In this paper, members of the Virality of Racial Terror Project (VRT) explain our data-driven work and discuss both the inherent challenges and opportunities of making our large-scale data serve research questions about individual cases of racial violence. These challenges raise a longstanding issue in the digital humanities (DH): how best to utilize data at scale for questions about the nuance or specificity of individual cases so often the focus of humanities research? Some scholars have pointed to this issue as a fundamental flaw of computational methods. Da (2019), for example, argues that such methods reveal a “mismatch between the statistical tools that are used and the objects to which they are applied” (p. 601). Through our work, however, we observe that our large-scale data can serve as foundational context from which further work on individual cases may be conducted. As Kirilloff (2022) suggests, “the information produced from computational methods should be understood as a type of context” (p. 2). While she speaks to literary text analysis, we find her argument equally relevant to our work where the data represents real people–victims of racial violence–and where specific cases involve their stories. In this regard, we have found promise in our data. In some instances, it provides opportunities for serendipitous discoveries that can lead to new information about individual cases.

Our paper begins with a description of our data-driven work. By utilizing two authoritative records of lynchings–the Tolnay-Beck-Bailey Inventory and the Seguin-Rigby Dataset–we have identified over 50,000 newspaper reports of lynchings published between 1865 and 1921. All of these reports are tied to specific cases. Our processes included text-mining the Chronicling America archive for lynching victim names, conducting light OCR correction, approximating the beginnings and ends of reports within the full texts of newspapers, and hand-keying a training set of lynchings reports then used to fine-tune BERT for the binary classification task of labeling further reports across our newspaper data. The result of these processes is the Dataset of U.S. Lynching Reports (DUSLR). Using this new resource, we have been able to analyze the larger system of newspaper reporting on lynchings and assess how particular cases were narrated through this system.

Our paper then notes some limitations of this data. While DUSLR provides an incredible amount of information about racial violence in America, it tends to only include a small amount of information about each case. We demonstrate this fact by highlighting specific examples. These include cases with minimal reporting, but also cases with seemingly large amounts of coverage from dozens of reports from across the country. In these latter instances, DUSLR still tends to provide only basic details largely because these reports spread through newspaper exchanges wherein editors copied each other’s content–a common practice throughout the period and a phenomenon studied in projects like the Viral Texts Project (Cordell, 2015). This aspect of the data informs our research on newspaper exchange systems, newspaper attitudes toward lynchings, and the anti-lynching movement, but it creates unique challenges when trying to investigate individual cases.

In addition, many reports in the data often describe cases from a racist perspective. This is an issue caused by racist newspaper editors as well as the entire information system created in part by newspaper exchanges. These exchanges, in other words, often iterated and reinforced racist biases on the issue of lynchings. In these examples, they tend to lack investigative elements. They do not identify the perpetrators of lynchings. We have noted instances where victim names were incorrectly reported. We have also identified reports that attempt to justify the brutality and violence of lynchings. These aspects of the data reflect what historians and activists have long said about the period. It is what Ida B. Wells (1895) sought to rectify in A Red Record. More recently, it demonstrates what Williams (2021) has described of the period, that “many of the individuals and institutions recording and narrating these events stigmatized victims” (p. 150). Yet there are other reports in the data that emphasize the deeply unjust nature of lynchings, creating counterpoints to the more ubiquitous racist perspectives. These counterpoints often signal the labor of the anti-lynching movement. These variables in the data therefore require informed and conscientious engagement on the part of any scholars or activists utilizing DUSLR for research purposes. An awareness of these biases is especially important for any projects seeking to address the painful legacies of individual cases.

Finally, our paper outlines some examples and processes we’ve used to both address these challenges and identify research opportunities. In particular, we describe two related cases in the data: the murder of Will Godley in 1901 and the murder of Mumford Godley in 1902. By close reading reports of Mumford’s murder in DUSLR, we uncover that these two lynching victims were relatives. Their murders by mobs occurred in separate instances and places. These events impacted whole communities and families. In Will Godley’s case, it led to the expulsion of an entire Black community from Pierce City, Missouri, where he and Mumford had resided. In Mumford’s case, it was a continuation of mob violence following the family’s exile from Pierce City. By tracing their stories as told in the reporting captured by DUSLR, we demonstrate how interrogation of the data at the level of individual cases can reveal connections within the data otherwise obscured–an opportunity for research that we hope further scholars will pursue. We also describe how we have used other sources, such as census data, genealogical records, marriage records, and Black newspapers, to enrich our knowledge of these cases and build a fuller picture of their history from the foundational context provided by DUSLR. We argue that these mixed methods are necessary to overcome the limits of the information provided by DUSLR. By the end of our paper, interested scholars should have a better grasp on the development of DUSLR as well as suggested humanities approaches to utilizing DUSLR while investigating individual cases represented in the data.