We examine the relationship between the lynching of African Americans in the southern United States and subsequent county out-migration of the victims' surviving family members. Using U.S. census records and machine learning methods, we identify the place of residence for family members of Black individuals who were killed by lynch mobs between 1882 and 1929 in the U.S. South. Over the entire period, our analysis finds that lynch victims' family members experienced a 10-percentage-point increase in the probability of migrating to a different county by the next decennial census relative to their same-race neighbors. We also find that surviving family members had a 12-percentage-point increase in the probability of county out-migration compared with their neighbors when the household head was a lynch victim. The out-migration response of the families of lynch victims was most pronounced between 1910 and 1930, suggesting that lynch victims' family members may have been disproportionately represented in the first Great Migration.
Lynching was ubiquitous in the postbellum American South. Although estimates vary, more than 2,500 African Americans were murdered by lynch mobs between 1882 and 1930 (Beck and Tolnay 2004). Lynchings, which could range from clandestine nighttime affairs to grand daytime public spectacles of ritualistic torture and dismemberment (Smångs 2017), led historian Edward Ayers (1992:158) to state, “young black men [and women] learned early in their lives that they could at any time be grabbed by a white mob . . . dragged in the woods or a public street to be tortured, burned, mutilated.” Ayers' words draw attention to the fact that lynchings were woven into the southern Black consciousness, where tales of lynchings were shared among the Black community and children were instructed not only on the proper use of the plow but also on how to avoid the wrath of the mob.
Lynching—a brutal and socially disruptive tool used by southern Whites in attempts to subjugate the Black population—has been studied for generations (Mintz 1946; Raper 1933; Reed 1968; Smead 1988; Wells 1892). Quantitative social scientists have contributed salient and insightful scholarship that explains the individual and contextual characteristics that predict lynching (Bailey and Snedker 2011; Beck et al. 2016; Tolnay and Beck 1995), while also documenting how lynching's effects continue to reverberate in a variety of contemporary contexts related to racial inequity, including school-based corporal punishment (Ward et al. 2021), voting patterns (Abbott and Bailey 2021), and church burnings (Howell et al. 2018). Historians too have added significant value to our understanding of lynching through tomes on distinct lynching events (Clegg 2010; Wexler 2004) and the broader ethos of the Jim Crow South that was partially created and maintained through terror lynching (Dorr 2004; Feimster 2009; Williams 2018). Despite our collective understanding about when and where lynching was likely to occur in the southern United States (Bailey and Snedker 2011; Beck et al. 2016; Tolnay and Beck 1995) and the characteristics that put people at risk of victimization (Bailey and Tolnay 2015), data limitations have constrained our ability to ascertain how this form of racial violence shaped the lives of the Black families that were directly impacted by mob violence (Tolnay and Beck 1990, 1992). In particular, there exists no quantitative research systematically assessing the migration patterns of Black families after a member of their household was lynched.
To address this limitation, we have constructed a novel individual-level longitudinal dataset that links Black lynch victims and their families in the U.S. Census between 1880 and 1930 using machine learning and a large, public wiki-style family tree (see Price et al. 2021). We leverage these data using linear probability models to assess the county out-migration patterns of Black lynch victims' surviving family members. The results of our investigation provide suggestive evidence on the consequences of lynching for those who were directly affected by this form of racial violence—the families of lynch victims. In addition, our study has implications for our understanding of lynching specifically, and racial violence broadly, as we conceptualize it to function as a proximate driver of spatial mobility for Black Americans fleeing as refugees from violence before and during the first Great Migration.1
Background and Theory
A Brief Summary of Southern Racial Social Control and Black Migration
The enslavement system in the United States was designed to extract the maximum amount of labor from enslaved people for the purpose of profit. To ensure the efficient functioning of this economic system, southern Whites instituted all-encompassing methods of social control. These included slave codes—rules constraining the freedoms of enslaved Black people with such policies as not being able to marry, quarrel with a White person, inherit property, or leave a slaveholder's property without permission. Lasting for generations, slave codes formally concluded in 1865 with the end of the Civil War (Middleton 2020). In the postbellum era, the southern economy was in a state of uncertainty, with African Americans having increased economic options and enhanced geographic mobility (Ruef 2014). These circumstances prompted southern Whites to adapt the postbellum social and economic structure to maintain—as closely as possible—the same societal arrangements present in the antebellum era through new forms of slavery and social control.
Potent incarnations of neoslavery were instituted by White elites through sharecropping (Mandle 1983) and convict leasing (Blackmon 2009), both of which constrained the upward social and economic advancement of Black individuals, families, and communities. Black economic exploitation was buttressed by a Jim Crow system of formal and informal policies that structured the entirety of the southern milieu (Woodward 1974). Under the regime of Jim Crow, quotidian forms of segregation separated Whites from Blacks in all areas of public life, from cemeteries to interstate railroad travel. Blacks were also disenfranchised through poll taxes, literacy tests, and grandfather clauses, which limited their ability to challenge the southern power structure. Undergirding this system of racial domination was the threat of extreme violence, typified by racial terror lynching.
Despite the frequency of racial violence during the antebellum era, lynching became a much more common, and widely publicized, practice in the postbellum South. During the antebellum era, there was little financial incentive for southern planters to kill enslaved Blacks since the enslaved population was of high economic value (Baptist 2016). Consequently, White elites were motivated to exercise a relatively strong influence over lower-class Whites and their desire for retribution over the perceived crimes of Blacks. Once Black and White workers became direct labor market competitors, the terroristic violence used to enforce White supremacy became more lethal. Indeed, from 1882 to 1930, roughly one Black person was lynched each week in the South (Tolnay and Beck 1995).2
The incidence of lynching peaked in the 1890s and declined precipitously over several decades. Since 1930, lethal mob violence has been exceptionally rare. Scholars attribute the decline in lynching to several factors, such as increased objection to the practice by individuals and the state (Ames 1942; Beck 2015; Dray 2003), a more organized criminal justice system (Pfeifer 2004), and migration (Tolnay and Beck 1992). Starting in the 1910s and ending by 1970, the Great Migration saw millions of southern Blacks migrate for better economic prospects and to escape the highly segregated system of Jim Crow and the racial violence that supported it (Tolnay 2003). This immense internal migration largely saw Black populations migrating to the North along with nontrivial shares migrating within the South (Collins 2021).
In assessing patterns of Black migration during this era, scholars have stressed economic factors such as financial opportunities in northern industry (Collins 1997) and the boll weevil epidemic that spread through the region, which sharpened economic hardship (Higgs 1976). Research has also indicated that lynching may have spurred migration. Tolnay and Beck (1992) found that Blacks' patterns of county out-migration in the South between 1910–1920 and 1920–1930 was strongest from counties with more pronounced levels of lynching, suggesting that mob violence was a powerful mechanism driving Black families to search for safer climates. Although this finding provided substantial insight toward understanding the link between racial violence and migration, the analysis is limited to investigating county-level migration rates for the entire Black population. Focusing on this level of geographic aggregation masks whether lynchings were potentially associated with the subsequent migration of the families of specific lynch victims.
There is also striking anecdotal evidence that the families of some lynch victims were forced to flee as refugees from racial violence. As a case in point, in 1916 in Abbeville, South Carolina, Anthony Crawford, a highly successful Black businessman, was lynched by a mob for arguing with a White merchant over the price of Mr. Crawford's cottonseed (“Anthony Crawford” 1916). Shortly after Mr. Crawford's murder, the White residents of Abbeville voted to expel his family from the area and seize their 427-acre farm. The Crawford family subsequently fled the county (Equal Justice Initiative 2017). We hope to establish whether this migration response was typical for those who were related to victims of terroristic violence—in essence, whether familial migration might be an unacknowledged outcome of lynching and, therefore, whether the link between mob violence and migration before and during the first Great Migration might be more individually causative than has been quantitatively understood thus far.
The connection between violence and migration is well established. Individuals are increasingly likely to move away from their homes and communities as they experience threats to their personal safety (Davenport et al. 2003). There is also evidence that those who experience violence report they are more likely to intend to migrate. For instance, an analysis of Honduran crime victims found that they were more than two times as likely to intend to migrate relative to nonvictims (Hiskey et al. 2016). Similar to contemporary migrants, southern Black people appear to have been responsive to violence and the threat thereof during the Great Migration era, such that escaping terrorism has been conceptualized as a proximate determinant of Black migration during the period (Tolnay and Beck 1990, 1992).
We know, however, that not all who are targeted by or exposed to violence choose to migrate. Households vary widely in their responses to the threat of violence (Verwimp et al. 2009). A household's decision to flee is only one in an array of responses employed by those under threat to secure their safety (Barter 2012). Indeed, migration from areas experiencing widespread violence appears motivated by a combination of factors, including both the fear of violence as well as more established factors associated with migration (Czaika and Kis-Katos 2009). One established factor related to migration is widowhood. Evandrou et al. (2010) found that female widows have a higher probability of migration relative to their male widowed counterparts, with the differential likely being driven by male partners often functioning as primary earners in the labor market. Another established factor of migration is human capital. Silva and Massey (2015) showed that individuals with higher levels of human capital resources are more likely to migrate. Contrary to predictions of traditional economic-based theories of migration, however, those fleeing violence may move even when doing so damages their economic standing (Atuesta and Paredes 2016). Yet, the link between migration and household economic resources appears to be complex and may be highly sensitive to local context. The relationship between household economic situation and the decision to flee a violent environment is strongly conditioned by local exposure to violence and perceptions of the way that violence has impacted the household (Engel and Ibáñez 2007; Jampaklay et al. 2017).
Considering previous research on the association between lynching and county-level out-migration of Black Americans, and the scholarship assessing the relationship between violence and refugee migration, we investigate a series of research questions. Our central question asks whether lynch victims' family members differ in their county out-migration from other Black individuals living near them who did not experience a lynching in their household. In light of prior work on the link between widowhood and migration, we also ask whether patterns of county out-migration among surviving family members when the head of household was lynched differ from those of individuals living near them who did not experience a lynching in their household. Answers to these questions will help provide clues as to how a population we know little about—the families of Black lynch victims—was impacted by racial terror lynching—an institution of social control that has had lasting effects into the modern day (see, e.g., Abbott and Bailey 2021; Durso and Jacobs 2013; King et al. 2009; Ward et al. 2021).
Data and Methods
For our analysis, we assess the county out-migration patterns of family members of Black lynch victims by comparing each family member's place of residence in the nearest census before and after a lynching event.3 To accomplish this, we use the Tolnay, Bailey, and Beck Database of Southern Lynch Victims (2019). In these data, we found household census records for 357 Black people who were lynched in the American South between 1882 and 1929, who were living with at least one relative during the time of enumeration, and whose relative is found in the ensuing census.4 After locating lynch victims' census records, we observe 772 of their family members living in the same household at the time of census enumeration who can be linked to the subsequent census.5 Since the Tolnay, Bailey, and Beck database covers the years of 1882 through 1929, we observe them coresiding with their family members in the census years of 1880, 1900, 1910, and 1920.6
The control group for our analysis consists of 104,733 Black individuals living in the same census enumeration district as the household of a lynch victim.7 This allows us to compare the county out-migration of lynch victims' family members with the out-migration of individuals living in similar racial, social, and geographic circumstances at the level of the enumeration district.8 Census enumeration districts themselves were geographic areas that census enumerators were assigned to cover over a period of weeks, typically making them smaller than counties. The average population of enumeration districts in our sample is 2,112, with a maximum population of 6,481.
Our analysis of the county out-migration of the families of Black lynch victims as compared with their neighbors across 10- and 20-year periods requires that we link census records over time. To accomplish this, we follow a process similar to that of Price et al. (2021) by combining several methods of historical record linking.9 We first use links that were created by individual contributors to the public wiki-style family tree at FamilySearch.org, whose decisions are informed by private information about their family members (Price et al. 2021). Second, we create probabilistic links using machine learning models that are trained on millions of links from the family tree at FamilySearch.org (Price et al. 2021). Third, we include a set of links that are established on the basis of individual characteristics that often remain constant over time, such as name, gender, and birthplace (referred to as Abramitzky, Boustan, and Eriksson links, or ABE links; see Abramitzky et al. 2019). Fourth, we include additional probabilistic links from a machine learning method that uses existing ABE links to help link remaining family members in the same household (Helgertz et al. 2022; Helgertz et al. 2020). Finally, we use record hints generated from a machine learning algorithm from FamilySearch.org that uses information from their Family Tree and from various historical records to curate possible census links.10
Combining census links from various sources allows us to increase the coverage of individuals in longitudinal datasets for each pair of successive census years between 1880 and 1930. In addition to increasing the fraction of individuals linked between census years, aggregating links from various methods allows us to improve their combined accuracy through employing a technique called “sheet checking” (Price et al. 2021). Each sheet in the census includes up to 50 individuals, and each census has millions of sheets and trillions of possible combinations between sheets of contiguous censuses. Correct matches across censuses frequently have other pairs of family members and neighbors on the same sheet. Hence, sheets with relatively more matches across censuses are more likely to contain valid links than those with only one match.
After creating our final set of intercensal links, we observe that family members of lynch victims are linked across censuses at somewhat lower rates than the comparison set of individuals living in the same enumeration districts.11 Lower linkage rates could be due to a higher rate of migration, since temporal consistency in an individual's location can help establish a link. In addition, the death of a family member reduces the available information for establishing census links, particularly for linking methods that draw upon information about household members. Moreover, owing to the frequent lynching of men and boys, lynch victims' surviving family members are more likely to be female. Thus, the lower rate of linkage for family members may be partially driven by more frequent surname changes from marriage (for daughters) or remarriage (among women who became widows when their husbands were lynched)—an assumption supported by the fact that a larger percentage of the matched surviving family members are male than is true of the comparison sample. All of these differences are also key determinants of selection in our focus sample, yet it is unclear whether they contribute to selection bias.
The coverage of census links for the control sample is also lower than expected when compared with the general population (Price et al. 2021). Since we select only Black individuals in our sample, we attribute some of the lower linkage rate to relatively high rates of migration to the North and to urban centers within the South in the early twentieth century. Furthermore, challenges in linking arise from higher rates of underenumeration, age misreporting, and name homogeneity for Black people (Goeken et al. 2011; Rosenwaike et al. 1998).12 Other than migration, however, we do not suspect that any of these potential challenges in linking will vary systematically between family members of lynch victims and Black Americans living in the same enumeration district who did not experience a lynching in their household. Despite these obstacles, we obtain census links across 10-year time spans for 34% of lynch victims' surviving family members in 1900, 1910, and 1920, and for 38% of Black people living in the same enumeration districts. Linkage rates from 1880 to 1900 are 18% for lynch victim's family members and 20% for Black individuals within our comparison sample.
To consider whether an individual migrated out of their county of residence, we follow a procedure to link counties between censuses inspired by a method developed by Otterstrom et al. (2022).13 While counties have relatively stable boundaries over time, a small proportion of counties experienced boundary changes over the half century of our analysis. We address the issue of linking counties over time by embedding individuals within the population of their enumeration district in the first census where we observe them. If, in the later census, they are found residing in the same county as most of the individuals within their initial enumeration district, we classify them as having not migrated. Moreover, each enumeration district is likely to be much smaller in area than the county that contains it, so any changes in county boundaries will only divide a subset of enumeration districts.
We proceed by using our full set of individual-level census links to create a crosswalk for each period of possible migration that links each enumeration district in the starting year (e.g., 1920) to a county in the ending year (e.g., 1930). Using these links across census years, we count the number of individuals in the starting enumeration district who are residing in each possible destination county. The enumeration district is then matched to the destination county with the highest frequency of links. In very rare cases, we assign a single enumeration district to two destination counties that have similarly high rates of linkage.14 We do so to avoid falsely classifying some individuals as migrating when they did not actually move.
Variables and Analytic Strategy
Our dependent variable of interest is a dichotomous indicator for whether a Black family member of a lynch victim or a Black individual living in the same enumeration district as a person who was lynched lived in a different county 10 or 20 years after they are first observed.15 In our primary analysis, the focal independent variable is a dichotomous measure of whether a Black individual was living with a family member who was a lynch victim in the period beginning with the census immediately prior to the lynching and ending one year prior to the next available census. Additionally, we estimate a secondary analysis where the focal independent variable is whether the head of household was a lynch victim. It is possible that the lynching of a household head will have a relatively pronounced association with out-migration given that the household head will likely have been a strong economic contributor to their family.16 Also, our aim is not to establish a causal connection between lynching and familial out-migration; however, we do include a set of sociodemographic and socioeconomic variables to help isolate the net association between out-migration and the death of a family member to lynching.17 These sociodemographic and socioeconomic variables of the family member or individual living in the same enumeration district include gender, age, age squared, marital status, farm residence, living in the birth state, household size, homeownership, and literacy status.18
We use linear probability models to estimate the association between county-level out-migration and whether an individual lived with a family member who was a lynch victim. We investigate this association across several time periods. The first period includes lynchings that occurred between 1882 and 1895, where we measure migration between 1880 and 1900. We measure migration between each 10-year census interval between 1900 and 1930 because full-count census records are available during that time span. Our sample for each of these decade-level analyses includes the individuals in the same household or enumeration district as victims who were lynched over the 10-year period starting in the first census year (e.g., 1900) and ending nine years later (e.g., 1909). We also estimate a linear probability model with data pooled over each period from 1880 to 1930, as well as a pooled model over each decade from 1900 to 1930. The pooled results from 1880 to 1930 provide a long-range view of the association between lynching and the county out-migration of families across what some scholars have deemed the “lynching era.” Considering the potential measurement error introduced because of the loss of the 1890 census, we include separate estimates from 1900 to 1930. Estimated coefficients for our focal independent variables in the pooled regressions are interpreted as weighted averages of the increased probability of county out-migration across multiple time periods.
For each of our models, we apply clustered standard errors at the household level because migration is highly correlated within families that reside together. We also include fixed effects for census enumeration districts in our models. Enumeration district fixed effects allow us to control for time-invariant unobserved local characteristics that are related to both lynching and migration, such as corruption in law enforcement and judicial systems, presence of former slaveowners, proximity to urban areas, economic opportunities for Black people, history of racial violence, and history of northward and urban migration. In the pooled models, enumeration district fixed effects also control for the starting census year because districts are separately defined for each census.
Besides the analyses described in the foregoing, we estimate two supplemental analyses. First, since each migration outcome is measured over a 10- or 20-year period that includes the date of lynching, it is possible that some families migrated before the lynching of their relative and did not migrate afterward. We attempt to account for this by conducting an analysis that is restricted to family members of victims who were lynched in the same county where they were living in the previous census.
Second, statistical inference poses a challenge in isolating the net association between lynching and out-migration because the number of individuals affected by a lynched family member in their household is quantitatively small. We introduce an alternative approach in which we include the full sample for each census enumeration district and categorize those we are unable to link as if they had migrated between census waves. This approach is supported by the fact that unlinked individuals are more likely to have migrated. Note that because we are able to link family members of lynch victims between censuses at a lower rate than the comparison sample, these “inferred movers” constitute a larger proportion of the surviving sample members in our analyses. This introduces an unknown degree of measurement error into our sample, but it should provide a conservative lower-bound estimate of effect size at the cost of attenuating the precision of our models. Despite the limitations of this alternative specification, we expect that the increased sample size has the potential to supply supportive information to our study.
Table 1 highlights demographic characteristics and the general circumstances of the 357 Black lynch victims with at least one family member who is linked across subsequent censuses. At least 90% of victims in each period were male and at least half were killed in the same county where they lived at the time of their census enumeration. Most victims were in their late 20s when they were lynched. Furthermore, a sizable share of lynch victims were reported in the census as being married, ranging from 31% between 1882 and 1895 to 63% between 1920 and 1929. Approximately 80‒90% of Black lynch victims were the head of household or a son of the household head.
In Table 2 we describe the characteristics of the coresident family members of Black lynch victims along with Black individuals who resided within the same census enumeration districts. Several important differences are apparent when comparing these groups. Family members were more likely to be female and younger and less likely to be married; they had larger families and were more likely to live on a farm and to own a home or have a mortgage at the time of enumeration. Most importantly, across each period, family members of lynch victims had a higher likelihood of migrating from their counties than other Black individuals who lived in their surrounding area; this migration peaked between 1920 and 1930, when individuals who did not experience a lynching in their family migrated at a rate of 33% between censuses compared with 53% of individuals who had a lynch victim in their family.
Despite the descriptive differences in county-level out-migration rates, it is unclear whether this pattern holds when controlling for relevant characteristics that are associated with migration. To address this, in Table 3, we estimate a set of nested linear probability models to assess the probability of county out-migration among those who had a relative killed by lynching compared with individuals in their census enumeration district that did not, net of applicable characteristics.19 During each of the decade-specific analyses, our reduced-form models, controlling only for whether the individual was a lynch victim's coresident family member and enumeration district fixed effects, suggest a higher probability of out-migration among surviving family members compared with their neighbors. Although the direction of the association is positive for all decades, the relationship does not meet traditional thresholds for statistical significance in the earlier decades (1880–1900 and 1900–1910). In 1910–1920, however, relatives of lynch victims had a 13-percentage-point increase in the probability of county out-migration (p < .05) compared with those who did not experience a lynching in their family. During the 1920–1930 decade, shortly after the start of the Great Migration, the relative rate of county out-migration between those who were relatives of lynch victims during this time and those who were not increased to 17 percentage points (p < .05).
The series of full models controls for individual- and household-level characteristics and indicates that relatives of lynch victims had a higher probability of county out-migration relative to their local-area counterparts who did not experience such an event. However, again, those differences do not emerge as statistically significant until later decades. In the 1910–1920 decade, the probability of out-migration was 14 percentage points higher among surviving family members compared with their neighbors, with the addition of individual- and household-level controls (p < .05). During the period 1920–1930, shortly after the start of the Great Migration, surviving family members had an 18-percentage-point increase (p < .01) in the probability of county out-migration compared with their local peers, net of controls.
In addition to the models that assess differences in county out-migration between individuals who are relatives of lynch victims and those who are not across the 10- to 20-year time periods shown in Table 3, we also investigate models that pool time periods across two ranges: one across the entire time range of study from 1880 to 1930 and another from 1900 to 1930. For the pooled models using all observations from 1880–1930, the probability of county out-migration for relatives of lynch victims emerged as positive and statistically significant (p < .01), registering 10 percentage points higher than for their local peers in both the reduced-form and full models. The pooled results for 1900‒1930 show a 12-percentage-point increase in the probability of out-migration (p < .01) for relatives of lynch victims compared with those who did not experience a lynching in their family, and that probability increases to 13 percentage points (p < .01) upon the addition of individual- and household-level controls.20
In Table 4, we use linear probability models to estimate the county out-migration of relatives of household heads who were lynched compared with the out-migration of their Black neighbors. The reduced-form and full models for 1880‒1900 and 1900‒1910 indicate that relatives of household head lynch victims had a higher probability of out-migration compared with people within their enumeration district that did not have a family member who was lynched, yet similar to the findings in Table 3, these results are not statistically significant.21 During the beginning of the Great Migration between 1910 and 1920, in comparison with individuals in their area who did not experience a lynching in their family, relatives of lynched household heads had a 20-percentage-point (p < .05) increase in the probability of county out-migration in our reduced-form model and a 22-percentage-point (p < .05) increase in our full model. Between 1920 and 1930, the probability of county out-migration for individuals who had a household head lynched was 22 percentage points higher than for those who did not, when including only enumeration district fixed effects, yet this result is marginally significant (p < .10) and should be interpreted with some caution. After controlling for individual and household characteristics, the probability of county out-migration among surviving family members who had a household head lynched slightly attenuates to 19 percentage points (p < .10).
The reduced-form pooled model for 1880‒1930 indicates that over the lynching era, relatives of lynch victims who were household heads had a statistically significant 13-percentage-point increase (p < .01) in the probability of county out-migration compared with their neighbors who did not have a family member killed by lynching. This association slightly diminishes to a 12-percentage-point difference in the full model (p < .05). Similarly, for the pooled 1900‒1930 model, family members of household head lynch victims had a 15-percentage-point increase in the probability of migration relative to those in their enumeration district who did not have a family member killed by terror lynching, in both the reduced-form (p < .05) and full models (p < .01).
We recognize that since lynchings are recorded for a range of years between census waves, our measure of county out-migration will inevitably include moves that occurred before certain lynching events. Restricting our sample to the lynch victims who were killed in the same county where they were enumerated should reduce the likelihood of making this kind of measurement error.22 In Table 5, we provide estimates of county out-migration of family members of victims who were lynched in the same county where they and their family resided. The results of these analyses highlight that in all decades of our observation, the relatives of Black people who were lynched within their county of residence have a more pronounced probability of out-migration than their local peers who did not have a lynched relative. However, those differences do not reach conventional levels of statistical significance in the decades between 1880 and 1900 and between 1910 and 1920. From 1900 to 1910, surviving family members were 13 percentage points more likely to have migrated in our reduced-form model (p < .10) and 15 percentage points more likely to have migrated in our full model (p < .05). When assessing this same relationship for 1920 to 1930, both results emerged as positive and statistically significant at the p < .01 level (b = .30 and .31, respectively). Similarly, pooled results from 1880‒1930 and 1900‒1930 are also positive and statistically significant (p < .01).
Our measures of county-level out-migration are only possible for individuals who are linked between census years, and census linking methods are better poised to link individuals who do not migrate. As a consequence, our linked sample is likely to understate the amount of out-migration between decennial censuses. We investigate this potential issue by including the full sample for each enumeration district and treating every unlinked individual as if they had moved out of their county of residence between census waves.23 In Table 6, we find that estimates from this analysis are largely consistent with those that use only the linked sample to measure the association between county out-migration and the lynching of a family member, specifically Table 3. Using this alternative sample biases our effect size estimates downward since there are unlinked individuals included in the sample who did not change counties, however, many of the estimates of county out-migration are statistically significant, lending potential validity to our previous estimates.
In this article, we have constructed a new individual-level longitudinal dataset using the U.S. Census and a well-established database of victims of racial terror lynching to assess the county out-migration patterns of the families of Black lynch victims in the southern United States between 1880 and 1930. Although earlier research (Tolnay and Beck 1992) has indicated that county-level lynching patterns were associated with county-level out-migration flows of Black households in the periods 1910–1920 and 1920–1930, no research has determined how the lynching of a family member was possibly associated with the out-migration of the affected family during a large portion of the Jim Crow era and the first Great Migration. Most prior quantitative studies of lynching and victimization focused on the victims themselves, and rightly so. Yet the consequences of lynching extended beyond individual victims to potentially have lasting and consequential reverberations on the families left behind. This study quantitatively addresses that fact.
Our findings suggest that families who experienced a lynching between 1880 and 1929 had a higher probability of migrating than individuals in their local community who did not have a family member killed by a lynch mob. This finding is robust to the inclusion of a broad set of individual and household characteristics, census enumeration district fixed effects, and various sample specifications. The observation that the death of a family member from extreme racial violence potentially prompted surviving household members to desert the counties where they lived aligns with extant contemporary studies of refugees who flee violent social circumstances for more sanguine environments (Bohra-Mishra and Massey 2011) and helps inform our understanding of the proximate inputs of the Great Migration. We also find relatively large estimates of migration for families whose household head was lynched relative to those families who did not experience the lynching of a family member. Under these circumstances, the loss of a central economic provider in the family could leave the remaining individuals in the household particularly susceptible to economic insecurity, causing some to strongly consider a move. It is also possible, and perhaps likely, that a portion of these families were forcefully compelled to migrate, like the family of Anthony Crawford, who we referenced earlier, because their paterfamilias was killed by lynching.
Our decade-specific results suggest that during the first Great Migration—which was prompted by various forces such as the boll weevil epidemic, improved economic opportunities outside of the South, and racial injustice—the families of Black lynch victims escaping the psychological and economic aftermath of direct experiences with racial violence were potentially highly represented among those migrating. As it relates to this, we observe that the highest probability of migration for the families of lynch victims happens between 1920 and 1930, shortly after the beginning of the Great Migration. This finding is congruent with what Tolnay and Beck (1992) observed more than 30 years ago: that higher concentrations of lynchings within a county were related to Black populations being more likely to migrate out of that county. Our analysis, however, demonstrates that the actual families who were struck by a lynching were more likely to flee their counties than their local peers who were not as directly affected by this form of violent racial terror. Moreover, witnessing a relatively high probability of migration of lynch victims' families during this period brings into relief how some of these families perhaps benefited from migration-based resources that were established in the prior decade (e.g., spatially dispersed social networks, employment prospects), while those who had a family member killed by lynching before the Great Migration were possibly left with scant opportunities to flee to safer locales owing to limited social and informational resources.
Future research would be enhanced by a series of efforts. For scholars interested in studying the life course outcomes of the families of lynch victims, it would be fruitful to increase the number of links across census years. We obtained intertemporal census links for up to 35% of victims' family members, and increasing that sample would improve the ability to obtain more precise estimates. Future research might also address the migration destinations of the families of lynch victims. Although we have provided evidence that the families of lynch victims were more likely to migrate than their neighbors who did not have a family member murdered by mob violence, which likely contributed to the Great Migration, our findings suggest only a marginally significant increase in the probability of migrating across state lines (see footnote 19). Given that participation in the Great Migration frequently unfolded in a multistep process (Alexander 1998)—households first moving to urbanized areas within the South and then to rapidly growing industrial cities in the North and West—tracing the migration trajectories of these families across multiple decades and perhaps generations may uncover similar findings about the probability of victimized families to ultimately leave the South.
And, of course, migration is but one factor potentially affecting the life course trajectories of people who lost a family member to mob violence. Linking surviving family members to multiple decades of census enumeration, as well as to a variety of administrative records, would enable us to examine in much greater detail how individual lives were shaped by the experience of vicarious victimization for those who left their locales and those who remained. Scholars might fruitfully investigate occupational and economic trajectories of the families of lynch victims, including their family formation patterns, involvement with the criminal justice system, and life expectancy and cause of death, to name only a few potential avenues of inquiry. Of particular interest to the ongoing public conversation about reparations would be the economic losses incurred by landowning Black families who were forced to flee violent environments. Gaining a fuller understanding of how the lives of people who survived the violent, unjust death of a close family member have unfolded in ways that are similar to and different from others living in their communities can help us better comprehend the consequences of inherited trauma, as well as the long-term impacts of structural violence.
We thank the editors and reviewers for their helpful comments on an early version of this article. Ryan Gabriel and Adrian Haws contributed equally to this article.
The Great Migration saw approximately 6.5 million African Americans migrate out of the South and is frequently conceived as occurring in two distinct waves. Following Tolnay (2003), we conceptualize the first wave to be from 1910 to 1930 and the second wave to occur between 1940 and 1970.
Scholars agree that the tempo of lynching and other forms of terroristic violence likely increased following the close of the Civil War and grew substantially throughout the withdrawal of federal troops at the close of Reconstruction. The marking of the “lynching era” as beginning in the early 1880s probably reflects increasing media coverage and the ability to document specific events more than it does a shift in the incidence of violence at this time (Perloff 2000; Weaver 2019). In fact, it was not until January 1883 when the Chicago Tribune newspaper began to publish annually a list of legal executions and lynchings in the previous year.
See the online appendix for a flowchart that illustrates the inclusion and exclusion of individuals into our sample.
This database includes victims of lynching in 10 southern U.S. states: Alabama, Arkansas, Florida, Georgia, Kentucky, Louisiana, Mississippi, North Carolina, South Carolina, and Tennessee.
We exclude from the analysis household members who were listed with nonfamilial relationships to the household head, such as “roomer” or “employee.”
Individuals who were lynched between 1890 and 1899 are unable to be matched to the 1890 census because most of this census was destroyed in a fire in the U.S. Department of Commerce Building in 1921. People lynched between 1890 and 1895 are matched to the 1880 census. People lynched between 1896 and 1899 were not matched to the 1880 census because this longer time span might include more false matches (see Bailey et al. 2008).
Similar to the sample restriction for households containing the victim of lynching, we restrict the control group to individuals with a familial relationship to the household head.
We do not include the 0.6% of family members of Black lynch victims who were not Black. This allows for a more congruent comparison with the control group.
The online appendix contains additional information about the creation of our linked sample.
See the online appendix for a discussion of how our record linking approach compares with another established record linking approach known as the IPUMS Multigenerational Longitudinal Panel.
Census linkage rates are reported in Table A2 in the online appendix.
One reason for our lower match rates in comparison with Price et al. (2021) is that some individuals in each starting census may have died between censuses and cannot be linked forward. Conversely, the match rates in Price et al. (2021) account for mortality by calculating the fraction of individuals linked to an earlier census among those who are alive in the subsequent census.
See Otterstrom et al. (2022) for an exposition of their method for linking individuals across consistent geographies over time.
We retain two destination counties for one of the 556 enumeration districts in our sample. This adjustment is applied if the most frequently observed county does not have a strict majority of links from the enumeration district and if the second most frequently observed county has at least half of the number of links as the most frequently observed county. For example, consider that 50% of individuals linked from enumeration district A are residing in county B in the later census. County C is included in the crosswalk if at least 25% of individuals linked from enumeration district A are later residing in county C.
We measure migration over 10 years for individuals whom we first observe in the 1900, 1910, or 1920 census. For individuals in the 1880 census, we measure migration over 20 years because the 1890 census records were destroyed in a fire.
We investigated supplemental models with separate interactions between whether a Black individual was living with a family member who was a lynch victim and gender, age, and homeownership. These analyses did not emerge as statistically significant. Results are included in the online appendix.
We gratefully acknowledge IPUMS Ancestry Full Count Data in our composition of control variables (Ruggles et al. 2021).
Homeownership is unavailable in the 1880 census, so, for consistency, this variable is omitted for our decade-specific analysis in 1880–1900 and the 1880–1930 pooled analysis.
We also investigate whether relatives in the household of a lynch victim are more likely to migrate from their state (results shown in online appendix). The model coefficients of interest across all periods are positive, with statistical significance (p < .05) emerging only for models that pool multiple periods. We consider our empirical approach to be best suited to address questions of county out-migration given the lower frequency of longer distances of migration (with just 34% of all movers crossing state lines and 13% moving from the South).
Although we believe that our focus on enumeration districts facilitates a robust empirical analysis, we also investigate the migration of family members of lynch victims as compared with same-race individuals living in the same county. We find results similar to those in Table 3 using this broader geographic control group, which are shown in the online appendix.
In each decade, this restricted sample ranged from 50% to 67% of all lynch victims, as shown in Table 1.
The sample for Table 6 is based on 2,905 family members of lynch victims living in the same household at the time of census enumeration and 593,344 Black individuals living in the same census enumeration district as the household of a lynch victim. For summary statistics for this sample, see Tables A1 and A2 in the online appendix.