The Legacy of Lynching? An Empirical Replication and Conceptual Extension

ABSTRACT

Research shows enduring impacts of lynching on a variety of modern outcomes. For instance, Messner, Baller, and Zevenbergen found that lynching is associated with contemporary white-on-black homicide. We propose a model describing how events from the past can have effects on events in the present. Essential to our framework is the notion of social forces of “resistance” that can impede or facilitate the temporal transmission of collective memories. We test “indicators of resistance” that influence the transmission of a collective memory supportive of a “legacy of lynching.” Analyses reveal that the positive and significant association between lynching and white-on-black homicide observed by Messner et al. is attenuated and becomes statistically nonsignificant with the inclusion of these indicators. Our results suggest that the temporal transmission of a racist cultural schema manifested through lynching is more likely where resistance is low. These findings have implications for how researchers can study historical legacies.

Notes

¹The 10 southern states referred to are Alabama, Arkansas, Florida, Georgia, Kentucky, Louisiana, Mississippi, North Carolina, South Carolina, and Tennessee.

²The lynching era roughly extends from 1880 to 1930.

³These southern states are Alabama, Arkansas, Florida, Georgia, Kentucky, Louisiana, Mississippi, North Carolina, South Carolina, and Tennessee.

⁴Jacobs et al. (2012) eliminated Nebraska from their analysis because the nonpartisan legislature in the state makes their variable for Republican strength immeasurable.

⁵We are not suggesting that the study of social phenomena such as historical legacies lends itself to the application of lawlike propositions and predictions from the physical sciences. We fully recognize the probabilistic nature of the social world. Still, we do believe that it can be beneficial to borrow conceptually from the logic of such laws.

⁶Indeed, the possible relationship between these two forms of “resistance” is open to question. For example, would higher levels of resistance from the local black population strengthen or weaken the transmission of historical racist schemas? This is an interesting topic for future research but beyond the scope of our current study.

⁷This discussion draws heavily from Messner et al.’s (2005) Data and Methods section due to the scope of our research questions.

⁸These counties, as identified by their Federal Information Processing Standards codes, are not formed into clusters as determined by the variable “id1880” within the Horan and Hargis (1995) County Longitudinal Template: 5023, 5063, 5141, and 5145; 12087; 21007 and 21039; 37063, 37135, and 37183; 37051, 37093, and 37155; and 45019, 47049, 47133, and 47137.

⁹Upon adding decade-specific migration data to the original Messner et al. data set, Chattahoochee County, GA, was dropped because it did not exist in the county migration data, thus decreasing our data set from 661 to 660. Analysis reveals only an extremely small change in model results using the 660 sample compared to the 661 sample, not challenging the validity of our results.

¹⁰We thank Steven Messner, Robert Baller, and Steven Zevenbergen for generously sharing these data.

¹¹On this point, we replicate the same methodology used by Messner et al. (2005) through application of the weight “WTIMPST2.”

¹²To stay consistent with Messner et al., the lynching variable is operationalized as lynchings of both blacks and whites. In their analysis, limiting the lynching measure to only black or white victims is inconsequential to their results.

¹³We adopt Jeremy Porter and Frank Howell’s definition of a WFSA and thank them for generously providing these data.

¹⁴The migration data from 1950 to 1970 is produced by Bowles et al. (1990), and the 1970 to 1980 migration data is created by White et al. (1987), with both being distributed by Inter-University Consortium for Political and Social Research.

¹⁵SHR received homicide data from Florida only for six years and Kentucky only for nine years of the study duration; hence, population figures for Florida and Kentucky counties are multiplied by six and nine, respectively.

¹⁶Some of the models estimated by Messner et al. (2005) also included the percentage of churches in the county in 1850 that were Presbyterian and level of agricultural production in 1880 to tap the local strength of a Scots-Irish tradition and cultural code of honor. However, inclusion of those variables in their analysis of white-on-black argument homicides reduced by nearly 50% the number of cases available. Furthermore, neither variable attained statistical significance, and the relationship between lynching and homicide remained statistically significant. We chose to replicate their model (Messner et al. 2005:650) that did not include those two predictors.

¹⁷The factor analysis is done in Stata 12.

¹⁸Percentage urban and percentage whites ages 15 through 29 are not loaded on these factors, and thus they are not included in the factor variables, but they are included as separate control variables in the analysis. Refer to Appendix Table A1 for detailed results of the factor analysis.

¹⁹This actually represents a very minor change from the original results reported by Messner et al. Although not attaining statistical significance by conventional thresholds, the t-value for the coefficient representing white family structure in their analysis is 1.79 (.195/.109).

²⁰We experimented with an alternative measurement for the WFSA variable by dividing the number of academies by the total county white population. We found that the alternative measurement strategy had no effect—either on the statistical significance of the coefficient for the variable itself or for the partial relationship between lynching and homicide.

²¹See Appendix Table A2 for bivariate correlations of variables utilized in the analysis. In addition, we investigated VIF and Tolerance values for all predictors in the multivariate equations. Only values for the absolute value net-migration rates for 1950 to 1960 and 1960 to 1970 show relatively high VIF values at 4.25 and 5.08, respectively. The remaining covariates in our analysis have VIF values below 2. When the decade-specific migration measures are combined into a single variable—either through the addition of standardized scores or through a factor created with confirmatory factor analysis—the essential findings reported in Table 3 remain unchanged.

²²We thank Rory McVeigh for generously sharing these data.

²³We test for the interaction between lynching and the migration variables to see if there is any evidence that population churning moderates the effect of lynching on homicide. None of the interactions terms are statistically significant. Moreover, we estimate models that use black population churning and none of the decade specific absolute value net-migration rates are significant.

²⁴The results of these alternative analyses are available from Ryan Gabriel upon request.

Additional information

Notes on contributors

Ryan Gabriel

Ryan Gabriel is Assistant Professor of Sociology at Brigham Young University. His research focuses on issues related to race, ethnicity, and stratification in the United States. In addition to the legacy of historical lynching, he has investigated the patterns of residential mobility and locational attainment for mixed-race couples.

Stewart Tolnay

Stewart Tolnay is S. Frank Miyamoto Professor of Sociology at the University of Washington. His research has focused on the history of racial violence in the American South and the Great Migration of southerners to the North and West. He is the author of The Bottom Rung: African American Family Life on Southern Farms (University of Illinois Press, 1999), coauthor with E.M. Beck of A Festival of Violence: An Analysis of Southern Lynchings, 1882 to 1930 (University of Illinois Press, 1995), and coauthor with Amy K. Bailey of Lynched: The Victims of Southern Mob Violence (University of North Carolina Press, 2015).