Legacies and Extensions of Charles Tilly: A Relational and Dynamic Event Analysis of Contentious Repertoires in Great Britain1

Takeshi Wada

Abstract This paper introduces and attempts to extend Charles Tillys legacies in the field of event analysis. The first part explains three of his specific innovations in developing BRIT, a database of political events, or contentious gatherings, occurring in Great Britain in the 18th and 19th centuries. One of Tillys innovations is that he collected and recorded information about sequences of contentious interactions within events. The second part of the paper takes advantage of the availability of such intra-event information and employs association rule analysis to document empirically performances and repertoires of contention from the same time period and thereby re-evaluate Tillys theory of repertoire change. The paper concludes that Tillys action sequence data has the potential to generate a great deal of interesting and fruitful research. Keywords Social movements, contentious politics, popular contention, Charles Tilly, Great Britain, sequence analysis, association rule analysis, event data

This work was supported by JSPS KAKENHI Grant-in-Aid for Scientific Research (C) #22530531.

1 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Introduction The late American sociologist Charles Tilly made significant contributions in a variety of research fields, including historical sociology, democratization, and contentious politics (Tilly 1997). For many scholars, his work has been, is, and will continue to be an important source of intellectual stimulation, challenge, and guidance. This paper has two major goals. The first is to introduce Tillys specific legacies and innovations in the field of event analysis. The innovations are connected specifically to BRIT, a database created by Tilly of political events in Great Britain in the 18th and 19th centuries. As I will explain in the next section, BRIT stands out from other event data projects because of three unique features. First, BRIT permits researchers to perform not only quantitative analyses of political events (as all other event projects permit) but also qualitative analyses. Second, BRITs relational data structure enabled Tilly (as it continues to enable us) to preserve information about the complexities of contentious relations among contenders. The third unique feature of BRIT is that it collects information about detailed sequences of interactions among contenders within events. BRIT is thus a database of events as well as of intra-event interaction sequences. I argue that Tilly was able to exploit research opportunities opened by the first and the second unique features of BRIT but, because of an untimely death, did not take advantage of the third feature to produce another original study. I will discuss this issue further in this paper. The second goal of the paper is to extend Tillys legacies by utilizing the third feature: the availability of information regarding intra-event action sequences. Specifically, I employ a method called association rule analysis to examine dynamic interaction sequences and uncover more about repertoires and performances in British contentious politics. 2 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Why is a concentration on Tillys specific legacies in the field of event analysis important? I see three reasons. First, these methodological breakthroughs and their potentials have not been acknowledged as widely as Tillys other theoretical and empirical innovations. Second, I am fortunate to have had an opportunity to work with Tilly on a research project using BRIT, first in 2002, and again, from 2006 to 2008, a period in which I was able to learn about the structure of the database. During our collaboration, I contributed to the development of an updated version of BRIT, BRIT2007, in a Microsoft Access format (Tilly and Wada 2007). I use BRIT2007 for the present paper. The third reason for concentrating on Tillys legacies is that Tilly himself wanted to take advantage of BRITs third feature. As a copy of an email message I once received from him (reprinted below) clearly indicates, he thought an examination of intra-event action sequences a worthy endeavor to advance our understanding of repertoires of contention. His untimely death in 2008 prevented him from pursuing this line of research further. As I will discuss, a study of dynamic processes of interaction sequences has potential to both enhance our knowledge about repertoires of contention as well as enable researchers to make better predictions regarding future actions. In the next section, I introduce and explain Tillys legacies. Tillys legacies: BRITs three unique features BRIT consists of machine-readable descriptions of 8,088 events that occurred in four London metropolitan counties (Kent, Middlesex, Surrey, and Sussex) from 1758 to 1820, or anywhere in Great Britain (England, Scotland, and Wales) from 1828 to 1834. Tilly called these events contentious gatherings. A contentious gathering is an occasion in which 10 or more people gathered in a publicly accessible place and visibly made claims which, if realized, would affect the interests of at least one person outside their group (Tilly 1995:393). In principle, contentious 3 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

gatherings include almost all events that authorities, observers, or historians of the time termed riots or disturbances, as well as events that fall under headings such as public meetings, processions, and demonstrations (Tilly 1992:4). The data sources documenting these gatherings are newspapers and other periodicals such as The Annual Register, Gentlemans Magazine, The London Chronicle, The Morning Chronicle, The Times, Hansards Parliamentary Debates, The Mirror of Parliament, and Votes and Proceedings of Parliament. Three major innovations in BRIT are as follows:

1) BRIT IS GOOD FOR QUALITATIVE ANALYSIS A unique feature of BRIT is that researchers using it can conduct qualitative in addition to quantitative analysis of political events. In stark contrast to a majority of event databases created in order to quantify political events reported in news articles for quantitative investigations, Tillys project facilitates qualitative investigation as well. Why is Tillys database more flexible and open to both quantitative and qualitative methods? It is because Tilly employs a bottom-up approach to convert textual information in news and periodical texts into numbers in the database. Other databases mostly use a top-down approach. The top-down approach prepares a codebook with a list of codes, i.e., a logically organized system of abstract categories derived deductively from a theoretical system (Shapiro and Markoff 1998:81). So-called coders read articles, find an appropriate code in the codebook, and enter a number assigned to the code in the computer. This approach has been predominant in the field of protest event research. A disadvantage of the top-down approach is that only highly aggregated theoretical categories can be listed in a codebook. Moreover, coders must impose these a priori categories 4 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

on the textual data. This raises three problems. First, it is difficult to predetermine abstract codes when this requires having precisely the same detailed knowledge about the subject matter that one is trying to learn in the first place (Koopmans and Statham 1999:208). Second, if researchers realize that they should have used two separate categories instead of one, they must redo the entire coding. This makes a revision of codes impractical, restricting the use of the data set to only those research questions that are explicit at projects start. Thus, the top-down approach makes it difficult to shift attention to other questions in the course of the same study (Horn and Tilly 1986:15; Koopmans and Statham 1999:208). Third, it is coders, not researchers, who interpret texts and make important theoretical decisions in finding abstract codes. As Tilly states, [T]he more abstract the code, and the greater the judgment it requires, the worse the code will be. A bad code has multiple disadvantages. It is unreliable, its results are often inaccurate, and it places crucial analytic decisions in the hands of coders rather than in the hands of those who will actually carry out the analysis (Tilly 1981:75). Tilly opted for the bottom-up approach over the top-down approach (Horn and Tilly 1986). The bottom-up approach distinguishes two stepstranslation and aggregationin the process we customarily call coding (Shapiro and Markoff 1998). In the first step, translation, coders create codes as they go through newspaper articles. In its extreme form, this approach does not even prepare a codebook. Coders preserve in codes as much concrete description in the articles as possible. The translations involve a simplified transcription derived inductively from original texts rather than use the conventional reduction to a priori categories (Franzosi 1989; Horn and Tilly 1986:3; Shapiro and Markoff 1998:81). In the second stage -- aggregation -- researchers aggregate these concrete codes into abstract summary categories derived from theoretical ideas. I will illustrate the translation process with a hypothetical example. Consider the following 5 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

text: Fifty female operators of a telephone company Vodafone occupied the company building in London and made a demand on the management for a 20% increase in salary. The first task of a coder is to identify the basic elements to be coded. Any coder will be able to distinguish a phrase describing the subject of action (Fifty female operators of a telephone company Vodafone), the object of action (the management [of a telephone company Vodafone]), the action itself (occupied the company building [of a telephone company Vodafone]), the location (London), and the claim (made a demand for a 20% increase in salary). The next task is to assign a code to each element. Look at the subject and claim above. A coder does not have to (and should not) worry about making a decision over whether or not female operators should be coded as workers or women. Instead, he or she will retain fifty, female operators and Vodafone from the original text. Likewise, rather than having to search for an a priori claim code in a codebook (e.g., economic interest or labor demand), a coder will invent a concrete code, such as demand for a salary increase. Tilly is not suggesting copying the original text. The richness of the vocabulary and the complexity of the syntax of the natural language will produce an infinite number of codes. What he has done is to create codes that are concrete enough to retain the original texts main story and flexible enough to apply to a variety of similar texts. The code demand for a salary increase is applicable to texts such as made a demand for a 30% increase in salary, requested to double their salaries, asked for a wage increase, wanted to earn more money for work, and more. Coders can invent new codes when they feel it necessary to preserve the concrete information in the original text. Since the creation of abstract theoretical codes or categories is the researchers task at a later stage, coders can use their best codes, i.e., the one that the coders are most comfortable with, the code that they can remember, and in which they can organize and record 6 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

essential distinctions; this is most likely to be one they make up themselves (Shapiro and Markoff 1998:86). Coders no longer make substantive theoretical judgments as the highly abstract scheme obligates them to do (Franzosi 1989:290-291). Instead, their task is to translate, that is, to find information about Who does what to whom, why, when, where? Coders perform semantic tasks using their natural language competence, not tasks of theoretical interpretation. Here, coders become simply an instrument of measurement and, as a result, we can improve intercoder reliability (Franzosi 1989). In short, the coders task is an act of translation, or a many-toone transformation from the complex human language into one simplified and artificial code (Shapiro and Markoff 1998:75). In this manner, Tilly is able to produce data that are much richer than the data generated by top-down approaches. Researchers can easily reconstruct narratives from BRIT because it retains the textual information of the original sources. The descriptions of political events (contentious gatherings) stored in BRIT are not only machine-readable but also human-readable. FIGURE 1: BRIT SCREENS ENABLE USERS TO READ WHAT HAPPENED IN CONTENTIOUS GATHERINGS 1-A)

7 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

1-B)

Source: BRIT2007 (CGID=830072102) (Tilly and Wada 2007).

Examples of a qualitative reading of BRIT events are shown in Figure 1. In 1-A, BRIT demonstrates that officer arrested little boy in Middlesex on July 21, Wednesday, 1830. The detailed information about this event, such as location, duration, number of participants, issue, 8 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

and outcome, is easily accessible by clicking one of the relevant tabs shown in the figure. A researcher who wants to read the original text needs to click the Open SOURCE button to obtain the reference. Taking advantage of these features of BRIT, Tilly has been able to carry out qualitative analysis and extract an extraordinary number of rich narratives on British contentious events. We can find these narratives everywhere in his writings (Tilly 1995; Tilly 2008).

2) BRIT IS GOOD FOR RELATIONAL ANALYSIS The second unique feature of BRIT is that, unlike most event data projects, it preserves accurately information about complex relations among event characteristics (actor, target, form of action, issue, location, etc.). As a result, Tilly was able to conduct relational analyses of political events using network models. In the previous example in Figure 1-A, officer is related to little boy and the nature of the relationship is expressed by the verb arrest (which Tilly aggregated into a broader theoretical category, CONTROL). FIGURE 2: AN EXAMPLE OF THE SPREADSHEET DATA FORMAT

Source: European Protest and Coercion Data compiled by Ronaldo Francisco. The figure shows a part of the protest event data from France 1993-1995. The file was downloaded from the website (http://web.ku.edu/~ronfran/data/index.html) on August 4, 2011. 9 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

A common way of organizing the codes in a database is to use spreadsheet-based software such as Excel or SPSS, storing the information in a table with columns and rows. An example of such a data format is shown in Figure 2. This example is a part of the European Protest and Coercion Data database compiled by Ronaldo Francisco. Protest event is the data unit here and thus constitutes row entries. Column entries represent each event characteristic: who (Protester), does what (Action), to whom (State/target; Agent), when (Event date), where (Country; Location), and why (Issue). The spreadsheet format like the one in Figure 2 works when each element in the event happens to have only one entry. For instance, the spreadsheet format can handle the example in Figure 1-A because there is only one actor or, using BRIT terminology, one subject (officer), one form of action (arrest), one target or object (little boy), one location (Middlesex), and one date (July 21, 1830). Typical newspaper articles, however, are likely to report a more complex event structure such as the one shown in Figure 1-B. In this figure, there are two subjects (mob and prisoner) for one action (push) and one object (officer). This structure complicates the coding task. The spreadsheet format would handle this by creating an additional Subject (or Actor) column to store the second actor reported in the text. This solution looks acceptable, but what are we to do if we have many subjects, say 10? Should we create 10 columns for the subject entry? How about another 10 objects, 20 forms of action, or 30 claims? What are we to do if we want to record additional information about the subject, for instance, its political affiliation or place of origin? How many columns should we prepare? Facing this problem, even advanced research projects have so far made only arbitrary but practical decisions to limit the number of entries. For instance, a comparative study of four 10 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

European countries by Kriesi, Koopmans, Duyvendak, and Giugni decides [I]f the action was directed at more than one goal simultaneously, the most important of these was coded and [I]f more action forms were used in the same event, the most radical of these should be coded (Kriesi, Koopmans, Duyvendak, and Giugni 1995:266). German PRODAT research specifies that when coders find different forms of action in one event, they should code both a primary and a secondary form. PRODAT accepts up to two action forms (Rucht and Neidhardt 1998:68-69). Fillieule decides in his French study that [I]n my data base when an event includes multiple modes of action, it is recorded under up to three such modes (Fillieule 1998:202-203). In this manner, however, precious relational information is lost for technical reasons. Tilly solves this problem by using a relational data format (Horn and Tilly 1986). Rather than entering all the information in a single table (spreadsheet), BRIT creates many interconnected tables, as shown in Figure 3. An event characteristicwho (Subject) does what (Action) to whom (Object)has its own spreadsheet or table. An episode of protest is chopped into elements, which are stored separately in corresponding tables. FIGURE 3: STRUCTURE OF THE BRIT DATABASE

11 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

This figure is a simplified presentation of BRITs data structure. It describes the way in which information about contentious gathering CGID=830072102 (shown as the example in Figure 1-A and 1-B) is preserved. Each rectangular box represents a spreadsheet-like table. Two tables linked directly by a line are connected by shared field(s) or variable(s). For instance, ACTION and SUBJECT tables are linked by CGID and PHASE.

Figure 3 displays the BRIT database structure using the information shown in Figure 1-A and 1-B as an illustrative example. The Event table on the upper-left contains the list of political events (contentious gatherings). This table keeps information proper to the event such as the events identification number (CGID), the year of the event, and the location of the event, among many other descriptors. The Event table is linked to the Action table by its CGID. We see two entries in the Action table. Both actions have the same CGID (830072102), which we know from the Event table happened in Middlesex in 1830. Two actions thus belong to the same event. Look 12 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

at the third column VERB in the Action table. This column is linked to the Verb table. The first action (VERB=427) is arrest, and the second (VERB=82) is push. The Verb table contains a list of action verbs. This table is linked directly to the Action table because one action accepts only one action form (verb). The data structure becomes complicated for actors (subjects) and targets (objects) because one action accepts more than one subject and more than one object. In our example, the second action (PHASE=1001, push) involves two actors, mob and prisoner. In such cases, we cannot link the table containing a list of actors (the Formation table) directly to the Action table. We need an intermediary table, and the Subject and Object tables are examples of such. Look at the Subject table. It shows that for the first action of the event (PHASE=401, arrest), there is one subject, officer (FORMATION ID=2). For the second action (PHASE=1001, push), there are two subjects, mob (FORMATION ID=1) and prisoner (FORMATION ID=4). As Horn and Tilly have asked, Why use such a complex database structure? Wouldnt it be simpler to put all the information about an event into a single record? Their answer is, Simpler yes; more useful, definitely not. (Horn and Tilly 1986:15). Using a relational data format rather than a spreadsheet format enabled Tilly to preserve relational information more accurately and efficiently. This innovation also enabled him to go beyond the traditional research questions in protest event research, that is, the questions about levels of protest or of violence. As Tilly puts it, Little by little, nevertheless, the increasing availability of event catalogs that follow connection and variation within events (however conceived by their originators) will almost certainly reduce the popularity of protest and violence as answers to the question Exactly what are you studying? Thus, theories of the phenomenon at hand and theories of measurement continue to influence each other (Tilly 2002:252). 13 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

FIGURE 4: NETWORK ANALYSIS OF POLITICAL CONTENTION APPLIED TO BRIT DATA

Source: Tilly (2008: 56, Figure 2-4). Taking advantage of this second innovation, Tilly was able to ask new questions and carry out a series of brilliant network analyses (Tilly 2008; Tilly and Wood 2003). An example is shown in Figure 4. Here, Tilly shows us visually Who does what to whom? in Great Britain. It is evident that a mob engages in violent and conflict actions (attack) mostly against state 14 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

forces (constables, police, and troops) or merchants (trade); inhabitants prefer negotiation and bargaining when they interact with authorities (aldermen, local officials, members of parliament (MP), mayor, etc.); and various social groups (society, electors, inhabitants, etc.) express their support for a variety of political elites (members of parliament, royalty, ministers). Tilly dedicated substantial effort to explaining such variations in the patterns of contentious relations (Tilly 2008). This shift in Tillys research focus from levels of protest and of violence to shifting patterns of contentious relations represents more than a change in strategies of data analysis. The shift reflects a broad epistemological shift toward relational analysis in sociology (Emirbayer and Goodwin 1994 ). Tilly pioneered such an intellectual movement in the field of event analysis and contentious politics.

3) BRIT IS GOOD FOR DYNAMIC ANALYSIS The third unique feature of BRIT is that, unlike most event data projects, it records information about protests at two levels: the event on the one hand and the action on the other. Traditionally, most protest event studies have used a single level -- event (Beissinger 2002; Spilerman 1970; Taylor 1983). These studies typically assign a broadly defined category describing an events principal form of action such as vigil; strike; demonstration; and civil disobedience, as shown in the example of the European Protest and Coercion Data database in Figure 22 (Francisco 1995). This single-level coding strategy is inconvenient because it cannot deal with intra-event
2

Column C of Figure 2 shows the main characteristic of the event to which I am referring. Francisco named this column Action to denote the main forms of action used in each event. The level of data recording (the rows of the spreadsheet) is event. Readers should not confuse Franciscos use of Action, meaning a form of action, with my use of action, meaning a level of data recording.

15 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

changes and sequences (Rucht and Neidhardt 1998:68; Tarrow 1989). Within an event, protesters may employ a variety of action forms. New actors may join in the course of an event. The total number of participants changes accordingly. Newcomers may bring in new issues. A segment of actors may move to other locations. A countermovement may mobilize. Tilly notices the importance of recording these sequences of intra-event interactions and dynamics. His Great Britain database uses action as the lower level in addition to event, so that it can preserve information on intra-event changes in action forms, actors, locations, etc. (Horn and Tilly 1986). Tilly exploits the fact that most news articles on intra-event interactions are written in a simple grammatical structure, namely, Subject-Verb-Object3. The Subject identifies the actors who undertake an action; the Verb identifies the action; and the Object is the target of the action. In the database, Tilly has organized the basic elements of protest action: who (Subject) did what (Action Form) to whom (Object) where (Location) and when (Time)? Event, a higher observational level, is composed of a series of actions. Every event has at least one action. The previous example (Figure 1 and Figure 3) belongs to an event (contentious gathering) comprised of a sequence of 13 intra-event interactions. The entire action sequence is shown in Table 1. Figure 1-A corresponds to the fourth action and Figure 1-B to the eleventh. By reading the table from the first action to the last, we can easily tell that a mob and a prisoner interacted (followed, refused, demanded, pushed, etc.) with an officer over a little boy who had committed a robbery. Due to this innovation, Tillys narratives of contentious events often contain detailed descriptions of interaction sequences (1995, 2008). TABLE 1: AN EXAMPLE OF A SEQUENCE OF INTRA-EVENT INTERACTIONS FROM THE BRIT DATABASE (CGID= 830072102)
3

See Roberto Franzosis advanced coding strategy on this issue (Franzosi 1998; Franzosi 1999; Franzosi 2004).

16 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Sequence 1 2 3 4 5 6 7 8 9 10 11 12 13

SUBJECT (actor) Little Boy Mob Officer Officer Officer A Person Mob Officer Prisoner Officer Mob Prisoner Mob Prisoner Prisoner

VERB (form of action) Rob Gather Arrive Arrest Inquire Direct Follow Refuse Demand Refuse Push Rescue Bring before

OBJECT (target) ---Little Boy A Person Officer Officer Little Boy Mob Officer Prisoner Officer Little Boy Magistrates

In sum, the first innovationBRIT is open to qualitative exploration due to a bottom-up coding strategyallowed Tilly to produce rich narratives on British contentious politics in his seminal works, such as Popular Contention in Great Britain, 1758-1834 (1995). The second innovation the decision to use a relational data format to preserve all the complex relations among event characteristicsenabled him to apply network analysis to the data and discover important shifts in contentious networks in the form of who (actor) does what (action form) to whom (target) in eighteenth and nineteenth century Great Britain (Tilly 2008; Tilly and Wood 2003). Unlike these first two innovations, however, Tilly was not able to make the most of the third innovation, that is, his decision to record intra-event sequences in BRIT. I believe that this third innovation has potential to open up new inquiries and generate plenty of interesting research on political dynamics. In the following section, I take a step toward formulating such new inquiries, a step which, as I will indicate, Tilly also intended to take. 17 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Extension of Tillys innovations

AN EMAIL MESSAGE: SOMETHING TO THINK ABOUT

In the previous section, I have noted that Tilly did not make the most of his third innovation in BRIT (i.e., recording the intra-event dynamics of action sequences). Tilly saw the possibility of doing something original by analyzing these intra-event action sequences. The following is a copy of an email message I received from him on January 19, 2007, with the subject line Something to Think About (a minor change was made by the author for readability). In the message, he mentioned numerical categories called BVCAT (Broad Verb CATegories), which were about verbs representing forms of contention. Lets say we code BVCAT Move = 1, Attack = 2, Control = 3, Meet = 4, Negotiate = 5, Support = 6, and Other = 7. Then, for the contentious gatherings of 1758, we find the following sequences: 758010701 758012401 758012402 758012801 758020601 758030801 758060801 758071201 758090101 758100201 758100901 758102001 758111001 758121801 13322123 12 1231 12 31323 11233 11631 123 321323353 1332 117 1546 46 13523

Looks abstract, but its rather interesting. Even if we didnt already know a lot of 1758, we could see that it produced a high proportion of violent, high-conflict events, and guess that 1834s sequences would look entirely different. We begin to see the possibility of classifying events by both mixes and sequences as well as formulating hypotheses such as attack and control alternate, meet leads to negotiate and support, control as first action means opponents or authorities started it, et cetera. The numbers also make frequency distributions and numerical treatments of the data manageable. Worth thinking about. 18 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Chuck Worth thinking about, indeed. Tilly found some regularity in the action sequences (represented by the numbers) in the 1758 contentious gatherings. He expected different patterns of action sequence between the 18th century repertoires (1758) and those of the 19th century (1834). He also anticipated specific sequences of action verbs such as attack-control and meet-negotiate-support. The second to the last sentence clearly suggests that Tilly was interested in pursuing quantitative investigations of intra-event action sequences. Unfortunately for us, he was not able to carry out this line of inquiry further. In the remainder of the paper, I try to extend his research intentions as suggested in the email message above.

TILLYS THEORY OF PERFORMANCES AND REPERTOIRES What can we uncover by examining the patterns of intra-event action sequences? Tillys email message suggests that a study of patterns of action sequence will provide us with muchneeded knowledge about what are referred to in the literature of contentious politics as performances and repertoires of contention. In Tillys definition, a performance is a continuous sequence of actions by which an actor makes a claim (Tilly 1992:6). Elsewhere he has spoken of performance this way: Presenting a petition, taking a hostage, or mounting a demonstration constitutes a performance linking at least two actors, a claimant and an object of claims (Tilly 2006:35, the emphasis original). On the other hand, a repertoire is defined by Tilly as: a limited set of routines that are learned, shared, and acted out through a relatively deliberate process of choice. Repertoires are learned cultural creations, but they do not descend from abstract philosophy or take shape as a result of political propaganda; they emerge from struggle. People learn to break windows in protest, attack pilloried prisoners, tear down dishonored houses, stage public marches, petition, hold formal meetings, organize special-interest associations. At any particular point in history, 19 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

however, they learn only a rather small number of alternative ways to act collectively. (Tilly 1995:42) Furthermore, putting performance and repertoire together, Tilly has noted that all performances that characterize the interaction among a specified set of collective actors constitute that sets repertoire of contention (Tilly 1992:6). In short, a repertoire is a collection of performances, while a performance is a sequence of actions. As we know from Tillys various publications during the 1990s and the 2000s, he advanced a theory of repertoire change in Great Britain. According to this theory, the emergence of the strong central statethe national Parliament in particularand the expansion of capitalism during the 18th and 19th centuries caused a major shift in popular politics from what he labeled an 18th century repertoire to a 19th century repertoire (Tilly 2008:43-44). The 18th century repertoire was composed of various ways of claim-making, including food riots, blocking, touring, threatening, parading, forced illuminations, attacks on public figures, breaking windows, taking advantage of authorized public ceremonies to voice preferences, and so on (Tilly 2008:79, 133). The 19th century repertoire included public meetings, petitions, and demonstrations and, according to Tilly, became modular by the 1830s (Tilly 2008:79). As these forms of contention were widely used by social movement organizations in Britain and in other countries, Tilly also called this repertoire the social movement repertoire (Tilly 2008:79). To verify the theory of repertoire change from the 18th to the 19th century variety, it is necessary to gather empirical evidence of repertoires of contention from those time periods. As I have already discussed above in the section on his legacies, Tilly offered a variety of narrative analyses and network analyses of the BRIT database to provide such evidence. However, the kind of evidence we need to verify repertoire change is a bit more complicated. If a repertoire is 20 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

defined as a collection of performances, then we need evidence of performances to talk about repertoires. And if a performance is defined as a sequence of actions, we need evidence of action sequences to talk about performances and, thus, repertoires. This is why Tilly showed his strong interest in examining BRITs action sequences and why I regret that Tilly did not have enough time to take full advantage of BRITs innovations. And, this is why I have attempted to seek empirical evidence of the repertoire shift by analyzing intra-event action sequences.

METHODOLOGY: ASSOCIATION RULE ANALYSIS There is a methodological problem. It is difficult to quantify performances and repertoires. A reason why Tilly opted for the theatrical metaphors of performance and repertoire, rather than a more conventional concept such as forms of contention, is that he wanted to emphasize the cultural and habitual aspects of contention. People choose action forms not randomly, and not necessarily rationally, but do so culturally. As the above-cited definition of repertoire indicates, performances and repertoires are something to be learned from a history of previous struggles. Tillys choice of performance and repertoire indicates that people can adopt and modify a learned script more or less flexibly. They do not employ the script as something to be repeated and enforced strictly in the same order, as is the case in military drills. Tilly explained this flexibility thus: Claim-making usually more resembles jazz and commedia dellarte than the ritual reading of scripture. Like a jazz trio or an improvisatory theater group, people who participate in contentious politics normally can play several pieces, but not an infinity (Tilly 2006:35). In short, performances are similar but not identical patterns of action sequences. Any two sequences of actions belonging to the same performance are similar, but they are still different from each other. Then, where do we draw the boundary by which we classify similar 21 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

performances into the same group and dissimilar performances into a different group? I argue that the flexible nature of performances should be handled as a matter of transition probability from one action A to the next action B. If a particular performance looks more like a military drill and if the adopters of such a performance repeat an identical action sequence A-B over and over again, then the probability of action A followed by action B should be close to 100%. If a performance is more flexible and the adopters of the performance play slightly different versions, then the probability of action A followed by action B should be high enough so that we can find a durable pattern, but one not too close to 100%. In this paper, I employ association rule analysis to deal with these nuances because it is a method based on transition probability and thus suitable for action sequence analysis. Association rule analysis is a popular method used to discover relationships between variables in large databases in the field of data mining and text mining (Srikant and Agrawal 1996; Wang, Shi, Bai, and Zhao 2009). The primary objective of association rule analysis is to detect rules of associations or hidden patterns of relationships between specific values of categorical variables in large data sets, such as customers who buy magazine A often also buy magazine B. Association rules take the form of If X then Y and are written as X => Y. The left hand side (lhs) of the rule, X, is called the antecedent or rule body, and the right hand side (rhs), Y, is called the consequent or rule head. X and Y can be either single categorical items or conjunctions of categorical items. For example, suppose that a travel agent wants to discover relations between a variable on customers preferred travel destinations and a conjunction of two categorical variables about customers (Occupation and categorized Age). The travel agent might obtain the following association rules: if (Occupation=student and Age=20s) then 22 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

(Destination=beach) and if (Occupation=teacher and Age=50s) then (Destination=archaeological ruin). Once extracted, the knowledge about the travel preference rules can help the agent make special travel offers attuned to customers profiles (the information about Occupation and Age). TABLE 2: A DATASET OF ACTIONS WITH LAGGED ACTIONS USING A BRIT CONTENTIOUS GATHERING AS
AN ILLUSTRATIVE EXAMPLE (CGID= 830072102)

ID t=-4 t=-3 t=-2 t=-1 t=0 ----Rob 1 ---Rob Gather 2 --Rob Gather Arrive 3 -Rob Gather Arrive Arrest 4 Rob Gather Arrive Arrest Inquire 5 Gather Arrive Arrest Inquire Direct 6 Arrive Arrest Inquire Direct Follow 7 Arrest Inquire Direct Follow Refuse 8 Inquire Direct Follow Refuse Demand 9 Direct Follow Refuse Demand Refuse 10 Follow Refuse Demand Refuse Push 11 Refuse Demand Refuse Push Rescue 12 Demand Refuse Push Rescue Bring before 13 In this example, up to four previous actions are recorded as lagged actions. It is not hard to understand why association rule analysis has become popular in the field of data mining and, in particular, marketing research. But, how can we apply the method to a study of sequences of contentious interactions? My approach is to generate a dataset of actions with lagged actions and apply association rule analysis to the dataset. Table 2 shows an illustrative example of such a dataset. This example shows how the action sequences in one event (CGID=830072102) in Table 1 are modified before association rule analysis is applied. I use the categorical variable t=0 as the rule head, Y, and one or more of the remaining variables (t=1, t=-2, t=-3, and t=-4) as the rule body, X. In this way, I argue that the association rule 23 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

analysis is applicable to sequence analysis. Association rule analysis has not been used widely in the field of contentious politics or in sociology in general. In international relations, Kovars research team applied this method to a large event data set named the Kansas Event Data System (KEDS), a database covering interactions between countries in the Persian Gulf region. Kovars team attempted to discover robust event sequences and sequential rules (Kovar, Furnkranz, Petrak, Pfahringer, Trappl, and Widmer 2000).

ANALYTIC PROCEDURE Procedures of association rule analysis are best illustrated by an actual example. The main goal of the analysis is to identify performances in the BRIT database by uncovering frequently occurring sequences of two actions (verbs). There are four steps to follow. In the first step, a dataset of actions with lagged actions like the one in Table 2 must be constructed. We need only two variables related to actions (t=-1 and t=0) because we intend to find durable patterns of two action sequences in this exploratory analysis. Also, it should be noted that the first action of every event (e.g., ID=1 in Table 2) should be removed from the data set because it does not have a preceding action at t=-1. Furthermore, a variable related to the subject (actor) of an action is included in the data set. I do this because it is important to understand not only the actions but also the actors who undertake these actions. In the second step, the variables in the lagged-action data set are coded or aggregated if necessary. In this analysis, information about verbs as well as subjects is aggregated. As for verbs, BRIT tracks every verbal phrase describing actors interactions in reports from the periodical texts. There are 1,584 verbs recorded in a table named VERB, including 24 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

words/phrases such as abuse, accept, gather, hammer, ignite, and more (see VERB in BRIT2007 database). Tilly grouped these verbs into 45 verb categories based on frequency, linguistic similarity, co-occurrence, general knowledge of British contention, and sheer intuition [emphasis added] (Tilly 2008, p.50). Appendix 1 shows a list of main verbs in each of the 45 verb categories.4 I use VCAT not VERB for this paper. Information about subjects is coded in two categories: authorities (state actors) as A and population (social actors) as P. In a small number of cases, authorities and population act together as the subject of action. A code AP is assigned to these cases.

TABLE 3: OUTPUT OF ASSOCIATION RULE ANALYSIS EXAMINING RECURRENT ACTION SEQUENCES IN

POPULAR CONTENTION IN GREAT BRITAIN DURING THE EIGHTEENTH CENTURY

lhs 1 2 3 4 5 6 7 8 9 {SVCAT1=P:FIGHT} {SVCAT1=P:ENTER} {SVCAT1=P:ATTACK} {SVCAT1=P:ENTER} {SVCAT1=A:ATTACK} {SVCAT1=P:OPPOSE} {SVCAT1=A:CHAIR}

rhs => {SVCAT=P:ATTACK} => {SVCAT=P:ATTACK} => {SVCAT=P:ATTACK} => {SVCAT=P:CONTROL} => {SVCAT=A:ATTACK} => {SVCAT=P:ATTACK} => {SVCAT=P:RESOLVE} => {SVCAT=P:CONTROL} => {SVCAT=P:ATTACK} => {SVCAT=P:MOVE} => {SVCAT=P:CONTROL} => {SVCAT=P:ATTACK}

support confidence lhs.support 0.007034632 0.009199134 0.002705628 0.054653680 0.002164502 0.007575758 0.001623377 0.002705628 0.004329004 0.022727273 0.001623377 0.001623377 0.006493506 0.003787879 0.4594595 0.020021645 0.4166667 0.006493506 0.3568905 0.153138528 0.3333333 0.006493506 0.3000000 0.005411255

lift 2.432897 2.206304 1.889781 2.425197 1.588539

{SVCAT1=P:RESOLVE} => {SVCAT=P:RESOLVE}

0.6190476 0.011363636 36.903226

0.3111111 0.024350649 11.273203 0.3000000 0.005411255 29.178947 0.2941176 0.009199134 17.533207 0.2857143 0.015151515 0.2857143 0.079545455 0.2727273 0.005952381 0.2727273 0.005952381 0.2608696 0.024891775 0.2500000 0.015151515 2.078740 1.512894 3.549296 1.984252 3.394979 1.323782

{SVCAT1=P:PETITION} => {SVCAT=P:DELIBERATE} 0.001623377

10 {SVCAT1=P:RESIST} 11 {SVCAT1=P:MOVE} 12 {SVCAT1=P:MARCH} 13 {SVCAT1=P:MARCH} 15 {SVCAT1=P:RESIST}

14 {SVCAT1=P:DISPERSE} => {SVCAT=P:MOVE}

lhs: left hand side (antecedent or rule body), X, showing a combination of the subject (coded as A, P, or AP) and the action at t=-1. rhs: right hand side (consequent or rule head), Y, showing a combination of the subject (A, P, or AP) and the action at t=0. support: p(X, Y) or joint probability of X and Y.
4

For a complete list of verbs in each category, see VCAT in BRIT2007 (Tilly and Wada 2007).

25 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

confidence: p(Y|X) or conditional probability of Y given X. lhs.support: p(X) or probability of the rule body. lift: p(y|x)/p(y) or the ratio of conditional probability of Y given X to probability of Y. The 15 association rules (subject & t=-1=>subject & t=0) are extracted as the most important action sequences in 18th century Great Britain. The following three criteria are used to extract these rules: (1) lhs.support or p(X)>.005; (2) confidence or p(Y|X)>=.25; (3) lift > 1. The analysis is done using the arules package in R. Data Source: BRIT2007 (Tilly and Wada 2007).

In the third step, association rule analysis is applied to the lagged-action data set using the a priori algorithm that enables rapid processing of huge data sets. Specifically, the data set is imported into the statistical software R, and the a priori algorithm is implemented using an R package called arules (Hahsler, Grun, and Hornik 2005). The algorithm computes the probabilities of all association rules (all X => Y combinations). In the fourth and final step, all the association rules computed in the previous step are inspected, and important ones are extracted based on the criteria specified by the researcher. An example of the output is given in Table 3. This output shows the results of association rule analysis applied to the data between 1758 and 1781. In this analysis, I try to detect what Tilly called the 18th-century repertoire and its performances. Look at the first rule. Its lhs (the left hand side of the rule, X) is SVCAT1=P:RESOLVE and its rhs (the right hand side, Y) is SVCAT=P:RESOLVE. This means, if the previous action at t=-1 is resolve by a population (social actor), then the subsequent action at t=0 is likely to be resolve by a population. The a priori function in the arules package gives three probabilities in the next three columns. The first, support, is a joint probability of the items in X and Y, p(x,y). The first entrys support is about .007, meaning that a sequence of the two actionsa populations resolve followed by a populations resolvehappens about .7% of the time. The second probability, confidence, is a conditional probability (or transitional probability) of the items in 26 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Y given the items in X, p(y|x). If X is a populations resolve (t=-1), then the probability of Y, the subsequent action (t=0), being a populations resolve is .619 (61.9%). The third probability, lhs.support, is a probability of each of the items in X (actions at t=-1), p(x). A populations resolve happens about .011 (1.1%) of all the actions at t=-1. Finally, the last column of the output is a statistic called lift, p(y|x)/p(y). I will discuss this shortly. How should we set the criteria for rule extraction? The association rules in Table 3 are obtained by using the following three criteria: 1) lhs.support or p(X)>.005; 2) confidence or p(Y|X)>=.25; 3) lift > 1. First, I exclude the conditions (X) that happen less than .5% of the time because it is not fruitful to extract a rule body X which rarely occurs in the first place. Second, I set the threshold of confidence at 25%. Confidencethe conditional probability of Y happening given X, p(Y|X)conveys a sense of robustness of the association rule, X => Y. The higher the conditional probability of the rule X => Y, the more likely the rule is actually realized. Thus, we can find durable and repetitive sequences of two actions using this conditional probability criterion. Third, it is important to select rules when lift is greater than 1. The lift, p(y|x)/p(y), is the ratio of the conditional probability of Y given X to the probability of Y. The conditional probability p(y|x) needs to exceed the unconditional probability p(y). Even when the conditional probability of a rule X => Y appears high, say 70% (thus satisfying the second criterion), it is not meaningful to talk about the rule if Y in itself happens frequently, say 80%. This is because, to begin with, we can expect the occurrence of Y 80% of the time. By using a filter lift>1 we can find the rules that actually increase the likelihood of Y happening. The findings of the association rule analysis are reported in the next section.

27 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Findings This section compares action sequence patterns in Great Britain during the 18th century with those during the 19th century. The main goal is to find out whether association rule analysis can FIGURE 5: A COMPARISON OF EIGHTEENTH-CENTURY AND NINETEENTH-CENTURY PERFORMANCES 5-A) Eighteenth-century performances (1758-1781)

5-B) Nineteenth-century performances (1832-1834)

28 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Ovals represent actions by a population. Squares are actions by authorities. The arrows indicate the association rules {X => Y} with respective conditional probabilities, p(Y|X). Data source: BRIT2007 (Tilly and Wada 2007). really detect distinct patterns of verb sequence for the 18th century contentions and for the 19th century contentions, as Tilly anticipated in his email message. As we have already seen in Table 3, the 15 association rules (subject & verb t=-1 => subject & verb t=0) are extracted as the most important action sequences in 18th century Great Britain (1758-1781). A graphical presentation of the same results is shown in Figure 5-A. The ovals represent social actors actions, and the squares are state actors actions. The arrows indicate the association rules {X => Y} with respective conditional probabilities, p(Y|X). Each arrow in the graph represents an association rule. Here it is important to point out five issues. First, there are four clusters of actions. Since the clusters are disconnected from (not linked by arrows to) each other, a sequence of two actions belonging to different clusters (i.e., petition => fight, move => resolve) rarely happened. I conclude that such an action sequence does not qualify as a performance. Each of the four

29 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

clusters qualifies as a unique performance. In other words, association rule analysis finds that the 18th century repertoire of Britons contains four distinct performances. Second, one of the four performances includes many action sequences (the large cluster at the top). We can see that the performance often started with verbs indicating outdoor movement of people (disperse, march, enter, and move) and then led to verbs indicating conflictive and sometimes violent interactions (control, resist, fight, oppose, and attack). Such action sequences are the kind of performances Tilly expected to find in 18th century contention. Tilly noted in his email message, Even if we didnt already know a lot of 1758, we could see that it produced a high proportion of violent, high-conflict events. Association rule analysis adds to his observation that outdoor movement actions such as public gatherings, marches, or demonstrations likely preceded the violent, high-conflict actions (such as attack). It is likely that the intensity of contention would have escalated from outdoor gatherings to violent exchanges, and not the other way around. Third, the 18th century repertoire includes not only outdoor-conflictive performance which is, according to Tilly, typical of this century but also the other three sequences: (1) {state actors attack => state actors attack}; (2) {state actors chair => social actors resolve => social actors resolve}; and (3) {social actors petition => social actors deliberate}. Tilly would classify the first as a typical 18th century performance by authorities. The second and the third sequences are interesting. The second pattern {state actors chair => social actors resolve} is a typical action sequence in public meetings. A person holding an authoritative position presided and chaired a meeting and the participants resolved a conflict, a dispute, or a problem during the meeting. The third pattern {social actors petition => social actors deliberate} is a typical action sequence in petitions. A group of people petitioned a 30 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

government official for help, aid, or assistance, and the group appointed its representatives or deliberated a possible solution. Tilly claims that these two patterns or performances constituted an important part of the 19th century repertoire. Association rule analysis demonstrates clearly, however, that public meeting and petition performances were already a part of the 18th century repertoire. Fourth, I suggest that the complexity of the action sequence cluster possibly reveals the extent of the performers mastery of the performance. In Figure 5-A, outdoor-conflictive performance is the most complex of all. During the period between 1758 and 1781, Britons were most familiar with this performance and, therefore, were able to use a lot of improvisation, performing different versions of the performance following a multiplicity of action sequence paths. Look at Figure 5-B. The graph shows the results of association rule analysis based on the 1832-1834 data in BRIT. There are four unconnected clusters. The largest cluster at the top now combines petition performance {request => assemble => petition => disperse => authorities hear petition} and public meeting performance (the action sequences including communicate, meet, chair, and resolve). This cluster represents Tillys 19th-century repertoire. A four-verb cluster at the bottom right of the graph (meet, resolve, cheer, and disperse) is public meeting performance by state actors. Compare the degrees of complexity between the 18th and 19th centuries. Simple and undeveloped action sequences of public meeting and petition performances in the 18th century graph imply that these performances were at an embryonic stage. The structure of action sequences of public meeting and petition performances in the 19th century graph is very complex, signaling that Britons had learned, refined, and applied these performances at a more 31 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

sophisticated level. Outdoor-conflictive performance still exists in Figure 5-B (the three-verb cluster of fight, block, and attack), but its outdoor-related verbssuch as march, disperse, and move which are important components of performance in the 18th centuryare removed. Also, its structure of action sequences is considerably simpler than the one in the 18th century. Thus, it is important to acknowledge that although what Tilly calls the 18th century performance of outdoor violence still constituted a part of the 19th century repertoire, its flexibility, richness, and sophistication as a performance had declined. We also find the birth of a new performance. It is the fourth cluster {Receive => Cheer => Cheer}. In a gathering, a group of people received someone or another groupfor instance, politicians, social or religious leaders, members of royalty, or other types of elitesand cheered, greeted, and expressed support for those person(s). Emulating the performances employed in electoral campaigns, such public demonstrations of support became an important component of the 19th century repertoire in British politics. In short, association rule analysis provides quantitative empirical evidence of the existence of the typical 18th century performance of outdoor-conflictive action sequences and the typical 19th century performances of both public meeting and petition action sequences. This finding supports Tillys arguments. In addition, the analysis discovers both that the 18th century repertoire already contained the 19th century performances of public meetings and petitions, albeit in embryonic form, and that the 19th century repertoire included the 18th century performance of outdoor-conflictive action sequences. Conclusion 32 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

The first part of this paper has discussed Tillys legacy by introducing three specific innovations he made in the field of event analysis. I have argued that Tilly developed an event data set that worked well for qualitative analysis, relational analysis, and dynamic analysis of action sequences. First, differently from a majority of event data sets in the field of contentious politics, BRIT is readable because Tilly invented the bottom-up approach. This approach is able to convert a multiplicity of textual expressions in news and other periodical articles into codes of standardized expressions organized by a standardized grammar of who did what to whom, when, where, and why. As a result, BRIT permits not only quantitative analyses of contentious events but also qualitative examination of these events. Second, I have also pointed out that Tillys data set was designed for a relational analysis of contention. This is because he organized information about contentious events in the standardized grammatical form of who did what to whom. The form reveals the nature of a relationship (did what) between actors (who) and targets (to whom). Tilly demonstrated the utility of such a data organizing system by employing network analysis of contentious relations (Tilly 2008; Tilly and Wood 2003). Third, I have emphasized the originality of Tillys decision to preserve information about action sequences within each event (contentious gathering). The time and labor costs of recording such information must have been enormous, but his decision paved the way for analyzing more fully the dynamics of contentious events. His untimely death prevented him from pursuing such a line of research. I have devoted the second part of this paper to an attempt to follow through with his research intentions. In order to identify performances and repertoires in 18th and 19th century Great Britain as empirical evidence with which to evaluate Tillys theory of repertoire change, I have employed 33 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

association rule analysis to examine action sequence data in the BRIT database. Association rule analysis distinguishes probabilistically between likely action sequences and unlikely action sequences and successfully identifies performance clusters within a repertoire. Given the difficulty of documenting performances and repertoires empirically, I believe this is a noteworthy accomplishment. An advantage of association rule analysis is its ability to show the complex structure of action sequences. Moreover, such analysis clearly demonstrates the likely order in which specific actions are carried out within each performance. Another advantage of this method is that conditional probabilities are computed for each sequence (rule) of action, thus giving researchers ideas about the likelihood of the actual sequence. Charles Tillys three innovations in his dedicated work with BRIT have opened up a new research possibility. Anyone who plans to launch an event data project should learn from his remarkable contributions. As this paper has discussed, Tillys action sequence data has the potential to generate a great deal of interesting and fruitful research. I wonder if we can realize the potential without him. That is the problem.

34 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

APPENDIX 1: VERBS IN VCAT (VERB CATEGORY) AS CLASSIFIED BY TILLY

VCAT ADDRESS ADJOURN ASSEMBLE ATTACK ATTEMPT BLOCK BRACKET CELEBRATE CHAIR CHEER COMMUNICATE CONTROL DECRY DELEGATE DELIBERATE DEMONSTRATE DIE DINE DISPERSE DONKEY END ENTER FIGHT GATHER GIVE HEAR PETITION HUNT MARCH MEET MOVE NEGOTIATE OBSERVE OPPOSE OTHER PETITION PROCEED RECEIVE REQUEST RESIST RESOLVE SMUGGLE SUPPORT THANK TURNOUT VOTE VERBS Address Adjourn Assemble, Convene, Increase Assault, Attack, Beat, Break, Burn, Demolish, Destroy, Fire, Pelt, Stone, Strike, Throw at, Wound Attempt, Endeavor Close, Interfere, Prevent, Surround Commence, Continue Play Chair, Preside Acclaim, Applaud, Cheer, Greet Advertise, Announce, Cry, Declare, Notify, Read, Shout Arrest, Collect, Compel, Force, Order, Rescue, Secure, Seize, Sentence, Take, Threaten, Throw, Try Assail, Decline, Hiss, Hoot, Insult Depute Appoint, Commit, Conclude, Propose Salute Die Dine Break-up, Depart, Disperse, Dissolve, Escape, Leave, Retire, Retreat, Separate Donkey End Enter Arm, Fight, Riot Congregate, Gather Give Hear Petition Poach March, Parade Meet Appear, Arrive, Attend, Come, Follow, Go, Join, Move, Pass, Pursue, Return, Rush Agree, Insist, Offer, Promise Discover, Recognize, See Oppose Find, Open, Swear Petition Proceed Receive Ask, Call, Call-out, Demand, Request Refuse, Resist Resolve Smuggle Assist, Nominate, Present, Support Thank Turn-out Elect, Poll, Vote

Some of the verb categories (VCAT) such as CONTROL contain many verbs. For these categories, the most frequent verbs appearing more than 50 times in the data and the other 35 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

important verbs are listed. For a complete list, refer to BRITs VCAT form. Data Source: BRIT2007 (Tilly and Wada 2007).

36 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Bibliography Beissinger, Mark R. 2002. Nationalist mobilization and the collapse of the Soviet State. Cambridge and New York: Cambridge University Press. Emirbayer, Mustafa and Jeff Goodwin. 1994. Network analysis, culture and the problem of agency. American Journal of Sociology 99:1411-1454. Fillieule, Olivier. 1998. Plus ca change, moins ca change. Demonstrations in France during the nineteen-eighties. In Acts of dissent: new developments in the study of protest, edited by D. Rucht, R. Koopmans, and F. Neidhardt. Berlin: Edition Sigma. Francisco, R. 1995. The relationship between coercion and protest: an empirical evaluation in three coercive states. Journal of Conflict Resolution 39:263-282. Franzosi, Roberto. 1989. From words to numbers: a generalized and linguistics-based coding procedure for collecting textual data. Sociological Methodology 19:263-298. . 1998. Narrative analysis--or why (and how) sociologists should be interested in narrative. Annual Review of Sociology 24:517-554. . 1999. The return of the actor: interaction networks among social actors during periods of high mobilization (Italy, 1919-22). Mobilization 4:131-149. . 2004. From words to numbers: narrative, data, and social science. Cambridge and New York: Cambridge University Press. Hahsler, Michael, Bettina Grun, and Kurt Hornik. 2005. arules -- a computational environment for mining association rules and frequent item sets. Journal of Statistical Software 14:125.

37 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Horn, Nancy and Charles Tilly. 1986. Catalogs of contention in Britain, 1758-1834. Working Paper Series, Center for Studies of Social Change, New School for Social Research, New York. Koopmans, Ruud and Paul Statham. 1999. Political claims analysis: integrating protest event and political discourse approaches. Mobilization 4:40-51. Kovar, Klaus, Johannes Furnkranz, Johann Petrak, Bernhard Pfahringer, Robert Trappl, and Gerhard Widmer. 2000. Searching for patterns in political event sequences: experiments with the KEDS database. Cybernetics & Systems 31:649-668. Kriesi, Hanspeter, Ruud Koopmans, Jan Willem Duyvendak, and Marco G. Giugni. 1995. New social movements in Western Europe: a comparative analysis. Minneapolis: University of Minnesota Press. Rucht, Dieter and Friedhelm Neidhardt. 1998. Methodological issues in collecting protest event data: units of analysis, sources and sampling, coding problems. In Acts of dissent: new developments in the study of protest, edited by D. Rucht, R. Koopmans, and F. Neidhardt. Berlin: Edition Sigma. Shapiro, Gilbert and John Markoff. 1998. Revolutionary demands: a content analysis of the cahiers de dolances of 1789. Stanford: Stanford University Press. Spilerman, Seymour. 1970. The causes of racial disturbances: a comparison of alternative explanations. American Sociological Review 35:627-649. Srikant, Ramakrishnan and Rakesh Agrawal. 1996. Mining quantitative association rules in large relational tables. In '96 Proceedings of the 1996 ACM SIGMOD international conference on Management of data. New York: Association for Computing Machinery.

38 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

Tarrow, Sidney. 1989. Democracy and disorder: politics and protest in Italy, 1965-1975. Oxford: Oxford University Press. Taylor, Charles Lewis. 1983. World handbook of political and social indicators. Volume 2: Political protest and government change. New Haven: Yale University Press. Tilly, Charles. 1981. As sociology meets history. New York: Academic Press. . 1992. How to detect, describe, and explain repertoires of contention. New School for Social Research, New York. . 1995. Popular contention in Great Britain, 1758-1834. Cambridge, MA: Harvard University Press. . 1997. Roads from past to future. Lanham, ML: Rowman & Littlefield Publishers. . 2002. Event catalogs as theories. Sociological Theory 20:248-254. . 2006. Regimes and repertoires. Chicago: University of Chicago Press. . 2008. Contentiouis Performances. Cambridge: Cambridge University Press. Tilly, Charles and Takeshi Wada. 2007. BRIT2007: Contentious Gatherings in Great Britain, 1758-1834. An Event Database in a Microsoft Access format. Tilly, Charles and Lesley Wood. 2003. Contentious connections in Great Britain, 1828-1834. In Social movements and networks: relational approaches to collective action, edited by M. Diani and D. McAdam. Oxford: Oxford University Press. Wang, Pei-ji, Lin Shi, Jin-niu Bai, and Yu-lin Zhao. 2009. Mining Association Rules Based on Apriori Algorithm and Application. In IFCSTA '09 Proceedings of the 2009 International Forum on Computer Science-Technology and Applications, vol. 01: IEEE Computer Society. 39 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2

40 The International Journal of Conflict & Reconciliation Fall 2013, Volume 1 Number 2