households in the metropolitan area), metropolitan area fixed effects, and state-by-year fixed effects. Standard errors are clustered by metropolitan area and reported in parentheses. Standard errors are clustered by metropolitan area and reported in parentheses. * = significant at the 5 percent level or better. Median value and rent are measured using the actual city boundaries rather than the constant 1950 city boundaries (see footnote 21). Table 4: White departures and black homeownership in the central city, 1940-80 Dependent variable = Number of black owner-occupier households in city White departures from city = Change in # of white households < 0 (1) (2) (3) (4) Method OLS OLS OLS IV # white HH in city (1000s) -86.658* -74.939* -107.595* (7.351) (9.199) (36.005) # white own in city (1000s) -137.56* (15.41) # white rent in city (1000s) -51.15* (11.93) Share whites in city Median family income -0.096 (0.061) Median value/rent ratio -154.552 (93.295) First stage: Highway rays -7.795* (1.893) 488 390 488 488 N Notes: Cells contain the coefficient of interest from equation 2 in text; all regressions contain vector of population controls (the number of black households in the central city, the number of black households in the metropolitan area, and the number of white households in the metropolitan area), metropolitan area fixed effects, and state-by-year fixed effects. Standard errors are clustered by metropolitan area and reported in parentheses. * = significant at the 5 percent level or better. The control variables in column 2 (median family income and price/rent ratio) are measured using the actual city boundaries, rather than the consistent 1950 boundaries (see footnote 21). The first stage coefficient in column 4 reports the coefficient on predicted highway rays from equation 4, the dependent variable of which is the number of white households in the central city (in 1,000s). Table 5: White departures and the probability of homeownership in a sample of black central city residents, 1940 and 1980 Dependent variable = Indicator equal to one if household is owner-occupier White departures from city = Change in # of white households < 0 (1) (2) (3) (4) # white HH in city (1000s) -0.00055* -0.00078* -0.00073* -0.00039+ (0.00015) (0.00021) (0.00021) (0.00022) Weighted N Y Y Y Individual controls N N Y Y Price/rent ratio N N N Y 45,137 45,137 45,118 45,788 N Notes: The sample of black households living in central cities is drawn from the 1940 and 1980 IPUMS. Cells contain the coefficient of interest from equation 3 in text; regressions contain a vector of population controls (the number of black households in the metropolitan area and the number of white households in the metropolitan area), metropolitan area fixed effects, and year fixed effects. Standard errors are clustered by metropolitan area and reported in parentheses. * = significant at the 5 percent level or better. + = significant at the 10 percent level or better. Column 1 weights each individual equally, while columns 2-4 weight each city equally. Individual controls in column 3 include a quadratic in the age of the household head, dummy variables for the head’s gender, marital status, and educational attainment, and the logarithm of household income. The price/rent ratio in column 4 divides the median housing value in the central city by the median annual contract rent. Table 6: Heterogeneity in the relationship between white suburbanization and black homeownership by city type Dependent variable = Number of black owner-occupier households in city White departures from city = Change in # of white households < 0 (1) (2) (3) Full sample of 45 cities that gained 53 cities that lost cities population, 1940-80 population, 1940-80 # white HH in city 25.440* -31.471 103.503* (18.075) (31.039) (14.553) # white HH * Share owner occ, 1940 -366.947* (60.006) 65.329 (82.329) -710.148* (97.895) # white HH * Share black, 1940 -289.105* (132.409) 128.970* (44.876) -836.531* (233.251) 488 224 264 N Notes: See notes for Table 4 for details on specification. Standard errors are clustered by metropolitan area and reported in parentheses. * = significant at the 5 percent level or better. The number of white households in the city is entered in 1,000s. The number of white households is interacted with the share of housing units in the central city that were owner-occupied in 1940 (mean = 0.357) and the share of the city population that was black in 1940 (mean = 0.105). Columns 2 and 3 subdivide the sample into cities that experienced a net gain/loss in white population between 1940 and 1980. Figure 1: Variation in the estimated number of new black homeowners for every 1,000 white departures by city 500 Birmingham, AL 450 400 Baltimore, MD 350 300 Detroit, MI 250 200 New York City, NY 150 100 50 0 Notes: Predicted number of new black homeowners per 1,000 white departures for 53 cities with net white population loss based on coefficients in Table 6, column 3. Figure 2: Variation in the estimated number of new black homeowners generated by white departures from the city, 1960-70 20000 Detroit, MI Oakland, CA 15000 Chicago, IL 10000 5000 New York City, NY 0 Canton, OH -5000 Notes: Number of new black homeowners generated from 1960 to 1970 in response to white departures. Predictions for 53 cities with net white population loss based on coefficients in Table 6, column 3.