A Method for Direct Systematic Observation of Collective Violence and Public Order Policing

Introduction

In his 1972 article on analyses of collective violence, the American sociologist Berk noted that the social-scientific literature about the behavior of crowds and the behavior of people in crowds was mostly based on restricted information and unsubstantiated interpretations and speculations. Berk (1972) indicated that existing investigations into “crowd behavior,” based as they were on the measurement techniques commonly employed by social scientists, concentrated on the conditions immediately preceding the events and the subsequent consequences and not on the more difficult to examine “mob process” itself. Data collection is rendered difficult because a great number of things happen in quick succession, simultaneously and spread out over a large area while they are difficult to anticipate. At the same time, participants have a clear interest in the interpretation of events, and their (usually retrospective) accounts, if participants are willing to cooperate at all, are especially vulnerable to conscious and unconscious distortions. When present, investigators keep their distance because of their own safety (to give a recent example, in a large ongoing comparative project that gathers data around [including during] street demonstrations in eight European countries, demonstrations expected to be violent are excluded out of concern for the safety of the data gatherers: Klandermans et al. 2011). According to Berk, this leads to interpretative pitfalls (he distinguishes five kinds) and he called for social scientists to collect far more data on what happens in crowds. This sentiment was echoed almost 20 years later for collective behavior in general by McPhail (1991:149) who stated that “Students of collective behavior have persisted in their preoccupations with developing explanations to the exclusion of establishing the phenomena to be explained” and went on to say that explanations for collective behavior cannot be developed until it is first specified, observed, and described what is to be explained. Schweingruber and McPhail (1999) developed a set of criteria and procedures for systematically observing and recording collective action across temporary gatherings such as political demonstrations. Their method uses trained observers, distributed across a gathering, who complete a code sheet recording the extent of participation in 40 more elementary forms of collective action by members of six actor categories during time interval samples. ¹ Although the code sheet has categories for violence, their method is not suited for recording and analyzing violent interactions in detail. Their methodology did not specifically focus on collective violence, as this is infrequent in temporary gatherings.

There have been several observational studies of collective violence or riots and public order policing in the last 25 years especially in the context of football matches and protest demonstrations. This article discusses a number of them and the methodological issues encountered by the respective researchers. Following that, a methodology is presented and discussed for the direct, systematic observation and recording of collective violence and public order policing, as it has been used to observe 225 volatile crowd events.

Observational Studies of (the Policing of) Football Matches and Protest Demonstrations

The focus here is on those published studies of football matches and protest demonstrations that have observation as their main (if not their only) research method. Studies were selected on the basis that they used observation to answer questions related to violent interactions between police and football supporters or protestors and, as a crucial condition, on the basis that they paid explicit attention to methodological issues related to the use of observation. No attempt is made to present an exhaustive review of all studies that used observation as part of their methodology, yet it is felt that the studies selected are representative of those that use observation in the study of (the policing of) football matches and protest demonstrations. Most of the studies selected are frequently cited in other publications on (the policing of) football and protest. The discussion of these studies will be restricted to issues that are directly related to observation. Obviously, studies on the (policing of) football or protest that do not employ observations and rely solely on other types of data (be they newspaper clippings, after-the-fact official or unofficial reports, surveys, interviews) are not included in this discussion.

P. Waddington (1994) studied the policing of protest in London over approximately a one-year period. He observed over 60 protest events in the streets and, connected to these events, numerous preparatory police meetings, police briefings, and negotiations between police and protest organizers. His access to these meetings was based on a relationship of trust with the police officers concerned. His participant observations focused on the operations considered to be more problematic. Even then, in most instances, no disorder or violence occurred. Waddington hesitates to call his research ethnography because his observations took place during formal meetings (and in the street) and not in informal settings. During operations, he usually accompanied the most senior police commander and was supplied with a police radio. As Waddington is careful to point out, this means his observations were made from the perspective of commanders, not lower ranking officers. He took notes (no literal quotes) contemporaneously, which he transcribed on a PC as soon as possible usually within a few hours. Like all such observational research, according to Waddington, his research suffers from a lack of transparency of the methods used and the rendering could justifiably be called impressionistic because it is impossible to say, for example, how frequently is “frequently.” He noted that it is virtually impossible in this type of research to obtain informed consent from all research subjects. The Metropolitan Police had given its consent, but then it is difficult, if not impossible, for subordinates to object. Obtaining informed consent from every participant in a meeting is practically not feasible. On meetings where protest organizers were present, Waddington was introduced by the senior officer as a researcher who “is looking at us.” On operations, there was no opportunity to secure informed consent from all of the numerous officers deployed. Waddington suggests that social scientists should reflect more upon the extent to which any group of individuals can genuinely consent to the research that is being conducted upon them. Waddington clearly states that his account is necessarily partial as just a perspective—albeit an important one—among others. Waddington also makes some interesting points with regard to reactivity: On the one hand, his presence encouraged officers “to get it right”; on the other, there was the fact that he could witness things going wrong. The impression Waddington got was that officers were proud of their professionalism and saw him as an “insurance”: an independent witness who could confirm the propriety of police action.

Giulianotti (1995) discussed methodological issues related to his participant observation fieldwork experiences on football hooliganism during a number of years with two rival groups of the so-called “casual” or “hooligan” Scottish football fans. His experiences are similar to those documented by Armstrong (1993) who studied a group of hardcore football fans from Sheffield (United Kingdom). Giulianotti (1995) deals with two “key problems”: those that emerge from attempted entree into hooligan subcultures and the everyday safety risks of comparative research with violent fans. He specifically indicates he does not discuss two other issues with participant observation because he sees them as less relevant to his studies: representative and generalizable qualities of the data and reactivity. To gain entrée, establishing rapport with the research subjects is essential, becoming an insider without becoming a native and Giulianotti stresses that he has never defined himself or was regarded as a casual. Flexibility was required in dealing with criminal activities of his subjects, and Giulianotti chose not to get involved in fighting or become a go-between for the two gangs in organizing fights. At no stage did he seek to intervene. As a relative insider of one of the groups of casuals, he felt he retained sufficient personal, professional, and ethical distance to eschew any identification as a native. At the same time, it was important to blend in (or pass) among the hooligans without exciting attention from rival casuals or police to avoid significantly influencing the latter’s actions toward the research subjects generally. Giulianotti points to the risks of harassment by his research subjects (especially when being perceived as a police “spy”), becoming the target of violence in a confrontation with rival hooligan groups (or the police) or being arrested by police.

Drury and Stott (2001) did several case studies on intergroup conflict during protest events in the United Kingdom and argue that partisanship in participant observation research may be necessary in research settings involving intergroup conflict because it enables access to a far greater amount and finer quality of data than do other frameworks (with which they especially mean experiments and questionnaire surveys). They recognize that, for a “single researcher, in situations of intergroup conflict, the reverse side of the coin of this enhanced access to one side is, restricted access to the other group involved” (p. 53) but feel that “what was lost in data from the out-group was more than made up for with data from the in-group.” Hong and Duff (2002) make a similar point discussing the dilemma of “distance” in fieldwork more generally.

Drury and Stott also recognize that the researcher’s observations of an incident of intergroup conflict are likely to reflect the perspective of the in-group (crowd participants) rather than those of the out-group (police) and that in the case of partisan research on intergroup collective conflict, the analysis might be shaped by the researcher’s wish to produce results in line with the in-group perspective. They relate two examples to show how they dealt with these biases.

Using a similar approach, Stott went on to conduct participant observation case studies of the behavior and policing of English fans at different football World Cups (Italia 90: France 98; Stott, Hutchison, and Drury 2001; Stott and Reicher 1998b). Pearson (2009), who has worked with Stott (e.g., Stott and Pearson 2007), takes partisanship a step further and discusses the dilemma of whether or not to commit offences in the course of his sociolegal ethnographic study into legislative and policing responses to crowds of “risk” English football supporters. In order to gain entrance to the field and maintain a useful research position within it, he found himself both witnessing criminal offences and being put under pressure to commit them personally. He feels little formal guidance is provided to researchers in the social sciences who wish to carry out ethnographic research within “criminal” fields.

O’Neill (2005) studied the policing of football in one of Scotland’s football grounds in the course of one year and conducted participant observations with police officers and stewards. The aim of her research was to study the relationship and interactions between police officers and domestic football supporters, focusing on the police perspective. She decided not to extend data gathering to football supporters, as it proved impossible to build rapport with supporters because of her association with the police and she did not want to risk losing the trust she had established with police. Her observations included attending briefings and joining officers on their breaks. As she puts it, she “did not engage in policing activity directly but was able to share in their experiences empathetically.” Wanting to get as wide a picture on football policing as possible, she joined several officers in the course of her study and did not just stick to a few. Rapport was therefore limited. She found tape-recording or open note-taking of her observations and interviews impractical and wrote down her field notes from memory following each match at which she participated (just as Armstrong 1993). Walking with the police, safety was not an issue for her.

A group of Swedish researchers (e.g., Granström et al. 2009) did several participant observation case studies on crowd incidents around protest events in Sweden. They provide little detail of their observation methodology except for the fact that they used multiple observers, some to focus on protesters, and some on police. That the use of multiple observers is no guarantee for reliable observations was made clear when they went out of Sweden to study the policing of a football match during the 2006 World Cup in Dortmund, where they reported a “mainly peaceful and orderly major football event” (Hau 2008), while in fact there were over 400 arrests and large-scale disturbances (Schreiber and Adang 2010).

Wahlström (2011) studied the interaction between police and political protesters in Denmark and Sweden through a series of case studies. His main focus was on the activists’ and police officers’ reflections on protest and protest policing rather than the actual interaction itself, and observations of demonstrations were initially of secondary importance. For him, observations of police and protester interaction are notoriously difficult especially if the aim is to explore group processes and/or the outbreak of violence. Close proximity to confrontations is potentially unsafe in terms of the risk of being struck by a bottle or a truncheon, and the researcher also risks arrest for involvement in the activity that he observes. Since single observers frequently find themselves in the wrong place to get an overview of crucial episodes, systematic research of police/protester interaction would require coordinated efforts of multiple observers. Wahlström nevertheless carried out a few field observations of demonstrations where he judged that there was a risk of conflict, with the primary aim to gather material that would enable him to pose better questions in interviews and to use as background knowledge for interpreting other data sources. He experienced how difficult it was to get access to both police and activists at the same time. He was told “that openly fraternizing with police officers during demonstrations was not good for my reputation in the activist groups that I was interested in” and he accordingly chose not to include any close ethnography of police officers during demonstrations. Still, gaining access to the activist groups and networks turned out to be problematic, and he notes as a crucial difference between activists and police that activists have no legal possibilities to have a researcher sign a binding agreement of secrecy. Wahlström took field notes of his observations contemporaneously and typed them out afterward. Wahlström also discusses the danger of “going native,” the risk of not establishing sufficient distance to the data during analysis and overly adopting the perspectives of the respondent, “a risk that appears to be particularly prominent when one carries out extensive participant observation and/or one enters research with an initial sympathy for the research subjects” (p. 60). Groups where there is a strong sense of us/them mentality toward their environment, such as police and many radical political groups, may exert a stronger pull on the researcher toward going native. Wahlström argues that it is actually one of the main advantages of studying two opposing groups that confrontation with contradictory perspectives on the same events can be used by the researcher as a corrective for becoming caught in either lifeworld.

Overview of Studies Using Observation

Also in the eyes (and words) of the authors themselves, all of these studies that used observation as one of their methodologies are participant observation studies (rather than studies using direct systematic observation—I did not find any study that used direct systematic observation of collective violence). Still, the level of participation differed considerably between studies, ranging from the explicitly partisan approaches of Drury, Stott, and Pearson to the more neutral and detached observations of Granström and his coworkers, representing three of the four observer roles distinguished by Gold (1958): the complete participant, the participant as observer and the observer as participant. All observations were conducted by a single observer (with the exception of Granström et al.), with Pearson the only one conducting his observations covertly. Some studies were conducted from the police perspective (Waddington, O’Neill), others from the perspective of fans or protestors (Giulianotti et al.). Wahlström experienced that it is difficult for a single observer to cover both perspectives simultaneously, where Granström et al. used multiple observers to achieve that goal.

Because most studies involved just one case or a limited number of case studies and the more prolonged studies involving multiple cases were limited to one group or two groups of research subjects (Waddington: Metropolitan Police, O’Neill: one Scottish police force, Giulianotti: two groups of Scottish casuals, Pearson: Manchester United fans and England fans, Armstrong: Sheffield United fans), representativeness of the data and generalizability of the findings are real issues. In all of these prolonged studies, collective violence was a rare occurrence. Wahlström found observing police–protester interaction to be notoriously difficult, especially at the outbreak of violence. Issues of physical safety are mentioned by several authors who did witness violent interactions. Unsurprisingly, this was not the case for researchers who attached themselves to police (O’Neill, Waddington).

Several authors mentioned ethical aspects of their research, especially the issue of informed consent. Waddington and Pearson in particular state that it is practically impossible to obtain consent from every individual involved in collective settings. Also finding the right balance between blending in and keeping professional distance was an issue in participant observation, with Stott and Drury and Pearson actually embracing a partisan approach. How to deal with witnessing crime or becoming involved in crime is another recurring issue for those embedding themselves with violent subgroups.

The closeness of participant researchers to their study subjects makes participant observation vulnerable to reactivity. This is recognized by several authors (especially Waddington), although it is remarkable that it is downplayed by those authors who keep least distance from their study subjects: Giulianotti dismisses it as less relevant to his research, as do Stott and Drury, who state that taking sides in research on intergroup conflict may minimize the bias of reactivity.

Several authors chose to write their field notes after conclusion of their observations rather than contemporaneously (Armstrong, Giulianotti, O’Neill, Pearson), a practice that as Johnson and Sackett (1998:304) note, cannot be a substitute for direct observation, as recall data “are about cultural pattern (the informant’s or the researcher’s), not about observed behavior.” It is not always clear how the researchers used the interviews they conducted as part of the participant observations: as supposedly accurate reporting of observed events or to extract meaning, interpretation, and subjective experience (cf. Johnson and Sackett 1998:302). If used as the basis for accurate reporting, Johnson and Sackett’s (1998:302-303) warning applies, “anthropologists who rely uncritically on their research suspects for descriptions of behavior are more likely to be wrong than right.”

Direct Systematic Observation of Collective Violence and Public Order Policing

As valuable as they are, the participant observation studies dealt with above also show the limitations of ethnography. P. Waddington (1994) stated that all of these types of studies (including his own) suffer from a lack of transparency of the methods and definitions used, and this was true for all of the studies mentioned above. They also tend to suffer from “exampling”: the practice whereby a researcher addresses a few examples but fails to explain how these examples represent a broader data set or why they were chosen (Gephart 2004).

According to Johnson and Sackett (1998), direct rigorous systematic observation is our best approach to developing trustworthy accounts of people’s behavior and deserves a more prominent place in the ethnographic tool kit (next to participant observation and interviews). At the same time, they note that measuring the ongoing stream of behavior in its natural detail, complexity, and context is so daunting as to be a practical impossibility. This is of course even more true when studying crowd events where numerous people gather and interact, especially if they are chaotic and violent.

By direct systematic observation, Johnson and Sackett explicate, they mean those ethnographic methods that—in contrast to interviews—rely primarily on the researcher’s first-hand observations and that—in contrast to participant observation—are seriously attentive to problems of sampling and measurement. Johnson and Sackett argue (“in the strongest terms”) that interviews and participant observation are by themselves inadequate to the task of constructing trustworthy accounts of activity patterns and that there is an irreducible need for other, more rigorous observational methods.

Accurate descriptions of behavior matter because, from the most theoretically abstract to the most immediately applied concerns, our understanding of the causes and consequences of human behavior depends on them. We would ordinarily have little respect for theories or policies based on data that had error rates ranging from 50%–80%. Yet participant observation typically produces descriptions of behavior with such error rates, unless the observations are carefully recorded immediately as the behavior occurs. (Johnson and Sackett 1998:328)

Methodology for Systematically Observing Collective Violence

The observation methodology was used to gather comparative data around 225 mostly preplanned crowd events all over the Netherlands where collective violence was thought to be likely (mostly football matches and protest demonstrations assessed as constituting risk by authorities), some of which led to disorder or turned into a riot whereas others did not. ² The first goal of the study was to provide a structured and contextualized description of violent interactions around protest and football events in the Netherlands. The second aim was to analyze contextual and interactional factors that are associated with the initiation and escalation of collective intergroup violence. The word interactional is important here, as the starting point for the research was that violence always involves at least an actor and a target and that the initiation and escalation of violence cannot be understood without paying attention to the interaction between those (this is in line with the participant observation studies dealt with above which make clear that collective violence is in essence an intergroup phenomenon). The distinction between initiation and escalation was made because it cannot be assumed that the factors influencing initiation of violence are the same as those influencing escalation.

At the beginning of the study, the observation methodology was developed by studying the literature on collective violence, watching video material of riots and by conducting a number of test observations at protest demonstrations and football matches. An important part of this was the development of clear definitions and behavioral categories. For the description and analysis, those variables were selected that could relatively well be observed in a systematic way. The assumption was that only those variables that can be discerned by the participants in an intergroup setting as well can have a direct impact on the intergroup interaction. These variables include violent, provocative, and coercive intergroup actions (including police actions) and reactions; the number of people acting; the distance between those acting and the immediate context of interactions. The study sought to examine the influence of these independent variables on the occurrence of violence as the dependent variable. For this study, direct behavioral measures were used, such as the kind of violence (e.g., throwing objects, hitting), the number of individuals that act violently, the duration of episodes of collective violence, and the frequency with which violence is used.

To this purpose, events were selected specifically on the basis of the likelihood for violence to occur based on (1) the fact that they were labeled as increased risk by authorities or police, (2) the deployment of specialized police units such as riot police, arrest units, horses, or dogs right from the start, and/or (3) explicit calls for violence made before the event by fans or protestors. Additionally, protest events with at least 1,000 expected participants were included in the observations, irrespective of the estimated likelihood of violence. As it turned out, during the research period (May 1986–September 1989), 50 percent of all protest actions occurring in the Netherlands with at least 1,000 participants reported in national media ³ and 80 percent of all violent protest events reported about in national media were, in fact, observed. ⁴ The observed football events were selected based on whether one or more of the five clubs perceived by police and media as being accompanied by the so-called risk-fan groups were involved. ⁵ In the course of the study, observations were conducted in all stadiums in the Netherlands that hosted matches in the highest football league. Some form of collective violence was actually observed in 53% of observed protest demonstrations and 92% of observed football matches (Adang 2011). ⁶

The observations on protest demonstrations started at the meeting point communicated by the organizers/initiators of the protest some time before the protest was planned to start. The observations lasted until after the event had ended and the dispersal of the participants had clearly started. The football observations started at the arrival of a group of away fans (usually by special train) in or near the city where the match was to be played and ended when they had left. ⁷

In all cases, observations were made in public (or semi-public) places that were in principle accessible to anyone (i.e., on public roads and in stadiums). Observational positions were chosen to obtain an overview of events (thus usually not in the middle of a group). The observer made no attempt to “blend in” by adapting his clothing or behavior. Observations were recorded contemporaneously on a portable audio recorder ⁸ while listening into police channels via a scanner. ⁹ In total, 700 observation hours were recorded on audio tape (this is exclusive of time spent in preobservation reconnaissance and information gathering and postevent data gathering). ¹⁰ All observations were divided into three different phases that constitute the life cycle of all events and temporary gatherings (cf. McPhail 1991; Wright 1978):

“convergence”: meeting phase where the individuals gather or assemble;

At the beginning of every observation and with any changes, the broader context of the events were recorded, including

weather conditions;

An “all occurrences” sampling method (Altmann 1974) was used to record every observed instance of a limited number of predefined acts by protestors, fans, or police (especially violence, annoyance/provocative behavior, and coercive measures); the contexts and sequences in which they occurred; the identity of actors and targets; and the course of the interaction that followed, with time recorded in seconds (using a wrist watch). As the study focused on intergroup interaction, only behaviors directed outside of the group to which the actor(s) belonged were recorded (it would have been impossible to record within-group interactions). When there were a large number of individuals gathered, it was obviously impossible for a sole observer to oversee a very large or spread-out group. Whenever the size or spread of the group concerned made this necessary, the systematic observations were limited to (focused on) interactions involving a specific subgroup thought to be most likely to become involved in a violent interaction. In virtually all situations the circumstances were such that this was possible, most events were spread over a broad time span and were nearly always accompanied by audible vocalizations or other sounds. This, combined with the mobility of the observer, made it possible to meaningfully apply the all occurrences sampling method. ¹² In recording the observations, the following definitions and criteria were used.

Violence: Every directed behavior that may cause damage and/or bodily pain, divided into:

physical violence (hitting, kicking, and biting) with or without use of an object;

So, for example, the undirected throwing of fireworks was not classified as violence, while throwing fireworks directly at a person was classified as violence. The climbing of gates or fences was classified as violence only if accompanied by systematic gate pulling in an attempt to damage them. It is imported to note that the word “violence” is used throughout according to this behavioral definition and what is usually called in English use of force by police is included. ¹³ A police charge where there was no use of a baton (or no other form of violence was applied) was not noted as violence. The arrest of an individual was not on its own classified as violence but only if accompanied by hitting or kicking, and so on.

Police charges and arrests were recorded as a form of coercive measure by police. Every act by police officials that compelled individuals or groups to read with a certain behavior was recorded as a coercive measure. In applying coercive measures, police officers could or could not use violence (according to the definition given above). Three forms of frequently applied coercive measure were distinguished:

arrest: (irrespective of the specific judicial qualification) Individuals must accompany the police and are detained/taken into custody;

Arrest and moving are easy to observe behaviors. Denial is more complicated: There are many places and moments where individuals or groups are denied access to a specific location. For the purpose of the observations, denial was classified as such only if a group was actively hindered in their progress or if a group exercised pressure and did not accept that an entrance or street was blocked to them by police.

In addition to violence and coercive measures, some behaviors were recorded as annoying or provocative:

words or gestures directed at offending another person ¹⁴ ;

For each interaction, the following factors were recorded:

the point of time (in minutes and seconds) and order of behaviors executed;

In determining context, it was important to distinguish between acts that form the start of a violent interaction and those that do not but form a continuation. The decision what acts form part of one and the same interaction or bout was not made completely arbitrary. On the basis of results of analysis of data gathered during test observations, interactions were considered ended when actors and targets had not interacted for a period of two minutes. After two minutes, an action against the same target was classified as a new initiative (with a new context being determined), while a display within this period was considered as forming part of the same interaction.

As fast as possible after finishing each observation, the cassette tape was transcribed and coded into computer files. In addition, I wrote a diary where observations and impressions were recorded in an unsystematic and qualitative manner. Besides this, other information (pamphlets, articles from papers, etc.) related to the events were placed in separate archive. Any national television broadcast related to the events were recorded onto videotape. Wherever possible, the observations were checked postobservation with video material, police accounts, and reports in the press or published material from protestors or fans.

Challenges

Similar to most of the participant observation studies mentioned above, I acted as a sole and overt observer, focused on intergroup interaction in context, and recorded my observations contemporaneously, without intervening in the proceedings. In other respects, the methodology differed considerably from all of the participant observation studies mentioned above, especially through:

the rigorous procedures for sampling and recording behavior;

It can be argued that it is impossible to conduct completely “neutral” observations of actions during (potentially) violent collective interactions according to clearly defined categories. The point is, and this is what clearly sets apart direct systematic observation from participant observational studies, to be as explicit and transparent as possible about observational procedures, categories being distinguished, and definitions used. Of course, as Johnson and Sackett (1998) indicate, direct systematic observations have to deal with similar challenges as participant observations, especially issues of access and personal safety, reactivity, and ethical challenges.

As an outside observer making observations in public places, access was not an issue during the research. When needed, access to the police did not present a problem (this was helped by the fact that the study was funded by the Dutch Ministries of Justice and Internal Affairs). Because of the specific focus on collective violence, personal safety was more of an issue during the observations than during most of the participant observation studies described above. Because there was no comparable direct observational study available at the start of the study, the best way to deal with this issue was developed gradually from the perspective that safety was the first priority. At the start of the study, special insurance was taken out at an additional premium. Whenever possible, I familiarized myself with the location where the event was to take place before the start of the observation. I carried a letter from my university explaining the research. During the observations, I made sure in my demeanor that I posed no threat to anyone, and I was open about my activities by not recording secretly and by only recording my own voice and not recording conversations involving others. I did not pretend to be someone I wasn’t. I maintained a high level of awareness for danger throughout the observations and chose my observation positions accordingly, preferably in such a way that I could exit when needed (this was not always possible inside football stadiums). Running from police baton charges or staying out of the way of missile throwing protestors or fans was a regular part of the observations. I was once hit by a police baton (inside a football stadium in a situation where I could not get away) and once arrested (when all football fans in a specific bus I was on were arrested. I was released within an hour). Whenever I felt uncomfortable, I would increase my distance. These measures are similar to those proposed by others for social scientists conducting fieldwork in a variety of dangerous settings (e.g., Bloor, Fincham, and Sampson 2010; Paterson, Gregory, and Thorne 1999; Sluka 1990; Williams et al. 1992). Having said that, in the course of the research, it proved possible to observe violent interactions from up close facilitated by mental readiness and increasing experience to anticipate violence.

During the observations, I moved between protestors/fans or in their immediate surroundings. I did not actively contact them to reduce the risk of investigation effects and to avoid every possibility of being seen as someone who could have exerted an influence on events by passing along information. In approximately half of the cases, I made contact with local police before making an observation to acquire information about the planned police operation, to obtain entrance where necessary/desirable or for reasons of personal safety.

In direct systematic observations, reactivity is less of an issue than in participant observations, and there are other reasons to assume that the methodology had little or no undesirable investigation effects. The observations always concerned very public events that already drew a great deal of interest. There was usually, in addition to those directly involved, a great number of observers present (such as press and onlookers): One observer more did not make much of a difference. In cases that local police was contacted beforehand, this was always done at such a late stage that their planning process had been completed. Having contacted police beforehand did not mean that every officer involved in the operation knew about the investigation; in one case only was I introduced to the complete police personnel. Analyzing the data, there proved to be no difference in the frequency of police use of force or the type and frequency of coercive police measures comparing observations where I had or had not contacted police beforehand.

Compared to participant observations, in the direct systematic observation methodology ethical challenges were less of an issue. Any issues that arose in the course of the study were discussed with the accompanying committee that was chaired by an independent researcher. The funding ministries put no restrictions on the research, and I was not asked to sign a secrecy agreement in relation to the contacts I would have with police. Even more than during participant observation, it is impossible to obtain informed consent from all participants. Ethical guidelines do not require informed consent for these types of observations. Some participants may have thought I was a journalist and others that I was a plainclothes police officer (as I was working alone, this interpretation was less likely: Police officers do not work on their own in a crowd). On occasion, I would be addressed by individual protestors or fans asking what I was doing, to which I responded that I was a researcher studying the way police dealt with them. The data I recorded never included details on individuals (most of whom were unknown to me anyway), and when asked, I made clear I was only recording my own voice. By definition, when observing collective violence, one is witnessing crimes being committed. At the same time, police officers were almost invariably present (as were many other witnesses). I was never asked to act as a formal witness. Except for one occasion where I threw aside a metal bar that could have been used as a dangerous throwing object, I did not intervene in the events I observed in any way.

Advantages of the Methodology in Data Analysis

As far as the accuracy or validity of data was concerned, for protest events in some cases, observational data could be compared afterward with police reports, although police did not keep systematic records related to protest events. The cases where comparisons could be made generally gave no reason to doubt the accuracy of the observational data. If coverage of events in police records, in the media, or in action reports by protestors gave reason to suspect that significant events were not included in the data, observations were excluded from quantitative analysis (this was done in seven cases). With regard to the football observations, observational data could be compared to data gathered by the Dutch National Football Information Point. A comparison of these data with observational data for the19 games with 92 violent incidents observed by me during the first year of the study reveals that the police data on these games contain eight violent incidents (excluding incidents during interurban transport) that were not observed by me. These incidents concerned four incidents among home fans outside the stadium and not in connection with away fans (the sort of cases that by definition is not included in the observations), two incidents between police and home fans out of view of the away fans after the finish of the game (this type of incident can also not be included in the observations), and two occasions where a bus carrying away fans was attacked on the way to or from the stadium. This was not observed by me because of the clustered transport arrangements of away fans: The attack occurred in a cluster that was separate from the one I was traveling with. This meant that just 2 percent of the incidents that should have been included in my observations, were not. In addition, the observational data systematically underestimate the frequency of violence performed by away fans traveling individually, by home fans, or by the police outside stadiums that was not directly connected to the presence of package group traveling away fans. On the other hand, the comparison of observational data with police data on football revealed that the police data systematically underreport the use of force by police and the use of violence by fans inside the stadium (especially smaller incidents were missing from police data). Police reports were routinely described in vague language as “problems,” “irritation,” “a confrontation occurred.” The observations were also checked against newspaper reports (if available, most of the smaller incidents were not reported in the media), and no examples were encountered of incidents reported in media that were not included in the observation of the event concerned.

With the data gathered with the use of the direct observation methodology, several types of questions can be answered in a quantitative and comparative manner in a way that was not possible before. The results of the study have been published elsewhere, here some examples of the types of questions and insights gained are discussed (Adang 2011). For instance, the observations made clear that even in highly escalated situations, the relative number of participants from a specific group taking part in collective violence never exceeded 10% of those present in the group and was usually much less (around 1%). This clearly shows that also in these types of situations, individuals make choices. The riskier the behavior, the less people participated: Less people acted violently than the number of people participating in running charges which was less than the number of people who participated in verbal and nonverbal threats and provocations. This result is a clear confirmation of the absence of unanimity in behavior (the “illusion of unanimity”; Turner 1964). The results also show that, on the whole, targets of violence were not chosen randomly but were related to the situational context: Violent fans directed themselves at rival fan groups, at inanimate objects when these could not be reached, and at police mainly when police took some kind of measure against them. Violent protestors usually targeted inanimate objects that stood in some relation to their protest or police taking measures against them. Around 50% of collective violence around both football matches and protest demonstrations were preceded by an identifiable immediate context. In the other cases, no immediate context could be observed that could have served as a trigger or “flash point” (D. Waddington, Jones, and Critcher 1989). Although police actions sometimes triggered collective violence, contrary to the assertion by Stott and Reicher (1998a), crowd conflict did not characteristically arise when police intervened against unofficial mass action.

Based on an analysis of newspaper records, Martin, McCarthy, and McPhail (2009) indicated that the size of a (protesting) crowd differentially influences the likelihood of violence. They found that as the number of participants increases, the likelihood of violence against private and public property highly increases, while violence against the authorities slightly increases and those against civilians slightly decreases. No such correlation was found in this study. Rather, violence was more likely if groups of football fans with a history of rivalry met (irrespective of the number of fans present) or if masked protestors were present at the start of a protest (irrespective of the number of protestors).

“Bouts” of collective violence lasted on average for around three minutes both in a football and in a protest context. The way police reacted did not—on average—shorten (or lengthen) this duration, so police reactions did not seem to be effective in the short term. The sequential analysis made possible by the data showed that police reactions to (fan or protestor) violence did have an effect in the medium term: The interval to the next bout of collective violence increased after a police intervention.

Conclusion

The direct systematic observation of episodes of collective violence around football matches and protest events in the Netherlands led to a wealth of valuable and unique data. This article does not argue that direct systematic observation is under all circumstances a superior methodology to study collective violence. It goes without saying that for any study, the methodology used should fit the aim of the study and should provide the data that help answer the specific research questions being addressed. Direct systematic observations will not allow to answer certain types of questions, for example, a limitation of the methodology is that it does not address more structural causes or the psychology of collective violence. The methodology instead concentrates on short-term interactional processes. Observations also do not provide information on any behavior-affecting substances (alcohol, drugs) participants may have ingested, and any reader may come up with other questions that cannot be answered using direct systematic observations. Also, there are limits to the level of detail of the observations (although video recordings, which have their own limitations, could be of benefit here). Having said that, direct systematic observations do address the “irreducible need for other, more rigorous observational methods” (Johnson and Sackett 1998) not to replace but to complement interviews and participant observation. Given the limitations of the different forms of participant observation, there is a real need for direct systematic observations to be added to the toolbox to help “establish the phenomena to be explained” (McPhail 1991).

In this way, theory can be advanced, and existing theories may be challenged or developed: The results of the direct systematical observational study led Adang (2011) to combine different theoretical approaches and propose the initiation/escalation model for collective violence, with initiation of collective violence coming about either as a response to specific frictions or triggers in the situation or as the results of young males actively seeking opportunities to start a confrontation. Escalation of collective violence is mediated by the perceived risk of retaliation and an increasing “us versus them” antagonism.

The methodology puts high demands on the observer both as regards the skills needed to record observations in complex and dynamic situations and with regard to the experience to assess correctly where to position oneself or move to be able to be at the right place at the right time. The method will be more difficult to apply (especially for a single observer) in settings where there are very large crowds, where things are happening simultaneously at a number of different locations, where there is no obvious location where violence might erupt, or where there is a clearly identifiable (sub)group that is more likely to become involved in violence (as is often the case in celebratory situations or community riots). Also, when events are unplanned, it will be more difficult for observers to be present. At the same time, it should be clear that to gather meaningful data, it is not necessary to be complete but to use a sampling strategy that is representative. In fact, variations in the methodology have already been used to study public order management in the more complicated setting of football championships in different European countries, where large numbers of fans gather at different locations in host cities. Using the definitions and criteria described in this article, Adang and Cuvelier (2001), Stott et al. (2008), and Schreiber and Adang (2010) trained multiple observers in a simplified methodology. Rather than an all occurrences sampling method, a fixed-interval instantaneous sampling combined with a one-zero sampling method was used that is less demanding on observers (cf. Johnson and Sackett 1998). In addition, ethnographical methods and surveys were used.

The results of the study also led to a number of policy recommendations with regard to the policing of football and protest and public order management in the Netherlands and elsewhere. Adang (2011:65-68) documents some concrete examples of application of the friendly but firm “low-profile” approach resulting from the research, especially but not exclusively in connection with the policing of football (see also Adang 2001; Bassam 2001; Frosdick 2005; Nordqvist 2008; Schreiber and Adang 2010).

The main conclusion to be drawn from the study is that it is practically possible to conduct meaningful direct systematic observations of episodes of collective violence in their natural complexity and context in a reliable way (more complete and more detailed than police records or newspaper reports) without compromising the physical safety of the observer. Even though violence at gatherings is relatively rare, it is also possible to specifically sample events with an increased likelihood for collective violence. The direct systematic observation of collective violence yields data that cannot be obtained by other means (newspaper reports, surveys, interviews, participant observation) that are crucial to an understanding of the initiation and escalation of collective violence.