From Side Street to Ghetto: Understanding the Rising Levels and Changing Spatial Pattern of Segregation, 1900–1940

Abstract

A standard interpretation of the intensification of segregation in the early twentieth century is that residents of Northern cities reacted against a growing African American presence, using segregation as a tool of social control that was less needed in the South. Evidence from newly available data for 134 cities in 1900–1940 puts this interpretation in question in several ways. We find that segregation was already high in 1900 at the neighborhood scale. Not only was it rising, but it was changing its spatial scale as clusters of Black settlement in side streets and alleys disappeared from White districts while expanding into large Black zones. Finally, multivariate analyses show that trends were similar in the North and South, and in neither region was Black population size (i.e., “Black threat”) a significant predictor of increasing segregation. The general trends of rising segregation and increasing spatial scale became a nationwide pattern.

It is now well established that Black-White segregation was high in North American cities as early as 1880. Observations based on ward-level data had led urban researchers to conclude that segregation was quite modest at that time, and that it was only after the initial wave of the Great Migration that the urban Black population became highly segregated (Massey and Denton 1993: 26–42; see also Cutler, Glaeser, and Vigdor 1999; Lieberson 1963). Grigoryeva and Ruef (2015, see also Logan and Parman 2017) tabulated data from 1880 on the race of next-door neighbors, revealing very high segregation at that micro-scale. Logan, Zhang, and Chunyu (2015) reinforced this finding in a study of Chicago and New York (1880–1940) with data at the scale of enumeration districts (EDs), which are comparable to contemporary census tracts. Logan, Zhang, Turner et al. (2015) replicated their study in 10 large Northern cities for 1900–1930. The conflicting results from earlier and more recent research draw attention to how much our impression of segregation hinges on the scale at which we measure it.

We extend these studies to a large national sample of cities. We show that segregation increased continuously from 1900 through 1940 not only in large migrant destinations but also in both the North and South, and in cities with few or many African American residents. We demonstrate that the spatial scale of segregation was also changing in this period. At the end of the nineteenth century, the predominant form was clustering of African Americans in specific small areas that were in close proximity to the homes of Whites. This was the usual pattern in the South in 1880, where Logan and Martinez (2018) used geocoded census data to show that Blacks often lived in alleys, backyard housing, and side streets with few White residents while larger neighborhood areas were racially mixed. In Philadelphia, at the same time, African Americans lived disproportionately in alleys and short streets (Logan and Bellman 2016). But by 1940, we will show that these dispersed small settlement areas had been replaced by large and growing areas of cities that were becoming nearly wholly Black, a pattern that urban scholars began to refer to as Black ghettos.

THE CHANGING SPATIAL SCALE OF SEGREGATION

Contemporary studies measure segregation at the census block scale (often around 300 residents) or census tract scale (typically around 3,000 residents). Disproportionately Black “neighborhoods” usually cover much larger zones of racial clustering, zones of a magnitude that observers have described as “Black ghettos” like Harlem or Chicago’s Black Belt (Drake and Cayton 1945; Myrdal 1944). If by definition segregation refers only to separation at that scale, then cities were only modestly segregated at the beginning of the twentieth century. That is the conclusion of initial studies of this early period (Cutler et al. 1999; Massey and Denton 1993). These researchers knew, of course, that segregation indices would be higher if measured for smaller units. For this reason, they added a fixed amount (.15) to the Index of Dissimilarity ($D$) that they calculated at the ward level. They then treated ward-level indices as a proxy for tract-level segregation, which is the scale that they wished to study. This technical adjustment relies on the untested assumption that the difference between tract and ward measures was (1) constant over time and (2) invariant across cities. Philpott (1978) raised early objections to the use of ward data. In the case of Chicago (pp. 120–121), he noted that the 1900 ward map showed Blacks “scattered over” many areas of the city. Census tract data, he argued, showed that, in fact, “the residential confinement of the Blacks was nearly complete at the turn of the century.” One reason to study segregation trajectories with data at multiple spatial scales, as we do here, is to test Philpott’s observation in other cities.

Another motivation is the conjecture that the predominant spatial scale of segregation changed in the 1900–1940 period. Others have argued that different cities are segregated at different scales (Reardon et al. 2008), and evidently the scale can also vary over time. Possibly, for example, what was once a pattern of segregated side streets and alleys was replaced by a pattern dominated by large Black or White districts, possibly not only in the South but more generally across the country. We examine this question by gathering data for a large sample of cities for each decade and at varying spatial scales. We also carry out a spatial decomposition of segregation, examining how much of total segregation is uniquely contributed at each spatial scale and in each year. Such questions are arising more frequently in recent studies. For example, Reardon, Yun, and Eitle (2000) ask to what extent school segregation is due to differences between school districts or to segregation within school districts. This distinction matters because courts have generally held that only within-district segregation can be challenged on constitutional grounds. In a similar way, Parisi, Lichter, and Taquino (2011) ask to what extent residential segregation in the entire United States is attributable to separation at a macro level, such as between regions or between central cities, suburbs, and rural fringe areas, and to separation at the scale of blocks within cities or suburbs.

If the scale of segregation within cities was rising, this change could have multiple sources. One is that the outer boundaries of cities were expanding, and former suburban communities were becoming integrated into the urban fabric (Warner 1978). This process was well documented before 1900. By the 1910–1920 decade, population was growing more rapidly in the suburban fringe than in central cities (Bogue 1953). As the walking city was transformed by new modes of public and private transportation, large new areas were opened for residential construction, and many of these were segregated through mechanisms such as restrictive covenants. Another possible source is urban redevelopment within existing neighborhoods. For example, the alleys in Washington, DC near the Capitol were once used as cheap housing for African Americans in the nineteenth century (Logan and Martinez 2018). As automobiles were increasingly used by commuters, these inner areas of large blocks became more profitable to landlords as parking lots for commuters, pushing the Black population out of the alleys and, consequently, out of that part of the city. Yet another contributor could be the usual process of expansion of Black settlement areas. As congestion in established Black neighborhoods reached a breaking point, and despite being hemmed in by White community opposition to racial integration in surrounding areas, they often pushed their boundaries outward in a block by block fashion, establishing an ever-growing spatial footprint (Massey and Denton 1993: 37–38). As a result of changes like these in the spatial structure of cities, even as separation at a more microlevel was becoming more acute, it could also become visible at a much larger scale.

Previous studies have not emphasized variability in the spatial scale of segregation. They could not because they lacked data at multiple scales in this period. Yet, Duncan and Duncan (1955: 215) observed long ago that “it is common knowledge that in some cities—e.g. Chicago—the non-white population is predominantly clustered in a ‘Black Belt,’ whereas in other cities nonwhite occupancy takes the form of scattered ‘islands’ or ‘pockets.’” We posit that these two spatial forms may have tended to appear in a sequence over time in many cities, where the initial “pockets” are replaced by “Black Belts.” In this sense, the ghetto may be attributable not only to the rising levels of segregation in U.S. cities after 1900 but also to a qualitative change in the spatial configuration of Black settlement.

SOURCES OF THE RISING LEVEL OF SEGREGATION

Much research has focused on institutional sources of rising segregation, including restrictive covenants, exclusionary zoning, and redlining (Gotham 2002; Massey and Denton 1993; Rothstein 2017). Lacking systematic information on these phenomena for the pre-1940 decades, we do not address them here. Instead, we focus on the predictors that have been invoked in previous studies of this period. One is the difference between the South and non-South, which does raise the question of institutional differences associated with the South’s Jim Crow regime. The other is the size of the Black population, which has been interpreted in terms of the threat that minorities pose to White dominance (Blalock 1956).

Regional Differences in Segregation

The main substantive questions here are (1) whether the South, compared to the North, was less segregated (possibly because it relied more on Jim Crow boundaries in other spheres of life), and (2) whether the South was more segregated at a micro-level but less segregated at larger scales. The evidence is not conclusive. Some analyses are available only for one point in time; others span a 40- to 50-year period but they rely on varying measures and varying spatial scales.

Massey and Denton (1993, in a retabulation of data from Taeuber and Taeuber 1965) show that block-level White-non-White segregation in 1940 was higher in Northern cities ($D$ = 89) than Southern cities ($D$ = 81). However, the original table for 109 cities presented by Taeuber and Taeuber (1965: 44) reports that the average for Southern cities in 1940 was 84.9, compared to 83.2 in the Northeast, 88.4 in the North Central, 82.7 in the West, and 85.2 in all regions combined (see also Cowgill 1956). Taeuber and Taeuber point out, in addition, that the Southern city average remained similar to the national average in 1950 (a single point higher) and had reached more than four points higher by 1960. For the 1940–1960 period, therefore, there is no empirical support for the hypothesis of Southern exceptionalism.

More relevant here is the experience in earlier decades. Based on ward data, Cutler et al. (1999: 464) reported almost no difference in average $D$ for their samples of matched cities in 1890 (averaging 24 in the South verses 24 in the Northeast and 28 in the Midwest) or in 1940 (46 in the South, 45 in the Northeast, and 49 in the Midwest).

Other studies bear on the question of whether the scale of segregation was distinctive in the South. Grigoryeva and Ruef (2015) found average values of $D$ at the ED scale for 1880 were lowest (25) in the South, and higher in the Northeast (47) and in the Midwest and West (37). In contrast, in the same year but using an alternative “next-door neighbor” measure of segregation in 1880 (the Sequence Index of Segregation, SIS), they reported an average value of 53 in the South, 43 in the Northeast, and 32 in the Midwest and West combined. In their view, the ED measure was unable to pick up the finer “street-front” segregation that was typical of Southern cities, but the SIS revealed that segregation at this scale was actually highest in the South.

Logan and Parman (2017) also calculated a measure similar to the SIS that could reveal segregation at a next-door neighbor scale. This measure increased between 1880 and 1940, but in both decades it was higher in the South (50 in 1880, 87 in 1940) than in the North (39 and 75, respectively). They also conducted analyses at the more usual ED scale. They reported a mean value of $D$ equal to 29 in the average Southern city in 1880, compared to 52 in the Northeast and Midwest. That is consistent with the expectation of lower segregation in the South. But by 1940, the Southern average had jumped 45 points to 74, while increasing in the Northeast and Midwest only 32 points to 84, which suggests substantial regional convergence in this period.

“Group Threat” as a Predictor of Rising Segregation

Another explanation of rising segregation is the hypothesis of group threat on which Massey and Denton (1993: 26–42) rely in their discussion of the construction of the ghetto during 1900–1940. In their view, the main causal force was White hostility to Black population growth. This was a new phenomenon in Northern cities (following Massey and Denton, we use the term “North” hereafter to refer to the Northeast, Midwest, and West regions), which historically had modest Black populations. Here, “as the size of the urban Black population rose steadily after 1900, white racial views hardened and the relatively fluid and open period of race relations in the North drew to a close” (Massey and Denton 1993: 30). Even in the South, where many cities had much higher Black population shares in the range of 30 to 40 percent after the Civil War, “whites were not completely immune to threats posed by Black urbanization. After 1910 Black populations also began to rise in southern cities … and whites similarly became alarmed at the influx of Black migrants” (1993, p. 41). They posit important regional differences, however. Segregation was lower and rose less rapidly in the South because “the Jim Crow system of race relations … guaranteed the subordination of Blacks and paradoxically lessened the need for a rigid system of housing segregation.” In addition, the existing “traditional grid pattern of white avenues and Black alleys [in Southern cities] kept segregation levels relatively low” as measured at the scale of city wards. In the South, unlike the North, Whites and Blacks were more mixed in every major city district, but highly separated at the scale of “avenues and alleys.” High segregation at a more micro-scale lessened the impulse for imposing greater residential separation in the South.

In short, in this reading, segregation was low in 1900 and increased in proportion to the threat of Black population growth. It was especially low and lagging behind in the Jim Crow South, despite the much larger Black populations in Southern cities, though here, too, it rose in response to Black urban growth. But “[a]lthough the walls of the ghetto were rising in the south by 1940, they had not yet reached the height of those in the north” (Massey and Denton 1993: 42).

The theoretical reasoning behind the group threat hypothesis is that racist behavior stems in part from a dominant group’s sense that their position is threatened by a subordinate group. One possible response is to strengthen the spatial boundaries between groups. Group threat has become a standard hypothesis in research in recent decades, but these studies have mixed results. A multivariate analysis of segregation that pooled data for all four decades in the 1980–2010 period (Iceland and Sharp 2013) found that Black-White segregation was higher in metros with larger shares of Black residents. Several other studies in this period found no such evidence. These include an analysis of metropolitan Black-White segregation in 1980 (Massey and Denton 1987), and also analyses of change in segregation between 1980 and 1990 (Farley and Frey 1994) and 1990–2000 (Logan, Stults and Farley 2004).

As applied to earlier decades, Iceland and Sharp (2013: 666) point to the coincidence in timing between rising segregation and the growth of the Black population. That is, “the racial group threat hypothesis is consistent with the finding that Black–White segregation increased in northern metropolitan areas in the first decades of the twentieth century as the Northern Black population swelled.” A stronger test is whether segregation increased more in cities with larger or faster growing Black population. Only one study (Cutler et al. 1999) reports such an analysis for the early decades of the century. They used ward-level measures to predict change in segregation between 1910 and 1940, showing a significant positive effect of Black population growth. In their view, the multivariate analysis showed that “the growth of the ghetto is strongly and significantly related to the migration of Blacks” (p. 469), and they concluded that “[s]egregation rose dramatically with the influx of southern Blacks, particularly in the industrial North” (p. 465). This conclusion is undermined by a strong negative interaction between the initial level of segregation and Black growth, which means that in highly segregated cities (which were mainly in the North) there was no net effect of Black growth. In addition, their cross-sectional models in 1910 and 1940 showed that neither Black population share nor region was significantly associated with city segregation.

Lieberson (1963) in an earlier study tested this hypothesis with data at the block level in the decades 1940–1950 and 1950–1960. In both decades, Taeuber and Taeuber (1965) found that White population growth was positively related to increases in segregation, but the association with non-White population growth was negative. In a multivariate model including several other city characteristics, neither White nor non-White growth had significant effects in the 1940–1950 decade. But in 1950–1960, the strongest predictor of changing segregation was the negative effect of Black population growth, a result that they note “is contrary to that usually assumed” (Taeuber and Taeuber 1965: 77). Hence, like the hypothesis of Southern exceptionalism, the hypothesis of group threat—while plausible—has not been clearly confirmed by past research.

RESEARCH DESIGN

This review of past studies identifies unresolved questions about the trajectory and pattern of segregation in the first half of the twentieth century. On key points the evidence is fragmentary, mixed, or missing. What is firmly established is that segregation was increasing in the early decades of the twentieth century. Our plan is first to fill out the record for most American cities in the 1900–1940 decades with descriptive analyses at multiple spatial scales. This step will test when Black-White segregation first reached a high level, and also assess the proposition that the changes during this period involved not only increasing segregation but also a tendency to replace the street with separation at a larger spatial scale as the modal form of Black settlement throughout the nation. Second, we conduct multivariate models that estimate the independent effects of region, time, population size, and the size of the Black population in cities on segregation at varying spatial scales.

This study draws on the 100 percent microdata samples from the Census of Population for every decade from 1900–1940, made available by the Minnesota Population Center’s IPUMS (Integrated Public Use Microdata Series) project (Ruggles et al. 2021). In each decade it includes all 134 identifiable cities that had a population of at least 30,000 and at least 1,000 Black residents in any one of these decades.

Sample of Cities and Residents

Because this was a period of rapid urban growth and redistribution of the Black population, cities that greatly exceeded the population criteria in one year may have been quite small or may have had few Black residents in an earlier year. An example is Flint, MI, with over 6,000 Black residents in a population of over 150,000 in 1940, but a small town under 15,000 (and with only 260 Black residents) in 1900. To describe the trends in the magnitude and spatial scale of segregation (Tables 1 and 2), we present average values across cities weighted by the number of Black residents, so cities like Flint would count negligibly in 1900 but moderately by 1940. We include all cities equally in the multivariate models of predictors of segregation.

Table 1.

Black Population, Mean Black Share, and Isolation (P*_bb), 1900–1940, Weighted by Black Population.

	Total Black population	City Black share (%)	Street	ED	Ward
All cities
1900	11,25,443	23.6	0.561	0.373	0.292
1910	15,26,255	22.2	0.617	0.409	0.288
1920	21,28,709	20.8	0.659	0.477	0.296
1930	32,54,172	21.3	0.737	0.571	0.351
1940	38,47,665	21.9	0.806	0.696	0.410
South
1900	7,14,365	34.4	0.662	0.477	0.402
1910	9,27,350	32.1	0.726	0.516	0.400
1920	11,38,008	29.3	0.751	0.570	0.390
1930	14,92,028	28.9	0.814	0.651	0.422
1940	17,56,236	29.4	0.861	0.739	0.476
North
1900	4,11,078	4.8	0.395	0.201	0.111
1910	5,98,905	5.0	0.455	0.250	0.120
1920	9,90,701	6.1	0.557	0.372	0.192
1930	17,62,144	8.0	0.673	0.506	0.291
1940	20,91,429	8.9	0.760	0.661	0.355

Note. ED = enumeration district.

Table 2.

Trajectories of Segregation: Dissimilarity, 1900–1940.

	Street		ED		Ward
	M	SD	M	SD	M	SD
All cities
1900	0.688	0.105	0.476	0.137	0.334	0.120
1910	0.758	0.082	0.555	0.123	0.358	0.127
1920	0.803	0.079	0.651	0.118	0.435	0.144
1930	0.861	0.063	0.742	0.109	0.517	0.167
1940	0.902	0.046	0.818	0.086	0.567	0.164
South
1900	0.640	0.094	0.407	0.098	0.284	0.101
1910	0.730	0.087	0.496	0.099	0.304	0.105
1920	0.772	0.084	0.588	0.094	0.355	0.107
1930	0.841	0.067	0.682	0.088	0.414	0.150
1940	0.883	0.051	0.769	0.076	0.473	0.165
North
1900	0.770	0.063	0.596	0.110	0.408	0.106
1910	0.801	0.050	0.648	0.097	0.440	0.113
1920	0.839	0.054	0.725	0.098	0.518	0.129
1930	0.878	0.052	0.794	0.098	0.598	0.130
1940	0.917	0.035	0.859	0.071	0.644	0.115

Note. Mean values and standard deviations for cities, weighted by Black population. ED = enumeration district.

Like most studies of residential segregation, we include persons of all ages in our tabulations, not only adults or household heads. Households and buildings were nearly racially homogeneous in all cities and years. An exception was when African Americans lived as domestic employees in White-headed households. Such cases do not represent “neighbors” in the usual sense, and we omit them from all counts. This was more common in 1900 (20.8 percent of employed Blacks in New York City were live-in domestics and in most cities the share was over 5 percent), but it was a negligible factor by 1940. The average value of $D$ at the street scale in 1900, if live-in domestics were counted, would be about five points lower in both the North and the South, compared to values presented here. The differential is smaller (three points on a 0–100 scale) at the ED level, and only two points at the ward level. It diminishes to nearly zero by 1940. Hence, studies that include domestic servants as “neighbors” results in a small downward bias in measures of segregation at finer scales and in earlier decades. In the multivariate analyses, however, exclusion of domestic servants had negligible effects on the model results.

We provide results for all cities combined, but to address questions about North/South differences and to control for the much higher Black population share in Southern cities, we also present results separately by region. We categorize 38 cities (those in former Confederate states plus Baltimore, Washington, and cities in the border states of West Virginia, Kentucky, and Oklahoma) as “Southern.” These cities stand out for their much higher share of Black residents than found elsewhere in the country. Of 32 cities with more than 10 percent Black residents in every decade, 29 are Southern (the exceptions are Atlantic City, NJ; Chester, PA; and Kansas City, KS). Of 25 cities that exceeded 20 percent Black in 1940, only two (Atlantic City, NJ and East St. Louis, IL) are outside the South. Five Southern cities exceeded 50 percent Black at some point (Savannah, GA; Montgomery, AL; Charleston, SC; Shreveport, LA; and Jacksonville, FL), and six had Black populations above 100,000 in 1940 (Washington, DC; Baltimore, MD; New Orleans, LA; Memphis, TN; Birmingham, AL; and Atlanta, GA). We combine all other cities into the North category.

Classification by Race

In past research for years before 1980, racial groups have most often been categorized as White, Black (combining Negro and Mulatto in the earlier decades), and “other race.” The “White” category poses a problem in urbanized areas with large Hispanic populations. We refine the White category to exclude Hispanics. This is a substantial correction, because Hispanics were a large share of the total enumerated White population in many large Western and Southwestern cities and in Northern cities like Chicago and New York by 1940. Because Black residents are generally less segregated from Hispanics than from non-Hispanic Whites, segregation measures using the White-Black dichotomy are biased downward, compared to what the values would be if Hispanics were removed from the White category (Sørensen, Taeuber, and Hollingsworth 1975; Taeuber and Taeuber 1958: 64–68). We also remove Hispanics from the Black category, which is a much smaller correction.

We estimated the Hispanic population at the neighborhood level in the following ways. Using the 100 percent microdata, we adopted the coding of Hispanics developed by Gratton and Gutman (2000) that considers several indicators such as whether they, their spouse, or parents were born in Latin America and whether they spoke Spanish at home during childhood. A non-Hispanic White or non-Hispanic Black, then, is a person of White or Black race who is not Hispanic by these indicators.

Spatial Scale

Recent advocates of “spatial measures of segregation” (e.g., Lee et al., 2008) introduce spatial scale by establishing distance gradients around every census tract and report to what degree races are separated at varying distances such as 100, 500, or 1000 m. We approach spatiality differently, using methods that do not rely on GIS maps. Instead we have calculated measures at qualitatively different scales: streets, EDs, and wards.

Street.

The lack of historical GIS street grids and address ranges for most cities in this study prevents measuring segregation at the scale of single street segments, a feature of studies like Logan and Martinez (2018). We created an alternative by identifying people who lived on the same street and in the same ED. This required considerable effort to standardize street names (dealing with multiple forms of misspelling and different formats to represent the direction, name, and street type) to determine who lived on the same street. Often residents on one side of the street were enumerated in a different sequence than those on the opposite side, and often the street name was originally written differently or transcribed differently in each sequence.¹ The “street” as defined here is the same as a traditional street segment in the case of alleys, which were often densely populated in this period. In the more typical case, it refers to a set of two to three connected street segments. Many major streets were boundary streets between EDs. In these cases, the people in one ED (on one side of the street) are treated as living on a different “street” than those on the opposite side. In these cases, the street is similar to a face block.

ED.

The ED (an area assigned to a single professional for enumeration) is the smallest administrative unit available in the IPUMS microdata. It is typically an area including two to six census blocks. Due to its ready availability and relatively small size, the ED is becoming the standard neighborhood unit for historical segregation studies, much like the census tract is used today.

Ward.

The largest areal unit in this study is the city ward. It is not available for all cities or years. Some cities did not have wards in any year, and some had wards in one year but not another.² In some cities, we replaced the ward with a comparable spatial unit. In New York City, for example, Assembly Districts were treated by the Census Bureau as its standard spatial unit larger than the ED, while in some other cities, the relevant unit was a precinct, voting district, or council zone. In 1900, we drew ward numbers from the IPUMS microdata. In other years, we created ward data (or data for the comparable large spatial unit) by combining EDs. For this purpose, we relied heavily on listings prepared by Ancestry.com that identify the relevant areal units in each city and year and provide a list of the EDs found within their borders.

Not reported here, we also calculated measures at smaller scales. Households and buildings are the fundamental units in people’s locations. IPUMS provides the household’s address (house number and street name) in all years, though some data cleaning was required.³ We found that households (aside from domestic servants) and buildings were both nearly 100 percent homogeneous by race in all cities and years in this study. Some researchers cited above have experimented with a measure of segregation that relies only on the sequence of enumeration to establish the race of people’s “next-door neighbors.” Having cleaned and imputed street addresses in the microdata, we are able to assess the racial composition of neighbors more precisely, knowing who lives in the same building and who lives in a building on either side. We treat as “next-door neighbors” all residents of one’s own dwelling plus residents in the buildings on the same street and in the same ED that were enumerated before and after it.⁴ These buildings could be thought of as “triads.” We found, as expected, that Black isolation was greater at this next-door neighbor scale than at the scale of the street. Because triads are overlapping, however, it is not possible to calculate other standard measures of segregation at this scale.

Measures of Segregation

We rely mainly on the most often used measures of segregation, the Index of Dissimilarity ($D_{\text{bw}}$) and the Isolation Index (P*_bb), both of which can be calculated without mapped data. These measures are not explicitly spatial. However, when calculated at multiple spatial scales, they reveal whether clusters of Black residents are relatively small and dispersed or relatively large and homogeneous. $D_{\text{bw}}$ measures how evenly Whites and Blacks are distributed across areas. It reaches a minimum if every area has the same share of the city’s total White and Black residents, and a maximum of 1 if there is no overlap at all in where Whites and Blacks live at a given scale. Values above .60, by convention, are described as “very high.” Values in the range of .20–.30 are considered modest and are often found in the contemporary metropolis for tract-level segregation between Whites of different European ancestry. P*_bb is a variant of exposure indices, specifically the percent Black in the area where the average Black person lives. By its nature P*_bb is higher in cities with larger shares of Black residents. This complicates comparisons between cities with different Black population shares or over time in a city with a substantially changing Black share, but it is nevertheless a meaningful descriptive statistic, and it has the advantage that it can be calculated for overlapping building groups.

Trends in Segregation and Its Spatial Decomposition

We examine the trajectories of segregation in cities over the period 1900–1940, and we compare patterns for Southern and Northern cities at multiple spatial scales.⁵ The first part of the analysis presents the average levels of segregation on indices of isolation and dissimilarity, weighted by the size of the Black population in a given year. In effect, these averages are the level of segregation in the city where the average African American residents lived. We also decompose segregation spatially to assess whether the spatial scale of segregation changed over these decades. The second part of the analysis estimates multivariate models that pool data across years to predict the level of segregation at each spatial scale. To be consistent with recent usage, all measures are calibrated to a 0–1 scale.

A key contribution of this study is to show that the spatial scale of segregation was changing substantially in the 1900–1940 period. This requires a method to decompose total citywide segregation into the shares that are uniquely attributable to segregation at each scale—the street, the ED, and the ward. Most studies have relied on an entropy-based measure of segregation, Theil’s $H$, for this purpose. We report the decomposition here for both Theil’s $H$ and the Index of Dissimilarity. Both are measures of the unevenness of the distribution of groups across geographic areas, and they are highly correlated with one another. We include the decomposition of $D$, following Logan et al. (2023), to be consistent with its use in other sections of the study and also to demonstrate that results of the decomposition are not dependent on the particular measure that is used.

Theil’s $H$ compares the diversity of the whole urban area to the diversity of individual neighborhoods, with 1 representing maximum segregation and 0 reflecting perfect integration. The entropy ($E$) of a given area, is defined as follows:

$$E=-\sum _{r=1}^n{\pi }_r\log ({\pi }_r)$$

where ${\pi }_r$ is the proportion of racial group $r$ in the area. Then $H$ measures how closely $E$ of smaller subareas aligns with the $E$ of the largest geography.

$$H=\sum _{i=1}^M\frac{t_i(E-E_i)}{ET}$$

where for $T$ and $t_i$ represent the total population of the largest geography and the subgeography, respectively. $E$ and $E_i$ similarly represent the entropy scores for the largest geography and the sub-geography. The difference in $E$ between the street and the city is simply the sum of the difference between the street and the ED, the ED and the ward, and the ward and the city. The total segregation $H$ in the urban area can then be partitioned into components that represent segregation between streets within EDs, between EDs within wards, and between wards (Reardon et al. 2000).

Wong (2003) described how the Index of Dissimilarity ($D$) can be decomposed in a parallel way. Let us denote the “contribution to segregation” (CD) by each geographic level as ${\text{CD}}_{\text{ward}}$, ${\text{CD}}_{\text{ED}}$, and ${\text{CD}}_{\text{street}}$, which must sum to the total $D$ (at the street scale) for the city. Now consider the relation between segregation at the ward and ED scales. At the extreme, in an all-White ward, all EDs must be all-White. If all wards were racially homogeneous, there could be no segregation within them. Conversely, if every ward’s racial composition were like the city as a whole, there would be no segregation at the ward scale, and all segregation would arise from differences among EDs within them. More formally, the additional contribution to segregation attributable for segregation among EDs is simply

$$C{D}_{ED}=D_{ED}-D_{ward}$$

Applying the same logic to streets within EDs,

$$C{D}_{street}=D_{street}-D_{ED}$$

We can rearrange each equation solving for $D_{\text{ED}}$, yielding,

$$D_{street}-C{D}_{street}=D_{ward}+C{D}_{ED}$$

which simplifies to

$$D_{street}=D_{ward}+C{D}_{ED}+C{D}_{street}$$

In sum, the “ward contribution” to city’s total $D$ is simply the observed value of $D$ at the ward scale. The “ED contribution” refers to the average segregation across EDs within wards. It can be interpreted in the following way: if Blacks and Whites were relocated within the city so that there would be even shares in every ward, what additional share would then need to be relocated across EDs within every ward to achieve even shares in every ED? The “street contribution” refers to the average segregation across streets within EDs, and it can be interpreted in the same fashion. The total $D$ is the observed value of $D$ at the smallest scale (the street), and it is equal to the sum of these three components (i.e., the total share of Blacks or Whites who would have to be relocated to have no segregation at the street scale).

RESULTS: RISING LEVELS OF SEGREGATION AT MULTIPLE SCALES

Both Black isolation (P*_bb, the Black share of residents in the location where the average Black person lives) and the Index of Dissimilarity ($D$) provide useful information about settlement patterns. The P* measure of isolation is reported in Table 1 for all cities in the sample and for the Southern and Northern cities separately. Since the value of P* depends heavily on the Black share in the entire city, Table 1 also provides descriptive information about the total size of the Black population and the average Black share in these cities (weighted by the size of the Black population).

In 1900, at a time when the United States was a predominantly rural society, there were about 1.2 million Black residents in the sampled cities, representing just under a quarter of all residents. By 1940, both the White and Black populations of American cities had grown substantially. The Black population in these cities more than tripled to 3.8 million, though the Black share actually declined slightly. Most relevant to interpreting the Isolation measures, Blacks were a much higher share of the urban population in the South than in the North throughout this period. In the Southern cities, where a majority of African Americans lived in 1900, there had been a marked upsurge in Black population after the Civil War (documented by Logan and Martinez (2018)). By 1900, Blacks were over a third of these cities’ population in the South, compared to less than 5 percent in the North. The disparity was moderately smaller by 1940, as Black growth in Southern cities was outpaced by White growth, while the early decades of the Great Migration increased the Black share in Northern cities to almost 10 percent by 1940.

Although the share of African Americans in the urban population was declining nationally, isolation increased dramatically. African Americans already lived in households and buildings that were nearly all Black in 1900 (not shown), and there was little change in these measures after that time. Table 1 shows that isolation increased substantially at every higher scale. At a national level, the average urban Black person already lived on a street where they were a majority in 1900 (P*_bb = .561). They were a majority in their ED by 1930 (P*_bb = .571). Meanwhile, isolation was not as high, but it was growing at the scale of whole city wards. In 1900, African Americans lived in wards where they were only about a quarter of the population—not much more than their citywide share—which explains why early studies reported minimal segregation across wards. But by 1940, they lived in wards whose Black share (41.0 percent) was nearly twice as high as the Black share in their city (21.9 percent).

Table 1 shows that Black isolation was much higher in Southern cities than in the North, a finding that is largely due to high shares of the city population. A more striking result is how isolated African Americans were even in Northern cities with modest Black populations. In the North in 1900, the average Black person lived in a city with an overall Black share of only 4.8 percent, but lived on streets that were 39.5 percent Black and in EDs that were 20.1 percent Black. Isolation increased across decades in both regions and at every scale. Increasing isolation appears to be disconnected from changing overall Black population share, since it occurs both in the South (where Black share was declining) and in the North (where it was increasing).

We turn now to trajectories for the Index of Dissimilarity, which is independent of the Black share. Like Table 1, Table 2 presents averages weighted by Black population size. The pattern runs parallel to Black isolation in several respects. At the street scale, the national average $D$ was already .688 in 1900, and it rose toward an apartheid level of over .90 by 1940. The same trend but at lower levels is shown at the ED scale. Ward measures (replicating previously published findings) show that segregation was moderate in 1900 (.308), but then rose steadily. By 1940 ($D=.567$), it was approaching the .60 level that most contemporary scholars consider to be “very high.”

Table 2 reports trends separately by region. There is a uniform pattern of increasing average segregation at all scales. Southern cities stand out for substantially lower segregation than found outside the South at every scale in 1900. Contrary to the finding by Grigoryeva and Ruef (2015) for 1880, using a distinctive “sequence of enumeration measure,” Southern cities were not more segregated than Northern cities in 1900 or any subsequent year; in fact, they were considerably less segregated than Northern cities. But the South-North difference was diminishing over time, especially at the street and ED scales. By 1940, the main remaining difference is that ward-level segregation was still only moderate in the South by 1940 (.473), while it had reached .642 in the North. While we cannot formally test an explanation for this phenomenon, we concur with Massey and Denton’s conjecture that it is partly related to the legacy of slavery in the urban form of Southern cities. Slave quarters were built near the homes of White slave owners, resulting in a geographic spread of Black residents across districts. Though the institution of slavery was formally abolished in 1865, the built environments and racial hierarchies of these cities endured, and Black urban residents of the South continued to live in former slave quarters and back alleys in most city wards in 1880, 15 years later (Logan and Martinez 2018).

Table 2 also reports the standard deviations around the average values. There was a slight increase in the variation across cities at the ward level in both the South and the North. However, in both regions, variation was reduced by nearly half at the street scale, while declining less sharply at the ED scale. At both the ED and street scales and in both regions, there was a leveling of regional differences. Variation was reduced by nearly half at the street scale. As segregation was increasing and differences between regions were diminishing over time, there was also a general convergence toward similarly high levels across the nation.

RESULTS: SPATIAL DECOMPOSITION

Based on the differences in segregation ($D$) at different spatial scales, we now turn our attention to the question to the restructuring of segregation to larger spatial scales. Our expectation is that the unique “ward contribution” to segregation tended to increase while the “street contribution” declined. Or to state the pattern differently, the share of total segregation attributable to wards increased, while the share of segregation attributable to separation at the street scale declined.

To illustrate this phenomenon, we begin with a single case, Philadelphia. Philadelphia was mapped at the address level by the Urban Transition HGIS Project in 1880 (Logan et al. 2011), and we have developed a comparably detailed map in 1940. The 1880 map aggregates persons to specific street segments based on their geocoded address. The 1940 map is also based on an approximation of the street segment scale. We created a historically accurate 1940 street grid and ED map (Logan and Zhang 2017), aggregated the 100 percent census microdata for Philadelphia to the scale of streets within EDs (as described below), and divided these populations into individual street segments by interpolation (based on the segment’s length).

Philadelphia was the Northern city with the largest Black population in the late 1800s with nearly 32,000 Black residents in 1880 (3.7 percent of the total). By 1940, the number had grown to nearly 250,000 (12.6 percent of the total). Figure 1 displays the location of disproportionately White and Black streets in 1880 and 1940 for the densely populated central core of the city, including the well-known Seventh Ward studied by Du Bois (1899). Street segments are categorized as being above or below the city’s overall Black population share in either year (3.7 or 12.6 percent, respectively). In fact, most segments below this threshold were below 1 percent Black in each year, and most identified as “Black streets” were close to or above majority Black. Using these simple dichotomies is necessary to create a readable map where the identification as a disproportionately Black street has the same meaning in each year. Parallel maps with multiple categories of Black share (not shown here) reveal the same pattern of change. Ward boundaries from 1940 are superimposed with thick Black lines.

Figure 1.

Location of disproportionately Black streets in Philadelphia’s central core in 1880 (over 3 percent) and 1940 (over 14 percent), with 1940 ward boundaries.

In the 1880 map, the segregation of Black residents at the ward scale is apparent. Some wards have very few Black streets, while Black streets are numerous in other wards, especially around the famous Seventh Ward. However, Panel A shows that even in this section of the city, the disproportionately Black street segments are intermixed with White segments. On closer inspection, Blacks tended to live on alleys and short streets in this part of the city, although they also occupied consecutive segments of some major streets (Logan and Bellman 2016). Most major street segments were nearly all-White. In the mostly White wards, there is a large predominance of White street segments but typically these have Black segments interspersed among them. This spatial configuration represents a city with considerable segregation at the ward scale but limited by the common presence of White segments in wards with a larger Black population and of Black segments in wards with a larger White population.

The 1940 map shows that this pattern was disappearing. By then wards in sections of the city with larger Black populations had many fewer White streets, while Black streets had nearly disappeared in wards with larger White populations. It became rare for a White street to be adjacent to a Black street within any ward. This pattern is even clearer if one takes note of clusters of Black streets within specific sections of the Whiter wards. The maps show that Philadelphia’s wards have moved in the direction of being more homogeneously either Black or White, making the ward a more important spatial scale of segregation.

We now confirm what we observed in Philadelphia by measuring this trend quantitatively. Not all cities had wards or a comparable unit in all years, and in some cities, the available unit varies from year to year. For this reason, the following analysis reports results for a smaller constant set of 102 cities for which a full decomposition can be made in every decade. Not reported here, we repeated the analysis with the maximum number of cities in each year, including every city for which we could identify a ward or ward-equivalent areal unit in that year. Results are nearly identical. Table 3 reports the average values of total segregation for each year in all cities, the South, and the North. As discussed above, because past research has relied on the entropy measure $H$ for the purpose of spatial decomposition, here we show results for both $D$ and $H$. The table reports the unique contribution of segregation at the street, ED, and ward scales to total city for each measure, as well as the percentage of total segregation accounted for by each scale.

Table 3.

Decomposition of $D$ and $H$ into Contributions by Ward, ED, and Street Scales, by Year and Region.

	Index of Dissimilarity ($D$)							Entropy Index ($H$)
	City total	Street		ED		Ward		City total	Street		ED		Ward
		Unique Contribution	Share of total (%)	Unique Contribution	Share of total (%)	Unique Contribution	Share of total (%)		Unique Contribution	Share of total (%)	Unique Contribution	Share of total (%)	Unique Contribution	Share of total (%)
All cities
1900	0.697	0.209	30.0	0.153	22.0	0.335	48.1	0.470	0.246	53.6	0.118	25.0	0.111	22.6
1910	0.765	0.194	25.4	0.202	26.4	0.369	48.2	0.564	0.265	47.8	0.176	31.2	0.130	22.2
1920	0.810	0.149	18.4	0.222	27.3	0.439	54.2	0.628	0.223	36.6	0.230	36.3	0.188	28.9
1930	0.866	0.115	13.3	0.225	26.0	0.526	60.7	0.721	0.202	29.1	0.275	38.0	0.270	36.4
1940	0.906	0.083	9.1	0.248	27.4	0.575	63.5	0.797	0.137	18.0	0.341	42.8	0.318	39.2
South
1900	0.643	0.227	35.2	0.128	19.9	0.288	44.8	0.451	0.260	58.6	0.091	20.6	0.100	20.8
1910	0.732	0.226	30.9	0.195	26.6	0.311	42.5	0.572	0.301	53.5	0.156	27.5	0.115	19.1
1920	0.773	0.186	24.0	0.231	29.9	0.357	46.1	0.624	0.259	42.5	0.218	34.6	0.147	22.9
1930	0.844	0.160	18.9	0.264	31.3	0.420	49.8	0.726	0.243	34.3	0.273	37.8	0.210	27.9
1940	0.884	0.122	13.8	0.295	33.3	0.467	52.9	0.795	0.196	25.3	0.359	44.9	0.240	29.8
North
1900	0.772	0.176	22.7	0.191	24.7	0.406	52.5	0.497	0.226	46.7	0.156	30.9	0.127	25.0
1910	0.804	0.152	18.9	0.211	26.2	0.441	54.9	0.554	0.223	41.1	0.200	35.5	0.146	25.9
1920	0.841	0.115	13.7	0.212	25.2	0.514	61.1	0.632	0.192	31.5	0.241	37.7	0.222	34.1
1930	0.880	0.087	9.9	0.200	22.7	0.592	67.3	0.718	0.177	25.9	0.276	38.2	0.306	41.5
1940	0.919	0.059	6.4	0.220	24.0	0.640	69.7	0.798	0.103	13.7	0.330	41.5	0.364	44.7

Note. ED = enumeration district.

There is a clear pattern at the national level. First, as shown in Table 3, “total” segregation as measured by $D$ (i.e., segregation at the street scale) increased steadily from .697 in 1900 to .906 in 1940. $H$ similarly increased from .470 to .797. Second, the unique contribution of segregation at the ward level to total city $D$ increased very substantially over time, from .335 to .575. The ward contribution to $H$ rose from .11 to .318. The unique contribution at the ED level also increased for both measures. At the same time, the street contribution declined from .209 to .083 as measured by $D$ and from .246 to .137 as measured by $H$.

Another way of thinking about these findings is to ask what share of total segregation (as it changed decade by decade) was attributable to every geographic scale (so that the sum of ward, ED, and street contributions would add to 100 percent in every year). Viewed this way, the ward contribution to $D$ was 48.1 percent in 1900, but 63.5 percent in 1940, while the ward contribution to $H$ was 22.6 percent in 1900 and rose to 39.2 percent in 1940. The ED contribution accounted for 22.0 percent of total $D$ in 1900, rising slightly to 27.4 percent in 1940, and 25.0 percent of total $H$ in 1900, increasing to 42.8 percent in 1940. The street contribution, in contrast, was declining. Its contribution to $D$ in 1900 was 30.0 percent but much less (9.1 percent) by 1940. The street contribution to $H$ similarly dropped from 53.6 to 18.0 percent.

This does not mean that micro-level segregation fell. The opposite is true. But much of the increase in segregation across streets could be understood as simply reflecting the changing racial composition of the wards and EDs that streets were located in. We see here that within each ward and ED, as the wards and EDs became more segregated among themselves, differences among streets contributed less to total segregation. There were fewer remaining Black residents in predominantly White wards and EDs, and fewer remaining White residents in predominantly Black wards and EDs.

Did the same trend of declining street contribution to segregation occur in all cities, both North and South? As discussed above, previous studies based on microdata (Grigoryeva and Ruef 2015; Logan and Martinez 2018) have argued that Southern cities were distinctive in 1880 in the importance of specific patterns of micro-level segregation. Yet, Logan and Bellman (2016) showed a similar result for the alleys and side streets of Philadelphia in 1880. The decomposition in Table 3 adds evidence for Southern distinctiveness, not in the level of street-level segregation but in the street share of total citywide segregation. In 1900, the street share of segregation in the South averaged 35.2 percent, while its share in the North averaged only 22.7 percent.

For a more detailed assessment, we inspected the scatterplot of the street share in 1900 by street share in 1940 (not shown here). Most cities in both regions fell to one side of the diagonal, indicating that the street share had fallen. At the extremes there is a regional difference—the eight cities that lie closest to the diagonal are all in the North, while five of the seven cities with the largest declines in street share are Southern. Analysis of variance of the 1900–1940 changes (not weighting for Black population size) shows that the North-South differential is statistically significant (p < .05), with an average decline of 26 percent in the street share in Southern cities, compared to a 19 percent drop in Northern cities. If the 15 outliers are excluded from consideration, however, there is almost no average difference in decline in the Southern cities (−21 percent) and Northern cities (−20 percent).

RESULTS: EFFECTS OF BLACK PRESENCE, SIZE, AND TIME TREND ON SEGREGATION

We turn now to an examination of the city characteristics associated with segregation at every spatial scale, relying here on the Index of Dissimilarity as the outcome measure. Table 2 reveals large increases over time, differences across regions, and also differences by geographic scale. The average values shown in the table reflect overall tendencies, but they do not take advantage of the considerable variation across cities or the different trajectories in different cities. Almost all cities experienced rising segregation, but they had widely varying starting points and degrees of change. The following multivariate analysis seeks to begin to account for this variation.

We have already seen that region itself is a major factor in the level, scale, and change in segregation. In 1900, there was little overlap between Southern cities (almost all below .70 at the street scale) and Northern cities (almost all above .70). But by 1940, cities in both regions had street-level segregation in the range of .85 to .95, as a result of the generally larger increases in the South. Analysis of variance using these unweighted measures of $D$ or weighting by the 1900 city Black population shows that the regional difference was statistically significant at the p < .001 level in 1900. The difference in average values narrowed by 1940. Using unweighted measures, the difference between the South (.88) and North (.89) was not significant in 1940. Weighting by the size of the Black population in 1940, the difference between the South average (.89) and North average (.93) was small but statistically significant.

The simple dichotomy between South and North invokes many differences associated with region. Among these are the history of slavery, the specific urban structures associated with Southern cities, the post-Reconstruction Jim Crow laws, the much larger presence of African Americans in the urban South and their growing presence in the North, experiences with international migration, economic base, and the relative position of African Americans in the labor market. Studies of metropolitan segregation in the 1970s and beyond cited above (Farley and Frey 1994; Iceland and Sharp 2013; Logan, Stults and Farley 2004; Massey and Denton 1987) apply multivariate methods to estimate the independent effects of such factors. We estimate simpler models, partly due to the difficulty of creating predictors for each of five historical time points and partly due to limited sample size in each region. However, the models include the critical factors that scholars have highlighted in that period: time itself, region, and the size of the city and of its Black population.

We measure Black presence by the size of the Black population. In doing so, it is necessary to control for total population size (which may have its own independent effect), because there is a tendency for larger cities to have larger subgroup populations. We do not present results for Black presence as a percentage of the total. In the ordinary least squares (OLS) models that include a region dummy variable, this is not feasible because the Black share is so highly correlated with region. Not shown, we did include Black percentage as an alternative indicator in the fixed effects models, but found no consistent significant effects.

Table 4 presents results from models at each spatial scale (ward, ED, street) that predict cities’ level of segregation (the Index of Dissimilarity) in the 1900–1940 period using region, time (a set of dummy variables representing year), population size, and Black population size as predictors. It is based on pooled cross-sections including Northern and Southern cities together and adding a dummy variable for region. In Table 5, the models are estimated separately for Northern and Southern cities.

Table 4.

Predictors of Segregation ($D$) by Black Population Share and Year, OLS, and City Fixed Effects Models for Street, ED, and Ward Scales.

	OLS coefficient	SE	t	City fixed effects coefficient	SE	t
Street
South	−0.041	0.007	−5.82***
Black population (1,000s)	−0.0001	0.0001	−0.77	0.0009	0.0004	2.55**
Total population (10,000s)	0.0002	0.0001	2.87**	−0.0010	0.0004	−2.63
Year (reference: 1900)
1910	0.039	0.009	4.56***	0.042	0.005	7.89***
1920	0.064	0.009	7.41***	0.068	0.007	10.37***
1930	0.100	0.009	11.54***	0.103	0.008	12.73***
1940	0.144	0.009	16.45***	0.144	0.008	17.85***
Constant	0.746	0.006	116.61***	0.743	0.005	136.64***
	F(7, 655) = 59.4			F(6, 132) = 129.1
	Adj. R2 = .382			R2 within = .634
ED
South	−0.044	0.011	−4.12***
Black population (1,000s)	−0.0010	0.0002	−5.06	0.0006	0.0005	1.33
Total population (10,000s)	0.0006	0.0001	4 81***	−0.0008	0.0005	−1.50
Year (reference: 1900)
1910	0.066	0.013	5.03***	0.069	0.010	6.66***
1920	0.132	0.013	10.05***	0.138	0.011	12.39***
1930	0.200	0.013	15.07***	0.207	0.014	14.81***
1940	0.278	0.013	20.80***	0.284	0.014	20.77***
Constant	0.489	0.010	50.01***	0.490	0.009	53.36***
	F(7, 655) = 96.8			F(6, 132) = 178.2
	Adj. R2 = .503			R2 within = .702
Ward
South	−0.073	0.018	−3.99***
Black population (1,000s)	−0.00001	0.00034	−0.03	0.0008	0.0006	1.42
Total population (10,000s)	0.0005	0.0002	2.39**	−0.0003	0.0007	−0.44
Year (reference: 1900)
1910	0.018	0.023	0.78	0.020	0.011	1.82
1920	0.029	0.023	1.30	0.032	0.014	2.23*
1930	0.072	0.023	3.16***	0.072	0.018	4.10***
1940	0.129	0.023	5.62***	0.127	0.020	6.22***
Constant	0.372	0.017	22.13***	0.356	0.011	33.29***
	F(7, 654) = 12.7			F(6, 132) = 11.4
	Adj. R2 = .110			R2 within = .150

Note. OLS = ordinary least squares; ED = enumeration district. ** p < .01; *** p <.001.

Table 5.

Predictors of Segregation ($D$) by Black Population Share and Year OLS and City Fixed Effects Models, North and South Separately.

	South						North
	OLS coefficient	SE	t	City fixed effects coefficient	SE	t	OLS coefficient	SE	t	City fixed effects coefficient	SE	t
Street
Black population (1,000s)	−0.0006	0.0003	−1.89	0.00001	0.00050	0.02	−0.0003	0.0002	−1.62	0.0003	0.0002	1.64
Total population (10,000s)	0.0018	0.0008	2.29*	0.0013	0.0014	0.92	0.0003	0.0001	3.54***	−0.0003	0.0002	−1.38
Year (reference: 1900)
1910	0.100	0.017	5.83***	0.096	0.011	8.49***	0.016	0.009	1.71	0.019	0.004	4.17***
1920	0.135	0.017	7.83***	0.129	0.012	10.61***	0.036	0.009	3.87***	0.040	0.007	5.92***
1930	0.192	0.018	10.93***	0.182	0.015	12.39***	0.064	0.009	6.95***	0.067	0.008	8.29***
1940	0.236	0.018	13.23***	0.223	0.015	14.52***	0.109	0.009	11.69***	0.111	0.008	13.37***
Constant	0.639	0.013	50.70***	0.631	0.011	59.27***	0.771	0.007	117.74***	0.778	0.005	141.69***
	F(6, 177) = 40.1			F(6, 36) = 139.8			F(6, 472) = 33.5			F(6, 95) = 104.2
	Adj. R2 = .562			R2 within = .881			Adj. R2 = .290			R2 within = .561
ED
Black population (1,000s)	−0.0008	0.0005	−1.69	−0.0001	0.0010	−0.07	−0.0003	0.0003	−0.98	−0.0002	0.0003	−0.61
Total population (10,000s)	0.0026	0.0011	2.25*	0.0024	0.0028	0.86	0.0007	0.0001	4.60***	0.0001	0.0004	0.33
Year (reference: 1900)
1910	0.135	0.025	5.49***	0.127	0.029	4.39***	0.039	0.015	2.62**	0.042	0.008	5.06***
1920	0.214	0.025	8.61***	0.202	0.029	7.00***	0.100	0.015	6.66***	0.105	0.010	10.19***
1930	0.299	0.025	11.85***	0.282	0.029	9.59***	0.161	0.015	10.65***	0.168	0.015	10.97***
1940	0.382	0.026	14.92***	0.360	0.030	11.90***	0.238	0.015	15.66***	0.246	0.015	16.68***
Constant	0.371	0.018	20.45***	0.362	0.019	19.11***	0.517	0.011	48.33***	0.526	0.010	52.06***
	F(6, 177) = 51.7			F(6, 36) = 139.0			F(6, 472) = 67.3			F(6, 95) = 147.5
	Adj. R2 = .624			R2 within = .805			Adj. R2 = .454			R2 within = .684
Ward
Black population (1,000s)	−0.001	0.001	−1.42	0.001	0.002	0.27	0.0005	0.0005	1.03	0.0004	0.0005	0.71
Total population (10,000s)	0.002	0.002	1.14	0.002	0.006	0.26	0.0003	0.0003	1.09	0.0001	0.0007	0.16
Year (reference: 1900)
1910	0.056	0.042	1.33	0.053	0.028	1.92	0.005	0.027	0.18	0.005	0.011	0.40
1920	0.072	0.043	1.68	0.061	0.029	2.11*	0.014	0.027	0.54	0.016	0.018	0.88
1930	0.126	0.044	2.90**	0.102	0.050	2.04*	0.051	0.027	1.91	0.054	0.019	2.82**
1940	0.188	0.044	4.26***	0.153	0.051	2.99**	0.107	0.027	3.95***	0.110	0.022	4.98***
Constant	0.274	0.031	8.79***	0.242	0.027	8.84***	0.386	0.019	20.31***	0.390	0.014	27.16***
	F(6, 176) = 4.0			F(6, 36) = 4.6			F(6, 472) = 7.6			F(6, 95) = 9.4
	Adj. R2 = .090			R2 within = .203			Adj. R2 = .077			R2 within = .134

Note. OLS = ordinary least squares; ED = enumeration district. ** p < .01; *** p < .001.

Every model, additionally, takes two forms.⁶ One version is an OLS regression. The other model introduces city fixed effects. Both models are informative, but they have different substantive meaning. The OLS models compare cities to one another (e.g., does a city with a larger Black population have higher segregation than a city with a smaller Black population?). In the fixed effects model, all comparisons are made in relation to the city itself (e.g., in this city, in years when the Black population is larger, is segregation higher?). Because both the city size and Black population size are generally rising in this period, the fixed effects models can be understood as estimating how increases in population or Black population are associated with increases in segregation. The fixed effects model is more complete, as it controls for all other city characteristics that are constant (or at least, on which the rank ordering of cities remains similar) between 1900 and 1940.

Consider first the models in Table 4 that test the direct effect of region. Across the period and consistent with descriptive results in Table 3, there is a negative effect of Southern location on segregation at the street, ED, and ward scales. This contradicts past research that argued for a special Southern spatial pattern of microlevel segregation in which African American residents were present in all areas of the city.

Most coefficients for Black population are not significant, and we conclude that there is little support here for the Black threat hypothesis. In the fixed effects model at the street scale, the Black population coefficient is significant. It estimates that if a city’s Black population increased by one thousand, its level of street segregation would increase by about one point. However, there would be no change in segregation at the ED or ward scales, and cities with larger Black populations in a given decade are not estimated to have higher segregation at any scale.

Finally, most of the explained variance in the OLS models is due to the effects of time. Segregation increased consistently from 1900 to 1940, and the differential over four decades was about .14 at the street scale, .28 at the ED scale, and .17 at the ward scale. A major unresolved question here is how to account for the temporal change that is not specific to any kind of city, but rather is a feature of the whole sample of cities across the nation. We return to this question in the conclusion.

Table 5 reports separate models for Southern and Northern cities. The net differences between regions can be inferred by comparing the constant terms in these models, which are consistently lower for Southern cities (a statistically significant difference of .10–.20). Black population is significant in none of these models. City total population, however, is positive and significant in the OLS models at the street and ED scales, in both South and North. Larger cities, then, were more segregated than smaller ones at these finer scales. Size is not significant in the fixed effects models, however. Larger cities were more segregated, but cities did not become more segregated as they grew larger.

Again, time is significant and has large effects in all models. There was a generalized rise in segregation in both regions and at all spatial scales in this period. There is, however, a difference between regions in the size of the increase. This is especially clear at the street scale, where the 1940 coefficient (compared to 1900) is twice as large in the South (above .20) than in the North (around .10) in both the OLS and fixed effects models. The difference between the North and South coefficients is statistically significant for all decades at the street and ED scale. At these scales, then, segregation in Southern cities was catching up to the higher levels in Northern cities. But at the ward level, the differences are not significant.

DISCUSSION

The descriptive tables and multivariate models together provide the most comprehensive findings to date on trajectories of segregation in U.S. cities in the 1900–1940 period when the Black ghetto emerged as an “American dilemma.” As found by all prior studies, these analyses confirm that segregation was rising and reached extreme levels by 1940. Beyond this simple fact, past research has led to conflicting results or provided no information on key questions that we have addressed here. These are (1) when did cities become highly segregated, (2) what was the spatial scale of segregation and how did it change over time, (3) what were the differences between cities in the South and the rest of the country, and (4) what was the role of higher and rising Black population in stimulating segregation?

On the question of timing, Tables 1 and 2 show that in the city where the average African Americans lived, segregation had reached very high levels in 1900 at the street scale. The average Black person lived on a street (within an ED) that was 56.1 percent Black, and in a city where the Index of Dissimilarity ($D$) at the street scale was as high as .688. These findings contradict earlier studies that reported low levels of separation at the beginning of the century and attributed the advent of segregation to post-1900 conditions. Residential segregation was built into the urban fabric early on.

Segregation ($D$) varied by spatial scale, necessarily higher at finer scales. It increased more rapidly at the ED and ward scales, and consequently, the contribution of segregation at these scales to total segregation in the city also increased. As illustrated in the Philadelphia case, cities that once had clusters of Black residents (in streets and EDs) in many large districts moved in the direction of a clearer separation of Whites and Blacks at the ward and ED scale. As a result, at the same time, as segregation increased at every scale, macro-scale segregation became more predominant. The creation of the Black ghetto, which combines a high level of segregation with concentration of African Americans in some large districts while excluding them from other parts of the city, is the joint result of these two processes.

The largest differences between South and North are in measures of Black isolation, which are mainly attributable to the much higher share of Black residents in Southern cities. But despite the overall decline in Black share in the South and the growing but still modest Black share in the average Northern city, levels of isolation increased in both regions and especially in the North. As a result, by 1940, the regional difference in average isolation was much diminished. The change was greatest at the ward level, where the 1900 values of P* were .402 in the South and only .111 in the North; by 1940, the Southern average was up moderately to .476 and much more in the North, to .355.

The opposite occurred with dissimilarity. Table 2 shows lower average $D$ in the South in 1900 at every spatial scale, even the street. But by 1940, segregation in the South was reaching levels much closer to the Northern average. The multivariate models in Table 4 suggest that Southern cities were less segregated than Northern cities at every scale throughout the period. Comparing the constant terms between models for South and North in Table 5, where coefficients for other predictors can vary between regions, suggests the same conclusion—at every scale, the constant term for Southern cities is lower than for Northern cities. Our general conclusion is that segregation was initially lower in the South than in the North, but tended to catch up over time.

With respect to the fourth question, results of the multivariate models are clear. Table 4 reports eight coefficients for Black population size, of which only one suggests that segregation rose in cities with larger Black populations. Table 5 reports no significant coefficients for Black population growth. White Americans may have been distressed by the growing Black population in Northern and Southern cities, and their reaction may have been more intense in cities where the Black presence became larger, but there is almost no evidence that this reaction resulted in higher levels of residential segregation within cities.

CONCLUSION

If regional differences diminished over time and at every scale, if the differences across cities at the street and ED scale were also declining, and if variations across cities are not associated with variations in Black presence, how can we interpret the overall changes in the level and spatial scale of segregation shown here? These findings lead in a similar direction as the study by Logan and Parman (2017) cited above, which examined the 1880–1940 period and considered segregation in both rural and urban areas. They concluded (p. 166) that the substantial rise in segregation that they documented was “far more general than previously thought. […] Rather than being the product of Black migratory patterns, regional differences in Black location patterns, or white population flows out of central cities, the increase in racial sorting was quite general.”

If not mainly a response to population shifts, why was there such a strong upsurge in segregation? We now know that race was a major determinant of the organization of urban space from an early time. As Lieberson (1980: 35) concluded from his comparison of the situation of Blacks and White ethnic groups, “Blacks in America were generally viewed as a biologically inferior population” whose efforts to reach equality were considered inappropriate. Yet, why did a housing market that was already highly racialized in 1900 experience such an intensification of racial separation in subsequent decades? We know that its spatial pattern was initially more localized, but that it shifted toward separation on a larger spatial scale during this period. We know that similar changes occurred in both the North and the South, and that it was experienced in almost all cities, both in the major destinations of the Great Migration and in cities where the Black population was never more than 1 or 2 percent of the total. Although there was variation among cities in these patterns, there is a very clear overall national direction of change.

This observation leads toward the conclusion that explanations for the rise of the Black ghetto need to consider how race and neighborhood were becoming interlocked at a national scale, even though the processes were carried at a local level, city by city, and neighborhood by neighborhood. An important direction for further research is to understand how housing markets were racially structured across urban America. There are strong reasons to be believe that institutional structures such as racially motivated zoning, restrictive covenants, and mortgage redlining influenced racial change in individual neighborhoods. Possibly, such public and private sector policies affected segregation at a citywide scale. Yet, if a similar outcome were found in cities with fewer restrictive covenants and where redlining maps were never published, the nature of their effects would need to be reconsidered. Were exclusionary practices already becoming so widespread and deeply rooted in the real estate market in the early twentieth century that they were having an effect in every city? Were there other changes in progress that could have affected cities throughout the country, such as the changing transportation infrastructure and relocation of manufacturing industries that encouraged White flight to newly built areas of cities and to suburbs? Were qualitative changes occurring in how people made decisions about where to live? How, one might ask, did the racial composition of neighborhoods become so relevant to their choices that Whites would so strongly avoid racially mixed neighborhoods and exert such pressure to block entry of African Americans into White areas? Looking to the future, studies that seek to explain the variation in segregation trajectories should proceed in tandem with efforts to understand the overall trend across all cities.