Abstract
This paper advocates an actor-centered, relational view of agency and proposes Quantitative Narrative Analysis (QNA) as a promising method for operationalizing and measuring agency. QNA organizes the information contained in narrative texts by exploiting the invariant linguistic structural properties of narrative—namely, sets of SVOs (Subject, Verb, Object) organized in predictable sequences and where in narrative S are actors and V are actions. The relational data made available by QNA are ideally suited for analysis with geographic information systems (GIS) tools, sequence analysis, or network analysis. These tools preserve the centrality of agency (actors and their actions) in social scientific explanation of social reality. An application of QNA to newspaper stories of lynchings in Georgia (1875–1930) will illustrate the power of this approach. The paper complements the illustration of this quantitative way of measuring agency with discourse analysis—another popular social science approach to texts. We will rely on this approach to illustrate how linguistic and rhetorical strategies can be used to hide agency in texts and the challenges (and solutions) this poses for measurement: How can we measure something that is not there?
1. The Study of Events as Event Counts: Where Are the Actors?
The Springfield Daily Republican of April 24, 1899, reported the following news:
NEWNAN, Ga., Apr. 23— Sam Holt [also known as Sam Hose, after assuming that name upon moving to Coweta County, Georgia], the murderer of Alfred Cranford and the ravisher of the latter’s wife, was burned at the stake, near Newnan, Ga., this afternoon, in the presence of 2000 people. The black man was first tortured before being covered with oil and burned. An ex-governor of Georgia made a personal appeal to his townspeople to let the law take its course, but without the slightest avail. Before the torch was applied to the pyre, the Negro was deprived of his ears, fingers and genital parts of his body. He pleaded pitifully for his life while the mutilation was going on, but stood the ordeal of fire with surprising fortitude. Before the body was cool, it was cut to pieces. The bones were crushed into small bits, and even the tree upon which the wretch met his fate was torn up and disposed of as “souvenirs.” The Negro’s heart was cut into several pieces, as was also his liver. Those unable to obtain the ghastly relics directly paid their more fortunate possessors extravagant sums for them. Small pieces of bones went for 25 cents, and a bit of the liver crisply cooked sold for 10 cents. As soon as the Negro was seen to be dead there was a tremendous struggle among the crowd, which had witnessed his tragic end, to secure the souvenirs. A rush was made for the stake, and those near the body were forced against it and had to fight for their freedom. Knives were quickly produced and soon the body was dismembered.
Perhaps not surprisingly, Louis P. Le Vin, a private detective hired by a group of prominent Chicago citizens spurred on by journalist and activist Ida B. Wells-Barnett to investigate Sam Hose’s lynching, would conclude his report with the words: “I made my way home thoroughly convinced that a Negro’s life is a very cheap thing in Georgia.” 1
Social scientists have studied events such as lynchings, strikes, and riots as “event counts” (i.e., number of events) and have employed multivariate statistical techniques to handle these counts—typically, econometric models in which a “dependent” variable Yi measuring the number of events (e.g., lynchings, episodes of political violence, or strikes) is regressed on a set of “independent” variables X1i, X2i . . . Xki, each measuring different effects. For instance, Beck and Tolnay, in one of the best quantitative analyses of lynching, related the number of lynchings that occurred in Jim Crow South between 1882 and 1930 to a set of independent variables: size of black population in a county, black crime rate, deflated price of cotton, etc. Interpreting the signs and statistical significance of the parameters in their model, Beck and Tolnay (1990:526) concluded:
Net of other factors, lynchings were more frequent in years when the “constant dollar” price of cotton was declining and inflationary pressure was increasing. Relative size of the black population was also positively related to lynching. We conclude that mob violence against southern blacks responded to economic conditions affecting the financial fortunes of southern whites—especially marginal white farmers. These effects were significantly more important in the decades before 1900, possibly because of the declining importance of agriculture, the “Jim Crow” disenfranchisement of blacks, and the increasing out-migration of blacks and whites from the Deep South.
For all the considerable knowledge generated by multivariate statistical models of the structural determinants of the temporal and/or spatial dynamic of lynching, where are the actors? Where are the interactions among those actors? Where is the mob capturing, torturing, emasculating and burning the Negro? 2 Where is the odd individual—the former governor of Georgia, William Y. Atkinson, in the case of Sam Hose’s lynching (Ellis 1992:110–12)—who would at times meet head-on the lynching mob and try to prevent summary justice? Where are the relic hunters who would scout for and sell body parts as souvenirs? 3 Where is the macabre ritual of men, women, and children—some 2,000 of them, in Sam Hose’s lynching, brought in from Atlanta by special trains (Ellis 1992:4)—applauding, cheering, shouting, and laughing, turning the ghastly happening into a gala party? 4
2. The Question of Agency: Theoretical and Methodological Considerations
The search for answers to the questions about actors and their actions (and their capacity to act) has generated a lively theoretical debate in the human sciences, often referred to as the structure versus agency question: To which extent do “human beings make their own decisions or make their own history” (Burke 2005:127) vis-à-vis social structures? The debate has cut across several disciplines, from history (e.g., Callinicos 2004) to sociology (e.g., Emirbayer and Mische 1998), economics and development studies (e.g., Sen 1985; Kabeer 1999), political science (e.g., Sibeon 1999), and international relations (e.g., O’Neill, Balsiger and VanDeveer 2004).
On one side of this classic debate, structuralists conceive of human behavior as largely under structural constraints, of history as a process where subjects or goals and social change are the outcome of structural contradictions. Marx famously championed this view in The Eighteenth Brumaire of Louis Bonaparte: “Men make their own history, but [. . .] not [. . .] as they please [. . .] under circumstances chosen by themselves, but under circumstances directly encountered, given and transmitted from the past” (Marx 1970:96). French historian Fernand Braudel was no less fond of structures and leery of events. History can only be understood (and should only be studied) in terms of structures, these temporal “expanses of slow-moving history”“these depths, this semistillness” (Braudel 1980:33). On the opposite side of the debate, rational choice scholars have adopted “methodological individualism”—the idea that “the elementary unit of social life is the individual human action” (Elster 1989:13)—to explain social actions and outcomes, emphasizing the autonomy and rationality of the individual (Coleman 1990). Against these starkly conflicting positions, Giddens (1976), Sewell (1992), Emirbayer and Goodwin (1994) have elaborated more nuanced views of the relationship between structure and agency. “Human agency and structure, far from being opposed, presuppose each other” (Sewell 1992:4). Structures “must not be conceptualized as simply placing constraints on human agency, but as enabling” (Giddens 1976:161). More recent theoretical discussions of agency (e.g., Archer 2003; Dépelteau 2008) aim to specify how structures affect actors’ interactions, as well as how social actions reshape social structures.
Rich as the theoretical work on agency has been, methodological development has lagged behind. To be sure, researchers in different disciplines have, implicitly or explicitly, proposed ways of measuring agency. Historians spend years in the archives to uncover precious few instances of agency, of actors’ actions. Indeed, the word agency appears repeatedly in two edited collections often cited in this paper (Brundage 1997; Dailey, Gilmore, and Simon 2000). Historians’ work, however, rarely translates in quantifiable measures that could provide synthetic, immediate grasps of the phenomenon. 5 Quantification is rather the domain of social scientists who, on the other hand, are remarkably silent about agency even when working with counts of historical events, from lynchings to riots or strikes. Nevertheless, attempts to provide measures of agency and related concepts have come from various social science quarters. Scholars working in the rational choice tradition and analytical sociology have used game theory (e.g., Ordeshook 1986), simulation models of action or agent-based models (e.g., Macy and Willer 2002; Gilbert 2008; Manzo 2009; see also the review by Manzo 2007) and experimental designs (e.g., Fehr and Gintis 2007) to analyze individuals’ strategies and decisions, as well as to investigate meso and macro interactions among groups, organizations, states, and aggregate social outcomes (e.g., Gambetta 1993; Hechter 1987; March and Olsen 1998). Development and empowerment scholars, relying on psychological notions of agency as self-efficacy and self-determination (e.g., Ozer and Bandura 1990; Deci and Ryan 2000), have proposed choice-making as a way to capture women’s agency (Kabeer 1999) and have used measurement scales in surveys to study actual instances of decision-making activities (Alkire 2005). Relational sociology has suggested a network methodological approach to measure agency (e.g., Emirbayer and Goodwin 1994:1442–46), although only a few studies have taken up the suggestion (e.g., Passy 2001; Stevenson and Greenberg 2000; see also Diani and McAdam 2003).
3. Measuring Agency: An Alternative
Our goal in this paper is to make a methodological contribution to the operationalization of agency and its quantitative measurement by exploiting the link between agency and narrative (as a sequence of actors doing/saying something—i.e., acting). Agency is not only “what the person is free to do and achieve in pursuit of whatever goals or values he or she regards as important” (Sen 1985:203) but also what the person actually does (action). And what the person does involves two components: the action proper and the capacity to give meaning to this action (Sewell 1992:18, 19; Emirbayer and Goodwin 1994:1443; Kabeer 1999:438). The close link between agency and action suggests another link, that between agency and narrative. After all, narrative is about action. It is the recounting of actors acting, of someone doing something. For some 2,500 years, rhetoricians have translated narrative into what is known to us as the five Ws of journalism: Who, What, When, Where, Why (and How) (in its Latin rendering: quis, quid, cur, quomodo, ubi, quando, quibus auxiliis) (Franzosi 2012).
It is by exploring that link (between agency and narrative) that we propose a way to operationalize the concept of agency through a set of detailed procedures known as Quantitative Narrative Analysis (QNA), a tool for the collection and analysis of large volumes of narrative texts based on computer-assisted narrative grammars. QNA delivers quantitative measurements that can be turned into variables for use in traditional statistical models (e.g., Franzosi 1994, 2004:113–8). Its real advantage, however, is that it provides a way to abandon a variable-centered approach to socio-historical reality in favor of an actor-centered, but still quantitative, approach. As an actor-centered methodological approach, QNA represents an ideal tool for measuring agency. Furthermore, by focusing systematically on actors, their actions, and, critically, their spatio-temporally situated interactions, QNA tackles different questions from variable-centered socio-historical research based on event counts (Gurr 1974; see also Abbott 1988; Tilly 2008; Franzosi 2010).
The limits of QNA lead us to discuss also a qualitative approach to measuring agency: discourse analysis (Brown and Yule 1984; Johnstone 2007). We have chosen discourse analysis not simply because this approach provides a qualitative alternative to QNA (other social science approaches to texts, such as conversational analysis, do that as well). We have chosen discourse analysis because it focuses on the linguistic and rhetorical strategies used in texts to hide agency (e.g., passivization and nominalization). As such, discourse analysis helps to highlight the danger of an uncritical and superficial application of QNA: How can we measure something that is not there? How can we measure agency if agency is hidden? But this awareness of the limits of QNA also helps to bring out solutions—namely, turning passive forms into active ones (e.g., “the negro was lynched” as “unknown lynches negro”) and turning nouns into verbs/sentences (e.g., “the lynching of the negro” as “unknown lynched negro”). But to the extent that discourse analysis privileges the exploration of the power and ideological dimensions of texts, discourse analysis also helps to put the issue of newspaper bias in a broader perspective, to understand the agency of newspapers. When it came to lynchings, Jim Crow South newspapers were not passive conduits of information but actors in their own right and makers of history, no different from the mob or the lynched alleged criminals.
3.1. Narrative and Narrative Grammars: Quantitative Narrative Analysis (QNA)
The lynching of Sam Hose reported by the Springfield Daily Republican of April 24, 1899, is an archetypal narrative text—a sequence of actions and events arranged in chronological order. Linguists and literary critics have identified action (by an actor) and the sequence of actions as the main features of narrative texts and defined narrative as “one method of recapitulating past experience by matching a verbal sequence of clauses to the sequence of events which [it is inferred] actually occurred” (Labov 1972:369–70), or, more simply, as “a succession of events” (Rimmon-Kenan 1983:2–3). Furthermore, in the unfolding of a story, some actions and events play a greater role than others in altering a narrative situation; they are consequential rather than simply sequential. Finally, not all sequences of events produce a narrative; the sequence has to make sense (to be characterized by semantic coherence) and produce a coherent story. Expressed differently, without a story linking the different sequential parts of a text, there is no narrative (Rimmon-Kenan 1983:15).
The newspaper’s account of Sam Hose’s lynching is, undoubtedly, a narrative, a meaningful succession of actions and events, some of which are consequential. Thus, the alleged slaying of Mr. Alfred Cranford and the raping of his wife Mattie Cranford by Sam Hose are consequential actions. The ensuing arrest of Sam Hose, his snatching from the authorities by a threatening mob (consequential action), the burning of the victim, and the relic hunters’ taking of body parts as souvenirs provide a mix of sequential and consequential actions within the larger story of Sam Hose’s lynching.
In the narration of an event—a story—not only are some actions sequential and consequential, but they also, at a deep-structural level, correspond to a limited number of “types of actions.” It was Propp who first drew attention to this characteristic of narrative, back in 1928. In his influential Morphology of the Folktale (first translated into English in 1958), Propp reduced all Russian folktales to an invariant set of 31 distinct functions, where a function is “an act of a character,” or an action (Propp 1968:21). Greimas reduced Propp’s 31 narrative linear functions to a set of six fundamental dichotomous invariant roles (“actants”): subject-object, sender-receiver, helper-opponent (Greimas 1971:798–99, 805). Greimas further argued that, at the surface level, stories are nothing but organized sequences of narrative units minimally constituted by actants and functions (basically verbs or processes of either “doing” or “being”) (Greimas 1971:799, 800, 802). This combination of invariant deep-structural and surface-level properties of narrative results, for Greimas, in a narrative grammar or story grammar, at the border between linguistics and general semiotics.
A story, then, can be conceived of as a set of distinct narrative units that, for simplicity, we can equate to the basic canonical form Subject-Verb-Object (SVO). In historical narratives, both the Subject and Object of an event are typically social actors, be they individual, collective, or organizational/institutional actors; the Verb refers to social actions characteristically indicating acts of doing something or saying something (Greimas 1971:800). Each of the three SVO elements can have a number of modifiers (i.e., the characteristics of actors and the circumstances of actions; for the history of this terminology, see Franzosi 2012). Thus, Subject and Object may have the following modifiers: the name and last name of an actor, its job, race, gender, religious or political affiliation, etc. The modifiers of a Verb include time and space—when and where an action occurred—but also the reason, outcome, or instrument of that action.
The SVO structure with its modifiers (a “semantic triplet”) functions as a “story grammar,” the “set of rules that provides the categories into which the various invariant elements of a story fall (e.g., actor, action, time, space), the nature of each category (e.g., a text, a number, a date; allowed to occur one or multiple times), and their reciprocal relationships” (Franzosi 2010:23). 6 Essentially, a story grammar is the key parsing tool employed to extract information on Who, What, When, Why, and Where—the 5 Ws of journalism + H, How—from narrative texts (Franzosi 2012).
Contrary to traditional content analysis coding schemes, the categories of a story grammar “are formally and explicitly related to one another throughout the coding scheme via a set of rewrite rules (e.g., subjects are linked to actions, actions to objects, and subjects, actions, and objects are linked to their modifiers)” (Franzosi 2010:35). A rewrite rule is symbolized by a right-pointing arrow (→) which indicates how an element to the left of the symbol can be rewritten in terms of the elements on its right (Franzosi 2010:23–24). A semantic triplet can thus be rewritten in terms of its constitutive components: 7
The elements of the semantic triplet can in turn be rewritten, down to their “terminal” symbols (those found in the language itself). For instance, <subject> could be rewritten as follows:
A verb and its modifiers (or, more precisely, its circumstances) would instead look like this:
The relational nature of a story grammar makes it possible to implement a story grammar in a Relational Database Management System (RDBMS). Franzosi developed a specialized software—PC-ACE (Program for Computer-Assisted Coding of Events)—which utilizes computer-assisted story grammars as the main tool to collect, organize, and store large bodies of narrative data. 8
Within PC-ACE, the Springfield Daily Republican story of Sam Hose looks like this (where the categories of the grammar appear in black and the information taken from the newspaper in italics):
(Semantic Triplet: (Subject: (Actor: ?)) (Verb: (Verbal phrase: tortured) (Time: (Time expression: today) (Article date: 04/23/1899) (Time of day: afternoon)) (Space: (City: (Spatial direction: near) (City name: Newnan))) (Object: (Actor: negro) (Personal characteristics: (Name: Sam Hose)))
(Semantic Triplet: (Subject: (Actor: ?)) (Verb: (Verbal phrase: deprived) (Time: (Time expression: today) (Article date: 04/23/1899) (Time of day: afternoon)) (Space: (City: (Spatial direction: near) (City name: Newnan))) (Object: (Actor: negro) (Personal characteristics: (Name: Sam Hose))) (Object: (Case: of) (Physical object: ears)) (Object: (Case: of) (Physical object: fingers)) (Object: (Case: of) (Physical object: genital parts)) (Triplet relation: while))
(Semantic Triplet: (Subject: (Actor: negro) (Personal characteristics: (Name: Sam Hose))) (Verb: (Verbal phrase: pleaded for life) (Action type (Adverb): pitifully) (Time: (Date: (Time expression: today)(Article date: 04/23/1899) (Time of day: afternoon))) (Space: (City: (Spatial direction: near)(City name: Newnan)))))
(Semantic Triplet: (Subject: (Actor: ?)) (Verb: (Verbal phrase: covered) (Time: (Time expression: today)(Article date: 04/23/1899) (Time of day: afternoon)) (Space: (City: (Spatial direction: near) (City name: Newnan))) (Instrument: oil))(Object: (Actor: negro) (Personal characteristics: (Name: Sam Hose))))
(Semantic Triplet: (Subject: (Actor: ?)) (Verb: (Verbal phrase: burned) (Time: (Time expression: today) (Article date: 04/23/1899) (Time of day: afternoon)) (Space: (City: (Spatial direction: near) (City name: Newnan))) (Object: (Actor: negro) (Personal characteristics: (Name: Sam Hose))) (Object: (Case: at) (Physical object: stake))
4. An Application of QNA: Lynchings in Georgia (1875–1930)
Sam Hose is one of the 462 victims of 392 lynching events (36 events were multiple lynchings, totaling 106 victims) that occurred in Georgia between 1875 and 1930. We relied on 1,332 newspaper articles and PC-ACE to build a large database of these Georgia lynchings yielding 7,070 semantic triplets. The analyses presented here are based upon 340 events and 388 victims coded thus far.
4.1. Newspapers as Sources of Historical Data
With such widely known titles as Bad News, More Bad News, Really Bad News, 9 it may seem preposterous to rely on newspapers as sources of data for a historical study of lynchings. Would we be studying historical lynchings or newspaper representations of lynchings? (For a study of newspapers as sources of socio-historical data, see Franzosi [1987].)
Certainly, with only a handful of black newspapers available, the representations we have mostly reflect a white perspective. Yet, for all the biases of newspapers, at least in the case of lynchings, for the most part, we have no alternative sources. As very few cases were ever brought to trial, all the information we have is from newspapers, and white newspapers, Southern newspapers at that. As historian Brundage (1993:296) writes, in one of the most authoritative studies on lynching in Georgia (and Virginia): “newspapers are the essential source for my study.” Brundage acknowledges: “Despite the serious limitations of white accounts, there simply is no other foundation upon which to base a comprehensive study of lynching” (Brundage 1993:294; see also the historians Matthews [1970:iv] and Ellis [1992:6]). Indeed, all available lynching counts upon which the quantitative scholarship on lynchings is based, by the National Association for the Advancement of Colored People (NAACP) and the Tuskegee Institute, come from the same source: newspapers (Ellis 1992:15; Tolnay and Beck 1992:259–63).
To keep bias in check, Brundage recommends “collecting accounts from as many different newspapers as possible” (Brundage 1993:294; see also Ellis 1992:15). Indeed, that is what we did. Our data come from 212 different newspapers (at the current state of data collection). The Atlanta Constitution, however, with a frequency of 473 articles, provides the bulk of the narratives (the next most frequent entry, the Macon News, has 46 articles). Most lynching events in our database are reported in one or two newspaper articles: 22 in one brief article, 87 in two articles, 112 in three, 63 in four, and declining after that (but even multiple-newspaper lynchings are often copies of the same brief story).
4.1.1. Questions about Agency for the Database: Some Caveats
In keeping with the kind of questions one typically asks of stories, let’s query this large dataset to find answers to questions about actors and actions in lynchings, questions about agency. Some words of caution about the answers you are about to hear. First, do not confuse the limits of QNA with the limits of the sources used for QNA (e.g., newspaper articles). QNA works independently of the validity of the narratives. Work on questionable data requires researchers to pay close attention to issues of data validation and data interpretation, regardless of the power of the method adopted.
Second, with data collection, 10 data cleaning, and data aggregation still ongoing, our answers are not so much intended as substantive answers about lynchings as answers to the question at the heart of this paper: Can QNA provide quantitative measures of agency? Nor are the analyses meant to exhaust the range of statistical techniques we can apply to QNA data, which can vary from simple exploratory data analysis (EDA) to traditional regression, and, notably, to tools based on the narrative underpinnings of the data: namely, sequence analysis, 11 GIS tools, and network analysis (on the range of techniques of analysis for QNA, see Franzosi [2010:107–141]).
Finally, the numbers (basically frequency distributions of words for specific categories) will vary depending upon what we count: events or individual verbal expressions of violence (which depends upon the richness of detail provided by the different newspaper articles). To avoid bias in numbers introduced by descriptions of different “thickness,” we rely here on counts of both events and individual violent actions. For illustrative purposes, we analyze these numbers with simple EDA, network graphs, and GIS maps (the tools most closely linked to agency, of actors operating in relation to other actors, in time and space).
4.1.2. Answers from the Database: The Broad Picture
Table 1 shows a list of the most frequent individual and collective actors found in the database. 12 African American males are the protagonists of the stories told by the newspaper articles, together with (black and white) females and mobs.
Aggregated Actors in the Lynching Database (Frequency >80)
The list of the most frequent actions (Table 2), certainly not surprisingly, shows that lynchings involved a great deal of violence (primarily against people), coercion, movement (going, searching, coming), and control. Acts of oral communication are also frequent.
Aggregated Actions in the Lynching Database (Frequency >110)
The real power of QNA is its relational properties. By relating actors to their actions and the targets of their actions, we can use directed network graphs to provide a visual map of the social relations of lynching. To construct network graphs, we extract all the triplets (Subject-Verb-Object) stored in our database. We then select different subsets of triplets for specific spheres of action (e.g., all the triplets involving verbs of violence or communication). Using a specific subset of triplets as input, we then use PC-ACE to create a k×k square weighted adjacency matrix
Figure 1 shows the network graph of “violence against people” (a sphere of action that includes verbs of violence such as “kick,”“punch,”“wound,”“torture,” and “kill”). The graph visualizes the violent relationships between the different sets of actors labeled in the graph (mob, law enforcement, lynched negroes, groups, white men, white women, white girls, etc.). In the graph, the size of the arrowheads and the thickness of the ties/arcs are proportional to the strength, or magnitude, of the relation between the two connected aggregated actors (as measured by the frequency of actions between those ordered paired actors; frequency reported in the graph near each node). Lynched African Americans stand at the center of this network, the objects of violence by mobs (338 individual actions of violence in the database), generic groups 13 (81 actions), whites (79 actions), and law enforcement agents (27 actions). Conversely, the most likely victims of violence performed by African American males are white men (more than 60 actions), law enforcement (56 actions) and white women (37 actions). The graph also shows that mobs did not target African Americans exclusively. They would also attack law enforcement actors to impose “lynch law” (43 actions). Sheriffs and deputy sheriffs were almost as likely to be the target of actions of violence by mobs as by the African Americans eventually lynched.
Network of violence against people (Georgia, 1875–1930).
That strained relationship between law enforcement, the mob, and the lynched negroes is further underscored by the network graph of coercion (a sphere of action of verbs such as “forced,”“dragged,”“took,”“carried”), as illustrated in Figure 2.
Network of coercion (Georgia, 1875–1930).
As expected, lynching events present a great many coercive acts against African Americans (see bottom right of the graph). Mobs (129 actions), whites (36 actions), and groups (20 actions) were all involved in coercing African Americans toward their grim destinies. Law enforcement agents also coerced African Americans (46 actions), but, differently from white mobs, they performed such actions within the law—namely, taking to jail captured African Americans accused of an alleged crime. Perhaps more surprisingly, sheriffs and deputy sheriffs were also targets of mob coercion (49 actions, the second most frequent interaction of coercion after mobs coercing Negroes). These attacks on sheriffs and deputy-sheriffs took different forms: A mob would ambush a sheriff on his way to jail right after the capture of an alleged criminal or on his way to a different town for “safe keeping” of a prisoner already in custody. The daring storming of jails was also not uncommon, at times forcefully battering doors down, but more often obtaining the keys from the jailers.
Rare are the recorded cases of a sheriff’s collusion with the mob (e.g., the rumored case of Sheriff Kendrick in the burning at the stake of Paul Reid and William Cato in Statesboro on August 16, 1904). Just as rare in Georgia are the cases of sheriffs’ serious attempts to resist the mob. 14
Needless to say, the numbers and measurements one gets depend upon what one counts. The story that network graphs tell depends upon the stories newspapers tell. How would network graphs change if newspaper articles provided more or less “thick descriptions” of lynching events? We know from the description of Sam Hose’s lynching, with which we opened, that he had his ears, fingers, and genitals cut off, that his body, heart, and liver were cut to pieces, his bones crushed, and that he was burned at the stake. This “thick description” would yield a larger count of individual violent actions than one where all we may know is that the lynched victim was hanged or shot. What difference does it make? If we are interested in a specific type of relationship between any two actors (e.g., violence), greater thickness of description within that specific sphere of action would not affect the structure of the network (i.e., the actors and their relations). After all, the underlying unweighted adjacency matrix
By systematically inspecting any difference in the elements of the unweighted and weighted adjacency matrices
Social actors do not only act in relation to other actors as visually brought out by network graphs. They also act historically, in time and space. The time plots of Figure 3 show the temporal dynamic of lynchings in Jim Crow South and Georgia. The plots of lynched victims in 10 Southern states and in Georgia reveal that the number of lynchings rose sharply after Reconstruction to peak toward the end of the nineteenth century and declined steadily thereafter until they virtually disappeared by the late 1920s (yet, with occasional spikes in the postwar period). Contrary to the rest of Jim Crow South, lynchings in Georgia remained at a steady high level well into the early 1920s.
Lynched victims and lynched victims burned (Georgia and 10 Southern states, 1875–1930).
The time plot of the percentage of lynched victims burned also shows that in the twentieth century, lynchings in Georgia, while declining in number, rose in brutality in the forms of punishment, with two spikes in the percentage of lynched victims burned in the early 1900s. In our data, shooting is the most common type of punishment (151 lynched victims were shot, or 39%), followed by hanging (64 victims; 16%) and burning (10 victims; 3%). A total of 148 lynched victims (38%) were subjected to more than one type of violent punishment (e.g., shooting and burning), while the remaining 15 victims (4%) endured other types of violence, like being beaten to death, strangled, or drowned.
Figure 4 shows a GIS map of lynchings in Georgia, by county. One thing is clear: only a handful of Georgia counties were lynch-free during the period 1875 through 1930, but some were more lynch-prone than others. Similar to Brundage’s geographic distribution of lynchings (see his chapter “The Geography of Lynching in Georgia” [1993:103–139]), lynchings were mostly concentrated across the central part of the state and in South Georgia, both characterized by cotton production and high percentages of African Americans.
Spatial distribution of lynching events in Georgia, by county (1875–1930).
The GIS map of Figure 5 of lynchings by burning, before and after 1907, the midpoint year in the frequency of this type of lynchings, adds further detail on lynching brutality: (1) Lynchings by burning were lumped together in a handful of counties; (2) there appears to be a diffusion of this type of lynching over time to neighboring counties.
Counties where African Americans were burned (in black, 1875–1907; in grey 1908–1935).
A direct comparison of the map in Figure 4 with that in Figure 5 further shows that “lynch-by-burning” counties were also the most lynch-prone counties in Georgia during the 1875 through 1930 period. Between 1889 and 1904 there were six lynchings involving extreme cruelty (e.g., scalping, skinning alive, cutting), cruelty seemingly linked to accusations of sexual violence (five out of these six events).
In ten lynchings in our database, the newspapers reported the content of warning signs pinned to the bodies of the lynched men or to the tree where they had been hanged. Typical are the words written on a piece of paper left hanging from the neck of Lige Strickland, the man accused by Sam Hose of paying him $20 to murder Alfred Candler, and lynched in Palmetto, Georgia, on April 23, 1899: “We must protect our ladies” (on one side) and “Beware all darkies. You will be treated the same way” (on the other side) (Newnan Herald and Advertiser, 4/28/1899). In nine out of ten events, such warning signs were reported by newspapers in connection with lynchings for alleged sexual assaults on white women (and signed as “WE ARE THE PEOPLE” or “BROTHER DOOLEY”).
4.1.3. Answers from the Database: Crimes and Criminals
The analyses of the QNA data based on simple EDA, network graphs, and GIS maps have revealed clear patterns in the data, allowing us to paint in broad strokes the social relations of lynching in time and space. But the analyses have only scratched the surface of what we can get out of QNA data. Of course, we can get more network graphs and paint a multidimensional picture of the social relations of lynching by focusing on different spheres of action. Or we can provide more detailed maps, zooming in on specific periods and places. The analyses presented here serve mainly as illustrations of what can be done with QNA data. But there is more that we can do.
QNA delivers detailed information on the individuals and groups involved (race, name, age, marital status, occupation, standing in the community, manner of dressing, whether armed or not, the size of a group, etc.) and on the actions performed (time and space, reason, outcome, instrument, type of action, etc.). This is true, of course, for as long as this information is provided by the source documents (in the case of newspapers, as we have seen, they are more eager to talk about some things than others). Furthermore, each single item of information stored in the database in the categories specified by the grammar of data collection is cross-referenced to specific documents—again, at least this is true for PC-ACE, the software used to carry out QNA. For each item of information queried in the database, the available cross-references make it possible to display the original document containing the information. We can toggle between QNA coded text and sources, thus linking text to context. Perhaps, we can squeeze more on agency out of QNA data.
The Crimes
The note pinned on Lige Strickland’s body reflects the late 19th century growing obsession of Southern whites: the raping of white women by black men. Lynching was justified as the only “bulwark between the women of the South and [. . .] a carnival of crime” by black men, as John Temple Graves, Southern editor, orator, and author, put it (New York Times, 8/12/1903, 8/28/1904; see also Cutler [1905:207]; Matthews [1970:155, 167, 175]; and Ellis [1992:26–27]). Rebecca Ann Latimer Felton, a Georgia lecturer and writer who went on to become the first woman to serve in the U.S. Senate, best expressed Southern attitudes toward lynching: “to protect women’s dearest possession from drunken, ravening human beasts [. . .] I say lynch a thousand a week if it becomes necessary” (in Grem 2006:41). Quite a different story from the antebellum period, when, according to Genovese, “rape and attempted rape of white women by black men did not occur frequently. [. . .] Slaves accused of rape occasionally suffered lynching, but the overwhelming majority [. . .] received trials [. . .] fair and careful. [. . .] Public opinion usually remained calm enough to leave the matter in the hands of the courts” (Genovese 1976:33–34).
Yet, this fear “that black men could think of little else but ‘ravishing’ white women”—a fear “of its [the South] own conjuring”—was not only new but also not justified by the data, even those available at the time (Ellis 1992:26). In one of the first systematic studies of lynching history and data, Cutler (1905:224) concluded: “The fact that not more than thirty four per cent of the negroes lynched in the last twenty two years [since 1892] have been lynched for that crime [rape] likewise vitiates such a plea of justification” (see also Raper 2003:37; Ellis 1992:28). Brundage’s later classification of alleged crimes (1993:263) in Georgia between 1880 and 1930 would further reduce that figure: 28% of alleged crimes involved a sexual assault (25% a minor offense, and 46% a murder).
The pie chart of Figure 6 of the distribution of lynchings by type of crime committed as extracted from our database lends support to Cutler’s and Brundage’s conclusions. Only 35% of the lynched victims (153 out of the 388 coded victims) involved accusations of sexual violence against women. In the newspaper accounts, and in our database, sexual violence was variously characterized as “assault” (74 times, 48.4%), “outrage” (16, 10.5%), “rape” (16, 10.5%), “attack” (4, 2.6%), “ravish” (2, 1.3%). In 19 cases (12.4% of the 153 cases of sexual violence), negroes were accused of having attempted to commit the “unspeakable crime” (see Figure 7).
Accusations against lynched victims.
Accusations of sexual violence.
At any rate, the term “assault” used by newspapers of the time could mean anything. As Crowe (1968:250) notes:
The statute of 1896 defined assault as “the attempt to commit violent injury” and the law on rape explained the crime as “the attempt to know a female forcibly against her will.” Moreover, the higher courts allowed assault and even rape convictions to stand when no physical contact had taken place, and Atlanta editors followed established custom by describing all incidents which involved black men and white women as “assaults.”
Crowe (1968:251) concludes: “A misinterpreted step, an unexpected presence, an unexplained word, a stare, a hysterical girl, a vengeful female, a woman with something to hide—all could lead to death.” As an enraged Alabama congressman George Hudleston, “a liberal friend of labor,” told Hollace Ransdall, the Chicago and Columbia young graduate who had come down to Scottsboro, Alabama, in 1931 to investigate the trial for alleged rape of nine African American boys: “[I don’t] care whether the boys were innocent or guilty. They were found riding on the same freight car with two white women, and that’s enough for me. [. . .] I am in favor of the boys being executed just as quickly as possible! You can’t understand how we Southern gentlemen feel about this question of relationships between negro men and white women” (Goodman 1994:45–46).
In any case, at least some of these alleged outrages and assaults may have been illicit 16 but consensual liaisons between black men and white women. We can search through our database for this type of relationships and, relying on the article cross-references, display the original newspaper narratives for context. Was it the relationship between Alex Brown, a “negro boy,” and the 12- or 13-year-old daughter of the man in whose house he was living consensual? After all, according to the boy, lynched in Newnan in 1882 for rape, “he was not guilty of rape, but that his connection with the girl was with her consent.” (Atlanta Constitution 8/11/1882) “The young lady’s story” (emphasis added), told only “five or six weeks after the rape,” when the case came into the open, was that she had been threatened at gunpoint. Uncharacteristically, for “good citizens” typically so incensed by rape cases, especially of such young girls, the paper noted: “It is thought by many good citizens that the governor should show his disapproval of lynch law by offering a reward for the parties guilty of lynching Brown.” Uncharacteristically, the paper also observed how “rumors” circulated about the case. “Various rumors” also circulated about another case, the lynching in 1903 in Cordele of the “25 years old” Mitchell Gilbert, found hanging from a tree, “dressed fairly well for a negro” (Vienna News 10/21/1903). “Deep mystery,”“the particulars of the lynching . . . kept very close” and the woman’s “name . . . withheld from the public”; but “Gilbert, it is said, wrote a loving note to the fair daughter of a prominent citizen . . . in which he expressed affection and a desire to claim her as his own and have her elope with him” (Vienna Progress 10/22/1903). Could the case of Mitchell Gilbert also be an illicit, but consensual, liaison?
What had been true of sexual relations between races in the antebellum South did no longer hold true in the Jim Crow South. 17 Genovese (1976:422) tells us that, during slavery, “white women of all classes had black lovers and sometimes husbands in all parts of the South.” And it was not uncommon for these women to refuse to leave their black lovers/husbands. In such cases of interracial consensual relations, “the black men did not suffer lynching; the whites apparently took these matters [. . .] in stride” (Genovese 1976:422). By the late 19th century, this type of sexual agency on the part of white women had become unthinkable. On November 8, 1887, the Columbus Daily Enquirer-Sun reported the news of a “vile and slanderous article” written by “Jesse Duke, the Negro editor of the Herald, a republican sheet” published in Montgomery, Alabama, “in which he mentioned the fact that negro men were frequently lynched for outraging white women and girls [. . .] because ‘the colored Romeo was becoming more and more attractive to the white Juliet.” Indignation among the white men of Montgomery had led to the immediate formation of a search party for the negro editor “who saved his neck by leaving the city as quickly as possible.” In 1890, Alabama senator John T. Morgan put it in these words: “The snows will fall from heaven in sooty blackness, sooner than the white women of the United States will consent to the maternity of Negro families” (cited in Apel 2004:27). Mary Phagan, the 13-year-old girl found dead in Atlanta in 1915, is a good case in point. Leo Frank, her young, Jewish, Yankee employer would be (unjustly) lynched for these crimes on August 16, 1915, while standing trial. Despite evidence that Mary may have been sexually active, public opinion could hardly conceive that this lily-white girl would be anything but pure and clung to the story that Mary preferred to die rather than consent to sexual intercourse (MacLean 1997:170–74, 159; Apel 2004:26–29). 18
Sooty-black snow may well have been seen over Southern skies, because, for all the indignation of white Southern gentlemen, miscegenation (as it was known) did happen throughout the South and throughout our period (e.g., Wells 1997; Brundage 1993:58–72; Hodes 1997:176–208; Odem 1999:355–56; Apel 2004:25). Even a quick search through the newspapers of the time will reveal year on year, case upon case of white women with black men appearing in court—yet, not leading to a lynching (but to a jail sentence, for sure) (e.g., Columbus Daily Enquirer-Sun 3/1/1876; 6/13/1880). John Duncan (October 1889) and Jake Davis (July 1922) were not that lucky, both lynched for living with a white woman, as found in our database. Jake Davis, “a well-known negro [. . .] about sixty-two years old” fathered a child with the 26-yearold Ethel Skittel. The Miller County Liberal in giving the news (7/19/1922) concludes: “Hundreds of the citizens throughout the county regret this lynching. Many have said the woman responsible for the black crime was guiltier than was Jake.” 19
White women’s agency may have been limited—but this also depended on class; working-class women, particularly factory workers in larger towns, had perhaps greater sexual agency, but upper-class women had an active role in a variety of charitable institutions that, bordering with the political, gave them some political clout (e.g., Brundage 2000). African Americans, however, did not fare much better. Lynch victims, of course, committed serious, but alleged, crimes of arson, murder, rape, along with many minor crimes (e.g., theft). We could perhaps, with Genovese, view these crimes as prepolitical, individual forms of rebellion and resistance against white oppression (see Genovese [1976:598, 597]; on “weapons of the weak” forms of resistance, see Kelly [1994, particularly chs. 1 and 2], Brundage [1997]). Under the political, legal, military conditions of the South, collective mobilization under slavery was out of the question.
But after the Civil War, freedom brought new forms of agency—social, legal, political, economic—for African Americans. For one thing, blacks could now marry. And marriage brought with it rights that could be defended in court (e.g., claims over children against the apprenticeship system of former slave owners; see Edwards [2000:15–18]). Blacks could vote; and be elected; and many were, to the whites’ dismay (e.g., Gilmore 1996:97–99). Churches became the organizing locales for democracy, meeting places for all sorts of collective efforts (e.g., Brown 2000:31). 20 When local white elites were deaf to their pleas, they would take their demands to the courts and even all the way to the president of the United States (Brown 2000:31–33). On election days, they would find clever ways to resist whites’ intimidation and electoral frauds. They became property owners. In their petition in 1865 to President Andrew Johnson, blacks from Richmond, Virginia, stated proudly that “among us there are at least 2,000 men who are worth $200 to $500; 200 who have property valued at from $1,000 to $5,000, and a number who are worth from $5,000 to $20,000” (in Brown 2000:31). There was more to blacks’ agency than both individual crime or “futile, pathetic, or even insane [collective] efforts doomed to defeat” (Genovese 1976:594–95).
What the Columbus Sunday Enquirer wrote on August 19, 1877, was unusual in two ways (the agency of black women, the comment by a white newspaper): “A negro woman in Troup county had a white man taken up for vagrancy the other day. This is the most hopeful sign of the new era.” These hopeful signs would not be long-lasting, as whites tried “to restore as much as possible of the world they had lost” (Kantrowitz 2000:67). By the late nineteenth century, initial political gains by blacks had been extensively scaled back through political and legal means, leading to blacks’ disenfranchisement. To blacks’ political and economic agency whites also reacted with lynchings, asserting their own agency.
We find in our database that still in 1930, “S.S. Mincey, a prominent negro of Montgomery County,” [Georgia], “was brutally flogged” to death by about 10 masked men for being “entirely too active in Republican affairs in this county and must resign his position as Chairman of the County Republican Committee” (Montgomery Monitor, 7/31/1930; see Kantrowitz [2000:70–73]; on the use of violence against Republicans, see also Dray [2003:83]). Blacks’ economic agency was no less fraught with potential problems. In 1901, “John Moody, a negro, was hung and his body riddled with bullets in Bryan county” for leaving a farmer to go work with another (Atlanta Constitution, 3/1/1901). As late as 1949, Hollis Riles was lynched for prohibiting whites to fish in his pond.
How did black communities react to the lynching of their people? A search through the database shows that, by and large, blacks kept quiet after a lynching (“all is quiet in this section”). In those cases when Georgia whites went on a rampage against the community, “the negroes in the vicinity locked themselves in their houses” (Charles Powell’s lynching, Atlanta Constitution, 2/5/1912) or fled to the swamps and woods (John Henry Williams, Early County News, 6/16/1921), or even packed up and left never to return (e.g., Dray 2003:82–83; Beck and Tolnay 1990, 1992; Tolnay and Beck 1992:202–38). The Atlanta Constitution of 1/27/1910 tells us that, after the lynching of Dan Lamkin, “Negroes leave Columbia County [. . .] afraid to continue farming there,” putting their farms up for sale with “Augusta real estate dealers.” Sometimes, as in the case of Owen Ogletree, 200 blacks who viewed the lynching alongside 200 whites “expressed gratification at his punishment” (Atlanta Constitution, 6/19/1894), or even “willingly assisted in the burning of the black fiend” (lynching of an unknown negro on April 13, 1893, near Fort Gaines, American Times-Recorder, 4/15/1893). At other times, they would kill a fellow African American for “squealing” to the whites (e.g., Alfred Thurman’s lynching). Rarely, did they organize to seek revenge (e.g., after the lynchings of Warren Powell and John Coleman) or simply to defend themselves (e.g., Will Atwater’s lynching). Rare it may have been, but they also mobilized successfully to avert a lynching. In Darien, Georgia, on August 24, 1899, some four hundred negroes “armed in every conceivable manner, surrounded the prison” where “Henry Delegal, a negro criminal,” was kept to prevent his removal to Savannah in fear he may be lynched on the way. “The wives and female friends of the rioters were encouraging them in every possible manner, calling them heroes” (Tuskegee, reel 221, p. 13). Until the militia stepped in and some 40 heroes were arrested and jailed.
The Criminals (and the Victims)
Of criminals there were two types in lynching events: (1) the lynched, the alleged criminals, and (2) the lynchers, although hardly ever characterized as criminals in the newspapers of the time.
The lynched, the alleged criminals, the victims of lynch law, were overwhelmingly African American men (91%, 357); 6% (21) were white males and the remaining 3% (10) African American women. Alleged criminals were on average 27 years old (median age 20). In some cases, lynching victims were described as “young” (4 victims), one was considered “aged,” another “elderly,” and two as “old.” The youngest victim was Warren Powell, a 14-year-old African American boy accused of assaulting a white girl in East Point, in the outskirts of Atlanta, in September 1889, and then hanged by a band of 15 to 20 masked white men. The oldest victim was 74-year-old J. R. Dorsey, lynched together with the 46-year-old African American Jane Wade, by three masked men in Chattooga in 1884. We have job information from the newspapers on only 34 lynched victims: almost half of them were farm hands or cotton pickers (15 of them) and in a few instances drivers (3), clergymen (3) and railroad workers (2), while the remaining ones had disparate jobs as chain gang workers, mechanics, soldiers, turpentine hand, painters, porters, etc.
The lynchers were overwhelmingly white. In a handful of lynching events (6), African American mobs lynched other African Americans (5 men and 1 woman) for crimes such as the rape of African American girls (2 events) and violent attacks and murder (2 events), but also for providing information on other blacks (2 events). Contrary to the alleged criminals who typically acted alone 21 lynchers acted in groups (posses, mobs, crowds). The size of these groups varied from a minimum of the three white men who lynched Sam Wright in Cochran in 1899 (Atlanta Journal, 8/15/1899) to very large groups. The mob that met Sam Hose at the train depot in Newnan is “described variously as numbering from 500 to an improbable 2000” (Ellis 1992:109). According to the Springfield Daily Republican, the crowd later present at the lynching numbered 2,000 people. The Atlanta Constitution tells us that a mob of “over 3,000 men wrested [John Lee Eberhardt] from officers at the Athens Courthouse” and then burned him near the place where he had allegedly murdered Mrs. Walter E. Lee (2/17/1921). Similarly, “four thousand unmasked citizens” gathered in Rome in 1902 to “take part in [the] lynching” of Walter Allen (Atlanta Constitution, 4/2/1902). For 53 mobs, we have a qualitative description of size (“crowd,”“small,”“large,”“very large”) and for 138 we have no size information. For the remaining 271 mobs for which we have quantitative information, nearly half fall in the range 0 to 100 individuals in size but much larger mobs of more than 400 individuals are not uncommon.
Contrary to the lynched alleged criminals, the lynchers went unpunished for their crimes. The verdict delivered by the coroner for the lynching of John Gilham in Macon on September 3,1918, was typical: “Death at the hands of persons unknown.” (Macon News 9/5/1918; Dray 2003). In our database, we find only a handful of events where the governor posted a reward for the capture of the lynchers (e.g., for the lynching of John Henry Moore in Columbia County in 1913, of Dan Ahern in Greensboro in 1894, and of John Ware in Franklin County in 1904) and even fewer instances of lynchers brought to trial (e.g., for the lynchings of John Ware in Franklin County in 1904 and of William Cato and Paul Reid in Statesboro in 1904).
Among the victims of the lynched, the vast majority of those murdered were white men (85, or 69.1%) and white women (30, or 24.4%); a few were African Americans (5 men, 4.1% and 3 women, 2.4%). The victims of sexual crimes were mostly white women and girls (93%; 7% African American). They could be as young as Dolly Woods, a 6-year-old white girl “ravished and horribly maltreated” by Jake Bradwell, who was then hanged and his body riddled with bullets near Savannah in 1886 (the average age of the victims of sexual violence or attempted sexual violence in our database is 17, with a median age of 12). Again, 14 victims were described as “young” and three as “little.” Interestingly enough, given the social costs of rape, and rape by a black man, in small communities steeped in Southern honor, newspapers throughout the period typically provided the name of the victimized woman. Of Mattie Cranford, the rape victim of Sam Hose, we know that “she never smiled again,” dying in 1923 “barely fifty years of age” (Ellis 1992:232–33).
5. But Where Are the Agents? Turning to Discourse Analysis for Answers
It may seem paradoxical that we should propose QNA as a method to measure agency. Because, after all, if the narratives QNA parses are anything like Sam Hose’s opening story, agency is only partially there. We do have Sam Hose. We also have an ex-Governor of Georgia. But where are the actors who tortured Sam Hose, who “deprived him of his ears, fingers, and genital parts of his body,” who covered his body with oil and applied the torch to the pyre, who dismembered his body, cut it to pieces, heart and liver, who crushed his bones into small bits? Nearly all the sentences in the newspaper article that reports Sam Hose’s lynching are in passive forms that end up hiding agency. This is not an innocent stylistic preference by the writer. It reveals deep-seated (albeit, perhaps, unconscious) ideological motives, which were not captured by QNA. It takes a much more fine-grained approach to tease these out in texts.
Discourse analysis is a qualitative approach to the analysis of text (Brown and Yule 1984; Johnstone 2007). In its applications to news discourse, the approach has focused on the linguistic mechanisms that contribute to the process of ideological production of news, i.e., to the role of news in maintaining unequal relations of power and in preserving the legitimization of the social order (Fowler et al. 1979; Trew 1979a, 1979b). Media news is a prominent locus of political and ideological conflict. Newspapers, like any other media outlet, “have a major ideological role” (Trew 1979b:156). 22 Passivization, nominalization, classification of processes and participants, and modality are some of the many linguistic devices utilized in discourse to reproduce ideologies (Trew 1979a:97). Here, we focus on passivization and nominalization since these linguistic processes refer directly to issues of agency. 23
There are several reasons why a writer/speaker might choose a passive, rather than active, construction of a sentence: “one is that it allows for the omission of the agent, though this may itself be variously motivated by the fact that the agent is self-evident, irrelevant or unknown. Another, political or ideological, reason for the use of a passive voice may be to obfuscate agency, and hence causality and responsibility (compare “police shot 100 demonstrators” with “100 demonstrators were killed”)” (Fairclough 1992:182 [emphasis added]; see also Trew, 1979a).
The foregrounding (or backgrounding) of agency, causality, and responsibility is crucial in the reconstruction of violent events like lynchings. It is no accident that “one category of events where this issue constantly arises [is] violence and violent death” (Fairclough 1992:181). Indeed, passivization occurs throughout the article on Sam Hose. The agents responsible for his lynching are simply not there. Discourse analysis would thus suggest that the writer of the article, consciously or unconsciously, failed to attribute responsibility to a specific agent, concealing the context of violent oppression and virtual impunity suffusing racial domination in the U.S. Deep South at the turn of the twentieth century. The word “white” never appears in the article. But “the absence of a sign can be significant too” (Barthes 1970:149). Significant, certainly, is the absence of Mr. or Mrs. prefixed to any personal name of African Americans.
Nominalization—the transformation of verbs into nouns—is another major linguistic tool used to deny agency in discourse (Fairclough 1992:179–82). 24 Thus, sentences like “Sam Hose was lynched by the mob,”“the mob stormed the jail” or “the mob attacked the negro” are nominalized as “the lynching of Sam Hose,”“the storming of the jail,” and “the attack on a negro,” where the agent “mob” disappears. Nominalization is commonly used in newspaper headlines to remove the perpetrators of despicable acts (Fowler et al. 1979:14; Billig 2008:785). Besides passivization, the article on Sam Hose’s lynching (our nominalization!) also makes use of nominalization: “He pleaded pitifully for his life while the mutilation was going on,” where the verb “mutilate” is transformed into the noun “mutilation”; as a result, agency is once again eliminated: The reader is unaware of who was mutilating whom. Like passivization, nominalization is not an inconsequential linguistic choice; it reveals underlying ideological practices aimed at maintaining unequal power relations (Billig 2008:786). These linguistic practices are not necessarily conscious. They may well be based on “taken-for-granted assumptions” (Tuchman 1973:127), the result of professional training and “years of craft apprenticeship” (Tuchman 1978:105), aimed at creating an aura of objectivity and at spinning a “web of facticity” (Tuchman 1972, 1978:82–103). But conscious or unconscious, linguistic practices end up having ideological consequences. 25
5.1. Newspapers and Agency
Discourse analysis puts the issue of newspaper bias in a broader context. Because newspapers—or better, the journalists writing for the white, Southern newspapers—now become agents in their own right in the social relations of Jim Crow South, rather than simple conduits of news. They expressed their agency in a variety of ways (Ellis 1992:21–22). Newspapers used different language, different representations for whites and blacks. The most typical ways of describing a lynched African American as found in our newspaper database are these: bad character, bad nigger, diabolic, without the fear of god, without the fear of the law, fiend, brute (Sam Hose is “a brute in human shape” for the Quitman Free Press). In contrast, whites are almost always represented positively: prominent, well-known, highly respected, popular in the city. Even mobs are described as controlled and silent, efficient in their business, and only occasionally as “frenzied” and “howling.” 26 “A more orderly set of men was never seen, for no one seemed excited or boisterous,” the Quitman Free Press (4/28/1899) described the crowd who met Sam Hose’s arrival in Newnan by train. “White southerners wanted to believe that mobs demonstrated tremendous self-control and restraint” (Brundage 1993:65), thus “embodying the supposed moral superiority of whiteness through their purposeful and controlled actions” (Wood 2005:374). Occasionally, the lynchers’ deeds would be deprecated, but even in these cases journalists would accomplish true rhetorical feats in providing justifications. Such is the article titled “A Carnival of Crime” in the Atlanta Journal (8/17/1904) of the lynching of Will Cato and Paul Reid, burned at the stake for the murder of Mr. Hodges, his wife and two children. The article concludes: “The friend of the negro is not the white man who implants within his breast the germ of dissatisfaction . . . but the one who teaches the negro that he can never rise to the plane of social equality, nor can he commit crime without expecting due punishment.” Such is the description of the Quitman Free Press (6/17/1921) of the murder of 12-year-old “little Lorena” from Autreyville, her “throat slashed open” and dumped into a pond where her “small white hand, the hand of Lorena Wilkes” protruding above the surface of the water, in the atmosphere of “inflamed sentiments of the people of the community,” must have served as added fuel to those sentiments. John Henry Williams, in jail and already sentenced to be hanged for the murder, would be burned at the stake. “An immense crowd [. . .] quietly looked on” as Williams “was burned to a crisp.” No mercy. No guilt. No reason for either. “A few moments before [Williams] told how he had cut a girl’s throat as she cried and begged him to spare her life.” (Quitman Free Press, 6/24/1921)
Yet, newspapers’ use of rhetoric extended well beyond the stylistic manipulation of language. Newspapers also used effectively epideictic rhetoric, the rhetoric of blame and praise. The Quitman Free Press (4/28/1899) closes the article that describes Sam Hose’s death with these words: “At 3 o’clock his [Hose’s] lifeless body was being consumed by the flames, and one hour later the special for Griffin was leaving town amid the shouts and cheers of the citizens of that place. Thus ended the life of the most awful criminal in the history of Georgia.” The Miller County Liberal (8/14/1918) similarly closes a story on the lynching of Ike Radney for allegedly raping a white woman: “So ends the story of a black brute, who has gone the route of all his class. A violent death at the hands of an angry mob. A warning to others.” Yet, not all Southern white newspapers endorsed a lynching. In leads and subleads, the Oglethorpe Echo of February 2, 1894, does not mince words in condemning the lynching of “Bob Collins [. . .] an inoffensive old negro”“BEATEN TO DEATH,”“A HORRIBLY ATROCIOUS ACT,”“It Brings Disgrace upon our County and One That Will not be Permitted to go Unpunished.” It opens: “It is with a feeling of great shame for honored old Oglethorpe county.”
Newspapers also took sides, giving voice to some and silencing others, carrying certain news but not others (e.g., Ellis 1992:81, 166; Brown 2000:32; Godshalk 2000:144, 152). Newspapers put up rewards for information leading to the apprehension of “negro fiends” ($500 by the Atlanta Constitution in the case of Sam Hose; Ellis 1992:81). They would run several daily special editions with updates on “expected” lynchings (Arnold 2009:145). In capital letters, the Atlanta Constitution (4/24/1899) announces: “DETERMINED MOB AFTER HOSE; HE WILL BE LYNCHED IF CAUGHT.” And in the sublead: “Assailant of Mrs. Cranford May be Brought to Palmetto and Burned at the Stake.” Newspapers would play a crucial role in spreading wild rumors of negroes “taking the town and killing every white man, woman, and child” (Columbus Daily Enquirer-Sun, 8/20/1875) and, worse yet, of “negroes [. . .] murdering the white men and makeing [sic] slaves of the white women” (Columbus Daily Enquirer-Sun, 9/22/1888). Daily news of a “rape epidemic” drummed up over months led to the 1906 Atlanta riot, when bands of white men went on a rampage leaving on the ground 25 black men dead, about 150 seriously wounded, and hundreds less critically injured; more than a thousand fled the city (Crowe 1969:168).
6. Agency without meaning? For an effective research strategy
In his 1922 Economy and Society Max Weber wrote: “We shall speak of ‘action’ insofar as the acting individual attaches a subjective meaning [sinn] to his behavior [. . .] Action is ‘social’ insofar as its subjective meaning takes account of the behavior of others and is thereby oriented in its course.” And “sociology [. . .] is a science concerning itself with the interpretative understanding of social action” (Weber 1978:4, emphasis added). But where is that meaning in the thousands of actions stored in our database? QNA, with its analysis of thousands of narratives, gives us the behavior of the actors, their actions, but, by and large, is silent about motivations and meaning. Silent, because silent, by and large, are the narratives we rely upon to reconstruct through QNA the social relations of lynching: newspaper articles (or official documents, court or police). Of course, these narratives do provide contingent and surface reasons for action: the brutish nature of the Negroes as the reason why they murder and rape; these murders and rapes as the reason why white mobs lynch them. But the deep reasons, the meaning of action is unlikely to be found in simple newspaper narratives of lynching (e.g., Southern gentlemen’s attitudes toward “relationships between negro men and white women”). Even discourse analysis would have no solution to the problem—a solution that after all would require working on different types of documents (e.g., letters, diaries, even newspapers, but newspaper editorials rather than event stories).
The problem of meaning extends to the tools best suited to analyze QNA data: network models. The diagram in Figure 1 is very clear about the social relations of violence in lynching, very clear about the actors involved, yet silent about meaning, silent about why actors resort to violence. As Gould (2003:267) notes:
Extant network methods are not very good at either incorporating or modeling subjective experience, meaning, or symbolic structures [. . .] Different methods are good at different things—and for the moment, at least, until someone figures out how to make structural analysis good at modeling meaning, we had better be content with the thought that network methods are meaningful for modeling social networks.
The problem, however, only partially rests with QNA. And the fact that QNA does have “technical” solutions 27 for the problems of agency highlighted by discourse analysis may be beside the point. If the value of QNA is to find patterns in large bodies of narrative data dealing with actors, their actions, and their interactions (agency, indeed, albeit perhaps with no meaning), such an in-depth investigation may be prohibitively costly at the current state of computer-aided, but still manual, QNA, although feasible for smaller research projects (e.g., fewer than 10,000 triplets). Similarly, could one carry out the in-depth, painstaking analysis on thousands of documents using discourse analysis? No. Again, the costs of such an approach would be prohibitive (and perhaps even unnecessary). Furthermore, they would require skills beyond the average college student typically involved as coders in content analysis projects (QNA or not).
Which then perhaps suggests a strategy for researchers wishing to use QNA to measure agency: perform QNA on thousands of documents, then sample a handful of these documents for a more in-depth investigation (marking perhaps any document that strikes the coder for its descriptive, evaluative content). Furthermore, events do not happen in a vacuum. And focusing on the event alone (albeit, perhaps thousands of events) may leave us in the end with a great deal of description and no explanation. Certainly, none of the newspaper articles on lynching in our collection ever mention the price of cotton. Yet, we know from the work of Beck and Tolnay (1990, 1992) that the state of the economy did have a significant effect on lynchings. Only a handful of those articles mention the flight of African Americans from a community after a lynching. Yet, we also know that outmigration contributed to the eventual decline of lynching in Jim Crow South. To understand the event, we need to go not only inside the event but also outside the event, and relate internal characteristics not just to each other but to external ones as well. Text and context must go hand in hand. Or, to put it differently, to understand agency (or the events in which agency finds concrete manifestations) we need to understand the larger underlying structure. Events may be “but manifestations of that larger destiny by which alone they can be explained” (Braudel 1980:3).
7. Conclusions
The main concern of this paper has been with ways of measuring agency. After all, methodological issues of operationalization of the concept of agency have not kept pace with the lively theoretical debate across various disciplines. In contrast to the hundreds of pages on the theory of agency, we have only fleeting remarks on how exactly to measure agency. By exploiting the links between agency and action, action and narrative, we have proposed a way to measure agency in socio-historical research based on QNA. More than ten years ago, Griffin et al. (1997:28) argued that events, as we know them through narrative accounts, are not “inexplicable random happenings” about which “all we can do is tell stories.”“To discern and understand that logic [of events. . .] storytelling must be transcended even as narrative, as the medium through which we know events, must be retained and analytically exploited. [. . .] To realize the analytical promise inherent in events, however, they must be systematically ‘unpacked’ and theoretically reconstituted as explicit interpretive and explanatory devices” (Griffin et al. 1997:28, 30).
And that is precisely what QNA does: “unpack” the events. QNA offers an approach to narrative that turns words into numbers by exploiting the invariant linguistic properties of narrative (namely, a sequential organizational structure of elementary narrative units based on actors, their actions, and the characteristics of both, a structure also known as story grammar). But while delivering numbers, QNA preserves much of the narrative richness of the original text. Hopefully, the analyses of our lynching database will have provided evidence of the power of the technique in its ability to measure agency. 28
Yet, discourse analysis has alerted us to the linguistic mechanisms that, consciously or unconsciously, may lead to the suppression of agency in a text (through passivization, nominalization, ex-nomination). QNA does offer technical solutions to these problems in the design of broader story grammars; yet, the extraction of additional information beyond the basic elements of a “semantic triplet” comes at a steep cost, at least for large volumes of narrative texts. As for discourse analysis, it too has its limits in its ability to deliver depth of interpretation: It can be reasonably applied to a handful of cases only. At the current state of the art, no methodology is able to deliver quantity and quality, breadth and depth, to be both telescope and microscope (Franzosi 2010:146).
With all its pros and cons, has QNA helped us to find answers to the question of agency? We hope to have shown, with analyses however exploratory and illustrative, focused on a subset of actors, on specific aspects of lynchings, and based on incomplete data analyzed mostly with descriptive tools, that QNA provides a way to operationalize agency and to obtain quantitative measures of agency. 29 Not only has QNA produced numbers dealing with actors, actions, and interactions, revealing meaningful patterns in these numbers; but since, in QNA, numbers are never too far from the words upon which the numbers are based, QNA has allowed us to rely on these words to flesh out the numbers, to use the text to understand the context and the meaning of the numbers. Not only are the “data” in the database the very words used in the newspapers to narrate a lynching event, but each one of these words, stored in the specific coding categories specified by the grammar of data collection, is cross-referenced to specific newspaper articles that can be easily displayed for context (at least in PC-ACE). Sections 4.1.3 and 5.1 provide an example of this interactive use of numbers and words, of placing the text found in our database in the context of the broader available historical record and scholarship. Further measures of agency also come out of this interaction.
Are these answers provided by QNA data different from those we could have found through either statistical analyses of event counts or in-depth historical analyses or social-scientific analyses? Analyses based on aggregate event counts would have allowed us to produce neither the network graphs of Figures 1 and 2, nor the time plot of lynchings based on burning of the victim (Figure 3), nor the GIS map of Figure 5, all based on event characteristics, characteristics internal to the event and not available with aggregate event counts. 30 Of course, QNA too counts. But those counts are based on event characteristics rather than events. Certainly, aggregate event counts would not allow us to zoom in on the details of agency, tease out the actions of specific actors, as illustrated in Sections 4.1.3 and 5.1.
Attention to detail is rather the hallmark of good historical work. Actors, actions, time, and space are the basic staples of narrative historians. But the very richness of historians’ narrative detail may raise questions about generalizability. Quantitative narrative analysis, with its close connection between text and numbers, never allows an investigator to forget where numbers come from, to forget that “there are people behind numbers,” as French historian Michelle Perrot reminded her readers (cited in Franzosi 2004:241). But at the same time, QNA will allow an investigator to record information as detailed and for as many of those people as made available by the narrative sources. 31
To the extent that it delivers fundamentally relational data, QNA is in line with various research programs, especially relational sociology (Emirbayer 1997). For relational sociology, agency is “agency toward something,” it is “interactions,”“a dialogic process” where “actors enter into relationship with surrounding persons, places, meanings and events” becoming “inseparable from the transactional contexts within which they are embedded” (Emirbayer 1997:287, 294, original emphasis; see also Emirbayer and Mische 1998:973). Relational sociology rebuffs explanations of “social behavior as the result of individuals’ common possession of attributes and norms rather than as the result of their involvement in structured social relations” (Wellman 1983:165; see also Burt 1986:206); a perspective that “questions the explanatory potential of all those conceptual strategies that emphasize the nonrelational attributes and/or purposive actions of individual or collectivities” (Emirbayer and Goodwin 1994:1416). It is a perspective that, by looking at historical events as we know them narratively (e.g., through newspapers), considers “an event as a particular happening that is constituted by a particular sequence of temporally ordered actions and occurs in a particular historical context. [. . .] Events are complex relational wholes . . . [to] be analyzed partly in terms of their particular contextual and temporal makeup rather than, as many sociologists do, simply aggregate isolated facts from narratives of lynchings and then explain the occurrence or frequency of the aggregate” (Griffin et al. 1997:29).
A metaphor by Dailey, Gilmore, and Simon (2004:4), in their introduction to an excellent collection of historians’ papers on forms of agency in the South “from Civil War to Civil Rights,” aptly summarizes our approach: “Jim Crow was at bottom a social relationship, a dance in which the wary partners marched their steps, bent, and whirled in an unending series of deadly serious improvisations” [emphasis added]. The step we have focused on—lynching—is certainly one such deadly serious improvisation in this dance. But to view all Southern social relations from the vantage point of this deadly step may lead to a gross misunderstanding of race and gender relations in Jim Crow South. Only a fraction of love affairs between black men and white women did lead to lynchings; most couples involved in miscegenation just paid the price of this form of illegality with jail sentences. Only a fraction of murders of whites by blacks did end in lynchings. What Batstone, Boraston, and Frenkel (1978:26) had written about strikes holds for lynchings as well: “If we want to understand strikes, we have also to understand non-strikes.” 32
QNA ultimately embraces what Charles Tilly called relational realism or “the doctrine that transactions, interactions, social ties and conversations constitute the central stuff of social life” (Tilly [2004:72]; on this relational program, dating back to Simmel and von Wiese, see Franzosi [2004:255–64]. Touraine was no doubt one such relational realist when he wrote about “a sociology that is concerned with agency” (1984:16), that “All approaches that reject the analysis of the relations between social actors are alien to sociology or even opposed to it. [. . .] The sociology of action lies at the center of sociological analysis” (1984:47). Analytical sociology, as well, shares with QNA a preoccupation with actors and relations. Although not explicitly concerned with issues of agency, “Analytical sociology explains by detailing mechanisms through which social facts are brought about, and these mechanisms invariably refer to individuals’ actions and the relations that link actors to one another” (Hedström and Bearman 2009:4; emphasis added). Indeed, it is the relational properties of story grammars and, more generally, of the set of tools used in QNA, that make this methodological program eminently compatible with a broad range of theoretical programs. “Relations! Relations! Relations!” no doubt (Franzosi 2010:51).
Footnotes
Acknowledgements
The authors acknowledge the generosity of E. M. Beck in making his data available, and the helpful comments on various drafts by the anonymous reviewers of Sociological Methodology and its editor, Tim Liao, and by Eduardo Bonilla-Silva, Aaron Cicourel, Sophie Doyle, Tyrone Forman, Darren Grem, Michael Hechter, Alex Hicks, Rogers Hollingsworth, Barbara Johnstone, Frank Lechner, Gianluca Manzo, Robert McCauley, and Denis O’Hearn.
Authors’ Note
The lead author wrote the final draft of this paper in 2011–2012 during a year fellowship at the Fox Center for Humanistic Inquiry.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Emory University (through the University Research Committee) provided the lead author with a grant in 2008. In 2011 a Digital Humanities Scholarship award from the Mellon Foundation provided additional help.