Media Disruption Exacerbates Revolutionary Unrest: Evidence from Mubarak’s Natural Experiment Navid Hassanpour∗ navid.hassanpour@yale.edu August 07, 2011
Abstract Conventional wisdom suggests that lapses in media connectivity - for example, disruption of Internet and cell phone access - have a negative effect on political mobilization. I argue that on the contrary, sudden interruption of mass communication accelerates revolutionary mobilization and proliferates decentralized contention. Using a dynamic threshold model for participation in network collective action I demonstrate that full connectivity in a social network can hinder revolutionary action. I exploit a decision by Mubarak’s regime to disrupt the Internet and mobile communication during the 2011 Egyptian uprising to provide an empirical proof for the hypothesis. A difference-in-difference inference strategy reveals the impact of media disruption on the dispersion of the protests. The evidence is corroborated using historical, anecdotal, and statistical accounts.
Keywords: Revolution, Social Networks, Learning, Media Disruption, Political Violence, Cascade, Egyptian Uprising 2011, Mobilization The author would like to thank Craig Calhoun, Alexandre Debs, Stefan Eich, Stephen Farrell, Stathis Kalyvas, Ellen Lust, Sergio Peca˜ nha, Nicholas Sambanis, Jason Stearns, Sekhar Tatikonda, Elisabeth Wood, and participants in APSA meetings of 2010 and 2011, EITM workshop 2011, and Berlin Summer School in Social Sciences 2011 for their helpful comments on this project. ∗
1
Following three days of unrest and to counter the growing urban protests across Egypt, in early hours of January 28th Mubarak’s regime shut down the Internet and cell phone networks across the country. The surprising events of the next day suggest the incumbent’s tactics were misguided. The protests in Cairo which were contained in Tahrir square and surroundings up to that day, proliferated across the city and flared in every corner of Cairo. By 6 P.M. on January 28th, the police forces were overwhelmed, and the military was called in to replace the police. In the following days a practically neutral military played a major role in the political developments of the country resulting in the ouster of Mubarak on February 11th. The expansion of the protests on the 28th questions common wisdom on the role of social media in civil unrest. The disruption of the media across Egypt at early morning hours of January 28th, proliferated the unrest and exacerbated the decentralized nature of revolutionary contention. In the course of this study I examine the role of media at the time of revolutionary unrest and argue that disrupting social and mobile media, contrary to Mubarak’s intent, fostered more contention of a decentralized nature. Disrupting media is a common characteristic of many revolutionary situations. Sometimes it is a byproduct of the paralyzing unrest; often it is the result of a governmental crackdown. In both cases, I will argue that the disruption acts as a catalyst of the revolutionary process and hastens the disintegration of the status quo. The recent Egyptian uprising provided a unique opportunity to put such a hypothesis to test. The disruption of media prior to major revolutionary upheavals is not limited to the case of the Egyptian Revolution of the 2011, but the evidence is not as clear-cut. On November 6, 1978, in solidarity with the other factions of the Iranian society and in opposition to the Shah’s newly appointed military government, Iranian journalists and newspaper staffers announced an indefinite strike plunging the country into an information blackout for two months till the reopening of the press on January 05, 1979. The largest demonstration of the Iranian revolution of 1979 took place during this news industry hiatus on December 10 and 11, 1978 (Historical New York Times n.d.). The information vacuum was filled with audio-cassettes, pamphlets and other decentralized means of face-to-face communication. Not surprisingly, during Tehran’s post-election protests of 2009, the authorities repeatedly disrupted mobile communications and Internet-based social media, but
2
never imposed a universal blackout. Similarly on the 25th of February 1917, in the midst of an urban revolt, Petrograd newspapers ceased publication just a few days before the Duma’s dissolution. They resumed circulation on the first of March, after which the Duma acted quickly to limit and control the actions of the press (Historical New York Times n.d.). The proliferation of protests happened on the 26th immediately after the disruption of normal communications across the city on the 25th. The violent response to the unrest on the 26th culminated in an army rebellion and the take over of the Duma on the 27th (Hasegawa 1981, Wade 2005). Both of the above examples hint at the disruption of media as a revolutionary catalyst but do not present a fully convincing evidence. Instead the abundance of microlevel information on the Egyptian case provides a reliable test for the hypothesis. In the following I argue that the disruption of media can fuel revolutionary unrest. Using a network stylization I detail situations in which higher connectivity can stall collective action. After proposing a social network formalization, I employ statistical evidence as well as micro-level data on the extent of contention during instances of revolutionary unrest to back the hypothesis. In particular, the Egyptian media disruption of January 28, 2011 is examined in greater detail to provide an account of underlying processes that increase the dispersion of protests.
1
Dissent and the Media
The debate on the role of the media in social unrest was revitalized during the Arab Spring and in the ensuing debates. In the past two decades, the rise of the Internet and decentralized online communication coincided with numerous instances of mass uprisings across the world. Many analysts have taken the social media to be an indispensable part of the mobilization process in mass protests in places as diverse as Ukraine, Iran, Moldova, Thailand, and Egypt. Various arguments have been put forward: some suggest that the new technology makes coordination easier, while others highlight the role the media play in broadcasting scenes of confrontation to the outside world, thereby encouraging the aggrieved population. Monitoring communication in the context of the social media also seems to be more difficult. The proponents of such arguments overlook several facts. Social media can act against grass 3
roots mobilization. They discourage face-to-face communication and mass presence in the streets. Similar to more traditional and highly visible media, they create greater awareness of risks involved in protests, which in turn can discourage people from taking part in demonstrations. In the following I will argue that lack of credible information at times benefits cascades of contention.1 Kern and Hainmueller (2009) cite similar processes as an explanation for why watching West German television broadcasts might have discouraged East Germans from applying for visas to travel to the West: once they saw it on television, they were less likely to embark on a personal exploration to see the unknown. Similarly knowing about the situation on Facebook and Twitter and having access to news propagation sources may make personal moves and physical presence unnecessary. The lack of the intermediary sources of communication between the state and the people fosters local news production and propagation on the individual level, deprives the state of a normalizing apparatus, and sets the stage for cascades of collective action. I would like to argue that the paralyzing mass demonstrations and widespread antagonistic uprisings in question would not have happened if the media had continued channeling their supervised, censored, and perhaps realistic narration of the events. In the absence of the mass media, information is communicated locally. Without state intervention, crowds shape an idea of risk that is independent from the government, testing their perceptions by staging demonstrations. The government’s response to the acts of public defiance signals either weakness or strength on the side of the state (Chwe 2001, Kuran 1991). If the demonstrators’ speculations about the weakness of the incumbent regime turn to be correct as it did in Tehran in December 1978 or Leipzig in 1989-90 (Lohmann 1994), the cascade of events can grow to unanticipated dimensions.2 1
In fact a number of mass uprisings in the Eastern Block were initiated by rumors. See section (1.1). On the difference between revolutions and riots: in this study revolution is defined as a mass violent act targeted at the governing body, intended to topple the incumbent regime. Mass uprisings under the title revolutions show distinctive common traits: they are large in scale (thousands involved if not millions); their major aim is to dismantle the political status quo and the ruling apparatus; the new rulers (in the case of successful execution) would be the prior underclass; the ruling elite would be confiscated from their political and economic power; and finally successful revolutions bring vast and far reaching changes in legal and judicial practice. Defined as such, revolutions are rare events. They are different from riots in several aspects. First, they affect lives of a sizable population inside the domestic polity, second the stakes are higher. Participants face higher risks, their ultimate goal being a stand off against the ancien r´egime. Both of these characteristics are in contrast with the defining 2
4
1.1
Revolutions and Misinformation
The Weberian definition of the state (Weber 1958) anticipates such destabilizing moments. The state is defined as the monopoly over physical force and bureaucracy. In addition to the military and police, the press act as a proxy for the state’s control bureaucracy, hence any interruption of this industry would alter the functionality of the state. Weber’s definition is also in line with the transformation of the press after the French Revolution. While decentralized pamphleteering was a common practice among the revolutionaries, the state they created moved to standardize and regulate the process. According to Schumpeter (1950) these notions extend to democracies and dictatorships alike. In a well functioning democracy the media are used to shape electoral opinion. Likewise in the Gramsci’s depiction of totalitarianism (Gramsci 1971), the media impose an aura of normalcy and oppressive calm under the cultural hegemony of the state. What is left out from Schumpeter and Gramsci’s accounts are the brief moments where the outreach of the media does not exist or is interrupted. When the normalizing force of the media collapses, production of opinion outside the reach of the incumbent regime can force the polity to change course under the pressure of an opposing public sphere. In a society on the verge of political unrest, the state’s control over news media prevents widespread dissatisfaction from turning into a united opposition. Media outlets are highly visible and not hard to control. The population that relies on the media for estimating the political atmosphere is provided with a view that is supervised by the ruling power. The elite use their influence to pacify the population or discourage sedition. Nevertheless, the widely acknowledged view of the role of the media in revolutions points at the opposite direction. Popkin (1995) among others, notes the positive impacts of the free media on bringing about revolutions. According to him print media disseminate knowledge and awareness. The constituents are informed about political possibilities and grievances; therefore they are more inclined to engage in a resurrection against oppression and incompetence. This argument is misguided on two grounds, first it overlooks the fact that most seditious communication is invisible to the ruling elite. If they were aware of it, factors of riots (Wilkinson 2009) as smaller gatherings, which although can be violent, are not of the magnitude of revolutionary movements and are not directed toward the demise of the principal political hegemony.
5
they would disrupt it. The centralized media, including semi-autonomous dailies, are too exposed to foster revolutionary violence. Second, those who engage in radical acts of dissidence usually are not primed by potential free discussions in the press. The free media excites the intelligentsia and is more likely to result in a political transition of more usual types. What incites mass revolts is “misinformation� properly defined.3 To confirm such hypotheses and to add to the descriptive and speculative statements on the positive or negative role of the media on mass protests, one needs to find situations in which social media coverage changes sharply, then gauge the impact of such a change on the level of unrest. In fact authoritarian regimes repeatedly provide such a scenario: in the course of street protests, mobile communications are often disrupted and internet access is restricted. However ideally the disruption should be ubiquitous and universal and the level of confounding factors kept to the minimum for the conclusions to be meaningful. In the following I provide a stylization using a dynamic social network model and present a recent example of media disruption in Cairo as evidence.4
1.2
Dynamic Models of Network Collective Action and Media Influence
Media disruption drastically changes the way information is transmitted in society. In response to such disruptions new links for imitation and deliberation are set up and new spheres of influence are created. In the following I propose a stylization of such dynamics. First I outline a dynamic and structural version of a well known threshold model of collective action proposed by Granovetter (1978) and Schelling (1978) and later expanded by Kuran (1989), Lohmann (1994), Gould (1993), and Siegel (2009) among others. According to the threshold model, individuals have different 3
Consider the case of the Czech Velvet Revolution. According to the recent accounts, the movement was ignited by false rumors of the brutal death of a 19-year-old university student, see http://nyti.ms/2tioTq. At the times of civil unrest, exaggeration tactics are known to be highly effective. In a similar manner the fall of the Berlin Wall started with a false and ambiguous statement at a news conference, see http://wapo.st/4wXDkC. A vaguely communicated decision on the Eastern German television prompted protesters to demand free passage to the Western side of Berlin. 4 The contingencies involved in a revolutionary event ask for very detailed accounts (Sewell 2005). Later in the empirics section I use journalistic reports to cope with such nuances in event research.
6
risk-taking habits represented with a personal participation threshold. If the percentage of one’s network neighbors engaging in action exceeds one’s threshold, she switches from inaction to action. Granovetter’s model can be improved upon by adding two components, first a model of social structure and second a model of threshold dynamics. Why structure? Taking the overall level of participation to be fully visible to all society members is an unrealistic assumption. An individual’s perceived levels of participation can be quite myopic. In fact a major disruption of media and mobile communications does exactly that: it reduces a globally connected network relying on a backbone of information propagating nodes to a multitude of smaller local networks barely connected to each other. These local micronetworks are strongly influenced by patterns of interpersonal links and spatial confines. Because of the disproportionate size of the core of mobilization, structural patterns also represent the relations between protest leaders and the rest of the population. In the context of mass demonstrations often one needs to imitate, observe, and update beliefs based on the neighbors’ acts in a local network. Such modes of behavior based on limited information can result in a fast-paced contagion of political participation. Why dynamics? In addition to action, personal thresholds are also in flux. It is plausible to think that (1) while interacting with others one would be influenced by his network neighbors’ beliefs (2) when participating becomes prevalent, one’s threshold could decrease; or as inaction becomes the norm, thresholds also increase. Kuran (1989) and Lohmann (1994) propose dynamics, while Gould (1993) and Siegel (2009) combine particular dynamic models with structural models. Hence the distinctive characteristics of mass collective action are products of two factors, first the structure of the social network and the characteristics of network relations underlying political action, and second dynamics of inter-personal learning, imitation, and influence. The following section contains a model for formalizing both components.
2
The Model
In this section I propose a formalization of the Granovetter threshold model for participation in collective action in networks, which takes both the network structure and belief updating into 7
account.5 In order to make verifiable predictions, I outline a graph theoretical model for threshold updating using DeGroot learning (see (Jackson 2008)). I demonstrate that full connectivity in a social network sometimes can hinder collective action. Later I will show that with some assumptions on the structure of the social network, repeated threshold updating takes the network to an equilibrium on the network graph; hence, the updating procedure acts as an equilibrium selection mechanism based on network parameters and initial participation thresholds. When these assumptions do not hold, cycles of participation and disengagement can occur. Unlike the Granovetter/Kuran model, this formalization predicts non-monotone participation levels and heterogeneous outcomes at the final equilibrium, where some individuals act and some do not. Hence, it provides a more realistic model of mobilization dynamics, which can explain the ebb and flow in large-scale political demonstrations. Later I examine network transformation as a result of media disruption and hypothesize an increase in protest dispersion resulting from such a transformation. The empirics in the next section confirm the plausibility of the model.
2.1
Formalization
Each individual decides to either join a collective act of dissent or to stay put based on a personal threshold and the percentage of his acquaintances who have already joined in. If the percentage is above that threshold, he would join in, otherwise he would not act. The model is based on two parameters, the personal thresholds of each agent, and the social network structure which dictates the details of interpersonal influence. There are radicals with very small thresholds whose acts start the process. The network is represented by a graph G(I, E), where I is the set of all nodes in the network i = 1, . . . , n, and E is the set of all edges connecting these nodes. Each node represents an agent, and each link is a social connection. Edges can be directed, i.e. some can not see others acting, while they can be seen by others.6 Each of the agents is deciding between taking (A) or not taking (N) action–this is a binary choice between N and A. The decision is made based on the proportion of the network neighbors 5 6
Those not interested in technical details may skim through Notes 1 to 4 and skip to section (2.4). There is always a self-loop, because everybody is aware of what he himself does, or what threshold one has.
8
who are acting, according to the following rules. Take pi (t) to be the proportion of i’s neighbors acting at time t = 1, . . . , T (self included), and Γi (t) to be the i’s threshold at time t. At each time t, i acts if pi (t) ≥ Γi (t), and does not act otherwise. This defines a game in which each agent, based on her threshold, has to choose between A and N.
2.2
Threshold Game’s Equilibria in Heterogeneous Networks
An equilibrium in this network game is defined similar to conventional games. Each agent should not have an incentive to deviate. There can be more than one network equilibrium. The case of networks with agents with equal thresholds (a homogeneous network) was studied by Morris (2000), (see (Jackson 2008) for a short summary). Consider the following homogeneous example. Note that I have not included self-loops in these figures, but it is implicit in the model. N
N
∀τ
N
N
N
N
A
A
1/3 ≤ τ < 1/2
A
A 1/2 ≤ τ < 2/3
A
∀τ
A
Figure 1: Equilibria for a network game, all agents have a common threshold τ
All players share a common threshold τ . Contingent upon τ , the game in figure (1) can have multiple equilibria. Note that for any value of 0 < τ < 1, all nodes acting (A, A, A), and none of them acting (N, N, N) are two equilibria of the game. There exist two other asymmetrical ones as well. For example when 1/2 ≤ τ < 2/3, there is another equilibrium (N,A,A), the third configuration from left in figure (1). The central player is in equilibrium because τ < 2/3 (A), the peripheral ones are as well, because for the first player τ ≥ 1/2 (N), and for the third τ < 1 (A). Homogeneity assumption is limiting. For instance, it is clear that the relation between media, the state, and citizens demands a threshold model with at least three classes of thresholds. Before modeling a tripartite case, consider the following heterogeneous example. The central actor has a negligible but larger than zero threshold ("). In other words she is a radical, while the two other 9
players have identical thresholds τ , i.e. the threshold triplet is (τ, ", τ ). In this case the equilibria of the game are different from the previous case. N
N
N
N
∀τ
A
N
A
A
τ > 1/2
A
N
A
τ > 1/2
∀τ
A
Figure 2: Equilibria for a network game, agents have thresholds (τ, ", τ )
Here when τ < 1/2, there are only two equilibria (N, N, N) and (A, A, A). When τ > 1/2, there are four, (N, N, N), (A, A, A), (N, A, A), and (N, A, N), see figure (2). Chwe (1999) also studied games with a similar threshold strategy. In addition to Chwe’s analysis, Gould (1993) and Siegel (2009) consider dynamic versions of similar network games. In (Hassanpour 2010a,b), the DeGroot learning scheme (Jackson 2008) is superimposed on the threshold model. Again the learning dynamics acts as an equilibrium selection mechanism. The network game is played repeatedly, and the thresholds are updated at each iteration. In this model the asymptotic values of the thresholds and the agents’ final choices between A (action) and N (non-action) are of interest. In the following subsection I outline a dynamic model to explain transitions between two distinct equilibria.
2.3
A Dynamic Model
One expects that at each action period, agents revise their action thresholds based on the history of previous actions. If the majority of their neighbors in the network are eagerly active in collective action and have low participation thresholds, the agents update their thresholds accordingly and will be more prone to participation in the next round. In the case of inaction, the same mechanism is at work. Lohmann (1994) takes threshold updates to be Bayesian. I instead adopt an averaging mechanism based on the DeGroot model. Political actors need to approximate the situation and quickly extrapolate about future. It is plausible to assume that they simply adopt a weighted 10
average of their own thresholds with their neighbors’. The weights are proportional to the level of interpersonal influences. The averaging weights between two agents is not necessarily symmetric. i can take j’s recommendation very seriously, while the opposite might not be true. An individual may closely watch the acts of an opinion leader, while the leader does not care as much about a follower’s beliefs. The weights individual i assigns to person j are taken to be δij s. I normalize ! the δs so that j δij = 1. Note that one’s neighbors’ thresholds are not always fully known, and are hard to exchange in the course of fast paced mobilization. Hence, actors have to infer the real value of thresholds from the acts of each of their neighbors. For example they can take i s threshold to be the the proportion of the times she has failed to act. Or if keeping a detailed history is implausible, one can make a coarse approximation and take i’s threshold to be 1 if i did not act in the previous round, and 0 if she did. In the following, I examine two updating mechanisms. One is based on averaging neighbors’ thresholds, and the other infers a neighbor’s threshold from his act in the previous round. I show that these two dynamics result in quite different asymptotic outcomes. 2.3.1
First Model, Full Threshold Knowledge
First type of dynamics is to replace one’s threshold with a weighted average of one’s own and neighbors’ thresholds; and act at each time t according to the threshold and the level of activism in one’s neighborhood. Because of the continuous nature of political action, I take this process to be repeated multiple times. Ideally the objective is to use the model to find asymptotic acts and thresholds of each individual in the network. Dynamics of Thresholds: At time t, i updates his threshold to be a linear combination of his neighbors’ thresholds and his own. Define Matrix ∆n×n = [δij ] equal to the weight that i gives to neighbor j’s threshold. Take Γn×1 (t) to be the vector of thresholds for individuals 1 to n at time t.
11
Γ(t) = ∆Γ(t − 1) " Γi (t) = δij Γj (t − 1) j:ij∈G
Dynamics of Action: take An×1 (t) as the vector of individuals’ action. 1 implies action, and 0 non-action. At each time t, individuals either join in collective action or refrain. The perceived level of participation for individual i, pi (t), can be a product of her personal network, or a number known to everybody and the same for all, pi (t) = p(t), ∀i. The decision to act or not act is made based on the comparison of pi (t) and Γi (t). At time t, person i acts if pi (t − 1) ≥ Γi (t) and would not act if pi (t − 1) < Γi (t). For the purpose of analysis in this paper I assume that pi (t) is the percentage of i’s neighbors acting at time t. Note that there could be various ways of modeling pi (t). For example we could assume a universally accepted p(t) or perceived participation levels, p˜i (t), that are different from real pi (t)s. We are interested in the dynamics of both Γ and A, specifically their asymptotic behavior when t→∞ Note that in this model, the action vector A is a derivative of the threshold vector Γ. Define
D ∼ Dij = ˙
1 if ij ∈ E, deg(i ∈ I) + 1
Adding +1 for the self-loop. D is fixed for all t,
A(t) = sgn(p(t − 1) − Γ(t)) = sgn(DA(t − 1) − Γ(t)).
sgn(t) is the sign function, sgn(t)=0 if t < 0, sgn(t)=1 if t ≥ 0. Therefore these two equations
12
together characterize the Markov dynamics of this system,
Γ(t) = ∆Γ(t − 1)
(1)
A(t) = sgn(DA(t − 1) − Γ(t)).
(2)
In the following examples we take ∆ = D, hence
Γ(t) = DΓ(t − 1)
(3)
A(t) = sgn(DA(t − 1) − Γ(t))
(4)
Initial conditions: Γi (0) is given. Ai (0) = 1 if Γi (0) = 0, otherwise Ai (0) = 0. Example: Consider the following two networks in figure (3). In this case, there is one radical agent with initial threshold 0 and two normal individuals with identical thresholds 0 < τ < 1. At each time unit, players update their thresholds and decide to act or not. The relations are symmetric. Which means δij = 1/(degree of i + 1). The initial thresholds and the configurations are as shown below. (τ, 2/3τ)
(0, 2/3τ)
(τ, 0.57τ )
(τ, 2/3τ)
(0, 0.57τ)
(τ, 0.57τ )
Figure 3: (initial threshold, steady state thresholds), full connectivity is not always helpful
Note that for the fully connected graph, threshold updating gives (2/3τ, 2/3τ, 2/3τ ) for time t = 1 onwards. For action set, if τ < 1/2 the final action profile will be (A, A, A) for t = 1 onwards; otherwise it is (N, N, N) for t = 1 onwards. For the other network, the dynamics are not trivial. Applying the dynamic model in equation (1) gives the following progression in table (1), The asymptotic thresholds for both networks are reflected in figure (3). Note that the fully connected graph has a larger final threshold compared to the other one, i.e. full connectedness is 13
Thresholds τ (1) = (2τ /3, τ /2, τ /2) τ (2) = (5τ /9, 7τ /12, 7τ /12) τ (3) = (31τ /54, 41τ /72, 41τ /72) .. .
Actions A(1) = (A, N, N) A(2) = (A, A, A) if τ < 1/2; o.w. = (N, A, A) A(3) = (A, A, A) if τ < 6/7; o.w. = (A, N, N) .. .
τ (∞) = (0.57τ, 0.57τ, 0.57τ )
A(∞) = (A, A, A) if τ < 0.875; o.w. = (N, N, N)
Table 1: Dynamics of the star network in figure (3)
not always helpful. Note 1: Connectivity does not always help collective action. It is usually assumed that full connectivity among participants in collective action is beneficial to the act of mobilization. Now consider cases where there are few radicals who aim at recruiting ordinary individuals. Further connections among ordinary individuals can foster inaction. The above example in figure (3) shows that establishing separation among ordinary individuals can effectively help mobilization. This is inline with similar observations in (Gould 1993) and (Siegel 2009) and provides an intuitive explanation for why disrupting media and mobile communications can assist mobilization instead of impeding it. Removing regular communication channels provides radicals with more effective venues for organization and encourages citizens to frequently engage in face-to-face communications. This all weakens the incumbent’s control and provides more opportunities for grass roots mobilization. Asymptotic Thresholds: Consider the thresholds at time t, Γ(t) = ∆Γ(t − 1) = ∆t Γ(0), with the condition that ∃i, Γi (0) = 0. If the network’s graph G is aperiodic and irreducible,7 the steady state thresholds will be the same for all the individuals in the network and is equal to v where v is the normalized8 unit left eigenvector of the matrix ∆ (the left eigenvector with eigenvalue equal to 1) and the final threshold for each individual is (Jackson 2008)
Γi (∞) = v T .Γ(0). 7
A graph is aperiodic if the greatest common divisor of all of its cycles is 1. This is true of all graphs discussed in this paper, because of the self-cycle (each individual counts her own threshold in her averaging). A graph is irreducible if there is a path from each node to any other node. Again it is true of all of the graphs in this study unless it is stated otherwise. 8 Such that the final vector is stochastic i.e. its elements add to one.
14
Further derivations on Γi (∞) in specific network configurations can be found in the appendix, section (6.1). Note that while the threshold perceptions are changing indefinitely (although converging), the actions might reach the steady state and remain the same. Here we have the private preferences changing while the actions remain the same. 2.3.2
Dynamics, Second Model, Only Action Knowledge
Dynamics: Consider a case where there is not enough information about personal thresholds of one’s neighbors. It is usually the case that the agent has to infer the neighbors’ thresholds based on their actions. Take the coarse estimation of a neighbor j’s threshold at time t to be 1 if j does not act at time t − 1, and 0 if he does act. To update the threshold one takes an average of his own threshold and an estimation of his neighbors’ thresholds based on their prior acts. The dynamics of this new updating mechanism is
Γi (t) = Dii .Γi (t − 1) +
"
Dij .(1 − Aj (t − 1))
(5)
j#=i
A(t) = sgn(DA(t − 1) − Γ(t))
(6)
The initial conditions are the same as the previous case in section (2.3.1). The section (6.2) in the appendix contains the details of the asymptotic thresholds and actions for star and fully connected networks. There are a number of noteworthy points in section (6.2). First, the dynamics based on inference from actions results in oscillations, even in conventional topologies such as star networks. The periphery and the central agent switch between action and inaction. In the case of threshold based updating, such oscillations do not happen. Note 2: Incomplete information brings about oscillation. Dynamics are not always monotone. Protests and mass political acts ebb and flow through time.9 An exact or approximate knowledge of thresholds is usually unavailable. Instead, the actions of one’s neighbors can give clues on the their proclivity for participating in mobilization. One expects that inference based on 9
Kuran (1989)’s model predicts a monotonically increasing or decreasing level of participation.
15
only previous actions (not the thresholds themselves) slow down convergence toward a final steady state. In other words, oscillations are more likely when updating is based on speculations instead of accurate knowledge. For example Ermakoff (2008)10 observes series of vacillations among French parliamentarians in the process of voting to establish the Vichy government in July 1940. He takes these oscillations to be a product of insufficient communication among the members of the French parliament. The above stylized examples qualitatively confirm Ermakoff’s speculations. Finally an observation on the size of critical mass (CM) in the 3-star and the triangle networks in figure (3). Note that for any 1/2 < τ < 0.875, the CM for the 3-star network is 1, i.e. one radical is enough to incite the whole network to act, while the CM for the fully connected network is larger than 1 because the asymptotic result of the dynamic is non-action for all.11 Note 3: Size of critical mass is contingent upon network structure. The importance of a body of radical actors who unconditionally engage in mobilization is well known (Marwell and Oliver 1993). A network model can predict the smallest size of such a group needed for engaging the whole population. Sudden changes in network structure, e.g. through media disruption, can change critical mass needed for universal participation. While before disruption radicals were not able to incite a rebellion, their influence grows in a highly connected news propagation/mobilization network on the local level. 2.3.3
Finding Steady State Thresholds and Actions
In the above equations (1), (2), (5), and (6) analytical expressions for the steady state vectors A and Γ sometimes can be found by putting A(t) = A(t − 1) and Γ(t) = Γ(t − 1). For example one can find the steady state Γ from equation (1) and substitute it in (2). Then one could simply try all of the 2n possibilities for A and find the ones that satisfy the identity A(t) = A(t − 1). Because the size of the possible As is well bounded, finding the equilibria is feasible. Same techniques can be applied to the equations (5), and (6). In the case of dynamics in (5) and (6) such trials are particularly helpful because both equations are non-linear and finding closed-form solutions is not 10
Ruling Oneself Out Ch.9 In the appendix (6.2) I show that in the fully connected network, the size of critical mass (CM) is at least half of the actors. In any network it is possible to find the minimum size of critical mass needed for inciting global action. 11
16
easy. Also note that any steady state is an equilibrium of the network game (excluding the cases leading to oscillation). If that is not the case, in the next period, all players choose actions that are in equilibrium, hence it is impossible to have a non-equilibrium outcome as the steady state. As it was mentioned before, the dynamics give us an equilibrium selection mechanism. At the same time, one can assume a particular action equilibrium A, and find δs that result in that equilibrium. Hence one could design a network to achieve a desired action equilibrium.12 Note 4: A dynamic model proposes an equilibrium selection mechanism. It is plausible to assume that there are multiple equilibria for the network game defined based on the threshold mechanism. Network dynamics act as an equilibrium selection mechanism. Along the same lines as the above mentioned notes, there are four ways through which a cataclysmic event such as disrupting connections across a society can change the levels of mobilization: through changing the patterns of news production and learning among individuals, through changing the amount of information one can acquire about the acts and intentions of one’s social network neighbors, through changing the size of critical mass on the local level, and finally through changing the asymptotics by altering the dynamics.
2.4
State, Media, and Citizens, A Network Threshold Game
In section (1), I argued that media have a pacifying role in the societies facing mass political turmoil and their sudden disruption can aggravate the situation. Before continuing with historical anecdotes and statistical evidence, I offer an explanation using the threshold network game described in the previous subsection. Consider a network model with three classes of actors, modeled in the context of a tree network. The sole central player is the “state,” the nodes connected to the state represent the media visible to the state and under its influence. Each of the media nodes provides a number of individuals with news and information on the revolutionary situation. See the topology in figure (4). Links here represent functional and behavioral influence. For example media are influenced by the state 12
Hence we can perform a structural mechanism design. Knowing the final outcome, we can find a network construct that induces the desired equilibrium.
17
policy, hence the links from the state to the media nodes. Individualsâ&#x20AC;&#x2122; level of risk taking is influenced by the content of the media, hence the links from the media to the individuals. Note that influence is not synonymous with control. A change in strategy on a Facebook page due to information security concerns amounts to a link between the incumbent and the social media, although the state does not dictate the pageâ&#x20AC;&#x2122;s content. Similarly the existence of the page provides citizens connected to it with a forum in lieu of face-to-face communication. Social media nodes that serve as news propagation forums often regulate news aggregation and oversee news distribution among the online population. More traditional venues such as printed press serve the same purpose but are not as affected by their audience as the novel social media. For the purpose of this study I examine two separate cases in turn. One is the traditional media such as the daily print press, where there is no established mechanism for channeling the opinion of the audience in real time. The second category contains new social media. Ideally they represent an aggregation of their audienceâ&#x20AC;&#x2122;s attitude toward risk besides serving as an information source for the users. The distinction between two media classes is not complete. There are cases in between, e.g. independent newspapers or a government sponsored blogosphere.
Figure 4: The state, centralized media and citizens, represented as a tree graph
18
Figure 5: The state and citizens in the absence of media, multiple local networks-each highly connected
Case One: Traditional Media Stylization: Individuals engage in binary acts of dissent. As a trivial assumption, the threshold of the state as the central node Γs = 1. The media is taken to have a threshold of Γm which is close to 1 and is larger than ordinary citizens’ threshold of Γc . Media threshold is not related to action, but is an indicator of the media node’s propensity toward the uprising. The smaller Γm is, the more disparate from the state the media node is. Γm ’s being close to 1 is a plausible assumption because during a revolutionary situation the traditional media are under direct supervision of the state. Often they act as the regime’s propaganda machine, and citizens discount their reports. Among the citizens at each locale there are a few radicals with zero thresholds. This means they would act irrespective of what others do. The government monitors media’s operations. The media on the other hand act as intermediaries between the state and its constituents.13 In the situation depicted in figure (4), citizens’ main source of news are the established media. Because of their ubiquity and their strong distribution network, the media-individual connections are much 13
Assume there are k citizens connected to each media source and there are m media sources, i.e. there are mk citizens evenly distributed among the news sources. Also Consider the following conditions over the thresholds of the media Γm and of the citizens Γc , 1 > Γm > (k + 1)/(k + 2) > k/(k + 1) > Γc .
19
stronger than the relations between the citizens. Hence comparatively speaking, interpersonal links are non-existent. In such a situation, the state and the media’s allegiances are static while the citizens’ thresholds can change. Ordinary citizens do not act, because all what they see are controlled sources of information. In the absence of connections with radicals, the majority of the population is not mobilized. Because of their isolation from the others, radicals’ action does not amount to much. At a stalemate, citizens learn from static media nodes and become more risk averse. The result would be a non-action equilibrium with a slight nuisance from the radical elements. After multiple updates, the thresholds tend to converge to the media’s risk aversion levels. Case Two: The Social Media Decentralized media, e.g. Internet forums for communication, influence citizens’ inclination to risk in ways different from traditional media. Controlling news propagation on a grass roots level is much more difficult than policing newspapers, television, and radio broadcast. The political inclinations of the social media are often in line with the population majority, not the incumbent. To stylize the situation I take the configuration of the network to be the same as in figure (4), but in this case the threshold of the media nodes is the average of the thresholds of their neighboring citizens. Assume that the social media represent the average threshold of their correspondents, Γm = (
!
i
Γmi )/N. In line with the dynamics presented in equation (5), take action and thresholds to
be reversely related Ai = 1 − Γi . In the stylization, each individual sees the average threshold at the social media node and decides to act if Γi < 1 − media threshold.14 Now according to this rule, at most half plus one individuals engage in action, while according to the threshold dynamics in equation (1), all of the thresholds are converging to the average threshold represented by the social media node. Note the difference between this scenario and a fully connected network of individuals. Here the social media node provides everybody with a view representing only the “aggregate” threshold. “Individual” thresholds are not visible to other members of the network. 14
The calculation behind this decision rule is straight forward: to act Γi < ! 1 − j Γj /N = 1 − average threshold.
20
!
j
Aj /N =
!
j (1
− Γj )/N =
Disruption of the Media, Stylization Equilibrium Now consider the situation in figure (5) in which the media nodes are not present. In this case, citizens have to rely on each other for gaining information about the political and social atmosphere. The incumbent regime is deprived of its propaganda tools; furthermore, it can not exert influence by supervising newspapers or manipulating the social media. In such a situation, citizens are influenced by their peers including their radical neighbors in the network. Dismantling preexisting links among the public and between the media and the population encourages building new connections. Individuals have to engage in exchanges with their immediate neighbors in order to hear the news, or to estimate the prospects of contention. Extreme conditions-brought about by disrupting mobile communications as well as the Internet-incite physical presence instead of online activity (more on the underlying processes in the next section), and produce new connections on the local level. In the transformed network depicted in figure (5), highly connected cells of contention start to take hold in different locations, increasing the dispersion of the protests and proliferating communal activity throughout the society. Unlike the situation in the fully connected tree network, action equilibria are possible.15 Dynamics: The static model does not outline the path to widespread participation. For modeling the transient part of the process, one could implement a dynamic model of threshold updating and action in a heterogeneous network. For examples during the first round, the instigators act; hence they motivate the rest with low thresholds, and this process is repeated. In each repetition the rebellion spreads more widely. This was impossible in the case of figure (4), because each agent was influenced by the media, hence the action of the radicals could not incite universal dissent. Each fully connected local network presents a dynamic situation similar to the scenario initially proposed by Granovetter (1978): acts of each are visible to all. Radicals exert influence and become opinion leaders, cascades become a possibility. In other words, a disruptive action meant to stop the rebellion turns to a catalyst for it. 15
Note the configuration in figure (5). In this set up, again let’s assume that there are k citizens per each subgroup, some of which are radicals with threshold 0, and all others have a threshold of Γc . Take r to be the minimum integer z for which z/(k + 1) is larger than Γc , r = min{z|z/(k + 1) > Γc }. In this case for all s, k ≥ s ≥ r, “s citizens acting” is an equilibrium in the network threshold game.
21
Dispersion hypothesis: According to the above stylization disrupting the media and mobile communications promotes local mobilization, increasing the dispersion of protests. In the following section I exploit reports from the Egyptian uprising of 2011 to test the dispersion hypothesis.
3
Media Disruption and Dissent, Empirical Evidence
In this section, I employ multiple methods to confirm the dispersion hypotheses. I start with a statistical analysis of revolutionary unrest in relation to media penetration followed by some suggestive archival evidence. Then I examine the case of the Egyptian uprising of 2011 in greater detail. In late January 2011, in response to demonstrations, Mubarakâ&#x20AC;&#x2122;s regime disrupted communication all together for a few days providing a unique opportunity for studying the role of media disruption in fostering unrest. The main criterion for testing the theory is the existence of a visible increase in the dispersion of protests after media interruption.
3.1
Statistical Treatment
The existing statistical studies of mass political violence do not directly address the link between media and dissent. For example the extensive study of Hibbs (1973) does not consider the media component among other economic indices. One way of linking media influence to the level of revolutionary activity is to compare the frequency of revolutionary unrest to the extent of the reach of centralized media using available country-year data. An index of media influence can be the number of newspaper copies per capita or prevalence of radio or television use in a country. According to the above argument I expect that the number of revolutionary resurrections to be a decreasing function of media penetration. I use country-year data from Banks (2009) to examine a potential inverse relation between media penetration and revolutionary unrest. Consider the following plots.16 16
Revolutions according to Banks (2009) are â&#x20AC;&#x153;any illegal or forced change in the top government elite, any attempt at such a change, or any successful or unsuccessful armed rebellion whose aim is independence from the central government.â&#x20AC;? This definition includes more than so called social revolutions and emphasizes the violent nature of political takeover. Nevertheless it contains most of what constitutes a social revolution: its grass-roots
22
Revolution count vs Newspaper copies per capita (Banks) 9 8 7 6 5 4 3 2 1 0
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
Figure 6: Domestic mass political violence for state takeover versus newspaper copies per capita
Now consider the following negative binomial regression (table (2)) of the number of revolutions over the number of radios, television sets, and newspaper copies per capita controlling for GDP per capita and a measure of democracy at each country-year point (see table (3) for more information on the range of the parameters and scaling data).17 As it was expected from the plots, the regression multipliers for printed media influence (newspaper copies per capita) are negative and strongly significant. Even after controlling for a democracy index and GDP per capita still the pacifying role of media penetration is evident and accounts for around at least 20 (and up to 40) percent of the dependent variable count (the average revolution count is 0.185, see table (3)). Low levels of civil disobedience is often linked to high levels of political and economic development. Controlling for GDP and a democratic index deals with the very same concern.18 nature, its large scale, and its aim at toppling the incumbent regime. 17 The democracy index is effectiveness of legislature according to the Banks data set. 18 Huntington (1968) took mass violent takeovers of the polity to be the result of a gap between social and economic advancements and political modernization. Such an explanation can be improved upon because it does not clearly specify what the elements of political modernization are. According to Huntington, political development is a source of stability, and political institutionalization is an antidote to major reversals of power such as revolutions. While
23
(1) (2) Radios per Capita (10 ) −1.553e − 05 (2.448e − 05) 5 TV sets per Capita (10 ) 1.055e − 05 (5.373e − 06)∗ Newspaper Copies per Capita −2.544e − 04 −2.151e − 04 4 ∗∗∗ (10 ) (5.549e − 05) (4.487e − 05)∗∗∗ GDP per Capita −1.206e − 04 −1.048e − 04 ∗∗∗ (2.237e − 05) (1.737e − 05)∗∗∗ Effectiveness of Legislature −5.093e − 01 −5.123e − 01 (4.099e − 02)∗∗∗ (3.851e − 02)∗∗∗ Observation Count 6278 6832 AIC(≈ -2 x log-likelihood) 5743.4 6357 . Signif. codes: 0 ∗ ∗ ∗ 0.001 ∗ ∗ 0.01 ∗ 0.05 4
(3) −6.927e − 06 (2.057e − 05)
−2.068e − 04 (4.824e − 05)∗∗∗ −1.029e − 04 (1.879e − 05)∗∗∗ −5.087e − 01 (3.908e − 02)∗∗∗ 6799 6346.8 . 0.1
1
Table 2: Revolution count, negative binomial count regression
Parameter Revolution Count Radio per Capita for each 104 TV per Capita for each 105 Newspaper Copies per Capita for each 104 GDP per Capita (Factor Cost) Effectiveness of Legislature, 0 to 3—3 most effective
Average Max Min 0.18528 9 0 1650.55 68406 0 7358.77 226400 0 1018.23 9000 0 2859.45 44797 18 1.67004 3 0
Table 3: Codebook, range of parameters
24
(4)
9.041e − 06 (4.875e − 06). −2.613e − 04 (5.438e − 05)∗∗∗ −1.220e − 04 (2.236e − 05)∗∗∗ −5.126e − 01 (4.033e − 02)∗∗∗ 6285 5743.9
Controlling for GDP and type of the state, the negative impact of newspaper penetration is robust and significant.19
3.2
Historical Precedents
Ubiquitous and abrupt changes in media coverage provide an opportunity for studying the role of connectivity or lack thereof in the course of mass protests. After the introduction of the new social media such disruptions have become more commonplace. In the face of civil unrest the authorities have frequently chosen to disrupt communication venues available to protesters. To gauge the impact of media interruption, I study examples of mass uprisings in which there is a sharp interruption in service and estimate the effects of the change in coverage. To mitigate the endogeneity concern, it is necessary to control for confounding parameters that alongside media disruption might have changed the level of protests. As it was mentioned above, the most visible change that can attest to the validity of the theory is an increase in the dispersion of protests. There are multiple reasons for adopting such a research design: first, media disruption has happened frequently during the recent years as a tactic against the the new social mediaâ&#x20AC;&#x2122;s mobilizing capabilities, hence we know of multiple cases that can be used for the purpose of testing the theory. Second, because of the extent and accuracy of reports in recent years we have very detailed accounts of protests preceding and ensuing breaks in media/mobile communication coverage. The same is true of confounding factors. In the course of my study of the Egyptian uprising of 2011, I will present a showcase of such a design, but before doing so I include some archival evidence suggesting that the same process may have been at work in other historical occasions as well. this is a plausible account, it fails to specify what these political institutions are and what constitutes a measure of political modernization. Taking development as a repellent of mass political instability is at best tautological. Its credibility is proved when one finds the elements of political development, demonstrating the mechanisms through which they contribute to political stability. 19 One caveat is effectiveness of the number of newspaper copies in relation to the rate of literacy in corresponding countries. If the link between illiteracy and low GDP is strong, the impact of illiteracy is counted in when GDP is included as a control parameter.
25
3.2.1
From Pamphlets to Centralized Media
One of the more interesting facts about the French Revolution is the prevalence of pamphleteering immediately before the revolution (Popkin 1990). While the established periodicals of the time, e.g. Gazette de France and Gazette de Leyde, mostly ignored the rebellion during the summer of 1789, many of the leaders of the Third Estate (and later revolutionary leaders during the struggle of summer 1789) published pamphlets to disseminate news and make their own take on the situation known to the public. At the beginning of the summer 1789, Louis XVI tried his best to block the growth of pamphleteering (Popkin 1990), but did not succeed. Later pamphleteers such as Mirabeau and Abbe siey`es lead the revolution. After victory, the revolution redefined the relation between the state and media; pamphleteering was discouraged. Instead the printed press with wider circulation replaced grass roots means of written communication and reporting. Prior to the culmination of the February 1917 unrest in Petrograd, the cityâ&#x20AC;&#x2122;s newspapers stopped circulation immediately before the Dumaâ&#x20AC;&#x2122;s dissolution on the 27th. They resumed on the first of March. Afterwards the Duma tried to maintain a monopoly on the press (Wade 2005, Hasegawa 1981, Historical New York Times n.d.). Wade (2005) believes the confusion caused by the absence of printed media might have hastened the collapse of the ancien r´egime.20 Similarly during the Iranian Revolution of 1978-79, the printed press stopped circulation on November 6th, 1978 and did not return to normal till two months later (Historical New York Times n.d., Historical Washington Post 1978, Kayhan 1978-9). The largest protests of the Iranian Revolution happened during the very same period when media coverage was at minimal levels and the scant broadcast of the state media was widely disregarded in favor of pamphlets, audio cassettes and foreign radio stations (Historical New York Times n.d., Kayhan 1978-9). Again in this case the spread of grass-roots rumors had a major impact on the success of revolutionary mobilization. Instead of the state setting the general political agenda across the society, the opposition encouraged repetitive cycles of rebellion. In both of the above cases it is difficult to estimate the impact of the absence of media because 20
It is necessary to mention that literacy rates among the urban Russian population around 1917 were around 70 percent (Mironov 1991), i.e. the printed press as the sole formal news propagation mechanism (prior to television and radio broadcast) played a major role in everyday news learning. Their absence might have acted as a catalyst for the revolutionary unrest in Petrograd.
26
of confounding factors. The evidence we have is not accurate enough to test the “dispersion hypothesis” (see section (2.4)). Specially because the treatment (disruption of the media and cutting down venues for communication) was not always complete, i.e. did not include all means of news propagation. In more traditional societies constituents relied on a wide array of traditional communication processes, hence stopping a number of state controlled channels, e.g. newspapers might not have had the same impact as cutting all mobile/Internet communications overnight. That’s exactly what Mubarak’s regime did at the early hours of January 28th, 2011.
3.3
Mubarak’s Natural Experiment in Media Disruption
In response to the opposition staging demonstrations for three consecutive days in Tahrir Square and promising a yet larger demonstration on a Friday of Rage, Mubarak’s regime shut down the Internet and cell phone coverage across the country at the early hours of January 28, 2011. Instead of stalling demonstration in Tahrir, the consequences caught the regime by surprise. Protests flared across Cairo and other Egyptian cities including Alexandria and Suez. The protests were unusually diffuse and widespread and overwhelmed Mubarak’s security forces by the end of the day (Historical New York Times n.d.). Around 7 PM on January 28th the military was brought into the scene to replace the dysfunctional police force. After deployment of the military, dynamics of the interaction among the political players (the incumbent, the military, and the opposition) changed. The military’s inaction, accompanied with unexpected implications of the regime’s bold experimentation with the mass media in the following days, put an end to Mubarak’s thirty year rule. At the turning point of January 28th, lack of cell phone coverage and Internet connection forced the population to find other means of communication, encouraging local mobilization. Meanwhile apolitical strata of the Egyptian society, aggrieved by the disruption, were pushed into joining the confrontation. Instead of protests only in and around Tahrir Square, sizable demonstrations appeared in many locations in Cairo (Historical New York Times n.d., Shehata et al. 2011). As mentioned above, the Egyptian case offers a unique opportunity to test the plausibility of the dispersion hypothesis in a context similar to a natural experiment (alas run by Mubarak’s
27
regime). The disruption was abrupt and its timing (1:30 AM on January 28th) rules out any preparation for countering the blackout on the previous day. Between 10 PM and 2 AM, SMS and Internet communications were shut down, more importantly cell phone communications were shut down during the 28th, see a timeline at table (5).21 While the regime had experimented with selectively disrupting network coverage and websites such as Twitter and Facebook, it was the first time a universal communication blakcout was imposed on the nation. Dismantling regular venues for communication incited Egyptians to find new ways of staying online or forgoing online communication altogether. During the social media hiatus, older mobilization tactics were used in conjunction with new means of mass communication. For example satellite television stations such as Al Jazeera broadcast news communicated to them via landline phones (Shehata et al. 2011). Al-Arabiya television station also started broadcasting informative tweets on radio. At the same time tweeting over the phone became a possibility using Google’s Speak2Tweet (Dunn 2011a). In addition to these spontaneous innovations, the protests proliferated through much more mundane means. On the 28th those worried about their friends and family members participating in protests, could not reach them via cell phones, and had to join the crowds in streets to find out about their acquaintances (Shehata et al. 2011). In hazardous conditions of the ongoing stand off across the city, focal local points became gathering locations. Many congregated in local squares, strategic buildings, and mosques instead of trying to reach Tahrir (Historical New York Times n.d.)22 . To summarize, the disruption of cell phone coverage and Internet on the 28th exacerbated the unrest in at least three major ways: it implicated many apolitical citizens unaware of or uninterested in the unrest; it forced more face-to-face communication, i.e. more physical presence in streets; and finally it effectively decentralized the rebellion on the 28th through new hybrid communication tactics, producing a quagmire much harder to control and repress than one massive gathering in Tahrir.23 21
sources: Ramy Raoof, Egyptian activist and blogger, the Lede Blog, Dunn (2011b), and Renesys.com among others (blog posts, video interviews, news reports)). 22 http://nyti.ms/gcHvjz 23 Even in Tahrir there was no single leadership (Shehata et al. 2011): “Nobody was in charge of Tahrir,” and “a lot of those who joined the protests on January 25th in Tahrir were not aware of the Facebook campaign, they had heard about it from the protesters in the square and surrounding streets.” According to Mourtada and Salem
28
3.3.1
Notes on the Dynamics of the Egyptian Unrest
Four important cycles of unrest are evident during the 18 days of the Egyptian uprising, each of which starts with a significant media event. There are four cycles: January 25-27, January 28-February 01, February 02-07, February 08-11. a) January 25-27: Initial mobilization on January 25th, the social media campaign b) January 28-February 01: Disruption of the media on January 28th and the proliferation of the protests, the military steps in and acts as a game changer during the following clashes between pro and anti-Mubarak crowds c) February 02-07: Provocative national address by Mubarak on late night February 01st (stating he will stay in power till September and will die in Egypt), ensuing clashes on the 2nd and 3rd, the military’s inaction emboldens prevailing opposition crowds, relative calm afterwards till the 8th d) February 08-11: Emotional appeal by Wael Ghonim on a late night television show on the 7th and his follow up speech in Tahrir square in the morning of February 08th initiates the final phase of protests in Tahrir, eventual announcement of Mubarak’s resignation by Suleiman on the 11th. The above parsing of the events emphasizes the pivotal role of media operations during the unrest. Setting aside the military’s involvement–itself brought about by the sudden disruption of all online communication means–the role of the media is in full display. Dismantling the cell network in particular influenced mobilization on the local level. Egyptians had to revert to local interactions for gaining information. New local mobilization networks were formed that were smaller and better connected. Radicals became more effective on the local level, because they could directly contact more people on the ground. Moreover the networks underlying collective action and news propagation became smaller and more diffuse, making it more difficult to contain the protests. To overcome the endogeneity critique (that the government disrupted the media in anticipation of a major increase in the size of the protests, or that the closing down of the media was simply (2011) majority of respondents to an online survey on the role of the media disruption on the protests, believed it had a positive effect on the demonstrations.
29
a by-product of the escalating unrest) one could show that the major prediction of the model, i.e. sharp increase in the dispersion of the protests is manifest and statistically significant even after controlling for other confounding factors. The fact that there were calls for attending a protest in Tahrir during the previous days does not account for a sharp jump in the number of â&#x20AC;&#x153;locationsâ&#x20AC;? in which the clashes on the 28th took place. The dependent variable here is the dispersion of the protests, not their total size. Furthermore, such proliferation never happened again during the course of the protests, in spite of the the return of the media and enduring contention in Tahrir, even when the numbers in Tahrir were unprecedented (e.g. upon the restoration of the Internet on February 2nd). In the following I outline the procedure for extracting protest location data using available journalistic reports (photos, wires, and blog posts), describe the media interruption timeline and present a statistical analysis for singling out the effect of disruption on protest dispersion. I show that the Mubarak experiment is yet the most convincing evidence on the dispersion hypothesis we have on file.
3.4
Protest Dispersion in Cairo and Media Disruption
In the following I outline the data on protest dispersion and media disruption during the 18 days of the Egyptian uprising of 2011. I include four variable in the OLS regressions. The dependent variable is protest dispersion defined as the number of locations in Cairo where protests were happening each day. As independent variables I include a dummy for the treatment, i.e. media disruption, controlling for Fridays and national addresses via television. Controlling for Fridays is necessary because Friday is the Muslim weekly holiday and more people have the time and latitude for demonstration. Friday prayers also act as focal events. Media announcements are included as a control to estimate and control for the influence of mainstream media reporting. The details of protest locations are included in table (4), a log of state interruptions of media/cell coverage is included in table (5). A difference-in-difference strategy Here I regress daily differences in protest dispersion in Cairo on media disruption, controlling for Fridays and media announcements. First note the
30
Date January January January January
Dispersion: Protest Locations 25 1: Tahrir 26 1: Tahrir 27 1: Tahrir 28 Friday 8: Tahrir-NDP headquarters-Egyptian National Museum/Kasr al-Nil bridge/ 6 October bridge/TV headquarters/Al Azhar mosque/ Mohandeseen/Mustafa Mahmoud Mosque/l Istiqama Mosque January 29 4: Tahrir-NDP headquarters-Egyptian National Museum/ Interior Ministry/Corniche al-Nil/Abu Zaabal January 30 3: Tahrir/Heliopolis/Abu Zaabal January 31 3: Tahrir/ Mohandeseen/Arkadia Shopping Center February 01 2: Tahrir/Kasr al-Nil Bridge February 02 3: Tahrir-Egyptian National Museum/Mohandeseen/Corniche al-Nil February 03 1: Tahrir-Egyptian National Museum February 04 Friday 1: Tahrir-Egyptian National Museum February 05 1: Tahrir-Egyptian National Museum February 06 1: Tahrir-Egyptian National Museum February 07 1: Tahrir-Mugamma February 08 2: Tahrir/ Egyptian Parliament February 09 4: Tahrir/ Zamalek/ Ministry of Health-Egyptian Parliament/ Dokki(organized labor protests) February 10 4: Tahrir/ TV Headquarters/ Egyptian Parliament/ Abdin Palace February 11 Friday 5: Tahrir/ TV Headquarters/Presidential Palace/ Mustafa Mahmoud Mosque/ Egyptian Parliament Table 4: Source: The New York Times, The Lede Blog, See Section (5.1)
31
Date January 25
January 26 January 27 January 28
January 30 January 31 February 01 February 02 February 05
Type of Disruption or Restoration Twitter.com blocked Bambuser.com (Live video streaming) blocked at 02:00 PM Activists’ mobile lines shut down Network coverage shut down in Tahrir Facebook.com blocked Blackberry services shut down 7:00 PM SMS shut down 10:00 PM Internet shut down-except one ISP, 1:30 AM Mobile phone calls shut down for one day Landlines shut down in some areas Al-jazeera Cairo bureau shut down Last ISP shut down State media campaign against protesters (text messages) Internet service restored 12:30 PM SMS restored 12:35 AM
Table 5: All times are Egyptian local time, i.e. EST+7 in January/February 2011–Source: Ramy Raoof, Egyptian Activist and Blogger, Alix Dunn Blogger, Renesys.com–23.51 Million Internet users (30% penetration), 71.46 Million Mobile subscribers (90% penetration) in January 2011 sharp jump in the number of protest locations on the 28th in figure (7). 28th was the first day of blanket disruption and the only day in which cell phone communications, Internet services, SMS, and occasionally landlines were unavailable (see table (5)). Lack of cell phone coverage is more far reaching among the population than the absence of the Internet. In January 2011, there were 23.51 million Internet users compared to 71.46 million mobile subscribers.24 Hence compared to Internet outage, any disruption in mobile communications directly implicates a much larger portion of the Egyptian population. Also mobile phones are more effective means of communication on the fly during protests in streets. On the other hand, Internet disruption complicates longer term planning and organization. Cutting cell coverage has an immediate impact on personal communication in streets. Considering these issues and the fact that dismantling Egyptian online network came as a surprise during the very first day of the outage, I code the treatment as a dummy on the 28th (discon-diff)-Later I consider an extended version of the disruption variable. National media addresses by the heads of the state and the opposition (see appendix 5.2) are also coded as a 24
23.51 Million Internet users (30% penetration), 71.46 Million Mobile subscribers (90% penetration) in January 2011, from report by Egypt’s ministry of Communications and Information Technology http://slidesha.re/mtrPuM
32
control variable in table (6). Using a difference-in-difference strategy, I regress the daily change in dispersion over the treatment (disruption) controlling for other variables. As can be seen in table (6) media interruption on the 28th had a statistically significant and relatively effective role in proliferating the protests in Cairo.25
Figure 7: Top: Protest Dispersion (Number of Distinct Protest Locations According to Table 4), Bottom: All Google Products, Egypt Traffic, from http://bit.ly/h3Q1FZ In addition to the difference-in-difference strategy in table (6), I also code the disruption across the 5 day period of 28th to 01st. Taking the relative penetration of mobile devices to the Internet into account, I implement the treatment vector [3 1 1 1 1]. Note that this is an underestimation of the importance of the disruption on the 28th, because on the first day full disruption came as an utter surprise, hence its effect was more pronounced in the first day compared to the following 25
Similar results are obtained when the variable “television announcement” is excluded.
33
Estimate Std. Error t value Pr(> |t|) (Intercept) −0.08333 0.40886 −0.204 0.84143 discon-diff 5.91667 1.96080 3.017 0.00923∗∗ fridays 1.16667 1.29291 0.902 0.38213 announc −0.58333 0.91423 −0.638 0.53373 . Signif. codes: 0 ∗ ∗ ∗ 0.001 ∗ ∗ 0.01 ∗ 0.05 . 0.1 1 Residual standard error: 1.416 on 14 degrees of freedom Multiple R-squared: 0.6382, Adjusted R-squared: 0.5606 F-statistic: 8.23 on 3 and 14 DF, p-value: 0.002109 Table 6: Protest Dispersion in Cairo, OLS over treatment (disruption on 28th), controlling for Fridays, and media announcements four days. In fact as it was mentioned above, Egyptians found their way around the blockage soon, hence the decentralizing influence from the type discussed above was the most pronounced on the very first day of disruption. Nevertheless even with the new coding, the impact of the disruption on protest dispersion is large and statistically significant. In this case, the dependent variable is the dispersion per se (not daily differences). The results are included in table (7). Estimate 1.2845 1.8223 0.6526 1.3610
Std. Error t value Pr(> |t|) (Intercept) 0.3980 3.227 0.006083∗∗ discon 0.4328 4.210 0.000873∗∗∗ fridays 0.9048 0.721 0.482635 announc 0.6818 1.996 0.065752. . Signif. codes: 0 ∗ ∗ ∗ 0.001 ∗ ∗ 0.01 ∗ 0.05 . 0.1 1 Residual standard error: 1.205 on 14 degrees of freedom Multiple R-squared: 0.6746, Adjusted R-squared: 0.6049 F-statistic: 9.676 on 3 and 14 DF, p-value: 0.001025 Table 7: Protest Dispersion in Cairo, OLS over treatment (disruption on 28th-01st), controlling for Fridays, and Television announcements Note the positive, large and statistically significant impact of social media/mobile communication disruption. In addition to media disruption, the influence of nationally broadcast television addresses is also manifest here (although not as statistically significant). Witnesses present at the unrest have repeatedly cited the role of satellite television stations in proliferating the rebellion (Shehata et al. (2011) among others). The above results confirm those speculations. Major television addresses preceded unrest in multiple occasions during the 18 days of the uprising (see 34
appendix (5.2)). The contrast with the common view on the singular importance of the new social media in generating unrest in Cairo is thought provoking. 3.4.1
Blocking the New Social Media for Stopping a Rebellion?
From the above it is clear that disrupting the media (and to a lesser degree the ever present satellite television coverage of the events) charged the population and overwhelmed the authorities; but what are the conditions under which such disruptions might not be as mobilizing? An overview of the press at revolutionary times reveals that there are varying degrees of affinity between the press and the people. At times the press completely ignore the protests even till the very moment the ancien r´egime is about to collapse (Petrograd Gazette 1917, Gazette de France 1789, Al-Ahram 2011). Sometimes they cautiously report on the unrest and sympathize with the protesters in a subtle tone (Kayhan, Iran 1978). Finally at times they fully support the protesters, reporting on their actions in detail, Al Jazeera Egypt 2011). There are multiple cases of failed rebellion during which access to the social media was highly restricted e.g. the post-election unrest in Iran 2009 and the Thailand Red Shirts rebellion of 2010. At the height of the protests in Tehran in June 2009, mobile communication was cut off and press reporters were banned. Opposition leadersâ&#x20AC;&#x2122; communication with the outside world was very limited (Historical New York Times n.d.). In spite of limited access to social media at the time of the unrest, there was never a blanket interruption, but limitations were targeted, local, and temporary. Similarly in Thailand, the supervised flow of the opposition press allowed the government to suppress the opposition. The largest fatalities in one single protest event happened when the Red Shirts tried to take over the building housing the Peoples Television Station (a television station sympathizing with the Red Shirts). After an initial retreat from the premises, the government allowed for the stationâ&#x20AC;&#x2122;s operation but under strict supervisory26 . In both cases a slow flow of supervised communication was allowed; on the other hand, face-to-face means of mobilization were actively disrupted. 26
Reports on the incident here http://nyti.ms/bP55zl and http://bbc.in/oMA7Wo
35
4
Conclusion: too many in too many places...
In the above I showed that the favorable portrayal of social media in fostering unrest in the context of heterogeneous networks should be reconsidered. In other words, in the presence of a risk-averse majority and a radical minority, adding more links among the majority does not necessarily help mobilization. In the absence of centralized media, crowds’ risk-taking behavior becomes independent of the state’s intentions. Note that even the most authoritarian regimes prefer not to systematically bomb their own population, they instead use a threat of forceful military action in order to deter. When it is impossible to communicate the possibility of a painful military retaliation, the state is unable to dissuade the crowds. In fact protests proliferate when such threatening measures fail. In reaction to a shock similar to the one exerted on the Egyptian society on January 28th, the population can overwhelm the incumbent apparatus. The consequences of such an action were evident in the aftermath of the Egyptian media blockage. The Lede Blog reported from Alexandria on January 28th (emphasis is mine): “It’s clear that the very extensive police force in Egypt is no longer able to control these crowds. There are too many protests in too many places... said Peter Bouckaert, emergencies director of Human Rights Watch, who observed the street battle in Alexandria on Friday[01/28/2011].”
5
Appendix 1: Event Data Collection: Egyptian Uprising of 2011
5.1
Event Data Collection, Methodology
In this appendix I outline the data collection procedure for the Egyptian uprising of 2011. In my reconstruction of the events, as for the location of protests, I used graphic designs from the New York Times’ website in addition to accounts from the Lede blog,27 for media disruption I have 27
Links to visual representations on The New York Times’ website: a. http://nyti.ms/gf4hzJ b. http://nyti.ms/g3LanI c. http://nyti.ms/fh5Ypb
36
used online information from Egyptian bloggers28 and coded the data after crosschecks. For confirming the hypothesis on the role of the Egyptian media shut down on the unrest and the eventual ouster of Mubarak, I used the closest log to news wires I could find i.e. the Times’ Lede Blog and the aforementioned graphic designs to reconstruct a fine-grained description of the events. News blogs often can be valuable sources of information on protests. My reconstruction of the Cairo events has been mostly based on my readings of various blogs and inferences from news photos. The real-time evidence provided by blogs, twitter and Facebook updates, in addition to event photos provide an unprecedented level of accuracy in event reconstruction and enforce new ways of studying decentralized mass protests.
5.2
Major Media Announcements During the Egyptian Protests
Announcements are major addresses to the nation, broadcast from the state television or cable TV channels with a national audience. I have collected the following accounts from The New York Times’ Lede Blog. All times are Egyptian local time (EST+7 in January and February 2011). 1. January 29th: at around 12:30 AM, Mubarak’s late night address (coded as 29th), he gives concessions, dismisses the government, appoints a Vice President/ Obama talks to the press immediately after Mubarak’s speech. 2. February 01st: at around 10 PM (earlier that day the largest protest to date in Tahrir), Mubarak gives another speech on the state television. He announces he intends to remain in office until the end of his current term. Obama again speaks after Mubarak’s speech. Note: Worst clashes start on February 02nd and last for two days (2nd and 3rd of February)by 5th or 6th the situation has almost returned to normal. 3. February 04th: Suleiman makes TV appearances the day before–total of two television appearances before Friday 04th. 4. February 08th: Wael Ghonim gives an emotional interview on Monday night (07th) on a 28
See a timeline at: http://bit.ly/pNRvEH
37
cable TV channel and appears in Tahrir the next morning (08th) to give a speech addressing the massive crowd in the square. 5. February 10th: Around 7 pm Egyptian state TV announces an address by Mubarak to be broadcast soon-Obama gives a speech around 8:30 PM, “Egyptians are making history.” Around 10 PM, Mubarak gives a speech on the state television, asserting that he is not stepping down and will remain in office till September. 6. February 11th: Around 6 PM Suleiman gives an address on the state TV announcing that “President Hosni Mubarak has resigned and handed over power to the country’s military.” Obama gives a speech at 10 PM.
6
Appendix 2: Dynamic Models
6.1
First Dynamic Model, Asymptotics
We would like to find final thresholds Γi (∞) for the first dynamic model. In any connected network, the elements of the stationary transition matrix are di + 1 . j (dj + 1)
∆∞ = !
In the case of the example in section (2.3.1) for the star network, v = [3/(3 + 2 + 2) 2/(3 + 2 + 2) 2/(3 + 2 + 2)]T = [3/7 2/7 2/7]T and the final thresholds will be [3/7 2/7 2/7].[0 τ τ ]T = 4τ /7 ≈ 0.57τ . Also as the number of elements in a star network increases, the threshold → 2/3: using the same technique as above, the final threshold for a k + 1-star (the above example was a 2-star.) is (2k)/(3k + 1) which is always smaller than 2/3, but approaches 2/3 when k becomes large. Interestingly this is the final threshold for the fully connected graph with the number of nodes n = 3 (i.e. triangle).
38
6.2
Action Based Dynamics, Examples
The general equilibrium solution was outlined in section (2.3.3). Here I include the details for a number of well-known topologies of networks under the action-based dynamics. 6.2.1
Star Networks
Starting at t = 0, the central node’s threshold is 0, the peripheral nodes’ thresholds are all taken to be τ . In a star network with n agents, at time t, • if t is odd, the central node’s threshold is
γ1 =
n−1 1 1 (1 + 2 + . . . + t−1 ), n n (n2 ) 2
and the central node does not act (N). The peripheral nodes will have a threshold of
γi =
τ 1 1 + + . . . + t−1 , t 2 4 4 2
and act, (A). • If t is even (larger than 0), then the central node’s threshold is
γ1 =
n−1 1 1 (1 + + . . . + t−2 ) n2 n2 (n2 ) 2
and the central node acts (A). The peripheral nodes will have a threshold of
γi =
τ 1 1 1 + (1 + + . . . + t−2 ), t 2 2 4 4 2
and do not act (N). Note the asymptotic cases, when t → ∞. When t is odd, the central node does not act (N), and the peripheral ones act (A). γ1 →
n , γi → 1/3 n+1 39
when t is even, the central node acts (A), and the peripheral ones do not act (N). 1 , γi → 2/3 n+1
γ1 →
6.2.2
Fully Connected Networks
Consider the case in which the network is fully connected. If there is only one radical agent, then there will be no action starting from t = 1, and thresholds are 1 1 1 n−1 (1 + + 2 + . . . + t−1 ) → 1 n n n n for the radical node. For the others, if t = 1, γk (1) = 1 nt−1
(
τ n
+
n−2 , n
and for t ≥ 2, γk (t) is
τ n−2 n−1 1 1 1 + )+ (1 + + 2 + . . . + t−2 ) → 1. n n n n n n
The size of critical mass in a fully connected network in this model is m > n/2. In this case at time t the thresholds are n−m ) n 1 τ n−1−m = ( + ) t−1 n n n
γk = γk
1
nt−1
(
for the critical mass and the ordinary agents respectively. Both of the above go to zero as t → ∞. From t = 1 onwards both groups act (A).
References Banks, A. S. (2009), Cross-National Time-Series Data Archive, Databanks International. http://www.databanksinternational.com. Chwe, M. (1999), ‘Structure and strategy in collective action’, American Journal of Sociology 105(1), 12856.
40
Chwe, M. S. (2001), Rational Ritual, Culture, Coordination, and Common Knowledge, Princeton University Press, Princeton, NJ. Dunn,
A.
(2011a),
‘How
the
internet
kill
switch
didnt
kill
egypts
protests’.
http://bit.ly/gAeqbu. Dunn, A. (2011b), ‘Unplugging a nation: State media strategy during egypts january 25 uprising’, The Fletcher Forum of World Affairs 35(2), 15–24. Ermakoff, I. (2008), Ruling Oneself Out, A Theory of Collective Abdications, Duke University Press, Durham, NC. Gould, R. V. (1993), ‘Collective action and network structure’, American Sociological Review 58(2), 182196. Gramsci, A. (1971), Prison Notebooks, International Publishers, New York, NY. Granovetter, M. S. (1978), ‘Threshold models of collective behavior’, The American Journal of Sociology 83(6), 1420–43. Hasegawa, T. (1981), The February Revolution: Petrograd, 1917, the University of Washington Press, Seattle, WA. Hassanpour, N. (2010a), Dynamic models of mobilization in political networks, in ‘Proceedings of the 2010 Political Networks Conference’, Duke University, Durham, NC. Hassanpour, N. (2010b), “structural dynamics of collective action in political networks”, in ‘APSA 2010 Proceedings’. Hibbs, D. A. (1973), Mass Political Violence: A Cross-National Causal Analysis, John Wiley & Sons, New York, NY. Historical New York Times (n.d.), The New York Times . Historical Washington Post (1978), The Washington Post .
41
Huntington, S. P. (1968), Political Order in Changing Societies, Yale University Press, New Haven, CT. Jackson, M. O. (2008), Social and Economic Networks, Princeton University Press, Princeton, NJ. Kayhan (1978-9), Kayhan Archives . Kern, H. L. and Hainmueller, J. (2009), ‘Opium for the masses: How foreign media can stabilize authoritarian regimes’, Political Analysis 17(4), 377–399. Kuran, T. (1989), ‘Sparks and prairie fires: A theory of unanticipated political revolution’, Public Choice 61(00), 41–74. Kuran, T. (1991), ‘Now out of never, the elements of surprise in the east european revolution of 1989’, World Politics 44(00), 7–48. Lohmann, S. (1994), ‘The dynamics of informational cascades: The monday demonstrations in leipzig, east germany, 1989-91’, World Politics 47(1), 42–101. Marwell, G. and Oliver, P. (1993), The Critical Mass in Collective Action: A Micro-social Theory, Cambridge University Press, New York, NY. Mironov, B. N. (1991), ‘The development of literacy in russia and the ussr from the tenth to the twentieth centuries’, History of Education Quarterly 31(2), 229–252. Morris, S. (2000), ‘Contagion’, Review of Economic Studies 67, 57–78. Mourtada, ments:
R.
and
Salem,
F.
(2011),
‘Arab
The impact of facebook and twitter’,
social
media
report,
civil
move-
Dubai School of Government 1(2).
http://www.arabsocialmediareport.com/. Popkin, J. D. (1990), Revolutionary News, The Press in France 1789-1799, Duke University Press, Durham, NC. Popkin, J. D. (1995), Media and Revolution, Comparative Perspectives, The University Press of Kentucky, Lexington, KY. 42
Schelling, T. C. (1978), Micromotives and Macrobehavior, W. W. Norton, New York, NY. Schumpeter, J. A. (1950), Capitalism, Socialism and Democracy, Harper Prennial, New York, NY. Sewell, W. H. (2005), Logics of History: Social Theory and Social Transformation, University Of Chicago Press, Chicago, IL. Shehata, D., El-Hamalawy, H. and Lynch, M. (2011), ‘Youth movements and social media: Their role and impact’, Proceedings of “From Tahrir: Revolution or Democratic Transition” . Videorecordings available at http://bit.ly/pOy1QJ. Siegel, D. A. (2009), ‘Social networks and collective action’, American Journal of Political Science 53(1), 122–138. Wade, R. A. (2005), The Russian Revolution, 1917, Cambridge University Press, New York, NY. Weber, M. (1958), From Max Weber: Essays in Sociology, Oxford University Press, New York, NY. Wilkinson, S. (2009), ‘Riots’, Annual Review of Political Science 12, 32943.
43