Transcript
Residential Segregation, Discrimination, and African-American Theater Entry during Jim Crow∗ Ricard Gil† Justin Marion‡ November 2015
Abstract We examine the role of residential segregation and racial discrimination in determining the entry of movie theaters serving African-American customers in the 1950s. These theaters provided an alternative to the segregated theaters of the Jim Crow era. Consistent with preference externalities in racial and ethnic enclaves, we find that a greater degree of residential segregation leads to more African-American theater entry. Using estimates from a Bresnahan and Reiss model of theater entry, we find that this effect is due to higher variable profits in residentially segregated cities rather than lower fixed costs of entry. The effect of racial bias among whites is found to be complex. Using several measures of racial discrimination, we conclude that bias leading to a taste for segregation leads to greater entry, while more generally racial bias results in fewer theaters.
∗
We appreciate the helpful comments of Bruno Cassiman and seminar and conference participants at ISNIE 2015, IESE Strategy and Entrepreneurship Winter Mini-Conference, Universitat Pompeu Fabra, Tilburg University, Universidad de Oviedo, Universidad Publica de Navarra, UC Santa Cruz, the University of Utah, and the Carey Business School at Johns Hopkins University. The usual disclaimer applies. † Associate Professor, Carey Business School, Johns Hopkins University. Email:
[email protected]. ‡ Associate Professor, Economics Department, University of California, Santa Cruz. Email:
[email protected].
1
1
Introduction
Many cities and states were slow to dismantle the institutions of segregation. Prior to the Civil Rights Act of 1964, segregation in public accommodation persisted in many parts of the country. Access to business and culture was significantly affected both by Jim Crow laws mandating segregation along racial lines, as well as by the practices of business owners that either explicitly or implicitly excluded blacks. Despite the prominence of public accommodation segregation in African-American life during this time, its effects on access to business and culture are little studied in the economics literature, possibly due to data limitations from the era. Businesses catering to African-American customers often entered the marketplace to fill the void resulting from the lack of equal access. In this paper, we examine what factors enabled entry of such businesses, particularly focusing on the effects of residential segregation and racial discrimination. We utilize a comprehensive annual census of African-American movie theaters in the years immediately following World War II to understand the role of residential segregation and racial discrimination on theater entry. While often smaller and without the full range of modern amenities, theaters catering specifically to African-American customers provided an alternative to theaters for general audiences, where black patrons were frequently segregated into inferior seating, denied admission entirely, or otherwise made unwelcome. African-American theater entry may have therefore expanded consumption opportunities for black patrons facing public accommodation segregation. The theoretical effects of residential segregation and discrimination on African-American theater entry are not clear, and mirror closely the modern implications these variables have for minority enterprise. As in Waldfogel (2003), increasing returns arise in this context from the fixed costs of opening and operating a movie theater. Residential segregation may therefore result in preference externalities, increasing the number of African-American movie theaters. Furthermore, residential segregation could protect minority businesses from competition by white-owned firms, as suggested by Cutler and Glaeser (1997) and Glazer and Moynihan (1963). Similarly, racial bias among the white population could increase the viability of African-American theaters by lowering black utility from attending white theaters, as it would if seating were segregated. Working in the opposite direction, both residential segregation and racial bias reduce access to key inputs such as credit, adversely affecting entry. Though explicit public accommodation segregation was a phenomenon largely associated with the South, African-American theaters were prevalent in all regions, representing 4.8 percent of U.S. movie theaters in 1950. Rather than region, the primary determinant of theater
2
location was the size of the African-American population. Across metropolitan statistical areas (MSAs hereafter), once a sufficiently large black population was achieved to support one theater, the share of African-American theaters increased in proportion to black’s share of the total population. Our main finding is that residential segregation leads to greater theater entry. At the MSA level, a one standard deviation increase in the index of segregation across census tracts is associated with an increase in the African-American theater share of 1.2 percentage points, a percentage increase of 15.6 percent compared to the mean MSA. To alleviate concerns that unobserved variables may be correlated with both the number of theaters and the degree of segregation, we propose a new instrument for residential segregation using the number of road bridges spanning natural barriers such as rivers, streams, and lakes built before 1920. This instrument are in the spirit of approaches taken in the literature, and yield estimated coefficients somewhat larger than those obtained using OLS.1 The estimated effects of residential segregation extend to rural counties as well. Using the segregation of African-Americans across wards within a county, we find that greater rural segregation is also associated with more African-American theater entry. This holds despite the larger area of wards, which makes the segregation measure less informative. We also consider the effect of racial bias by whites on African-American theater entry. We form several measures of racial bias - support for Strom Thurmond in the 1948 presidential election and responses to attitudes regarding residential and schooling segregation and interracial marriage from the General Social Survey (GSS). We find that the number of votes for Thurmond in a city is associated with fewer African-American theaters, while having Thurmond on the ballot is associated with more theaters. From the GSS, support for the rights of whites to residentially segregate is associated with more theaters in the area, while opposition to interracial marriage and support for schooling segregation both have small and statistically insignificant impacts. Taken together, these results suggest that any increase in the number of AfricanAmerican theaters that results from racial bias likely operates via the preference of the white population for residential segregation, while more general racial bias, as indicated by greater support for Thurmond, leads to fewer theaters. Preference externalities resulting from residential segregation increase the number of theaters by increasing demand for African-American theaters, conditional on market size. To provide support for this interpretation of our results, we apply the methodology of Bresnahan and Reiss 1
Cutler and Glaeser (1997) use the number of rivers, the number of local governments, and the share of local revenue comprised by intergovernmental transfers. Ananat (2011) uses the location of railroad tracks.
3
(1990, 1991a,b) to estimate a model of theater entry. Bresnahan and Reiss conjecture that in markets with an observed number of entrants, further entry is unprofitable. By examining what market size is profitable for a single entrant, and then subsequently how further entry is induced by larger market sizes, it is possible to separately identify fixed costs from variable profits. If discrimination and residential segregation deny access to key inputs of production, we expect these variables to alter fixed costs. If these variables instead increase per-person variable profits, it suggests that movie theater demand is higher. Our estimates indicate that residential segregation leads to theater entry by increasing per-person variable profits, while its effect on fixed costs is small and statistically insignificant. Votes for Thurmond are estimated to reduce variable profits rather than fixed costs. This finding narrows the possible channels for racial bias to impact theater entry. The paper contributes to the literature along two main dimensions. First, it uses new data to provide detail regarding an important historical economic institution. Second, it provides new evidence regarding a mechanism identified by the literature through which segregation and discrimination can affect minority well-being. The prior literature on the effects of residential segregation points toward conflicting forces on minority outcomes.2 Segregation has a negative impact since it may limit access to quality jobs, mentors and peers, and public goods. On the other hand, segregation could lead to a mixing of incomes within the racial groups. If there are spillovers across skill groups, this would benefit low-income blacks. Also, residential segregation may protect black-owned enterprises locating within the racial enclave allowing for greater entrepreneurship. According to the spatial mismatch hypothesis of Kain (1968), this could improve employment opportunities. The estimated effects of the impact of residential segregation in the empirical literature is similarly mixed across a wide range of settings and empirical approaches, as noted by Cutler, Glaeser, and Vigdor (2008). Related to consumption, Waldfogel (2008) examines the location of fast food restaurants. He documents differences in restaurant preferences by race and finds that residential location patterns by race are correlated with restaurant location. This confers a benefit to residential sorting. Cutler and Glaeser (1997) find that living in a segregated city leads to lower incomes, reduced schooling, increased idleness, and increased incidence of single motherhood for blacks compared to whites. Similarly, Ananat (2011) also finds adverse effects on rates of black poverty. On the other hand, Collins and Margo (2000) find that segregation may not have had negative implications prior to the 1970s. Cutler, Glaeser, and Vigdor (2008) 2
See Cutler and Glaeser (1997) and Borjas (1995) among others.
4
find that residential concentration has mixed effects for first-generation immigrants, but with a positive effect for the most highly educated. Borjas (1995) suggests that the socioeconomic background of residents in a child’s neighborhood may be responsible for apparent ethnic human capital spillovers. Fairchild (2008) studies segregation and entrepreneurship specifically, finding that the clustering of minority residents is associated with greater rates of self-employment, though greater interactions between races (a reduction in “exposure” segregation) also increases the likelihood of business ownership. Finally, Fischer and Massey (2000) also find that residential segregation reduces the likelihood of minority entrepreneurship. We are not able to observe the practices of general theaters with respect to segregated seating by race or the outright exclusion of African-American customers, though such practices may impact theater entry by driving black customers to African-American theaters. These practices could be related to either the discriminatory views of theater owners or their customers. Research by Holzer and Ihlanfeldt (1998) suggests that racial bias by a firm’s customers affects the hiring of minority workers by the firm. Theater owners may wish to exclude minority customers due to their own racial bias. An important question is the extent to which competition constrains supplier’s ability to do so. Li (2014) studies the commercial sex market in Singapore, finding that prostitutes charge different prices to customers based on race, though competition ameliorates this effect. On the other hand, Graddy (1995) examines discrimination in the Fulton fish market, finding that even in this highly competitive environment pricing according to race persists. Charles and Guryan (2008), studying labor market discrimination, point out that discrimination is only relevant for the marginal firm, so that discrimination can persist for the average firm even in the face of competition. The paper is structured as follows. Section 2 reviews the evolution of segregation and discrimination laws in the US from the early 20th century to 1950. This section also describes the US movie industry during those years. In Section 3, we describe the data used in this paper. Section 4 presents our reduced form results, while in Section 5 we introduce our structural model of entry a la Bresnahan and Reiss and describe our results. Finally, Section 6 concludes.
5
2
Background and Institutional Details
2.1
Segregation and Jim Crow laws
Segregation in public accommodation was an important feature of African-American life for much of the 19th and 20th century. While de jure segregation was particularly prevalent in Southern states, segregation was used in practice over much of the U.S. In the years following Reconstruction, even in the South integrated public accommodation in some types of facilities was common. For instance, Woodward (1974) describes integrated bars, soda fountains, and public transportation in Charleston, South Carolina, though this did not extend to facilities such as hotels and hospitals. In the late 19th century, segregationist practices became more widespread and often legislated. The landmark 1896 Supreme Court decision in Plessy v. Ferguson established the constitutionality of separate-but-equal accommodations for different races. There was a substantial degree of variation in segregation-related laws over time and across jurisdictions. However, this variation is unlikely to be useful in academic research studying the effects of segregation, as laws mandating segregation tended to be piecemeal and local.3 Many Northern states passed laws at various points in time banning segregation in public accommodation. However, these laws were often ignored, imperfectly enforced, or interpreted in such a way that allowed for the continued exclusion of blacks. Segregation in public accommodation was not only a Southern phenomenon. While Jim Crow laws mandating the segregation of races in public accommodation were more common in the South, informal and to a lesser degree formal segregation was routinely practiced outside of the South, as recounted by Woodward (1974).4 Despite several notable judicial victories for the civil rights movement, the practice of segregation remained entrenched in many areas of the country in the post-World War II years. In 1954, the Supreme Court in Brown v. Board of Education ruled unconstitutional the practice of segregation in public schools, though the practice continued both in education and public accommodation for several more years. The Civil Rights Act of 1964 marked the end of the segregation era. 3
For instance, one local statute may enact segregation in public transportation, while another may be later enacted that covers hospitals. An example cited by Woodward (1974) was a Birmingham law making illegal mixed race games of dominoes or checkers. 4 At the turn of the 20th century, Northern visitors to the South were often struck by the degree of mixing of races. In the years between the world wars, Ku Klux Klan membership rose dramatically, a phenomenon experienced outside the South to a greater extent. Also, as black residents of the south moved north during the great migration, their destination was often to residentially segregated urban areas in Northern cities.
6
2.2
Movie theaters
Segregation was still a relevant feature of movie theaters during the time of our data, the early 1950s. Movie theaters for white audiences often either completely barred admission to black customers, or would offer worse seating to only a portion of the screenings. While segregation was often met with resistance from civil rights groups such as the NAACP in some cities, the practice was still common, and often extended to barring entertainment with black performers.5 Movie theaters specifically targeting African-American audiences were differentiated both vertically and horizontally from theaters for white audiences. African-American theaters were often smaller and lower quality. It was rarer, for instance, for an African-American theater to be air conditioned, and in the early years of cinema it was less likely for an African-American theater to have sound.6 To some extent African-American theaters were also horizontally differentiated from white theaters. While the majority of films screened by African American theaters were movies shown in white theaters, African American theaters could also target movies with all-black casts, or movies featuring black protagonists. Prior to the 1950s several independent companies produced “race films” with partially or entirely black casts, and in the 1950s, Hollywood began producing movies with African-American protagonists. This could be particularly relevant since white theaters in some cases refused to show movies with mixed casts. Similarly, anecdotal evidence indicates that African-American theaters often refused to show movies depicting African-Americans in a negative light.7 Also, not surprisingly, Africa-American theaters were often located in neighborhoods with greater concentrations of black residents. Finally, theaters of the era often showed live music or plays, which provided a further source of horizontal differentiation between white and black theaters, and also among African-American theaters. 5
Even in major cities with significant shares of African-American population, segregation was practiced by theaters well into the 1950’s. For instance, Headley (1999, 2006) documents segregation lasting until at least 1953 in Washington, D.C. and around the same time or later in Baltimore. 6 Some of the African-American theaters located in the suburbs of DC used to be older reconverted white theaters that projected second-run movies and therefore vertically differentiated with neighboring general theaters. Further evidence of vertical differentiation comes from the construction cost of the theaters. Headley (2006) lists construction costs of theaters in Baltimore. Many were built in teens, twenties, and thirties. The Regent theater was built in 1915 for a cost of $10 thousand, and was remodeled in 1920 for $50 thousand, with an $18 thousand organ added in 1922. The Ritz, originally a white theater that was later converted to an African-American theater, cost between $75 and $100 thousand. Another theater was converted from a store at a cost of $5000. 7 As an example, The Regent Theater in Baltimore canceled showings of Belle Starr in 1949, and earlier had refused to show the Song of the South. (Headley, 2006)
7
3
Data
3.1
Movie theater location
We use an original data set comprised of movie theater information from yearly issues of the Movie Yearbook between 1945 and 1955. This Yearbook published annually a de facto census of theaters in the US as well as a directory of US theatrical firms with 4 or more theaters. Importantly for our purposes, the Movie Yearbook also contained a listing of all theaters by city and state that were designated for African-American customers. Because many of the theaters from the data no longer exist or were located in cities or towns that are no longer independent municipalities, we complemented the data with information from www.cinematreasures.com. The information on this site allowed us to find the approximate location of theaters and check whether changes in theater name that may have occurred during the sample period.
3.2
County and MSA characteristics
We combine theater location with county-level information from the 1950 and 1960 decennial U.S. censuses. We utilize demographic information on median income and education, and employment information including labor force participation, unemployment, and agricultural employment. Since theaters located in less densely populated areas are likely to enjoy lower profits per resident due to travel cost, we also utilize information on county area in square miles. To measure the access that potential entrants had to capital, we use the dollar volume held in 1950 in savings and loan association accounts located in the county. The introduction of television represents competition for movie theaters, and its importance likely differs for theaters catering to black customers. To measure the importance of television, we use county level data on the household TV penetration from Gentzkow (2006). For counties in metropolitan areas, we aggregate county-level data to the MSA level based on 1950 MSA definitions. For count variables, such as population and area, we sum across counties within the MSA. For measures such as median county income and education, we take a population weighted average of counties comprising the MSA.
3.3
Segregation
In our empirical analysis, a key variable of interest is the degree of residential segregation by race. We form measures at the MSA level as well as for counties located outside of MSAs.
8
Within MSA, we take census tract level data on the size of the black and white populations, and similar to Cutler and Glaeser (1997) calculate a dissimilarity index for MSA i of the form
si =
∑
1 bj wj N | − | 2 b wi i j=1
(1)
where bj is the size of the black population in census tract j and wj is the size of the white population. Due to data constraints, segregation can be calculated for only 56 MSAs in 1950. Hence, we also calculate a separate segregation measure using 1960 census population data, which allow us to observe residential segregation for 144 MSAs. For counties outside of MSAs, census tract-level population is not available, and we instead use population by ward. We observe 923 rural counties with more than one ward, and for these counties we are able to form a residential segregation measure using the same method as in equation 1. Of the counties with more than one ward, both the median and modal county has five wards, however the number of wards varies substantially across counties. Eight counties have only two wards, while 47 counties have ten or more. This leads to two sources of measurement error. First, the distribution of the population across a coarse geography understates the degree or residential segregation, since it masks segregation within ward. We anticipate that this will lead to attenuation of the estimated effect of segregation, and it will result in our instruments for segregation to be less informative. Second, more densely populated counties will have more wards, which will require controls for population and county area.
3.4
Instruments for segregation
Residential segregation may be correlated with unobserved factors that also influence theater location.8 We address this issue using instrumental variables. An approach taken by previous papers in the literature is to use the presence of barriers such as streams or railroads in a MSA, reasoning that these barriers form subareas across which the population will segment. We form a new instrument in this spirit. The National Bridge Inventory is a comprehensive list of roadway bridges in the U.S. This includes all structures carrying traffic across obstructions or depressions, such as water, railways, or other roads. These data report, among other things, the feature intersected and the year constructed. From these data, we sum the number of bridges crossing waterways and railroads by county and by whether the bridge was built prior 8 A second and less likely problem is the potential for endogenous residential sorting. Our preferred measure of residential segregation is from 1960, five years after the final year of our theater location data in 1955. If AfricanAmerican theaters represent a neighborhood amenity for black residents, this could lead to reverse causality.
9
to 1920.9 When applicable, we aggregate to the MSA level. We exclude from these counts bridges crossing roads. It is possible that infrastructure investment is endogenously related to neighborhood outcomes, so we also will control for bridges built in the years immediately prior to 1960 as well as those built after 1960. For each MSA, and for each non-MSA county, we then have one instrument, the number of bridges built prior to 1960. The presence of a waterway or railroad represents a constraint on the movement of population. Constructing a bridge relaxes this constraint. An area where a bridge will be built suggests that a binding constraint exists that has not yet been alleviated. Therefore, we expect that where bridges have already been built, we should see less segregation. By counting bridges built prior to 1920, we obtain a measure that precedes the Great Migration, the movement of African-Americans out of the rural South. The number of bridges provides two advantages over a simple count of rivers. First, bridges are less likely to be built across obstructions that do not constrain the population. A river that passes through a less populated area will not be associated with more density. Second, this instrument can be formed at any level of geography, since the location of the bridge is precisely indicated in the inventory.
3.5
Discrimination measures
Our empirical work places two requirements on the measures of discrimination. First, they must be timely, in that they reflect racial views held by whites during the 1950s. Second, they must contain sufficient geographic variation to correlate with theater location. We utilize multiple measures of the degree of racial bias in the white population that meet these requirements to varying degrees. One measure we employ is the number of votes received by Strom Thurmond in the 1948 presidential election, normalized by the size of the white population. This variable has several advantages. It meets the timeliness criteria, and it varies substantially across counties. It faces two main disadvantages. First, Thurmond was a segregationist, and indeed this was the defining component of his platform. A taste for segregation among the local population could have distinct affects on black business formation from racial discrimination, and if so this measure may have some difficulty separately identifying the effect of the latter. Second, Thurmond was not on the ballot in 32 states, and there is little variation in this measure outside the South. 9 We used the 1993 bridge inventory. This dataset has 664,123 bridge observations in the US. We scan the intersected feature for the terms (or their abbreviations) railroad, creek, river, branch, lake, stream, tributary, bayou, and run.
10
A second set of measures come from the 1972 and 1976 waves of the General Social Survey (GSS), where respondents were asked a series of questions regarding racial attitudes. The survey solicited respondents’ views on interracial marriage, segregation along racial lines in schools, and residential segregation.10 The scale of the potential responses varies across questions. Following Charles and Guryan (2008), we form an index of racial bias at the individual level by averaging the standardized responses to these questions. We then average this index across all individuals in the subregion. While these questions more directly measure racial bias, and potentially allow for distinguishing taste for segregation from racial discrimination, they are not ideal on the timeliness or geographic variation criteria. The GSS was initiated in 1972, which is well after the last year of our data on African-American theaters. This leads to measurement error if racial attitudes change over time, and the change over time in measured bias could respond endogenously to conditions in the 1950s. Furthermore, the finest level of geographic detail permitted by the GSS is the subregion level, of which there are nine in the US.11
3.6
Summary statistics
The location of African-American theaters was naturally driven by the spatial distribution of the African-American population. A substantial number of states had no African-American theaters, and many more had only one such theater statewide. These states were not simply parts of the country with a history of integration. Rather, these were exclusively the areas with few African-American residents. Of the nineteen states with the lowest black population share, fourteen had no African-American theaters, while the other five had only one. In total, fifteen states had no African American theaters, with the only one of these states not ranking in the bottom twenty of the black population share being Nevada. In Figures 1 and 2 we present map of U.S. counties, where counties are categorized by the number of African-American theaters and theaters per thousand black residents, respectively.12 In many markets in the country, no African-American theaters entered. Entry was concentrated in Southern counties where the black population was higher, and in other black population 10
The wording of the questions are as follows: “White people have a right to keep African-Americans out of their neighborhoods if they want to, and (Negroes/Blacks/African-Americans) should respect that right.” 1= strongly agree, 2=agree slightly, 3=disagree slightly 4=strongly disagree. “Do you think white students and Black students should go to the same schools or to separate schools?” 1=same 2=separate. ‘Do you think there should be laws against marriages between African-Americans and whites?” 1=yes, 2=no. 11 These are New England, the Middle Atlantic, East North Central, West North Central, South Atlantic, East South Central, West South Central, Mountain and Pacific. 12 The theater directory is at the city level, and some cities are spread over multiple counties. In these cases, we calculate the number of theaters per black resident for the entire MSA and assign this value for all counties covered by the MSA.
11
centers such as Chicago, St. Louis, and New York. Some areas, such as Chicago, have both a large number of theaters and also a relatively large African-American population, making the number of theaters per thousand black residents measure not stand out. Those counties with the largest values for this measure are Southern counties with smaller black populations, as is the case for several counties in Texas. Figures 3 and 4 provide evidence of the empirical relationship between theater entry and the black population at the MSA level. Figure 3 plots the share of black theaters against the share of black population at the MSA level for both Southern and non-Southern metropolitan areas. There is a linear relationship between the black theater and black population share. Furthermore, while African-American theaters are less prevalent outside the South, it is apparent from the figure that there is substantial overlap in the South and non-South distribution of the black theater share conditional on black population share. Much of the difference in black theater presence in the South is therefore simply due to the greater black share of the population. To better visualize the relationship between black population and African-American theaters in cities with a small black population, Figure 4 plots the relationship between the number of black theaters and the log of the total black population in 1950. Virtually no theater entry occurs for MSAs with less than three thousand black residents. Metro areas with less than 20,000 black residents rarely see more than one African-American theater. This fact is consistent with the model of entry that will be estimated, where a minimum market size is required for entry due to fixed costs and further entry is subsequently less profitable due to competition. Also, estimation of the entry model requires sufficient variation in the number of theaters on both the extensive and intensive margins, and this figure suggests that this is true in our setting. Next we provide summary statistics for both MSAs as well as non-MSA counties. We start with Table 1 where we provide summary statistics of all MSAs in our sample separately by whether they are located in the South. While the average MSA had 3.5 black theaters in 1950, the Southern MSAs had an average of 4.6 and non-South MSAs 2.9. The share of black theaters was also nearly five times as large in the South (17 percent versus 2.6 percent). The black population was overwhelmingly located in the South in 1950. Blacks comprised 21.3 percent of the average Southern MSA and 4.5 percent of the average non-Southern MSA. The average nonSouthern MSA, however, experience much faster growth in the 1950s in black population. The average value for the residential segregation index was 0.746 across all MSAs. This indicates that for the average MSA, nearly 75 percent of the black population would need to move census tracts to achieve an even population distribution. Despite the much larger level of black population
12
in the South, the degree of residential segregation experienced by blacks in the Southern MSAs was similar to the rest of the U.S. Strom Thurmond received 0.013 votes in the 1948 presidential election per white resident nationwide, though this was almost entirely a Southern phenomenon, partly as a result of Thurmond failing to make the ballot in most states outside the South.13 Television penetration was growing rapidly during this time. In the average MSA, 8.4 percent of households owned a television in 1950, which had grown to 84 percent in 1955. Throughout the period, TV ownership rates were higher outside the South. Among the other demographic variables, the South was characterized in 1950 by lower education attainment levels and lower levels of income. Table 2 provides the same set of summary statistics for non-MSA counties. The comparison between South and non-South non-MSA counties is not that different from that in Table 1. Southern counties have a larger number and share of black theaters than non-South counties, larger number and share of Thurmond votes per white person, larger black population share, lower black population growth, lower TV penetration, lower educational attaintment and median income and larger employment share than non-South counties. In Table 3 we present the distribution of the number of African-American theaters for both MSAs and non-MSA counties. The sample is restricted to observations for which no variables used in the structural estimation are missing. This information is important to understand the variation used to identify the model of theater entry, which requires variation in theaters across markets on both the extensive and intensive margin. Panel A of 3 shows the African-American theater distribution across MSAs. Of the 140 MSAs used in the structural estimation, 54 have no theaters, three of which are Southern MSAs while 51 are outside the South. In fact, the median non-South MSA has zero theaters. Importantly for the estimation of the model of theater entry, we observe a reasonable number of MSAs with zero, one, two, and three or more theaters. Examining the non-MSA county distribution, we see that it is far more common to observe no African-American theaters. Panel B shows that of the 914 rural counties, 786 have no theaters. As with MSAs this is particularly pronounced outside the South, where only five of 611 rural counties have one or more theaters while 40.4 percent of Southern counties have at least one black theater. For comparison, we also show in panel C the distribution of theaters for whites or general 13 Thurmond’s share of the vote was 8.6 percent in the average MSA in our data, and 22 percent in the average Southern MSA. We do not use the vote share since the race of the voter is not observed, and as a result the support received by Thurmond among the white population would be understated by Thurmond’s vote share in areas with many black voters.
13
audiences. All non-MSA counties have at least one theater, and all but 18 of the 914 counties have more than one. Consequently, the level of variation we are able to consider in this paper will not lend itself to estimating a model of theater entry for white theaters.
4
Reduced Form Results
In this section, we describe the reduced form relationship between the share of African-American theaters and our primary variables of interest, the index of residential segregation and measures of racial discrimination. The structural analysis that follows will identify the channels by which these variables operate, specifically the extent to which the variables are associated with market size, variable profits per person, or fixed costs. We begin by estimating the reduced form relationship
yi = α + β1 si + β2 di + BXi + ϵi
(2)
where yi is the share of black theaters in MSA or county i, depending on the geographic level of analysis. The primary variables of interest are the measure of segregation, si , and proxies for racial discrimination, di . The vector Xi contains covariates such as the share and growth of the black population, total population, MSA size, median income, share of employment, TV penetration rate and regional dummies.
4.1
OLS at the MSA level
The results of estimating equation (2) for African-American theaters is shown in Table 4. We first describe the results related to residential segregation, and then the estimated effect of racial bias. Across specifications, we see that the index of residential segregation is positively related to the African-American theater share. In the specifications shown in columns (1) and (2), we use the segregation dissimilarity index for 1950 to measure residential segregation within an MSA, while columns (3)-(7) employ the segregation index from 1960. Because the 1950 census data only allow for the residential segregation index to be calculated for 56 out of the 143 existing MSAs in our data, the specifications employing the 1960 measure are preferred. In the specification shown in column (3), which includes only a sparse set of covariates, we estimate a statistically significant coefficient on the 1960 residential segregation index of 0.11. This estimate is little changed as further covariates are included. In the column (4) specification, we include the size of the metro area, the income and education level, the employment
14
composition, the degree of household television penetration, and the rate of growth in the black population during the 1950s. The inclusion of these additional covariates has no effect on our point estimate of the effect of residential segregation. The inclusion of more detailed region dummies also has very little effect on the estimated effect of residential segregation, as shown in column (5). To give a sense of the magnitude of the estimated coefficient, consider a one standard deviation increase in the segregation index. This would be associated with an increase of 0.012 in the black theater share, a 15 percent increase for the average MSA. While residential segregation is positively correlated with the location of African-American theaters, we find mixed results across our measures of racial discrimination. First, the index of racial bias enters positively in the majority of specifications that we present. In our preferred specifications shown in columns (3) and (4), we estimate that a greater degree of racial bias is correlated with a greater degree of African-American theaters.14 A one standard deviation in the racial bias index is associated with a 0.11 increase in the black theater share. This could reflect two possibilities. Racial bias could drive black movie-goers to African-American theaters. Alternatively, the components of the racial bias index could reflect a taste for segregation among the white population, thereby easing the entry of African-American theaters. We will later describe the estimated effects of the individual components of this index, which will shed further light on this question. As we describe in the data description of Section 3, the racial bias index only varies at the subregion level, so may not be ideal in a specification of MSA-level African-American theaters. It is also measured in the 1972 and 1976 waves of the GSS, so is not as timely as would be ideal. We therefore also include in the regressions the support for Strom Thurmond in the 1948 presidential election, normalized by the size of the white population. This measure has the advantage of richer variation. We see consistent evidence that MSAs where Thurmond received greater support see fewer African-American theaters. This is true even if one controls for whether the state in which the MSA is located had Thurmond on the ballot. Thurmond ran on a platform of segregation, which would potentially exert a positive influence on black theater entry. That our estimated effect is negative points toward racial bias having a negative effect on African-American theaters. The South was undoubtedly an extreme environment for race relations, and it is therefore conceivable that the effects of our variables of interest differ between Southern and non-Southern 14
Since the index of racial bias is measured at the sub-region level, the controls for sub-region dummy shown in column (5) capture most of the variation in this variable, hence the large increase in standard error. The identification in this specification arises entirely from cities straddling borders between sub-region, where the racial bias index is a population weighted average of the sub-regions comprising the MSA.
15
MSAs. The specifications shown in columns (6) and (7) address this possibility by estimating equation (2) separately for the two areas. We see that the effects of both residential segregation and the racial bias index are concentrated in Southern states. The point estimate of the residential segregation measure is 0.24 in Southern states, compared to a statistically insignificant estimate of -0.033 in the rest of the U.S. Similarly, estimated effect of the index of racial bias is 1.02 in the South, compared to 0.033 outside the South.15 Finally, the black theater share is more responsive to the size of the black population outside the South. Each 10 percentage point increase in the black population share increases the black theater share by 0.052 in the South compared to 0.076 outside the South. We now investigate further the components of the racial discrimination index. The racial bias index is comprised of respondents’ answers to three questions related to residential segregation, interracial marriage, and segregated schooling. These different components may reflect views that have differing effects on African-American theaters. In Table 5, we present specifications that allow for these three responses to enter separately to uncover which of these components drives our results in Table 4. For comparison, we repeat the baseline results in column (1). The specifications of columns (2)-(4) allow each measure to enter separately, while in column (5) we include all three. Our results show that the impact of the index of racial bias is all driven by respondents’ views on residential segregation. This variable is positive and statistically significant in both column (2) when included alone and in column (5) when included together with the variables measuring views on interracial marriage and schooling segregation. It is worth noting that each specification shown in Table 5 includes controls for residential segregation, so that the estimated effect of variation in views regarding residential segregation do not reflect realized variation in this form of segregation. Rather, it likely reflects more fundamental racial bias or a taste for segregation in public accommodation more broadly.
4.2
MSA-level IV results
The empirical literature studying the effects of residential segregation has grappled with the problem of omitted variable bias, where an unmeasured attribute of an MSA leads to more residential segregation while also being correlated with the characteristics of the African-American population. The concern for most studies is that these unmeasured characteristics are associated with negative outcomes for blacks, since generally a negative correlation is exhibited between residential segregation and measures such as wages and education. In our setting, these un15
There is insufficient variation in the Thurmond variables outside the South to include it as a regressor in the non-South estimation.
16
observed city attributes must be positively correlated with both residential segregation and African-American theaters. As with many of the empirical studies in this literature, we take an instrumental variables approach to alleviate this concern. The instrument we use is the number of roadway bridges spanning natural obstacles such as rivers, streams, and lakes. We take the the number of bridges built in the city prior to 1920, normalized by the number of streams within the MSA. We hypothesize that bridges already constructed alleviate segregation, while those yet to be built reflect unmet demand for integrating areas on either side of a natural barrier. We discuss these instruments further in Section 3. To provide support for the validity of pre-1920 bridges as an instrument, we examine its correlation with population characteristics in 1920. In so doing, we can evaluate whether bridges were built in a manner uncorrelated with unobserved local economic or social conditions that may later affect African-American theater entry.16 We consider six different variables: log population density, manufacturing employment as a percent of the population, the non-white population share, the literacy rate for individuals greater than ten years of age, the share of the population urban, and the dissimilarity index of residential segregation in 1920.17 The results are presented in Table 6. We find that in each instance, the estimated coefficient is small in magnitude and statistically insignificant. To better understand the coefficient magnitudes, consider a one standard deviation change in the bridges-per-stream variable (a change of 0.69). The impact this has on the left-hand side variables is never larger than 1.6 percent of the mean of the variable.18 Having provided support for the validity of the instrument, we show the results of the instrumental variables estimation in Table 7. Panel A of this table reports the IV estimates of the effect of residential segregation on the black movie share, while Panel B reports the first-stage. As we expected, the number of bridges built prior to 1920 is negatively correlated with residential segregation. This relationship is robust to including further controls that might contaminate the instrument, such as population growth during the intervening years and more contemporaneous measures of bridge building, which may be correlated with both 1960 segregation and pre-1920 bridges. The IV estimate of the coefficient on the segregation dissimilarity index is 0.26 and 16
Ananat (2011), in evaluating historical placement of railroads as an instrument, examines this variable’s correlation with a similar set of city characteristics. 17 We use the segregation data for 1920 from Cutler, Glaeser, and Vigdor (1999). 18 Specifically, the estimated effects of a one standard deviation change in pre-1920 bridges per stream on 1920 population density, manufacturing employment as a percent of the population, non-white population share, literacy rate, share of the population urban, and residential segregation are respectively 0.016 (p=0.44), 0.013 (p=0.77), -0.014 (p=0.55), -0.001 (p=0.42), -0.011 (p=0.64), and 0.013 (p=0.66) stated as a fraction of the variable’s mean.
17
is remarkably stable across specifications. This estimate is around twice as high as the OLS estimates presented in in Table 4. If omitted variables are correlated with both residential segregation and the location of African-American theaters, the IV results suggest that they weaken the effects estimated in the OLS specifications.
4.3
Results for non-MSA counties
We next provide evidence regarding the relationship between residential segregation, racial bias, and African-American theater entry for counties outside of metro areas. Examining rural counties could be instructive for several reasons. Racial views could have different effects in these areas. The lower population density may lead to fewer day-to-day interactions with individuals of other races. On the other hand, the historical economic ties between the agriculture sector and the African-American population, particularly in the South due to the legacy of slavery, may complicate race relations. Furthermore, the entry decision in rural areas is likely to differ from urban areas, since fixed costs are lower in less dense areas yet the relevant market is more difficult to reach. In Table 8, we present the results of estimating for rural counties a similar set of regressions as that shown for MSAs.19 Our results in columns (1) to (3) all show that residential segregation and black population share are positively and statistically significantly correlated with the share of black theaters in non-MSA counties. As expected, the magnitude of the estimated coefficient on residential segregation is smaller than that obtained for metro areas. As with the MSA specifications, we find that the estimated effect of residential segregation is concentrated in the South. In columns (6) and (7), we show estimates from separate specifications for Southern and non-Southern counties. The estimated coefficient on the dissimilarity index variable is 0.093 and statistically significant. In non-Southern counties, the coefficient is -0.0014, and is precisely estimated enough to reject an effect more positive than 0.0014. Racial bias is not a statistically significant determinant of theater entry in non-MSA counties. The estimated coefficients on our measures of discrimination, including both the index of racial bias and the votes received by Thurmond, are smaller than in the specifications at the MSA level and are not statistically significant. The other difference worth noting is the effect of the black population share. The coefficient on this variable is attenuated substantially compared to that estimated at the MSA level. In the full specification in that model, the black population 19 Because we are only able to generate segregation dissimilarity index values for non-MSA counties that are large enough to contain more than one ward, our full sample with all controls include 866 observations compared to 2305 when we exclude the residential segregation measure from the estimation. To see how the other estimated coefficients are affected by this sample restriction, we also present estimates excluding the dissimilarity index in column (4).
18
share coefficient is 0.59 compared with 0.28 in the corresponding regression for non-MSA areas. The most likely explanation is that the black population is less dense in the larger geographic area, making travel costs to theaters higher.
5
Empirical Model of Entry
To quantify the impact of segregation and discrimination on firm entry, we employ the entry model of a static game with complete information by Bresnahan and Reiss (1990, 1991a,b) (BR hereafter). We estimate this model separately for metropolitan areas and counties outside of metropolitan areas. BR show that the combination of observed market structure and a reducedform profit function with an entry game is informative about firms’ profitability. As in BR, we model African-American movie theaters in the 1950s as a homogeneous-goods industry with identical potential entrants.20 Suppose we observe a discrete number of firms, N , in a given market, m, with the parameterized profit function for a firm in this market given by ΠN,m = VN (Xm ; α, β)S(Ym ; λ) − FN (Wm ; γ) + ϵm ,
where VN is a function describing the variable profits per consumer in a market with N entrants, S is the market size function capturing the number of customers in the market, and FN is the entrant’s fixed costs of entry. The first term, VN ∗ S, represents the total variable profits for a firm. The parameters α, β, λ, and γ are to be estimated, and Xm , Ym , and Wm are the marketlevel demand shifters, market-size shifters, and cost shifters. The term ϵm is a zero-mean iid normally distributed error term assumed to capture the factors that affect the profits and are unobserved by the econometrician. Because the variance of the error term is not separately identified from the scale of the parameters, we normalize the variance to one. We model the market size function S(Ym ; λ) as
S(Ym ; λ) = black popm + λ1 black pop growthm + λ2 neigh black popm .
To normalize the market size in units of current town population, we set the coefficient of black pop to one and let VN contain a constant term, which is α1 in the following Eq.(3). Black pop growth stands for growth in levels of the black population between 1950 and 1960 20
The entry literature has extended this baseline BR framework in several dimensions. Berry (1992) introduces firm heterogeneity and Mazzeo (2002) and Seim (2006) do so with product differentiation. We choose the original BR methodology because we do not observe theater-level attributes affecting profitability.
19
and neigh black pop is the black population in the neighboring counties. The variable profits per consumer in the market, VN (Xm ; α, β), is given as
VN = α1 + Xm β −
N ∑
αn ,
(3)
n=2
where α1 + Xm β stands for monopolist profits and αn is the degree to which variable profits decrease with the number of entrants. The Xm is a vector of variables that affect profits per customer. We include the segregation measure from 1960, the number of Thurmond votes per white resident, the number of white theaters in 1950 normalized by area in square miles, the share of households with television in 1950, log median income, log population density, and an indicator for the South. We hypothesize that white theaters represent competition for AfricanAmerican theaters, and may reduce variable profits. Population density, which will also enter into the fixed cost equation due to its likely relationship to land values, appears in the variable profit equation since the travel cost of the average movie theater customer is likely to be lower. Finally, we specify fixed costs FN (Wm ; γ) as
FN = γ + Wm γL ,
(4)
The variables we include in Wm consist of the segregation index, the number of votes for Thurmond per white resident, a South indicator, the savings and loan capital per capita, and the log of the population density. The S&L capital variable is meant to proxy for the access of capital, while population density in this equation reflects land costs. The two primary variables of interest are the measure of segregation and the number of votes received by Strom Thurmond, which will appear in both the variable profits and fixed costs equations. Doing so allows for a more informed interpretation regarding their effects on movie theater location. If preference externalities drive the effects of residential segregation, then residential segregation will affect variable profits rather than fixed costs. On the other hand, if loans are more difficult to secure in racially segregated cities or areas with stronger racial bias, then these variables will show up in the fixed cost equation. The BR model relies on an equilibrium condition: if we observe N theaters in market m, it must be that in equilibrium ΠN ≥ 0 and ΠN +1 < 0 for market m. For instance, the probability of observing markets with no firms equals Pr(Π1 < 0) = 1 − Φ(Π1 ),
20
where Φ(.) is the cumulative normal distribution function and Π1 = Π1 + ϵm .21 Assuming average profits decrease with firm entry in equilibrium (Π1 ≥ Π2 ≥ Π3 ≥ ...), the probability of observing N in equilibrium is Pr(ΠN ≥ 0 and ΠN +1 < 0) = Φ(ΠN ) − Φ(ΠN +1 ).
By assuming ϵm is an iid draw across markets, an ordered probit yields the model parameter estimates.22 BR defines the “entry threshold,” the minimum market size required to support the exact N firms, as SN =
FN , VN
which we obtain by equating ΠN to zero and solving for S. Essentially, this ratio establishes that the market size (number of consumers) necessary to meet the break-even point at which N firms are present in the market is directly proportional to the size of the fixed cost and inversely proportional to the magnitude of the variable profit per consumer. SN can increase for a given number of stations either due to a fall in variable profits per customer (VN ) or an increase in the fixed costs (FN ). We estimate the population entry thresholds by N ∑ γ b1 + W γc γ bn L+ ˆ FN n=2 ˆ SN = = , N ∑ VˆN α b1 + X βb − α bn n=2
where the bar over each variable stands for the sample mean of the variable. The estimated per-station entry thresholds are sˆN = SˆN /N . The separate identification of fixed costs from variable profits is based on two strong assumptions. First, we assume the Xm variables we choose only impact variable profits but not fixed costs, whereas the Wm variables affect only fixed costs. In reality, however, some variables may affect both, and in our specifications several variables are common to both Xm and Wm . The second assumption is the functional form of the profit function: the determinants of variable-profits enter the profit function as an interaction with the market size, whereas the fixed-costs determinants Wm enter the profit function in such a way that those determinants We estimate the profits of markets with three or more stations in a market by setting Pr(Π3 ≥ 0) = Φ(Π3 ). We assume each market is isolated both in demand and costs so we can treat each observation as an equilibrium outcome from the game. This approximation may not be appropriate in some local markets. Given the nature of the contents and movies by African-American theaters, however, we believe the market overlap is not likely to be an issue for the data. 21
22
21
will impact the fixed costs regardless of the population in the market. Although we believe the functional form represents a good approximation of the profit function, this form is to some extent an arbitrary one, and we cannot rule out the possibility that the model is misspecified.
5.1
Results
Table 9 shows the estimated structural parameters that correspond to the profit model above using data at the MSA level.23,24 This table contains three columns where we add more variables in the variable profit and fixed cost components as we move from column (1) to column (3). Of primary interest is the estimated effect of residential segregation, which we allow to enter both the variable and fixed cost equations. We see that residential segregation exerts a statistically significant effect on variable profit, having an effect on fixed costs not statistically distinguishable from zero. This provides support for the conclusion that residential segregation leads to preference externalities across consumers. Furthermore, it is not the case that residential segregation is simply correlated with an unobserved attribute of MSAs that eases the entry of African-American theaters. A second variable of interest is the votes received by Strom Thurmond in the 1948 presidential election. In the sparse specification shown in column (1), we estimate that metro areas with greater support for Thurmond see lower variable profits for black theaters. In the specifications with a full set of covariates shown in columns (2) and (3), this coefficient estimate is no longer significant, though this is due more to slightly less precision rather than a substantial attenuation of the estimated effect. Interestingly, the point estimate of the fixed cost effect of Thurmond support is also negative, though statistically insignificant. Together, these results suggest that the estimated reduced form effect of Thurmond support that we documented in Tables 4 and 8 is more likely due a variable profit effect rather than denying access to inputs needed for startup such as capital or land. This would be the case if areas with greater racial bias adversely impact variable cost of production such as wage rates, or reduce African-American demand for movies. Another notable determinant of theater profits is the concentration of white theaters in the MSA. A greater number of white theaters per square mile reduces the variable profits of African-American theaters, potentially through competition. Population density, as expected, 23
We repeat the same exercise in Appendix Table A2 separating our full sample into South and non-South MSAs, and we show the resulting entry thresholds in Appendix Table A3. Column (1) reproduces column (3) of Table 9 to ease the comparison with prior results. This exercise is made tenuous by the fact that only three Southern MSAs have no theaters, and there is little variation on the extensive margin to estimate an entry model. We present these results for completeness to complement the reduced form results that estimate separate by region. 24 In both the MSA and county estimations, we restrict attention to areas with fewer than 100,000 black residents.
22
is associated with both higher variable profits and lower fixed costs of entry. This is likely due to lower travel cost of movie theater customers and the higher cost of land in dense areas. The estimated competitive effects, α1 , α2 and α3 , imply that entry of each additional black movie theater will reduces the per-firm variable profits. This pattern suggests that variable profits becomes smaller with N, as we would expect. In Table 10, we present the analogous estimates of the entry model at the county level. In contrast to the MSA estimates, future changes in the black population enter positively and significantly in the market size equation. In contrast to the estimated entry model at the MSA level, the primary variable of interest, the dissimilarity index, does not enter significantly here in the variable profit equation. One other notable difference is that the number of white theaters enters positively in the variable profits equation. This could reflect that there are unobserved taste for movies that facilitates entry by both white and African-American theaters.
5.2
Entry thresholds
With the estimated structural parameters above, we can calculate the components of the entry threshold equation SN =
FN VN
for each value of the number of entrants N . The entry threshold SN
describes the market size, in our case the African-American population in thousands, required to support N African-American theaters. Similarly, we can also find at the per-firm entry thresholds, which are obtained by dividing the market size thresholds, SN , by the number of firms: sN =
SN N
. We assume that movie theaters are homogenous, which suggests that the
per-firm entry threshold should increase as the number of firms rises since greater competition drives down the per-person variable profits that a theater can earn. It is possible that this assumption is violated, and movie theaters are differentiated, which could actually result in the estimated per-firm thresholds to decrease in the number of theaters. We first describe the entry thresholds implied by the entry model estimated at the MSAlevel. Table 11 displays these thresholds for each specification in Table 9. We observe two clear patterns. First, the population thresholds in the non-South MSAs are always larger than in the South MSAs. Second, per-firm thresholds increase much faster in the non-South MSAs than in the South MSAs. While a black population of 11,931 is required for a monopoly to break even in a non-South MSA, a black population of 8180 is required in Southern MSAs. However, the per-firm threshold outside the South increases rapidly in the number of entrants and is in excess of 30,000 for three or more theaters compared to approximately 9,200 in the South. These two facts combined indicate that competition among black theaters in the Southern MSAs was softer
23
than competition among black theaters located elsewhere. This can potentially be explained by the fact that Southern MSAs had larger shares of black population and also that anecdotal evidence points out that white theaters were a poor substitute for black theaters in the South but less so elsewhere. We next provide the analogous results at the non-MSA level. In Table 12, we present the market size thresholds and per-firm thresholds derived from the parameter estimates presented in Table 10. We show the estimated entry thresholds for both Southern and non-Southern counties. The entry threshold estimates are not reasonable outside the South, however, owing to the distribution of theaters in this region. As documented earlier, only one non-MSA county outside the South has two theaters, and no counties have more than two. Consequently, we restrict attention to the Southern counties in our sample in this discussion. The market size threshold for a monopoly is fairly similar to that estimated for MSAs, requiring between eight and nine thousand black residents to support one theater. The results begin to deviate from the MSA results for the two- and three-firm entry thresholds. In the full specification, we estimate a per-firm duopoly market size threshold of 10,008 African-American residents, and a threshold of 12.5 thousand per firm for a triopoly. The larger estimated thresholds at the county level may be due to larger transportation costs and lower density in those areas. In summary, our findings are robust and show that black theaters face less competition in South MSAs and non-MSA counties than they did elsewhere. This may be driven not only by residential segregation but also the head-to-head competition with general theaters in areas where segregation and discrimination are less acute. Moreover, additional entry of black theaters in non-South regions increase competition more than additional entry in the South. This softer impact of entry on competition in the South may be caused by the fact that the South has larger shares of black population, stronger views on residential segregation matters, or stronger racial biases as measured by the Thurmond vote on the 1948 Presidential election.
6
Conclusion
In this paper we study African-American movie theaters in the early 1950s, presenting facts about theater location, and how theater location is influenced by residential segregation and racial bias. Little systematic empirical evidence exists regarding this important historical institution, and studying this setting allows for drawing broader lessons regarding the effects of residential segregation. To the extent that the movie theater industry is representative, the results in the paper shed light on the impacts of public accommodation segregation on African-
24
American life in the post-war era. We find that residential segregation leads to more African-American theaters than one would expect given the size of the black population and the socioeconomic characteristics of the area. Results from a structural model of theater entry are consistent with our hypothesis that residential segregation leads to preference externalities. Our instrumental variables specifications suggest that the effect of residential segregation is a causal one. That said, the formation of racial enclaves may arise endogenously, working against the adverse effects of segregation in public accommodation. The effects of racial discrimination are complex. Based on our results from multiple measures of racial bias, we conclude that racial bias in part exhibits as a taste for segregation, which tends to increase entry of black theaters. More generally, discrimination reduces African-American theater entry by reducing the variable profits of potential entrants. This points toward a channel for racial bias that reduces demand for movies or access to variable inputs rather than denying credit or access to fixed inputs such as land. A modern implication of the results for the post-Jim Crow era relate to the impact of residential segregation on African-American consumption opportunities. The results provide support for the possibility that residential segregation, via preference externalities, may improve consumption when goods are horizontally differentiated. However, these externalities may also impact vertical differentiation as well. Historical sources indicate that African-American theaters were lower quality, and residential segregation may have reduced access to higher quality theaters. Furthermore, the results have potential implications for understanding the forces that may impede business formation in minority neighborhoods.
25
7
References
Ananat, Elizabeth. (2011) “The Wrong Side(s) of the Tracks: The Causal Effects of Racial Segregation on Urban Poverty and Inequality” American Economic Journal: Applied Economics, 3(2): 34-66. Berry, S. T. (1992) “Estimation of a Model of Entry in the Airline Industry,” Econometrica, Vol. 60, No. 4, pp. 889-917. Borjas, George J. (1995) “Ethnicity, Neighborhoods, and Human Capital Externalities,” American Economic Review 85:3, p. 365-390. Boustan, Leah Platt. (2010) “Was Postwar Suburbanization ”White Flight”? Evidence from the Black Migration,” The Quarterly Journal of Economics 125:1, p. 417-443. Bresnahan, Timothy F. and Peter C. Reiss. (1990) “Entry in Monopoly Markets,” Review of Economic Studies 57, p. 531-553. ———- (1991a) “Entry and Competition in Concentrated Markets,” Journal of Political Economy 99:5, p. 977-1009. ———- (1991b) “Empirical Models of Discrete Games,” Journal of Econometrics 48:1, p. 5781. Charles, Kerwin Kofi and Jonathan Guryan. (2008) “Prejudice and Wages: An Empirical Assessment of Becker’s The Economics of Discrimination, Journal of Political Economy 116:5, p. 773-809. Collins, William J. and Robert A. Margo. (2000) “Residential Segregation and Socioeconomic Outcomes: When did Ghettos Go Bad?” Economic Letters 69:2, p. 239-43. Cutler, David M. and Edward L. Glaeser. (1997) “Are Ghettos Good or Bad?” The Quarterly Journal of Economics 112:3, p. 827-872. Cutler, David M., Edward L. Glaeser, and Jacob L. Vigdor. (1999) “The rise and decline of the American ghetto,” Journal of Political Economy 107:3, p. 455-506. ———– (2008) “When are ghettos bad? Lessons from immigrant segregation in the United States,” Journal of Urban Economics 63:3, p. 759-774. Fairchild, Gregory B. (2008) “The influence of residential segregation and its correlates on ethnic enterprise in urban areas.” Journal of Business Venturing, 23:5, p. 513-527. Fischer, Mary J. and Douglas S. Massey. (2000) “Residential Segregation and Ethnic Enterprise in U.S. Metropolitan Areas,” Social Problems 47:3, p. 408-424. Gentzkow, Matthew. (2006) “Television and Voter Turnout,” The Quarterly Journal of Economics 121:3, p. 931-972.
26
Glazer, Nathan and Daniel Moynihan. Beyond the Melting Pot (Cambridge, MA: MIT Press, 1963). Graddy, Kathryn. (1995) “Testing for Imperfect Competition at the Fulton Fish Market,” The Rand Journal of Economics, 26, 75-92. Headley, Robert K. (1999) Motion Picture History in Washington, DC: An Illustrated History of Parlors, Palaces and Multiplexes in the Metropolitan Area, 1894-1997, McFarland & Company, Inc., Publishers. Headley, Robert K. (2006) Motion Picture History in Baltimore: An Illustrated History and Directory of Theaters, 1895-2004, McFarland & Company, Inc., Publishers. Holzer, Harry and Keith Ihlandfeldt. (1998) “Customer discrimination and employment outcomes for minority workers,” Quarterly Journal of Economics 113, p. 835-867. Kain, John F. (1968) “Housing Segregation, Negro Employment, and Metropolitan Decentralization.” Quarterlv Journal of Economics 82:2, p. 175-97. Li, Huailu. (2014) “Does Competition Eliminate Discrimination? Evidence from the Commercial Sex Market in Singapore,” manuscript. Massey, Douglas S. and Nancy A. Denton. (1988) “The Dimensions of Residential Segregation,” Social Forces 67:2, p. 281-315. Mazzeo, Michael J. (2002) “Product Choice and Oligopoly Market Structure,” RAND Journal of Economics, Vol. 33, No. 2, pp. 221–242. Seim, Katja. (2006) “An Empirical Model of Firm Entry with Endogenous Product-type Choices,” RAND Journal of Economics, Vol. 37, No. 3, pp. 619-640. Waldfogel, Joel. (2003) “Preference Externalities: An Empirical Study Of Who Benefits Whom In Differentiated-Product Markets,” Rand Journal of Economics 34:3, p. 557-568. ——-. (2008) “The Median Voter and the Median Consumer: Local Private Goods and Residential Sorting,” Journal of Urban Economics 63, p. 567582. White, Michael J. (1983) “The Measurement of Spatial Segregation,” American Journal of Sociology 88, p. 1008-1019. Woodward, C. Vann. (1974) The Strange Career of Jim Crow, New York: Oxford University Press.
27
Figure 1: Number of black theaters
28
Figure 2: Black theaters per thousand black residents
29
0
.1
Black theater share .2 .3
.4
Figure 3: Black theater share in MSAs
0
.1
.2 .3 Black population share South
30
Non−South
.4
0
Average number of black theaters 20 40
60
Figure 4: Number of black theaters in MSAs
4
6
8 10 Log black population 1950 South
31
Non−South
12
14
Table 1: MSA Summary Statistics Non-South
South
All MSAs
Number of 1950 black theaters
2.889 (8.551)
4.647 (4.660)
3.525 (7.413)
Black theater share
0.0260 (0.0422)
0.169 (0.0858)
0.0776 (0.0922)
Seg. Dissim. Index, 1960
0.758 (0.0879)
0.726 (0.108)
0.746 (0.0963)
0.0000139 (0.0000810)
0.0350 (0.0351)
0.0127 (0.0269)
Thurmond on ballot in state
0.177 (0.379)
0.908 (0.279)
0.441 (0.493)
Black pop. share, 1950
0.0451 (0.0360)
0.213 (0.114)
0.106 (0.110)
Total population, 1950 (000s)
698.5 (1580.6)
312.7 (285.9)
558.9 (1285.3)
Log area in sq mi
6.917 (0.920)
6.757 (0.463)
6.859 (0.788)
% Black pop. growth, 1950-60
0.679 (0.401)
0.342 (0.284)
0.557 (0.397)
% HH TV penetration, 1950
0.104 (0.0993)
0.0490 (0.0833)
0.0843 (0.0972)
% HH TV penetration, 1955
0.881 (0.0736)
0.770 (0.148)
0.841 (0.119)
Median education level, 1950
10.28 (1.028)
9.606 (1.144)
10.03 (1.115)
Median income, 1950
3435.5 (318.6)
2894.1 (448.7)
3239.6 (452.3)
Ag share of emp., 1950
0.0480 (0.0400)
0.0522 (0.0400)
0.0495 (0.0399)
90
51
141
Thurmond votes per white
Observations
32
Table 2: Summary Statistics, Non-MSA Counties Non-South
South
All Counties
Number of 1950 black theaters
0.0114 (0.123)
0.594 (0.921)
0.212 (0.615)
Black theater share
0.00127 (0.0131)
0.101 (0.129)
0.0357 (0.0897)
Seg. Dissim. Index, 1950
0.414 (0.245)
0.242 (0.177)
0.354 (0.238)
0.00000890 (0.0000719)
0.0382 (0.0450)
0.0132 (0.0320)
Thurmond on ballot in state
0.0833 (0.277)
0.795 (0.404)
0.329 (0.470)
Black pop. share, 1950
0.0154 (0.0239)
0.218 (0.167)
0.0854 (0.139)
Total population, 1950 (000s)
39.77 (29.90)
38.84 (22.34)
39.45 (27.52)
Log area in sq mi
6.641 (0.666)
6.386 (0.546)
6.553 (0.638)
% black popgrow, 1950-60
0.406 (0.808)
0.0106 (0.222)
0.270 (0.693)
% HH TV penetration, 1950
0.0298 (0.0450)
0.0173 (0.0286)
0.0255 (0.0405)
% HH TV penetration, 1955
0.664 (0.217)
0.598 (0.204)
0.641 (0.215)
Median education level, 1950
9.525 (1.002)
8.182 (1.006)
9.062 (1.189)
Median income, 1950
2785.3 (456.2)
1898.9 (557.5)
2479.6 (648.8)
Ag share of emp., 1950
0.233 (0.131)
0.273 (0.148)
0.247 (0.139)
528
278
806
Thurmond votes per white
Observations
33
Table 3: Distribution of Theaters
Panel A: African-American Theaters,MSA-level
# Theaters: None One Two Three or more # of MSAs
Non-South 51 20 4 14 89
Region: South 3 9 12 27 51
All MSAs 54 29 16 41 140
Panel B: African-American Theaters, Non-MSA Counties
# Theaters: None One Two Three or more # of Counties
Non-South 606 4 1 0 611
Region: South All Counties 180 90 21 12 303
786 94 22 12 914
Panel C: All Theaters, Non-MSA Counties
# Theaters: One Two Three or more # of Counties
Non-South 10 39 562 611
Region: South All Counties 8 44 251 303
18 83 813 914
The sample is restricted to observations where no variables used in the estimation are missing.
34
Table 4: MSA-level black theater share, OLS specification
Seg. Dissim. Index, 1950
(1) All
(2) All
0.099∗∗ (0.041)
0.069 (0.043)
Thurmond votes per white Thurmond on ballot in state Black pop. share, 1950
0.24∗∗ (0.093) 1.02∗∗∗ (0.24) -0.53 (0.35) 0.040 (0.036) 0.52∗∗∗ (0.18) -0.035 (0.022) 0.013 (0.022) 0.032 (0.027) 0.011 (0.15) -0.0059 (0.012) 0.071 (0.11) -0.60∗ (0.32)
-0.033 (0.066) 0.033 (0.100)
0.76∗∗∗ (0.18) 0.0044 (0.0070) 0.0053 (0.0042) -0.0045 (0.011) -0.013 (0.047) 0.0030 (0.0058) -0.016 (0.041) -0.035 (0.070)
-0.15 (0.40) Yes
-0.74 (0.91) Yes
0.034 (0.32) Yes
141 0.82
51 0.72
90 0.58
143 0.80
141 0.81
Log median income Ag share of emp., 1950
Observations R-Squared
0.12∗∗ (0.055) 0.31 (0.20) -0.66∗∗ (0.32) 0.021∗∗ (0.0096) 0.59∗∗∗ (0.12) -0.0063 (0.0076) 0.0096∗∗ (0.0045) -0.010 (0.010) -0.030 (0.053) 0.0017 (0.0063) 0.019 (0.055) -0.21 (0.14)
56 0.93
Median education level, 1950
Detailed regions
0.11∗∗ (0.046) 0.11∗∗ (0.050) -0.78∗∗ (0.32)
56 0.92
% HH TV penetration, 1950
Constant
(7) Non-south
-0.0033 (0.032) -0.0071 (0.012) 0.00061 (0.0059) -0.073∗∗ (0.033) No
% Black pop. growth, 1950-60
West
(6) South
0.061∗ (0.032) 0.030∗∗ (0.013) 0.011 (0.0081) -0.10∗∗∗ (0.033) No
Log area in sq mi
Midwest
(5) All
-0.032 (0.067) -1.10∗∗∗ (0.32) 0.029∗∗ (0.012) 0.74∗∗∗ (0.15) 0.0010 (0.0082) 0.0062 (0.0056) -0.0055 (0.013) -0.058 (0.053) -0.0028 (0.0056) 0.018 (0.066) -0.44 (0.30) 0.071∗∗ (0.033) 0.039∗∗ (0.015) 0.013 (0.013) -0.24 (0.50) No
-0.0097 (0.055) -0.98∗∗∗ (0.26) 0.020∗∗ (0.0100) 0.75∗∗∗ (0.12)
Log pop. 1950
South
(4) All 0.11∗∗ (0.054) 0.11∗ (0.058) -0.81∗∗ (0.34) 0.022∗∗ (0.0090) 0.68∗∗∗ (0.10) -0.0037 (0.0073) 0.0056 (0.0042) -0.0074 (0.0090) -0.021 (0.052) 0.0013 (0.0047) -0.0091 (0.048) -0.20 (0.14) -0.014 (0.036) 0.0014 (0.016) -0.0035 (0.011) 0.0080 (0.36) No
Seg. Dissim. Index, 1960 Index of racial bias
(3) All
0.71∗∗∗ (0.085)
The dependent variable is the African-American theater share in the MSA, averaged across the years 1950, 1951, 1952, 1954, and 1955. The index of racial bias is the mean of the within-region average standardized responses from racial attitude questions in the 1972 and 1976 General Social Surveys related to interracial marriage, schooling segregation, and residential segregation index. Robust standard errors are in parentheses. *,**,*** denote significance at the 10%, 5%, and 1% level, respectively.
35
Table 5: African-American theater share and components of the racial bias index (1)
(2)
(3)
(4)
(5)
0.042 (0.058)
0.16∗∗ (0.065) -0.0040 (0.043) -0.029 (0.071)
141 0.80
141 0.81
0.11∗
Index of racial bias
(0.052)
0.14∗∗ (0.057)
Residential seg. views Interracial marriage views
0.059 (0.044)
Segregated schooling views Observations R-Squared
141 0.81
141 0.81
141 0.80
The dependent variable is the African-American share of theaters in the MSA. The index of racial bias is an average of three standardized responses to questions related to racial issues in the General Social Survey. These questions asked respondents their views on residential segregation, segregation in schools, and interracial marriage. These responses have been standardized and defined such that a higher number corresponds to more racial bias. Other controls are the same as those shown in column 4 of Table 4. Standard errors corrected for clustering at the detailed region level are in parentheses. *,**,*** denote significance at the 10%, 5%, and 1% level, respectively.
Table 6: Instrument Falsification Check, 1920 characteristics
Pre-1920 bridges per stream Observations R-Squared
(1) Log pop. density
(2) Mfg. emp./Pop.
(3) Non-white pop. share
(4) Literacy rate
(5) Urban pop. share
(6) 1920 seg. index
0.12 (0.15)
0.0020 (0.0068)
-0.0023 (0.0038)
-0.0020 (0.0025)
-0.011 (0.023)
0.0085 (0.019)
136 0.24
136 0.38
136 0.59
136 0.39
136 0.11
65 0.18
Each column shows a separate regression of the 1920 characteristic on the number of bridges built before 1920 and coarse region dummies (South, Midwest, West). The 1920 segregation index is from the Cutler-Glaeser-Vigdor data. The sample is restricted to those MSAs for which we observe 1960 segregation to match that used in the IV estimation. Robust standard errors are in parentheses. *,**,*** denote significance at the 10%, 5%, and 1% level, respectively.
36
Table 7: MSA-level black theater share, IV specification
(1)
(2)
(3)
0.26∗∗ (0.12)
0.26∗ (0.15) 0.00076 (0.015)
Yes Yes
Yes Yes
0.26∗ (0.15) 0.0021 (0.016) -0.0018 (0.0021) 0.00014 (0.00058) Yes Yes
Panel A: IV estimates Seg. Dissim. Index, 1960 Population growth, 1910-60 Bridges built 1940-60 per stream Bridges built after 1960 per stream Detailed regions Other covariates
Panel B: First-stage estimates -0.040∗∗∗ (0.013)
-0.034∗∗∗ (0.012) 0.073∗∗∗ (0.021)
Detailed regions Other covariates
Yes Yes
Yes Yes
-0.035∗∗∗ (0.013) 0.071∗∗∗ (0.022) -0.011∗∗∗ (0.0039) 0.0021 (0.0013) Yes Yes
Observations First-stage F-stat
136 9.53
135 7.25
135 7.61
Bridges built before 1920 per stream Population growth, 1910-60 Bridges built 1940-60 per stream Bridges built after 1960 per stream
The dependent variable is the African-American theater share in the MSA, where the number of African-American and white theaters are each averaged across the years 1950, 1951, 1952, 1954, and 1955. The instruments are the number of bridges spanning waterways built before 1920. The variables capturing bridges built after 1960 and between 1940 and 1960 are meant to control for contemporaneous infrastructure spending. The other right-hand side variables are identical to specification (5) of Table 4. Robust standard errors are in parentheses. *,**,*** denote significance at the 10%, 5%, and 1% level, respectively.
37
Table 8: County-level black theater share, OLS specification
Seg. Dissim. Index, 1950 Index of racial bias Thurmond votes per white Thurmond on ballot in state Black pop. share, 1950
(1) All
(2) All
(3) All
0.022∗∗∗ (0.0062) -0.075 (0.048) 0.23 (0.22) 0.0010 (0.0076) 0.36∗∗∗ (0.052)
0.018∗∗ (0.0073)
0.061∗∗ (0.029) 0.0093 (0.0089) -0.00094 (0.0043) -0.025∗∗ (0.011) No
0.019∗∗∗ (0.0073) -0.041 (0.049) 0.27 (0.21) -0.0011 (0.0078) 0.39∗∗∗ (0.050) 0.0031 (0.0043) -0.0025 (0.0049) 0.0025∗ (0.0013) -0.12∗∗∗ (0.041) -0.00057 (0.0026) 0.050∗∗∗ (0.015) -0.0060 (0.022) 0.052∗ (0.029) 0.0040 (0.0095) -0.0071 (0.0077) -0.40∗∗∗ (0.12) No
922 0.47
866 0.48
Log pop. 1950 Log area in sq mi % black popgrow, 1950-60 % HH TV penetration, 1950 Median education level, 1950 Log median income Ag share of emp., 1950 South Midwest West Constant Detailed regions Observations R-Squared
(4) All
(5) South
(6) Non-south
0.093∗∗∗ (0.035)
-0.0014 (0.0014)
0.32 (0.22) 0.00086 (0.0081) 0.39∗∗∗ (0.054) 0.0044 (0.0045) -0.0043 (0.0055) 0.0027∗∗ (0.0013) -0.13∗∗∗ (0.042) -0.00077 (0.0027) 0.050∗∗∗ (0.017) -0.0052 (0.022)
0.093 (0.12) 0.0071 (0.0053) 0.28∗∗∗ (0.033) 0.015∗∗∗ (0.0032) -0.0055 (0.0041) 0.00011∗∗∗ (0.000023) -0.22∗∗∗ (0.042) 0.0029 (0.0025) 0.048∗∗∗ (0.0093) -0.0063 (0.018)
0.16 (0.23) 0.0058 (0.016) 0.46∗∗∗ (0.058) 0.016 (0.014) -0.0040 (0.0088) 0.081∗∗∗ (0.027) -0.30 (0.19) -0.0012 (0.0098) 0.048 (0.042) -0.027 (0.079)
0.038 (0.023) 0.0017 (0.0011) 0.0014∗ (0.00078) -0.00062∗ (0.00033) 0.00011 (0.011) 0.00034 (0.00094) -0.0017 (0.0050) -0.0042 (0.0046)
-0.38∗∗∗ (0.12) Yes
-0.42∗∗∗ (0.070) Yes
-0.41 (0.33) Yes
-0.0046 (0.034) Yes
866 0.49
2305 0.26
303 0.35
563 0.028
The sample includes non-MSA counties. The dependent variable is the African-American theater share in the county, where the number of African-American and white theaters are each averaged across the years 1950, 1951, 1952, 1954, and 1955. The index of racial bias is the mean of the within-region average standardized responses from racial attitude questions in the 1972 and 1976 General Social Surveys related to interracial marriage, schooling segregation, and residential segregation index. Robust standard errors are in parentheses. *,**,*** denote significance at the 10%, 5%, and 1% level, respectively.
38
Table 9: Entry Model Estimates, MSA level
Market size
Change in Black population, 1950-60 (000s) Black pop., neighbor counties
Variable profit
Seg. Dissim. Index, 1960 Thurmond votes per white South
(1)
(2)
(3)
-0.12 (0.11) 0.048 (0.042)
-0.11 (0.17) -0.0039 (0.030)
-0.075 (0.17) 0.0071 (0.033)
0.20∗∗ (0.090) -0.57∗ (0.34) 0.040∗ (0.023)
0.24∗∗ (0.10) -0.59 (0.38) 0.025 (0.026) -0.14∗∗∗ (0.047)
-0.025 (0.072) 0.082∗∗∗ (0.015) 0.028∗∗∗ (0.0064)
0.046 (0.085) 0.15∗∗∗ (0.031) 0.033∗∗∗ (0.0079)
0.19∗ (0.11) -0.47 (0.39) 0.053 (0.034) -0.21∗∗∗ (0.081) -0.14 (0.095) 0.066 (0.061) 0.043∗∗ (0.019) -0.69 (0.48) 0.15∗∗∗ (0.031) 0.039∗∗∗ (0.0090)
1.82 (1.72) 0.27 (2.24) -15.8 (18.7) -0.41 (0.62)
2.01 (2.15) 0.62 (2.43) -12.8 (19.4) -0.80 (0.68) -2.45 (1.90) 0.061 (0.24)
-0.56 (2.11) -0.56 (2.44) -14.2 (19.9) -0.37 (0.73) -2.41 (1.92) 0.68∗ (0.35)
123 -85.6
123 -71.7
123 -66.8
1950 White theaters per sq. mi. % HH TV penetration 1950 Log median income Log pop. density α1 α2 α3
Fixed cost
γ1 Seg. Dissim. Index, 1960 Thurmond votes per white South S&L capital per capita Log pop. density Observations Log Likelihood
Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
This table presents estimates of an ordered probit model of entry. A unit of observation is the MSA. The categories for the ordered probit likelihood function are 0, 1, 2, and 3 or more firms in an MSA. The sample is restricted to cities with fewer than 100 thousand black residents. Each column represents a separate specification, with a different set of included covariates. Within each column, coefficients are grouped according to the equation in which they appear. Standard errors are in parenthesis. *,**,*** denote significance at the 10%, 5%, and 1% level, respectively.
39
Table 10: Entry model estimates, non-MSA county level
Market size
Variable profits
(1)
(2)
(3)
Change in Black population, 1950-60 (000s)
0.58∗∗∗
0.66∗∗∗
Black pop, neighbor counties
(0.17) -0.0010 (0.017)
(0.17) -0.0077 (0.017)
0.92∗∗∗ (0.15) -0.016 (0.014)
0.021 (0.054) -0.038 (0.18) 0.036 (0.036)
-0.037 (0.065) 0.028 (0.19) 0.030 (0.035) 0.16∗∗ (0.073)
0.12∗∗∗ (0.036) 0.11∗∗∗ (0.012) 0.032∗∗∗ (0.0064)
0.14∗∗∗ (0.036) 0.11∗∗∗ (0.012) 0.035∗∗∗ (0.0071)
-0.046 (0.064) 0.088 (0.22) 0.049 (0.046) 0.28∗∗∗ (0.086) -0.046 (0.16) 0.021 (0.022) -0.043∗∗∗ (0.017) 0.16 (0.18) 0.12∗∗∗ (0.012) 0.036∗∗∗ (0.0073)
2.85∗∗∗ (0.30) -0.50 (0.48) 0.49 (2.90) -1.18∗∗∗ (0.26)
2.46∗∗∗ (0.45) -0.59 (0.48) 1.08 (2.89) -1.23∗∗∗ (0.26) 0.00082 (0.0016) 0.10 (0.100)
3.09∗∗∗ (0.53) -0.66 (0.49) 2.44 (2.98) -1.17∗∗∗ (0.27) 0.00056 (0.0015) -0.045 (0.11)
914 -297.0
914 -292.6
914 -288.0
Seg. Dissim. Index, 1950 Thurmond votes per white South 1950 White theaters per sq. mi. % HH TV penetration 1950 Log median income Log pop. density α1 α2 α3
Fixed cost
γ1 Seg. Dissim. Index, 1950 Thurmond votes per white South S&L capital per capita Log pop. density Observations Log Likelihood
Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
This table presents estimates of an ordered probit model of entry. A unit of observation is the county. The categories for the ordered probit likelihood function are 0, 1, 2, and 3 or more firms in a county. Standard errors are in parenthesis.
40
Table 11: Entry Thresholds, MSA level
Market size thresholds S1 S2 S3
s1
Per-firm thresholds s2 s3
Panel A: Specification (1) Non-south South
15.82345 9.559705
44.47807 18.72077
116.3043 27.80902
15.82345 9.559705
22.23903 9.360383
38.76809 9.269674
17.42493 8.98632
21.11588 9.020917
19.64025 9.267644
32.26442 9.195664
Panel B: Specification (2) Non-south South
11.75546 7.285683
34.84986 17.97264
63.34764 27.06275
11.75546 7.285683
Panel C: Specification (3) Non-south South
11.93067 8.180129
39.28051 18.53529
96.79324 27.58699
11.93067 8.180129
This table displays the estimated required black population in thousands required to support the stated number of firms. These figures are obtained from the baseline model estimates displayed in Table 9.
Table 12: Entry thresholds, non-MSA county level
Market size thresholds S1 S2 S3
s1
Per-firm thresholds s2 s3
Panel A: Specification (1) Non-south South
21.09493 9.242759
159.9558 28.79274
-179.0244 73.53008
21.09493 9.242759
79.97789 14.39637
-59.67479 24.51003
110.7367 17.41131
-39.37656 68.78072
48.2551 10.00845
-114.0357 12.498
Panel B: Specification (2) Non-south South
21.28633 9.362823
221.4733 34.82262
-118.1297 206.3421
21.28633 9.362823
Panel C: Specification (3) Non-south South
18.47495 7.894196
96.51019 20.0169
-342.107 37.494
18.47495 7.894196
This table displays the estimated required black population in thousands required to support the stated number of firms. These figures are obtained from the baseline model estimates displayed in Table 10.
41
A
Alternative segregation measures
Some authors in the literature have criticized the dissimilarity index on the grounds that it does not distinguish an MSA with black census tracts that are scattered throughout the city (the so-called “checkerboard” pattern) from an MSA where the black tracts are concentrated in a broader “black belt.” To investigate the different implications the two geographic patterns of segregation, we also form the spatial proximity index suggested by White (1993) and the absolute concentration index described by Massey and Denton (1988). The former is meant to capture the average distance to other own-race individuals relative to the average distance between people generally. A value of 1.0 indicates that there is no differential clustering between races, while values greater than 1.0 indicate that individuals live closer to other own-group individuals than one would expect given population shares. The absolute concentration index measures the number of own race individuals in nearby tracts as a fraction of the population in nearby tracts. In Table A1, we present the results of regressing the African-American theater share on two alternative measures of residential segregation: the spatial proximity index and the absolute clustering index. For comparison, in column (1), we reproduce the main specification shown in column (5) of Table 4. As we introduce the alternative measures individually in the specifications in columns (2) and (3), we see that the spatial proximity and absolute cluster indices enter significantly and with a similar magnitude as the dissimilarity index. In column (4), we include both the dissimilarity index and the spatial proximity index. When included together, neither measure is statistically significant, though the magnitudes of the estimated coefficient on the dissimilarity index is similar as when introduced on its own. Lastly, in column (5), we include the dissimilarity index and absolute clustering index in the same specification. In this specification, the absolute clustering index is statistically significant at the 10 percent level, while the dissimilarity index is not. This provides some support for travel costs and preference externalities as the mechanism by which segregation operates.
42
Table A1: MSA-level black theater share, Alternative segregation measures (1) Seg. Dissim. Index, 1960
(2)
0.12∗∗ (0.055)
Spatial Proximity Index, 1960
0.070∗∗ (0.033)
Abs. Clustering Index, 1960 Observations R-Squared
141 0.82
141 0.82
(3)
0.13∗∗∗ (0.048) 141 0.82
(4)
(5)
0.099 (0.063) 0.033 (0.038)
0.079 (0.061)
141 0.82
0.096∗ (0.052) 141 0.83
The dependent variable is the African-American theater share in the MSA, averaged across the years 1950, 1951, 1952, 1954, and 1955. The other covariates are identical to those shown in Column (5) of Table 4. Robust standard errors are in parentheses. *,**,*** denote significance at the 10%, 5%, and 1% level, respectively.
43
B Estimation of ordered probit model separately by region
44
Table A2: Entry Model Estimates, MSA-Level Estimated Separately by Region
Market size
Black pop. change, 1950-60 (000s) Black pop., neighbor counties
Variable profit
Seg. Dissim. Index, 1960 Thurmond votes per white 1950 White theaters per sq. mi. % HH TV penetration 1950 Log median income Log pop. density South α1 α2 α3
Fixed cost
γ1 Seg. Dissim. Index, 1960 Thurmond votes per white S&L capital per capita Log pop. density South Observations Log Likelihood
(1) 1950 All
(2) 1950 South
(3) 1950 South
(4) 1950 Non-South
(5) 1950 Non-South
-0.075 (0.17) 0.0071 (0.033)
-0.38 (0.31) 0.13 (0.12)
0.39 (0.84) 0.33 (0.26)
0.066 (0.26) -0.012 (0.034)
0.050 (0.25) -0.010 (0.036)
0.19∗ (0.11) -0.47 (0.39) -0.21∗∗∗ (0.081) -0.14 (0.095) 0.066 (0.061) 0.043∗∗ (0.019) 0.053 (0.034) -0.69 (0.48) 0.15∗∗∗ (0.032) 0.039∗∗∗ (0.0090)
0.19 (0.15) -0.96∗∗ (0.47) 0.57 (0.99)
0.29∗ (0.16) -0.53 (0.40) -1.67∗ (0.96) -0.46∗∗ (0.18) 0.18∗ (0.099) 0.043 (0.034)
0.055 (0.21)
-0.0023 (0.22)
-0.11∗∗ (0.047)
-0.19∗ (0.11) -0.062 (0.12) 0.051 (0.16) 0.040 (0.027)
0.50 (0.31) 0.53∗ (0.28) 0.038∗∗∗ (0.012)
-1.44∗ (0.84) 0.26 (0.19) 0.054∗∗∗ (0.020)
0.16 (0.17) 0.12∗∗∗ (0.031) 0.023∗∗ (0.011)
-0.43 (1.27) 0.13∗∗∗ (0.034) 0.029∗∗ (0.014)
-0.56 (2.11) -0.56 (2.44) -14.2 (19.9) -2.41 (1.92) 0.68∗ (0.35) -0.37 (0.73)
7.77 (6.01) -4.57 (4.91) -37.4 (24.2) 2.82 (4.06) -0.31 (0.79)
5.60 (7.59) -2.97 (5.16) -17.8 (30.5) 12.3∗ (6.43) -0.24 (1.14)
1.59 (2.87) -0.059 (3.48)
-0.84 (2.73) -1.70 (3.64)
-3.92∗ (2.30) 0.26 (0.28)
-3.85∗ (2.33) 0.91∗ (0.47)
123 -66.8
43 -25.7
43 -19.0
80 -40.1
80 -38.2
Standard errors in parentheses ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
This table presents estimates of an ordered probit model of entry. A unit of observation is the MSA. The categories for the ordered probit likelihood function are 0, 1, 2, and 3 or more firms in MSA. The sample is restricted to counties with fewer than 100 thousand black residents. Each column represents a separate specification, with a different set of included covariates. Within each column, coefficients are grouped according to the equation in which they appear. Standard errors are in parenthesis. *,**,*** denote significance at the 10%, 5%, and 1% level, respectively.
45
Table A3: Entry Threshold Estimates, MSA-Level From Estimating Separately By Region
Market size thresholds S1 S2 S3 South Non-South
6.968761 13.20857
22.33745 41.07568
41.89898 77.48455
s1
Per-firm thresholds s2 s3
6.968761 13.20857
11.16873 20.53784
13.96633 25.82818
This table displays the estimated required black population in thousands required to support the stated number of firms. These figures are obtained from models estimated separately by region as displayed in Table A2.
46
