The Dynamics of the Demobilization of the Protest Campaign in Assam

ABSTRACT

This study highlights the role that critical events play in the demobilization of protest campaigns. Social movement scholars suggest that protest campaigns demobilize as a consequence of polarization within the campaign or the cooptation of the campaign leaders. I offer critical events as an alternative causal mechanism and argue that protest campaigns in ethnically divided societies are particularly combustible, as they have the potential to trigger unintended or unorchestrated communal violence. When such violence occurs, elite strategies change, mass support declines, and the campaign demobilizes. An empirical investigation of the dynamics of the demobilization phase of the anti-foreigner protest campaign in Assam, India, between 1979 and 1985 confirms this argument. A single group analysis is conducted to compare the dynamics of the campaign before and after the communal violence by using time series event data collected from The Indian Express, a national newspaper. The study has wider implications for the literature on collective action, as it illuminates the dynamic and complex nature of protest campaigns.

KEYWORDS:

• Assam
• critical events
• demobilization
• protest campaigns
• time series

Acknowledgments

An earlier version of this article was presented at the International Studies Association Annual Convention in New Orleans in February 2015. The author thanks Courtenay Conrad, Indiana Statistical Consulting Services, Scott Pegg, Karen Rasler, the journal editor, and the anonymous referees for their helpful suggestions.

Appendix

Explanation of coding procedures and the codebook

Coding procedures

I followed multiple steps to construct the codebook and weight the data. First, I identified the four broad event categories that I needed data for: (1) state accommodation, (2) state repression, (3) group collective action, and (4) external support to the group. These categories were followed by the lists of actors involved in the Assamese protest campaign.

For state actions, I relied on a majority of Krain’s (2000) list of repressive and accommodative events.²⁵ Krain includes a variety of repressive state actions, ranging from seizing assets to use of torture, as well as a comprehensive list of accommodative actions ranging from increasing security of target group to incorporation into government. Even though these lists cover a wide range of events, I discovered some additional categories that were essential to my four cases throughout the coding process. Therefore, I added violent arrests of opposition leaders, halt negotiations; curfews, declaration of President’s Rule or Governor’s Rule,²⁶ teargassing, violent disruption of group organization, beating up, clashes with group, damaging property, bombing, burning houses and bridges, and burning entire villages to the repressive events lists. These events came up with some frequency and provided detailed information about the exact form of events.

I then labeled the repressive events based on their objectives and decision-making agencies. Preventive repression corresponds to repressive acts imposed by higher-level authorities, such as the government or the judiciary, to demobilize the contenders and to prevent future acts of contention. The declaration of martial law and the restriction of the right of assembly, for instance, fall into this category of repression. Reactionary repression, on the other hand, stands for repressive acts of lower-level agents, such as the police, applied during the course of a contentious episode in an attempt to respond to collective action and maintain or reestablish order. Teargassing the protesters or damaging property are examples of reactionary repression. I consider arrests as a separate category because it is not always clear if arrests result from centralized policy directives or if they are consequences of ad hoc decisions made by the local police. Finally, I grouped all repressive events under six broader categories: Restrict; Seize; Warn; Judicial Actions; Nonjudicial Actions; and Use of Force.

Regarding the state’s accommodative policies, I expanded Krain’s list by adding organizing discussion meetings or committees and withdrawal of army. The Indian state used these two types of policies during the protest waves and insurgencies as a form of accommodation. Again, I grouped similar types of events under broader categories: Negotiations; Judicial Accommodation; Nonjudicial Accommodation; and Group Recognition. For the purpose of systematic coding, I arbitrarily assigned the broad categories a three-digit code and the specific event types a four-digit code that started with the first two digits of the broader category. All repressive events that involve violence were assigned a 9 at the very end (thus, violent actions are all five-digit codes ending with 9). All the codes for accommodative events start with a 1, and the repressive ones start with a 2. This code assignment was particularly helpful later throughout the data management process because it made it easier to identify similar types of events and their respective broader categories for aggregation purposes. It also helped me to detect miscoding, if any, more easily. The following are examples of categories of accommodation and repression with their respective codes.

For group actions, I used a majority of the Violent International Conflict Data Project (VICDP) events list (Moore and Lindstrom 1996). The VICDP is a data project that has generated conflict and cooperation data at the subnational level. Events from the Integrated Data for Event Analysis Framework (IDEA) (Bond, Bond, Oh, Jenkins and Taylor 2003) was also helpful in identifying specific events that were not included in the VICDP project, such as hijacking and taking hostages. For the purpose of distinguishing between situations in which violence is or is not used, I added the “violent” versions of demonstrations, strikes, and sit-ins to the list. I also created two separate categories for demonstrations based on size. Demonstrations with fewer than 200 participants are coded as small scale, and the ones with more than 200 are coded as large scale. Even though the cutoff point of 200 participants is arbitrary, there is a pragmatic reason behind it. The Indian Express usually reports massive demonstrations as featuring “hundreds” or “thousands” of demonstrators. I am conservatively interpreting “hundreds” as at least 200 and therefore, using it as my cutoff point.

After completing the list for group actions, I grouped those under five broader categories: Accommodative Group Actions, Low-Intensity Collective Action, Medium-Intensity Collective Action, and High-Intensity Collective Action. Again, the broader category was assigned a three-digit code, and the specific events were assigned a four-digit code. All the events that involved violence were assigned a 9 at the end. All collective action events start with a 5.

In addition to collective action, I included a “Splits” category for groups to code information about group cohesion. The list of events in this category does not reflect a wide range of group characteristics because the scope of the study is not on organizational aspects of the groups. Instead, it includes the events that are the results of splits within a group. These specific events are disagreement between factions, forming an organization as a result of split, expelling members, armed attack against other factions, and militants surrendering. These events also have codes starting with a 5 because they are related to groups.

For the list of acts of external support to the group, I built on Heraclides’ (1990:368–369) list of events that are specifically designed to fit cases with external involvement in secessionist conflicts. For example, some of the events in his list, including providing access to communication, providing sanctuary and training, or giving direct military assistance such as armed intervention, constituted the basis of the list for external support to the group. I also added firing across borders and granting asylum to leadership of group to the list to make it more comprehensive. All the events for external support to the group are arranged under two larger categories: Material and Diplomatic/Political Support. Material Support has three subcategories based on the level of involvement: Low, Moderate, and High. The three-digit codes for these main categories start with an 8, and the specific events have five-digit codes, just like the rest of the codebook.

While compiling the list of events for state, group, and external actor actions, I deliberately attempted to include similar policy options across the board. Obviously, the same exact policy options that are available to the state are not available for the group and vice versa. However, if the state has the option of putting down a demonstration, then demonstration is included on the list of group actions, and so is allowing demonstrations in state’s own territory for external support actions.

Finally, for the list of actors and targets, I used the actor/target list in Moore and Lindstrom’s (1996) VICDP as a starting point for my initial deliberations on what kind of actors to include. The specific party names and factions, names of governors, chief ministers, names of religious or ethnic groups along with broader categories of students, peasants, workers, and businessmen, for instance, are listed in the codebook. These specific actors were organized based on seven types of actors and targets: State, Social Actors, Political Parties, Religious Groups/Populations, Insurgent Groups, State Enforcement, and External Actors. All the broader categories have five-digit codes, and the specific actors and targets have six-digit codes. If there is more detailed information available, then those more-specific ones have seven-digit codes. For example, in Assam, the code for the State is 33100, for the Governor 331011, and for Governor Shri T. S. Misra, it is 3310113.

Once the entire list of events was compiled, I assigned scores to each event to reflect the intensity of these events more realistically. It would simply not be accurate to assume that the state’s restriction of assembly has the same impact as launching an armed attack. If both of these events would be assigned a score of 1 (or simply counted by frequency), it might very well be possible that restriction of assembly ends up having more weight in the statistical analysis than launching armed attacks against the group if its frequency is higher. In order to make the analysis more accurate, the weights of those events needs to be taken into consideration.

For assigning weights, my starting point was, once again, Krain’s (2000) scale of repression and accommodation. Krain’s scale is initially based on the rankings of 30 experts in the field. The experts were asked to rank the repressive and accommodative events on a 1 to 5 scale. The average scores for each event were calculated and then rounded to the nearest whole number. To be able to make finer distinctions, Krain took the square of the raw average score and assigned new scores to the events ranging from 1 to 25 (Krain 2000:42–43). So, for example, while arresting opposition leader and armed attack received a score of 3 by the expert rankings, Krain scored them as 7 and 12 respectively. By expanding the range of the weights, Krain gained enough flexibility to rank events that originally fell into the same score category.

As a wider range of scores allows me to rank the events with more precision and thus to see the increases or decreases in the intensity of the actions, I decided to use a range between 1 and 25 as well. I first assigned similar scores to the repressive and accommodative events that Krain had on his list. Then I ranked the events that I had added in the previous step that were not in Krain’s list. For example, while arresting opposition leaders is on Krain’s original list, I included violent arrests of opposition leaders into my list of repressive events. These additions inevitably led to the modification of several scores. The addition of a violent event to the list meant that it had to have a score higher than the nonviolent version of it. So, the score was assigned based on other violent events with similar intensity. As a general principle, the violent events were assigned a 4-point-higher score than the nonviolent versions.²⁷ Once this rule was established, similar events were modified based on the same principle. Finally, I made some adjustments to the scale based on case-specific characteristics. Throughout the data collection process, I read thousands of newspaper articles on the same subject, which gave me a good understanding of the actions of the state and the groups. I was, therefore, able to understand how the state used certain types of policies in what kind of context and used this insight to modify the scale. For example, while declaration of martial law has a score of 10 in Krain’s scale, it is 21 in my scale because the Indian state declared martial law in situations where other repressive measures had already been taken and martial law was seen as the only option left to halt the further deterioration of law and order. Therefore, declaration of martial law was assigned a dramatically higher score than violent or nonviolent putdown of demonstrations. Even with these various modifications, the correlation coefficient for accommodative and repressive events between the new scale and Krain’s scale is relatively strong (.92 and .78 respectively).

I then applied these similar guidelines to assign scores to group actions and external support events. The most severe forms of collective action and external support events were assigned scores closer to 25, and the least severe were assigned scores closer to 1. Actions included as options under both the state actions or group actions received similar scores. For example, armed attack receives a score of 20 both in the list of state’s repressive actions and the group’s most threatening actions list. Similarly, negotiating receives the same score of 6 in both the state’s and group’s conciliatory actions lists.

This systematic and consistent procedure of assigning weights was particularly important in ensuring some degree of balance and objectivity in this otherwise subjective process. Subjectivity is a legitimate concern both with Krain’s scale and also with my scale as, indeed, it is with any other weighting scheme. As there are no objective criteria used to rank such events, any ranking inevitably involves some kind of a judgment call on the researcher’s part. However, the application of the same principles across all types of actions will reduce the bias to a minimum and prevent it from affecting the analysis in a significant way.

Table A1. Codebook.

Descriptive statistics

Table B1. Descriptive Statistics for the Main Variables (Weighted).

Robustness checks

Table C1. ZINB and PAR Estimates for Number of Protest Activity in Assam (Unweighted Data).

Table C2. ZINB and PAR Estimates for Violent and Nonviolent Protest Activity in Assam.

References

Bond, Doug, Joe Bond, Churl Oh, J. Craig Jenkins, and Charles Lewis Taylor. (2003) Integrated Data for Events Analysis (Idea): An Event Typology for Automated Events Data Development. Journal of Peace Research 40(6):733–745.

Heraclides, Alexis. (1990) Secessionist Minorities and External Involvement. International Organization 44(3):341–378.

Krain, Matthew. (2000) Repression and Accommodation in Post-Revolutionary States. New York: St. Martinʼs Press.

Moore, Will H., and Ronny Lindstrom. (1996) The Violent Intranational Conflict Data Project (VICDP) Codebook. Department of Political Science, University of California–Riverside. Available at http://www.academia.edu/2662199/The_Violent_Intranational_Conflict_Data_Project_VICDP_Codebook.

Notes

¹ Communal violence refers to an intense, sudden, though not necessarily wholly unplanned, lethal attack by civilian members of one ethnic group on civilian members of another ethnic group, with the victims chosen because of their group membership (Horowitz 2001:1). The intention to kill and the civilian nature of the perpetrators and the victims are the defining characteristics of communal violence.

² My understanding of protest campaign is similar to that of social movement scholars in that it involves both types of actions: nonviolent ones as well as low-level forms of violence, short of organized military operations. While demonstrations, boycotts, or strikes are nonviolent tactics typically used during protest campaigns, the potential for a violent outbreak exists, as people might throw stones, break windows, or set vehicles on fire.

³ Demobilization refers to the process by which protesting decreases in scale and scope and eventually ends (Tilly and Tarrow 2007). While demobilization typically evokes an unsuccessful outcome, these two concepts are distinct. Although the demobilization of a nonviolent campaign often leads to its failure, demobilization might also precede a successful outcome. For example, the demobilization of the Baltic and Crimean Tatar protests in 1987 and 1988 under coercive measures eventually led to a successful outcome, as they triggered other protests, ending with the fall of the Soviet Union (Beissinger 2002).

⁴ Smaller-scale communal violence in 1980 in Assam is not considered as a critical event as it did not change the overall trajectory of the protest campaign. See Capoccia and Kelemen (2007) for a discussion of identification of critical events.

⁵ This figure is generated by using the count of events.

⁶ The codebook used for this data set is in the appendix.

⁷ Ethnic violence is distinguishable from protest violence as communal violence occurred in rural areas, whereas protest activity typically took place in major cities. Ambiguous cases were excluded.

⁸ Ambiguous events were excluded.</FN8

⁹ Every action of every actor was coded separately. For instance, if students shouted slogans and threw stones, two separate actions were coded. Clashes are coded for both protesters and the police.

¹⁰ Since only one coder was involved, potential issues of intercoder reliability do not arise.

¹¹ The correlation coefficients for all variables between the number of events and the weighted data range from .87 to .98.

¹² Descriptive statistics for the variables can be found in the appendix.

¹³ The variance of protest activity is greater than its mean, indicating overdispersion.

¹⁴ Vuong test statistics indicate that ZINB is preferable to a regular negative binomial.

¹⁵ A transfer function model assessing the impact of intervention on a time series is not suitable, as the data are based on event counts (Brandt and Williams 2001).

¹⁶ The AR parameters in larger lags were insignificant.

¹⁷ The same analyses were conducted on the unweighted data, and the results are similar. Also, the Critical Event Model was run on nonviolent and violent protest activity separately. Overall, the results still confirm the findings of the Critical Event Model. The short-term positive impact of communal violence on violent protests can perhaps be explained due to the radicals’ initial desire to show their continued determination to carry on. The results of robustness checks can be found in the appendix.

¹⁸ Although studies have demonstrated that elections are associated with high levels of violence (Dunning 2011), elections did not have a significant impact on protest activity. Parliamentary elections in December 1979 and the State Assembly elections in February 1983 were the only two elections held in Assam during the time period under study. The 1984 parliamentary elections were postponed and held in December 1985. A dummy variable coded 1 for three months before and after each election was statistically insignificant in the ZINB estimates of the Critical Event Model.

¹⁹ The full range of events are listed under the High Intensity Collective Action category in the Codebook in the appendix.

²⁰ Generating predicted values of observations that are not included in any of the analyses is another option for cross-validation.

²¹ The United Liberation Front of Assam (ULFA), a separatist insurgent group, was established in 1979 but did not actively challenge the state until the late 1980s.

²² I hold back the weeks after June 1983, approximately 33% of the observations, after the communal violence. These estimates can be found in the appendix.

²³ The AIC scores for the Critical Event and Cooptation models are 1427.15 and 1535.25 respectively.

²⁴ The Critical Event plot suggests that the variables present in the model are not sensitive enough to protest activity occurrence to be able to detect those spikes. This might be due to the inability of the model to account for the autoregressive component of the series.

²⁵ I did not include mass killings because some form of military armed attack, burning villages, or bombing already covered this category.

²⁶ Article 356 of the Constitution of India grants the President the right to dissolve or suspend a state legislature and place the state under direct federal rule if he is satisfied that there has been failure of the constitutional machinery in the state.

²⁷ This 4-point difference is enough to account for the difference in intensity, yet does not place the violent and nonviolent versions of the same event at the extreme ends of the scale.