Despite a well-established literature investigating race-related predictors of riot incidence, the racial aftermath of riots remains unexamined. In this study, I use the 1960s U.S. race riots to investigate trends in black residential segregation levels following large-scale riot activity in seven major U.S. cities. I use a novel approach—namely, synthetic control matching—to select a group of cities against which segregation trends can be compared. I find that levels of black segregation rose in 1970 for four of the seven cities, but these increases disappeared in 1980 and 1990 except in Detroit. These results mask differential trends at lower geographic levels: suburban neighborhoods in affected areas experienced larger and longer-term increases in segregation, particularly in traditionally hypersegregated cities in the Midwest and Northeast.
Scholarly research on the macro-level causes of civil disorder in the United States emerged after the 1960s (e.g., Downes 1968; Gurr 1968; Lieberson and Silverman 1965; Myers 1997, 2000; Olzak and Shanahan 1996; Olzak et al. 1996; Spilerman 1970, 1971, 1976), a period marked by an unprecedented rash of violent riots that erupted in cities across the country. Interest in determining the causes of riots continues today as instances of civil unrest, such as the recent riots in Ferguson, MO, and Baltimore, MD, continue to dot the urban landscape. Several studies have identified racial segregation—a measure often associated with racial economic deprivation and competition—as a key precipitant of large-scale rioting (Darden and Thomas 2013; Harris 1988; Olzak et al. 1996). This finding echoes the general response by the national government and media, which blames racial inequality and separation as the primary causes for urban civil unrest (National Advisory Commission on Civil Disorders (NACCD) 1968). Missing in these responses, however, is attention to segregation following riot activity.
In a report investigating the causes of the 1960s riots (NACCD 1968:1), an 11-member commission appointed by then President Lyndon Johnson recognized the impact that the riots may have on segregation by concluding that the “nation is moving toward two societies, one black, one white—separate and unequal. Reaction to last summer’s disorders has quickened the movement and deepened the division.” The commission recommended a policy that “combines ghetto enrichment with programs designed to encourage integration of substantial numbers of Negroes into the society outside of the ghetto” (NACCD 1968:11). Despite the report’s conclusions and recommendations, no study has empirically examined changes in segregation after riot activity, which is a significant omission considering that rising levels of segregation may lead to future outbreaks of civil disorder and harm the well-being and life chances of individuals, particularly African Americans (Cutler and Glaeser 1997; Massey and Denton 1993; Olzak et al. 1996; Quillian 2002). Furthermore, if large-scale race riots sustain or increase segregation, they may help explain why black segregation levels in certain cities remained stubbornly high after the 1960s while they decreased in other comparable areas.
In this study, I examine changes in black segregation after large-scale urban riots using data on the 1960s riots. I focus on the seven cities that experienced the most destructive riots: Baltimore (MD), Chicago (IL), Cleveland (OH), Detroit (MI), Los Angeles (CA), Newark (NJ), and Washington, DC. These cities account for more than one-half of the riot-related deaths, injuries, arsons, and arrests that occurred during the period. I use a recently developed method—synthetic control matching—to select an appropriate control group for each city (Abadie et al. 2010). I find that large-scale riots in the 1960s are associated with increased segregation in Detroit from 1970 to 1990 but only short-term increases in other cities. These results mask differential trends between suburbs and central cities: suburban neighborhoods experienced larger and longer-term increases in segregation, particularly in historically segregated cities in the Midwest and Northeast.
Segregation Levels After Large-Scale Race Riots
Analyses of black segregation in the twentieth century reveal two basic trends: (1) a rapid increase in the first half and (2) a gradual decline thereafter (Cutler et al. 1999; Iceland et al. 2002; Logan and Stults 2011; Massey and Denton 1987). The first-half increase in segregation was primarily due to legalized forms of housing discrimination in Northern cities that pushed migrating blacks from the South into poor urban neighborhoods. Scholars have attributed the decrease in segregation after the 1960s to a number of factors, including sweeping national civil rights reforms, changes in public attitudes toward blacks, increased economic opportunities for minorities, the growth of multiethnic metropolises, regional population shifts, and black suburbanization (Logan et al. 2004; Rugh and Massey 2014; Vigdor 2013). However, decreasing levels of black segregation were concentrated in smaller urban areas, cities in the West and South, and metropolitan areas with small black populations (Logan 2013; Massey and Gross 1991); segregation changed little in cities without these characteristics, particularly for certain cities in the Northeast and Midwest (Iceland et al. 2002; Massey and Denton 1987, 1993) described by Logan and Stults (2011) as America’s Ghetto Belt.
Scholars have identified the deindustrialization of central cities, new forms of housing discrimination, durable anti-black sentiment, and persistent white preferences for majority white neighborhoods as factors contributing to the persistence of high segregation levels (Charles 2003; Iceland and Sharp 2013; Massey and Gross 1991; Rugh and Massey 2014). Given their scale, severity, and highly racialized nature, the 1960s riots also may have played a role in maintaining high levels of segregation in affected areas. Recent empirical studies have made efforts in examining the impact of urban riots, particularly on black living conditions, but they have primarily focused on economic outcomes such as property values and taxable sales (Baade et al. 2007; Collins and Margo 2004, 2007; Johnson et al. 1997; Matheson and Baade 2004; see Button 1978 and Darden and Thomas 2013 for exceptions). A formal examination of changes in racial segregation following large-scale riots is missing in the literature.
In the following sections, I outline the three standard explanations for the presence of segregation: (1) neighborhood racial preferences, which stresses white and black preferences to live in neighborhoods where they are the dominant race; (2) barriers to entry, which emphasizes the informal barriers in the housing market that prevent blacks from moving into white neighborhoods; and (3) racial differences in income and wealth, which points to the significant economic differences between blacks and whites that sort each race into separate neighborhoods (Logan and Alba 1993; Massey and Denton 1985; Yinger 1995). Although not formally testing these explanations in the analysis, I consider how a large-scale riot may change segregation levels through these mechanisms. Although each explanation is presented separately, a complex set of reinforcing variables rather than a single dominant factor likely work in concert to generate observed changes in segregation after a riot.
Neighborhood Racial Preferences
Two theoretical explanations for the existence of racial residential preferences—specifically, white negative stereotypes of blacks, and black fears of discrimination from whites (Krysan et al. 2009)—offer potential pathways linking large-scale riots to subsequent changes in segregation. Whites may prefer to live in predominantly white neighborhoods because they hold negative stereotypes of blacks or the types of neighborhoods in which they live. In the latter case, whites choose to live in predominantly white neighborhoods not necessarily to avoid contact with blacks, but because black neighborhoods are often associated with inferior schools, lower property values, and general disorder (Ellen 2000; Farley et al. 1978, 1994). Both forms of racial stereotyping are linked to Blalock’s (1967) group threat perspective, which asserts that prejudice is a response to feelings that dominant-group (whites) privileges are under threat by the rising presence of a subordinate group (blacks). This perspective is closely related to the literature on racial residential tipping, which posits that whites relocate after a certain percentage of black residents is reached (Schelling 1971).
The media coverage of the 1960s riots often portrayed blacks in a negative light (Abu-Lughod 2007; Monroy et al. 2004; Myers and Caniglia 2004), which included images of black rioters looting stores, burning buildings, and committing other acts of delinquent behavior, likely forging negative stereotypes of blacks in the minds of whites, many of whom already harbored anti-black prejudice before the riots (Sugrue 1996). Indeed, surveys of white residents after the Detroit riots displayed increasingly negative and racist evaluations of blacks and had exaggerated estimates of crime and other problems in racially integrated neighborhoods (Bledsoe et al. 1996; Farley et al. 1978; Fine 1989; Quillian and Pager 2001). Many observers stressed “white fear, distrust, dislike, [and] anger toward blacks” as causes for strained race relations after the riots (Fine 1989:369). The evidence suggests that riots strengthened the existing implicit association of disorder with blackness, a form of bias that has been linked to white residential decisions independent of observed disorder and influences outmigration well into the future (Bobo 2001; Quillian and Pager 2001; Sampson 2012).
The riots were also portrayed as acts of black rebellion against the white majority for greater black representation (Button 1978; Campbell et al. 2004), which mainstream whites perceived as a serious threat to their established status over minority groups (Thompson 1999). In a survey conducted shortly after the 1967 Detroit riots, 70 % of white respondents believed blacks were pushing “too fast” for equal representation (Fine 1989). The increased racial prejudice and threats to established privileges may have lowered white tolerance for black neighbors. In an analysis of white population changes in metropolitan areas from 1970 to 2000, Card and his coauthors (2008) found that tipping points—the minority share after which white demand for a neighborhood falls sharply—were lower in areas experiencing greater levels of riot severity.
Riots can also change blacks’ perceptions of whites. In this case, blacks’ concerns about possible discrimination in white neighborhoods shape residential preferences (Krysan and Farley 2002). Given the high level of violence committed by a predominantly white local police force and the National Guard during the riots, as well as the perceived slow or indifferent response by historically white institutions (e.g., U.S. government) in addressing the racial inequalities presumed to have sparked the riots (Bergesen 1982; Button 1978; Darden and Thomas 2013; Fine 1989), blacks may become more wary of whites. They may perceive an increase in white hostility and prejudice during their everyday or community encounters, through confirmation biases or other social-psychological tendencies to seek out evidence that confirms emotionally charged suspicions. In this case, blacks carry the stigma of the riots with them even though they did not personally participate; thus, they will avoid living in white neighborhoods to protect themselves from discrimination and potential backlash (Wacquant 2008). For both whites and blacks, the movement into separate neighborhoods perpetuates mistrust and prejudice, maintaining if not increasing segregation in the long term (Du Bois 1903).
Barriers to Entry
The theoretical perspectives outlined in the preceding section are linked to the place stratification model (Logan and Alba 1993). According to this model, segregation emanates from structural forces tied to racial prejudice that preserve the relative status of whites, thus limiting the ability of socially mobile blacks to enter white neighborhoods. A key structural force limiting black residential mobility is discrimination in the housing and mortgage markets. Although discriminatory policies and laws have been made illegal since the 1960s, a large body of empirical evidence supports the continued existence of racial discrimination in the rental and housing markets (Galster 1988a, b; Yinger 1995). Practices that form barriers to entry include organized hostility from white incumbent residents, racial steering from real estate agents, landlord price discrimination, low-density zoning, and the provision and location of public housing (Charles 2003; Massey 2005; Ross and Turner 2005; Rothwell and Massey 2009; Squires 1994; Turner et al. 2002).
Riots affect barriers to entry for blacks into white neighborhoods in a number of ways. Whites may fear that the violence of the riots and its perceived repercussions (e.g., increased crime, depressed housing values) will occur in their neighborhoods if blacks move in. This fear is directed toward all blacks, even those who have the socioeconomic standing to engage in upward residential mobility (Farley et al. 1978; Krysan et al. 2009). Consequently, the barriers to entry into white neighborhoods increase as white homeowners—specifically those living in suburban communities—actively support local sources of racial stratification in housing, such as low-density zoning, and engage in organized community and political efforts to keep blacks from moving into their neighborhoods (Massey and Denton 1993; Rugh and Massey 2014; Sugrue 1996). For example, white Detroit suburbanites after the 1967 riots created “such a siege mentality that they hired auxiliary police to patrol their streets and organized a wave of gun clubs” (Darden and Thomas 2013:8). Exploiting these racial fears, lenders and realtors will indiscriminately steer white clients toward less racially integrated neighborhoods and push black clients toward the inner city (Ross and Turner 2005). Similarly, landlords increase their preferences for white tenants in order to avoid racial conflict amongst residents and prevent potential decreases in rental costs and property values (Collins and Margo 2007; Muller 1981). As a consequence of these practices, blacks in riot-affected cities are either pushed toward or prevented from escaping the inner city, while whites leave for the suburbs to join efforts with incumbent whites to bar similarly fleeing blacks from following them.
The final hypothesis is that segregation results from racial differences in income or wealth, which is best understood from the spatial assimilation perspective (Alba and Logan 1991; Massey and Denton 1985). Under this model, blacks and whites sort themselves into neighborhoods based on economic resources, such as income, household wealth, and equity. Because blacks, on average, are more socioeconomically disadvantaged, they will move into poorer areas while whites move into more advantaged areas, leading to racially segregated neighborhoods. Evidence from household-level studies suggests that blacks are consistently less likely than whites to convert income gains into moves to more prosperous white neighborhoods, suggesting that racial differences in income explain only a small proportion of total segregation (Denton and Massey 1988; Fischer 2003; Wilkes and Iceland 2004).
Despite this finding, racial differences in socioeconomic resources still play a role in influencing segregation. A weak version of the place stratification hypothesis predicts that blacks are able to convert socioeconomic resources into locational attainments but that they achieve gains only in comparison with other minority group members. In this case, although middle-class blacks reside in communities that are less affluent than those of their middle-class white counterparts, they tend to be more integrated than poorer blacks (Darden and Kamel 2000; Denton and Massey 1988). In a series of related studies, Collins and Margo (2004, 2007) and Collins and Smith (2007) found that the 1960s riots had significant negative economic effects on black residents. Median black family income decreased by approximately 9 % in cities that experienced severe riots relative to those that did not, controlling for relevant city characteristics. Between 1960 and 1980, severe riot cities had relative declines in black male employment rates of four to seven percentage points. These researchers also found that the average decline in black-owned property values in riot-affected areas range from 14 % to 20 %. These findings indicate that black socioeconomic standing in riot-affected cities diminished significantly, increasing the economic gap between whites and blacks and likely strengthening the association between racial differences in income and residential attainment. As higher-income whites move into surrounding suburbs, increasing the economic barriers for blacks to exit the inner city, black residents become trapped in neighborhoods with declining property values and diminishing amenities (Wilson 1987). The downward economic spiral continues as the negative decline reinforces itself through the declining quantity and quality of public services, increasing crime, and diminishing tax revenues.
For this analysis, I examine changes in the segregation levels of the seven cities experiencing the most severe riots during the 1960s: Baltimore, Chicago, Cleveland, Detroit, Los Angeles, Newark, and Washington, DC. I measure riot severity using an index constructed by Collins and Margo (2007) that summarizes the number of deaths, injuries, arrests, arsons, and days of rioting in cities affected by a riot during 1964 to 1970. Data on each component are derived from a data set of riot occurrences constructed by Spilerman (1970, 1971, 1976) and modified by Carter (1986).1 Figure 1 compares the riot severity for the seven cities to all other cities experiencing riot activity during the 1960s. The mean severity for all other cities is 0.01, which is nearly 16 times smaller than the severity in Cleveland. The seven cities are in the top 2 % of the severity distribution and the value at the 95th percentile (0.07) is less than one-half of Cleveland’s value. The total number of riot-related deaths, injuries, arrests, and arsons in the seven cities account for 72 %, 50 %, 56 %, and 53 % of the total amounts of each indicator of riot severity, respectively. Based on these statistics, the riots experienced by the seven cities examined in this analysis are among the most destructive in modern U.S. history.
In my analytic procedure, I compare each city affected by a large-scale riot to a pool of demographically comparable control cities. It is important to restrict the control pool to cities with outcomes that are driven by similar structural processes as for the riot-affected cities. Therefore, I limit the control sample to the 43 cities that have a population greater than 100,000 in 1960, are not in the same metropolitan areas as the seven affected cities, have a black population that is both greater than 1,000 and makes up more than 5 % of the total population in 1960, and are not missing data on the dependent variables and any of the control variables. A list of these cities and their pre-1970 characteristics can be found in Table S1 of Online Resource 1. Although many of these cities experienced some riot activity during the 1960s, their average riot severity is quite small (0.03).
The indices range between the values of 0 and 1, with greater values indicating higher levels of segregation. They were constructed using decennial census data going back to 1920 and up to 1990.2 A neighborhood is defined as a city ward from 1920 to 1940 and a census tract from 1950 to 1990. Ward-based indices tend to fall below tract-based indices because wards are larger. To correct for this discrepancy, Cutler et al. (1999) suggested using correction factors of 0.152 and 0.157 for ward-based dissimilarity and isolation values, respectively.
Segregation is defined at the city level from 1920 to 1950 and at the metropolitan statistical area (MSA) level thereafter. Cutler et al. (1999) found that when both city- and metropolitan-level data are available, city- and MSA–level segregation indices do not significantly differ. MSAs designated as consolidated MSAs (CMSAs) are broken up into their primary components (e.g., PMSAs). For example, the New York-Northern New Jersey-Long Island CMSA is broken up into the New Haven (CT), New York City, Newark (NJ), Trenton (NJ), and Jersey City (NJ) PMSAs. To control for boundary changes, MSA boundaries are normalized; however, the availability of city boundary maps going back to the 1920s is limited, and thus city-level boundaries are defined for each census year.
In addition to 1920–1960 values of segregation, I control for city-level variables that are commonly included in studies of riot causes and shown to produce the most straightforward findings. These variables should capture otherwise unobserved trends and factors influencing both riot occurrence and segregation levels. All variables are measured at the city level in 1960 unless otherwise noted. I control for variables that measure a city’s housing conditions (percentage dilapidated housing in 1950); absolute (nonwhite median income) and relative (nonwhite median income divided by white median income) deprivation; percentage foreign-born; percentage of 25-year-olds and older with a high school diploma; population density; a binary variable indicating whether the city resides in the South; and political structure (councilmen per 10,000 population, percentage of city council elected at large, presence of nonpartisan elections, and presence of mayor-council government). Following Farley and Frey (1994), I also control for a city’s functional specialization in 1960, characterized as manufacturing (proportion of employment in manufacturing industries), military (proportion of the male labor force enrolled in the armed forces), retirement (proportion of the population 65 years and older), and educational (proportion of 5- to 34-year-olds enrolled in college). I also control for population size and percentage nonwhite in 1920, 1940, and 1960.3 Demographic data are collected from official census bound periodicals, the City and County Data Books, and the Governmental Units Analysis Data (Aiken and Alford 1998). Table 1 presents summary statistics for the seven riot-affected cities.
To assess whether segregation in riot-affected cities changed in the post-riot period relative to pre-riot levels, I need to identify comparison cities that chart the counterfactual path for each of the seven riot-affected cities. In this study, I employ the synthetic control method developed by Abadie et al. (2010), a data-driven procedure that constructs a comparison group using pre-riot population characteristics and trends. The main idea of this method is to compare each city in the riot-affected group (treatment) with a weighted average of cities not affected by a large-scale riot (control). The weight attached to each control city is based on how closely the city resembles the treated city on the outcome and across selected demographic variables during the pre-riot period. The effect of the riot is measured as a function of the difference between the segregation levels of the city and its synthetic match after the riot relative to pre-riot differences. Given that the riots occurred between the years 1964 to 1969, the pre- and post-riot periods are 1920–1960 and 1970–1990, respectively.
Formally, the synthetic matching procedure is as follows. The observation pool consists of N cities separated into treatment and control groups, with J treatment cities and N – J control cities. Suppose we observe these cities at each decennial year t over T time periods, where Tpre designates the pre-treatment period, and t = 1970, 1980, and 1990 designate the individual post-treatment years. For treatment city j, let be a matrix containing pre-treatment covariates X and pre-treatment outcomes Y. Let contain the same variables, but for the entire set of N – J potential control candidates cr. The synthetic control method identifies a convex combination of the N – J cities in the control candidate pool that best approximates the pretreatment data matrix for the treated city.
The goal is to construct a N – J × 1 weight vector W = (w1, w2, . . . , wN – J), such that ∑i = 1N − Jwi = 1 and wi ≥ 0 for i = (1, . . . , N – J), that will be used to combine all N – J control units into a single unit, known as a synthetic match. The procedure chooses values for the weight vector W such that they minimize the distance between and . Specifically, W is estimated by solving , where the values V are set to minimize the mean squared prediction error of the outcome variable during the pre-riot period. The values of W yield a synthetic comparison group that best approximates the pre-riot period for the treatment city.
The synthetic control procedure has important advantages over standard regression and matching techniques (Abadie et al. 2010, 2015). First, the method controls for unobservable factors that have an effect on the common time trend of samples in the treatment and control groups. Second, the method is most useful for comparative case studies whereby the units of analysis are aggregate entities small in number with no natural comparison groups. Third, Hainmueller (2009) described three advantages of synthetic matching over popular matching methods like propensity score weighting: (1) it achieves the highest possible covariate balance; (2) it fully exploits all known features of the covariate distributions; and (3) rather than employing a rigid weighting scheme that forces units to be either fully matched or discarded, it frees the weights to vary smoothly across control units. Last, by using a weighted average of units as a comparison, where each weight is restricted to values between 0 and 1, the procedure precludes the type of model-dependent extrapolation on which regression analyses are often based.
Moreover, Abadie et al. (2015) argued that in contrast to regression analysis techniques, the synthetic control method makes explicit not only which cities are being compared with each treated unit but also the weight with which each of the control units are factored into the comparison, allowing researchers to assess the fit between the case of interest and its comparison units. Table 2 displays the cities receiving positive weights in the construction of a synthetic control for each riot-affected city and segregation measure. The weights represent the percentage of the synthetic control unit that each control city constitutes based on their similarity to each riot-affected city during the pre-riot period. For example, Milwaukee (WI) and New York City make up 60 % and 40 %, respectively, of Chicago’s synthetic unit for the dissimilarity index. Every other city in the control pool received a weight of 0. Milwaukee makes for an appropriate match for Chicago given that it is also a large Midwestern city with a long history of black segregation. Similarly, New York is an appropriate match given its comparable political, economic, and social characteristics. The synthetic groups for Detroit and Cleveland are largely made up of demographically similar cities in the Northeast or Midwest. Newark and Baltimore’s synthetic groups are similarly constructed, but with a larger presence of nearby Northeastern cities for Newark and Southern cities for Baltimore.4 As an area that borders the traditional South, Washington, DC, had a relatively large black population in the early half of the century. Hence, two historically black Southern cities, Memphis (TN) and Richmond (VA), make up most of its synthetic control. Although pre-riot segregation trends for Los Angeles and its synthetic control group are similar, many of the control cities are not intuitive matches. Potentially more suitable candidates (e.g., cities in the West and Southwest) are not a part of the control pool because they are missing segregation data for one or more years in the earlier portion of the pre-riot period. As a robustness test, I limit the pre-riot period to 1950 and 1960 in order to include these cities.
Similar control cities are often found across the two outcomes and the seven cities. For example, the four cities constituting Cleveland’s synthetic unit for the isolation index are also a part of its synthetic unit for the dissimilarity index. The seven cities share several control cities from the Midwest and Northeast that have substantial black populations and high segregation levels. In fact, the four cities that make up America’s Ghetto Belt that did not experience a large-scale riot—New York City, St. Louis (MO), Gary (IN), and Milwaukee (WI)—are common controls.
To formally test the significance of , I use the exact permutation test recommended by Abadie et al. (2010). I can map out the distribution of the null hypothesis of a no-riot effect by assigning a large-scale riot to each control city. I calculate the DID estimate in Eq. (3) for each control city and find where for each riot-affected city lies in that distribution. If it lies somewhere in the extreme, I can reject the null of a no-riot effect because the observed is too large relative to what I would see if control cities experienced a riot. Formally, the cumulative density function of the complete set of DID estimates, which includes the estimated effects of a 1960s large-scale riot on riot-affected city j and each control city, is given by Fjt(.). The p value from a one-tailed test of the hypothesis that > 0 for each riot-affected city j is given by , where t equals each decennial year in the 1970–1990 period.
I begin with a graphical presentation of the segregation trends for the seven riot-affected cities and their synthetic matches. Figure 2 graphs the dissimilarity index for each riot-affected city and their synthetic control group from 1920 to 1990. Focusing first on the pre-riot period 1920–1960, the figure reveals that the trend for each city closely matches the corresponding trend in their synthetic control group. The root mean squared error (RMSE) of the pre-riot difference between each city and its synthetic control is relatively small (0.026). Hence, the pre-riot segregation values for the riot-affected cities and their synthetic control groups match quite well.
In the post-riot period 1970–1990, Fig. 2 reveals very small differences in segregation levels between Chicago and Baltimore and their synthetic matches. The gaps between Los Angeles and Washington, DC, and their synthetic controls increase from 1960 to 1970 but decrease in 1980 and 1990, indicating short-term effects. The graphs for Cleveland, Detroit, and Newark indicate both short- and long-term effects.
Figure 3 shows trends in the isolation index. Similar to the dissimilarity index, I find that the cities and their synthetic controls match well in the pre-riot period (RMSE of 0.07) except for Chicago. The pre-riot gaps between Chicago and its synthetic match are large, particularly in the earlier portion of the period, because Chicago’s isolation levels in those years are substantially higher than the rest of the nation. As a robustness test, I limit the pre-riot period to 1950 and 1960 in order to match on years during which the gap between Chicago and other cities is much smaller. Results of this test and other robustness checks are provided in the following section.
In the post-riot period, I find small differences in the isolation index between Baltimore and its synthetic control. Chicago has noticeable gaps in all post-riot years; however, a large post-riot difference is not indicative of a large effect if the synthetic control does not closely reproduce the outcome of interest prior to the riot. For Los Angeles, the gap increases in 1970 but subsequently goes back down. Detroit, Cleveland, Washington, DC, and Newark show gaps in all post-riot years.
Rather than eyeballing differences, we can formally estimate them and test their statistical significance. Table 3 presents synthetic control DID estimates and their associated one-tailed p values of a large-scale riot’s effect on dissimilarity and isolation levels for each census year in the post-riot period. The p values are calculated from the ranks of the estimates for each city relative to the complete distribution of 44 estimates (one for each riot-affected city and 43 placebo estimates). For comparison, I also show DID estimates using the four cities in the Ghetto Belt (GB) that did not experience a large-scale riot (Gary, Milwaukee, New York, and St. Louis) as the control group. Detroit experienced a positive and significant effect in all post-riot years, growing from 0.055 in 1970 to 0.119 in 1980 and 0.122 in 1990. Los Angeles and Washington, DC, witnessed a 0.053 increase in their dissimilarity levels in 1970; however, the effects disappeared in 1980 and 1990. The other cities experienced no significant changes.
I find similar results for the isolation index. I find a significant effect on 1970 isolation indices for the four cities experiencing the most severe riots, increasing by 0.065, 0.088, 0.064, and 0.084 in Los Angeles, Detroit, Washington, DC, and Newark, respectively. Similar to dissimilarity, isolation increased in 1980 and 1990 only in Detroit. In comparison with estimates using the GB cities as the control group, the synthetic control estimates are generally smaller, particularly for the dissimilarity levels in Newark and Washington, DC; however, the general patterns and trends are largely consistent.
In sum, large-scale riots are associated with short-term increases in the segregation levels in Detroit, Los Angeles, Washington, DC, and Newark. The effects persist into 1980 and 1990 only for Detroit. The slight differences detected between Baltimore, Chicago, and Cleveland and their comparison groups in Figs. 2 and 3 are not statistically significant at conventional levels.
In this section, I probe the robustness of the main findings presented in Table 3 using a set of falsification exercises or placebo studies recommended by Abadie et al. (2015). The results of these robustness checks for dissimilarity and isolation are shown in Tables 4 and 5, respectively. The first two robustness checks explore whether the estimation results are sensitive to the definition of the pre- and post-riot periods. The top panel in each table shows results for a model that assigns the riots two decades before (1940s) their actual occurrence (1960s). Here, the pre-riot period is 1920–1940, and the post-riot period is 1950–1990. The model acts as a falsification test in that the validity of the main results would dissipate if the synthetic control method also estimated similar effects when applied to a period when the riots did not occur. Furthermore, starting the post-riot period in the 1940s tests for the effects of the 1943 race riot in Detroit. If we find an increase in segregation in Detroit, the significant effect uncovered in the main analysis may be partially or entirely due to the riot in 1943. I find that the effects across all cities, years, and outcomes are not statistically significant. That is, in contrast with the actual riots that occurred during the 1960s, the 1940s placebo riots have no perceivable effects.
The second robustness check determines whether the results are sensitive to a different grouping of the pre-riot comparison years in the DID specification: namely, 1950 and 1960. Similar to the main analysis, the synthetic controls are based on 1920–1960 characteristics; however, in calculating the DID estimates in Eq. (3), I replace and with and , respectively. By comparing against 1950 and 1960 values, the estimates are based on segregation levels temporally closer to the date of the riots. The middle panel shows results for these models. I find only slight differences between these results and the main findings.
The final robustness check expands the potential control pool by matching cities using just 1950 and 1960 characteristics. In the main analysis, several cities are excluded from the control pool because they are missing segregation values for at least one year between 1920 and 1940. Beginning in 1950, with the transition from ward- to tract-based measures of neighborhood and the creation of MSAs, a larger number of cities (54) have segregation data. More importantly, the expanded control pool includes potentially more intuitive matches for Los Angeles, the only riot-affected city located in the Western region of the United States, which is an important distinction considering the substantial declines in black segregation levels since the 1960s in the West relative to the rest of the country (Iceland and Sharp 2013). When matching only on 1950 and 1960 characteristics, the cities constituting Los Angeles’ synthetic match are New York, Milwaukee, Houston (TX), and Denver (CO) for dissimilarity, and Houston, New York, Kansas City (MO), and Denver for isolation. These control units are either large cities with a long history of segregation (e.g., Milwaukee) or cities located in the West or Southwest with similar demographic characteristics and trends (e.g., Houston). Limiting the pre-riot period to 1950 and 1960 also matches on years during which there is a closer fit between Chicago and its synthetic control, and restricts the analyses to segregation indices calculated only from tract-level data, which controls for the possibility that the results are merely a by-product of the differing levels of aggregation at which the segregation indices were measured. The last panel of Tables 4 and 5 show the synthetic DID estimates using the expanded control pool. Although the effect sizes are slightly different—specifically, larger effects for both indices in 1970—the results do not markedly differ from the main findings.
Segregation in Suburbs Versus Central Cities
The aggregate patterns presented in the main findings may be masking trends manifested at a lower geographic scale. In particular, the effects of a large-scale riot may differ between central cities and suburbs, which is a relevant concern given the rise in suburbanization after the 1960s. Many studies have reported lower average levels of segregation from whites experienced by blacks in suburbs relative to central cities (Adelman 2004; Logan 2001; Massey and Denton 1988b). However, segregation levels in suburbs compared with central cities have either been declining at a significantly slower rate or increasing since the 1960s (Fischer 2008; Lichter et al. 2015). Furthermore, Fischer et al. (2004) found that the distinction between central cities and suburbs has decreased in importance for black segregation since the 1970s, while the importance of distinctions among suburbs has increased. In this case, white suburbanization leads to movement into affluent, largely white neighborhoods, and black suburbanization leads to movement into older, inner-ring suburbs that are less affluent and white (Hall and Lee 2010; Logan and Schneider 1984). If whites in riot-affected areas are moving out of central cities and mobile higher-income blacks follow them but to poorer, more-black suburban neighborhoods, we should expect to see significant increases in suburban segregation levels. In contrast, the central cities of the seven affected areas may experience a smaller impact given that they were already highly segregated before the riots.
To test this hypothesis, I calculate separate synthetic control DID estimates for the central city and suburbs for each riot-affected area. The analysis is conducted at the MSA level and matches on the same pool of control units as in the main analysis.5 Following the U.S. Census Bureau, suburban locations are defined as tracts lying within the metropolitan boundaries outside the central-city core (U.S. Census Bureau 2002). Because MSA–level data are largely not available until 1960, each riot-affected area is matched to other MSAs using only 1960 characteristics.
Tables 6 and 7 show the effects of a large-scale riot on dissimilarity and isolation, respectively, in the central city (top panel) and suburbs (bottom panel) by post-riot year. Large-scale riots are associated with increases in suburban dissimilarity or isolation levels in all areas. For the dissimilarity index, I find that these effects occur later in the period (i.e., 1980 and 1990) except for Los Angeles, Washington, DC, and Baltimore, which experienced increases only in 1970; and Detroit, which experienced both short- and long-term changes. For the isolation index, segregation increased in every post-riot year for every city except Washington, DC, Baltimore, and Los Angeles, which experienced increases only in 1970. For central cities, increases in dissimilarity and isolation occurred only in Detroit and Washington, DC, with the latter experiencing changes only in 1970.
Discussion and Conclusion
In this study, I examine black segregation trends in the seven cities experiencing a large-scale riot during the 1960s. I find that riots were associated with increases in segregation in four of the seven cities in 1970, but the association disappeared in 1980 and 1990 except in Detroit. These results indicate limited effects, but they mask larger and longer-term increases in the suburbs, which accord with recent findings that reveal the growing importance of within-suburb segregation in explaining overall metropolitan segregation (Fischer et al. 2004; Lichter et al. 2015). In this case, suburban racial separation occurs when whites move into affluent outer-ring suburbs while blacks relocate into poor, increasingly black inner-ring suburbs (Logan and Schneider 1984). Central cities were already highly segregated before the riots, with blacks predominantly living in the ghetto and whites living in neighborhoods adjacent to the ghetto. The relocation of whites out of the central city was followed by a process of racial succession such that the ghetto expanded into inner-ring neighborhoods (Bader and Warkentien Forthcoming). Fischer (2008:492) concluded that these trends “may signal reconfigurations of suburban spaces in the face of rapid diversification, lending support to the place stratification model and related hypothesis of white tolerance thresholds for contact with Blacks.” Through the mechanisms described in the beginning of this article, the large-scale riots of the 1960s may have reduced these white tolerance thresholds, hastening white flight into suburban municipalities, which are ideal places for protection against real and perceived threats of increasing black presence and political representation because their political autonomy allows for practices that exclude minority groups, such as land-use regulations and zoning ordinances (Frey and Farley 1996; South et al. 2011). Moreover, the negative effects of the riots on black socioeconomic levels further limited the geographic reach of black suburbanization.
Note that the long-term increase in the suburbs applies to riot-affected cities located in the Northeast and Midwest. For suburban neighborhoods in Baltimore, Los Angeles, and Washington, DC, large-scale riots were exclusively associated with short-term changes. Several macro-level factors emerging after the 1970s may help explain these differences. First, Los Angeles and Washington, DC experienced rising levels of Hispanic and Asian populations (Frey and Farley 1996; Logan and Stults 2011; Manning 1998), which decreased black segregation by increasing exposure not only to nonblack minorities but also to whites through the production of “global communities” (Frey and Myers 2005; Logan and Zhang 2010). During the same period, Washington, DC’s central city rapidly gentrified, causing segregation in the urban core to subsequently decrease (Lee et al. 1985). Also, because of annexation and countywide governance, exclusionary land policies restricting blacks from moving into more-white neighborhoods were less prevalent in western cities, such as Los Angeles. Finally, scholars attribute suburban integration in Baltimore and Washington, DC, to the encroachment of white suburban development into rural black areas (Farley 1970; Schnore 1973).
The differential geographic trends accord with the prevailing distinction made in the segregation literature between Midwestern and Northeastern cities like Detroit and Newark (i.e., America’s Ghetto Belt) and the rest of the country. Since the turn of the century, these cities have housed a disproportionate share of poor urban blacks, have longer histories of racial animosity and unrest, and contain more racially entrenched neighborhoods. Large-scale race-related riots occurring in such hypersegregated environments may produce stronger effects. For this reason, the consequences of large-scale riots must be understood within the particular context and conditions from which they emanate. The case study approach employed in this article captures this exceptionalism.
In this regard, the large and long-term increases in Detroit may be a product of both its hypersegregated setting and the overall severity of its riots. Detroit is one of the most popular locations in the country for studying segregation, racial conflict, and related phenomena (e.g., Darden and Kamel 2000; Darden et al. 2010; Farley et al. 1978, 1979; Galster 2012; Sugrue 1996), including detailed attention to the city’s 1967 riots (e.g., Boyle 2001; Button 1978; Darden and Thomas 2013; Fine 1989; Singer 1970; Thompson 1999). The findings of this study only fortify the general consensus among scholars that the riots and, more broadly, other post–World War II factors, such as the deindustrialization of Detroit’s central city and the dismantling of its major employer, the automobile industry, had an immense impact on the city. Moreover, it is the interaction between contemporaneous conditions and the sheer destruction of its riots that further separated Detroit from the rest of the sample. Figure 1 shows that Detroit and Los Angeles experienced riots that were nearly twice as severe as the other riots examined in this study. Although the impact of the riots in Los Angeles may have been minimized by the factors outlined earlier, the effect on Detroit was enhanced by the conditions described in this paragraph.
The findings are subject to a number of caveats. First, the synthetic control methodology estimates only reduced-form aggregate associations of riots on segregation without identifying the channels of transmission. As a result, the findings cannot directly speak to the specific mechanisms through which riots influence segregation. Second, because individual years of data in between each decennial census are not available, the 1960–1970 period represents a uniform treatment across all seven cities. It might well be that other events that occurred during the 1960s are driving the results, which is an important consideration given that the decade was a period of significant change in the United States. Finally, given the sheer destruction and broad geographic reach of the 1960s riots and the unique circumstances surrounding each affected city during the period, we must be careful to not generalize the results to all types of urban race riots, particularly those that occur outside the United States or are generated through formalized demonstrations where a critical mass of individuals have already congregated. These caveats should be borne in mind when interpreting the results.
Despite these caveats, the findings provide several contributions to the current literature on segregation and civil disorder, offering insight on the recent urban unrest in Ferguson and Baltimore. The current study’s findings along with previous results showing the negative impact of riots on black socioeconomic outcomes and their limited effects on black political representation offer evidence against popular assertions (e.g., Cunha 2014) that violent civil disturbances are viable means for advancing black living conditions. Further examinations of other social outcomes, particularly those that precipitate riot activity, are needed in order to fully grasp the complete impact of large-scale race rioting. Relatedly, understanding the consequences of riots also contributes to the understanding of their causes. If the factors causing riots increase in magnitude, future civil unrest is likely not just in the affected city but also in other, similar areas (Myers 2000). For segregation, the findings reveal that riots have largely short-term effects, with the longer-term effects concentrated in the suburbs of historically hypersegregated cities, indicating that poor, black suburban neighborhoods, which have increased in prevalence since the 1960s, may be sources for future conflict. For example, although much attention was given to the high segregation levels in St. Louis (e.g., Jones 2014), the epicenter of the riots that began after the fatal shooting of Michael Brown was in Ferguson—a poor, predominantly black inner-ring suburb of the St. Louis metropolitan area. The increasing importance of suburban neighborhoods does not mean that segregation in central cities, particularly for the seven areas examined in this study, has diminished to ideal levels. Instead, the takeaway is that black suburban neighborhoods, as an additional source of racial stratification in the metropolitan landscape, warrant further attention.
The reaction to the riots in Ferguson and Baltimore primarily centered on understanding why they happened and not what they affected. The few studies that have examined the effects of riots have taken an economic perspective, which parallels the economic post-riot recovery efforts by communities and the local and national government after the 1960s riots. However, the results of this study suggest that the effects of a riot go beyond the local economy. Recovery efforts should also take into consideration the repair and improvement of frayed race relations in affected communities, which are often cited as likely factors sparking large-scale riots. Rebuilding efforts, therefore, must be deeper and wider in scope than simply hiring more police and repairing damaged physical infrastructure.
I thank Matthew Andersson, Shai Dromi, Michael Hout, Elizabeth Roberto, Sam Stabler, seminar participants at the 2014 Annual Meeting of the Social Science History Association, and anonymous reviewers for comments on earlier drafts of this manuscript.
Following the empirical literature on riots, I define a riot as a “spontaneous event” with at least 30 participants that resulted in property damage, looting, or other aggressive behavior.
I begin the pre-riot period in 1920 because many cities were missing segregation data in prior census years. Data from 1920 to 1950 come from Cutler et al. (1999). The 1960 tract-level data come from the Bogue Data Files (Bogue 2000). The 1970–1990 tract-level data were extracted from the GeoLytics Neighborhood Change Database (GeoLytics, Inc. 2002). For cities missing 1950 values, I interpolated values using 1940 and 1960 data.
I exclude values in 1930 and 1950 to prevent overfitting. Additionally, rather than matching on the segregation values for each pre-riot year, which can introduce bias and other undesirable results (Kaul et al. 2015), I use the average segregation for the first two periods (1920 and 1930) and each individual year thereafter, which maintains the integrity of using all pre-riot years in the matching process but places emphasis on the individual decennial years immediately preceding the riots.
Maryland and Washington, DC, are designated Southern states by the Census.
Two MSAs were omitted because all their tracts comprised their central cities in 1960.