What Drives Social Unrest?

Transcription

1 1 What Drives Social Unrest? Evidence from Brazil s Protests Arthur Aguilar (JPAL) and Claudio Ferraz (PUC-Rio) 1 July 2014 Abstract We investigate the effects of public service quality and corruption perceptions in the disruption of the 2013 June Protests in Brazil, where at least 2.8 million individuals attended more than 700 protests across the country. In addition, we test whether these effects differ in municipalities where social media usage is more widespread. We find that corruption perceptions and quality of public education and health are major determinants of the demonstrations. Nevertheless, effects are very heterogeneous: in cities with a high level of social media usage, poor quality of public services raises substantially the likelihood of a protest to happen whereas it can have barely no effects in cities with low levels of social media usage.(jel Classification: D72 D73 D74) Introduction We use data on protests to investigate the effects of public services quality and corruption perceptions in the emergence of the June 2013 Protests in Brazil. In addition, we test whether these effects differ in municipalities where social media usage is more widespread. We find that the quality of public education and health are major determinants of the demonstrations, and their effects are even larger when accounting for social media usage. In a regression of the interaction between satisfaction with public hospitals and social media usage, controlling for municipal characteristics, in manifest, a dummy equal to one if a municipality had a protest and 0, otherwise, we find that public hospitals quality has a considerable impact on protests. For municipalities in the 90 th centile of the social media usage distribution (.57), a standard deviation decrease in satisfaction with public hospitals (.272) increases the likelihood of a protest to happen in 27.9% compared to the mean of protest incidence. In the first two weeks of June 2013, the Movimento Passe Livre (Free Fare Movement; MPL) coordinated a couple of demonstrations in São Paulo claiming for the revocation of a R$ 0,20 raise (nearly U$ 8 cents) in the bus fares. The protests were violently repressed by the police and had their legitimacy questioned by both the Government and the press. However, protests spread rapidly to other states capitals, like Rio de Janeiro and Porto Alegre, and then to the rest of the country. Everywhere, Brazilians used social 1 We are grateful to Gustavo Gonzaga, Juliano Assunção, Ricardo Dahis and Tatiana Ruediger for comments and suggestions.

2 2 media to promote demonstrations, inviting their peers to the acts through Facebook events, and live broadcasting the protests on Twitter and Youtube for those who could not show up. The message that the giant has awaken was broadly disseminated. During June 17 th and 30 th, more than 775 protests happened in 433 cities, leading at least 2.8 million individuals to the streets. By that time, protesters had a much broader set of demands, questioning the quality of public services, the high incidence of corruption inside public institutions, the expenditures with the 2014 World Cup, as well as a variety of local issues. When thousands of protesters built a 20 meter fire in the front of the Presidential Palace in Brasilia, the national capital, and climbed the roof of the National Congress, President Dilma Roussef went on public to announce that her government was willing to hear the voice of the streets. The raise in the bus fares were revoked or postponed in several municipalities, and the government presented a wide package of policies, contemplating themes like public services, mobility, fiscal discipline and a political reform. The peak of protests eventually ceased, but since then Brazil has experienced a much more intense routine of strikes and demonstrations. Nevertheless, why has the giant awaken just now? And moreover, what has disturbed his dreams? In his new book, Networks of Outrage and Hope, Manuel Castells analyzes distinct episodes of social unrest like the Indignados in Spain and the Arab Spring. The author suggests that much of the contemporary cases of intense protesting follow a very similar pattern: they are headed by networks of individuals that coordinate themselves through comunication technologies and social media. These individuals are indignated with many aspects of the societies they live in, and since they actually believe that they have the means to change it, they decide to challenge the political status quo. In order to understand the Brazilian case, we use the following theorethical framework: Individuals feel entitled to a certain level of quality of public services, which is determined by their notion of fairness, given the observed state of the economy. They also have a self serving bias, which implies that they do not distinguish perfectly what is fair from what is economically convenient to them. If the level of public services quality implemented by the government deviates from the level they feel entitled to, they get aggrieved and decide to protest, which provides them a psychological reward, in order to punish the government (Passarelli, F. & Tabellini, G., 2013). The decision of whether to attend a protest depends as well on the private and public costs and bennefits associated with protesting (Tullock, G., 1971). As lower the opportunity cost (Campante, F. R., Chor, D., 2012) and the likelihood of being punished for protesting are, and as larger the agrievement and the likelihood of being able to influence public policy outcomes are, larger is the likelihood of a protest to happen. In addition, the available information and comunication technologies play a crucial role, helping the protesters to solve the collective action problem through information and coordination effects (Yanagizawa, Drott., 2010). In the figures I and 2, we show how satisfaction with public schools and social media usage are, respectively, negatively and positively correlated to protest willingness over Latin American cities (Latino Barometro 2010). With these descriptive evidences and the theoretical framework presented above on mind, we enunciate our main hypothesis about the determinants and the dynamics of the June Protests, which are the body of our

3 3 investigation: H1) Poor public services quality raises the likelihood of a protest happens. H2) The effects of low-quality public services on protests are more intense as the social media usage in a given municipality spreads itself. Our results suggest that both H1 and H2 hold in the case of the June Protests. First, we use the IDEB and IDSUS indexes as proxies of public education and health quality, and the proportion of households in each municipality that own at least a computer to access computer usage. We find that a standard deviation decrease in the IDEB 2011 raises the likelihood of a protest happening in 1.71 percentage points, or 21.3% compared to the mean of protest incidence (Table 2). The effects for IDSUS arises when we interact IDSUS 2011 with computer usage (Table 3). In the 90 th centile of the computer usage distribution, a standard deviation decrease in the IDSUS raises that likelihood of a protest happens in 28%, and a standard deviation decrease in the IDEB raises the likelihood of a protest happens in 39.88%, both compared to the mean of protest incidence. Next, we use data from Latino Barometro 2010 collapsed at the city level to estimate the effects of public services satisfaction and corruption perception on protest incidence, where we find no effects (Tables 4 and 6). Our results drastically change as we interact public services quality with social media usage (Table 5). We find expressive effects in all of our measures of satisfaction but satisfaction with the police, as the latter loses its significance when we control for municipal characteristics. At the 90 th centile of the social media usage distribution, a standard deviation decrease in each of the following measures of satisfaction, raise the likelihood of a protest happens in 27.9%, for satisfaction with public hospitals, 37.7%, satisfaction with public schools, and 26.8%, for satisfaction with public transport, all of them compared to the mean of protest incidence among the brazilian sample of Latino Barometro. Lastly, our results from corruption are more mixed, as the effects vary a lot depending of our measurement of corruption perception. To give a sense of magnitude (Table 7), at the 90 th centile of the social media usage distribution, a standard deviation increase in No Progress on Reducing Corruption raises the likelihood of a protest happens in 29.42%, compared to the average. 2. Related Literature In this work, we investigate the impact of public services quality in the emergence of the June 2013 Protests in Brazil. Over the next pages we present the literature related to this subject. What are the reasons and objectives that push people into demonstrations? We find at the social unrest literature three main answers to this question. First, individuals may protest to signal their preferences or to manipulate preferences from others (De Mesquita, E. B. 2010; Lohmann 1994), as in the case analyzed by Bueno de Mesquita (2010), in which vanguards use violence in revolutions to disseminate anti-government feelings through public opinion. Second, individuals may protest as a lobbying strategy. Yanagizawa Drott et. al., (2013) use rain to identify the causal impact of the Tea Party s Tax Day Rallies on several political and Tea Party movement outcomes. They find that

4 4 the larger the attendance to a rally in a county is, the more extensive is the media coverage about Tea Party, more funds are raised by local Tea Party committees and more the behavior of elected politicians converge into Tea Party s political view. Third, individuals may protest because they are aggrieved and feel that they have been treated unfairly, and hence decide to attend a demonstration in order to punish the government (Passarelli, F., & Tabellini, G., 2013; Woo, J., 2003; Romer, P. M., 1996). Tabellini and Passarelli (2013) develop a model to capture the relationship between social unrest and economic policy. The core of the model lies in three hypothesis: i) Individuals rationally weight costs and benefits to decide whether to participate of a protest. When they attend a protest, they feel a psychological reward, which is proportional to their sense of aggrievement. Moreover, the net benefit of protesting increases with the number of people that attend to the event. Thus, the larger the expected attendance, more individuals are willing to participate for a given level of aggrievement. ii) Individuals are rational on their expectations for/demands for entitlements, and may react emotionally if these are not met. A policy entitlement is a policy outcome considered fair, and this is endogenously determined and, therefore, predictable, although it can also be altered by exogenous shocks. Hence, if the government deviates from the fair level of a given policy outcome, people get aggrieved. iii) Individuals have a self-serving bias in moral judgments. This implies that material interests affect policy entitlements, so that individuals may confuse what they think is fair with what is economically convenient to them. Their model provides a rationale for understanding the dynamics of costly political action, and gives us several insights about the case we analyze. First, the idea that people may feel entitled to better public services in Brazil is a plausible one. Over the last 20 years, Brazilians experienced the universalization of access to very basic public services, like health, education and transport, which certainly provided a deeper sense of entitlement to those services. However, in the cases where the increase on the demand did not translate to an increase in the supply of those services, quality may have decreased. Furthermore, the complementarity between the expected attendance to a protest and the net benefit of protesting proposed by the authors may explain the relationship between the increasing usage of social media to organize and spread the protests, and the massive attendance to the events. Our work also relates to the literature that investigates the causes of protests, that is, the reasons, events and policies that make people attend to demonstrations. While some authors have focused on how ethnical heterogeneity (Esteban, J., Mayoral, L., & Ray, D., 2012; DiPasquale, D., & Glaeser, E. L., 1998) and institutional quality (Machado, F., Scartascini, C., & Tommasi, M., 2011) impact protest incidence, the majority of the works on this area investigates the role of budget and social expenditure cuts on generating social unrest (Passarelli, F., & Tabellini, G., 2013; Woo, J., 2003; Ponticelli, J., & Voth, H. J., 2011; Voth, H. J., 2012). Voth (2012) finds that budget cuts in South America are a much more critical determinant of protests than the deterioration of both macroeconomic conditions and personal income. Voth s work is the most related to

5 5 ours, in the sense that the author investigates the effects of a public policy decision. By addressing the effects of public service s quality on protests, we believe that our work is a relevant contribution to the literature of what causes social unrest. We have so far addressed the objectives and goals perceived by individuals that attend protests, as well as their demands. Now, we turn our attention to the dynamics of protesting. Here a natural first step would be to understand how the individuals react to the costs and benefits of protesting. Campante, F. R., & Chor, D. (2012) investigates how the evolution of schooling and the disparity between expected and actual returns on education lowers the opportunity cost of attending to protests, catalyzing the emergence of the Arab Spring. In his Paradox of Revolutions (Tullock, G., 1971), Tullock points out that the solution of the free ride problem on costly political action would be the extraction of private benefits by individuals who participate. DiPasquale, D., & Glaeser, E. L. (1998) test how the nature of payoffs, whether public or private, explains the incidence of the 1990 s riots in US, finding that private payoffs are much more important than public incentives. A large body of evidences shows how social ties matter for the dynamics of political action. Gerber, A. S., Green, D. P., & Larimer, C. W. (2008) and DellaVigna, S., List, N. J. A., Malmendier, N. U., & Rao, N. G (2013) show how peer pressure plays a crucial role on voting. Kleinberg et al (2011) find that individuals must observe several peers adopting a political behavior on Twitter before adopting it for themselves, a phenomenon known as complex contagion. In addition, as the June protests were largely organized and disseminated through social media, we think that the interaction patterns found in social media may influence the protests in several dimensions. For Gladwell (2010), political action organized through social media would lack both hierarchy and the strong social ties needed to engage in risky activities, two crucial features on costly political action. Granovetter (1973), however, emphasizes the importance of the bridging properties of weak ties, which are crucial in diffusion processes, political articulation and in the development of institutional structures. As we investigate the role of ICTs as a catalyst of the June protests, our article contributes to the literature that seeks to understand how new technologies impact political action. Gentzkow, M. (2006) uses differences in the timing of the introduction of television in the US to identify the impact of television on voting. He finds that the introduction of television has decreased the voting turnout on local elections in between a quarter and a half since In Politics 2.0, Campante, F. R., Durante, R., & Sobbrio, F. (2013) find that the introduction of broadband internet in Italy has increased the participation in alternative forms of political participation, like online protest groups, and after slight decrease, has increased the voter turnout, as new political entrepreneurs able to dialogue through the internet start to run in the elections. In Propaganda and Conflict: Theory and Evidence from the Rwandan Genocide, Yanagizawa Drott (2010) presents a model of collective violence in which mass media increases participation in conflict through two channels. First, there is an information effect: access to radio allows individuals to update their beliefs about a policy change

6 6 (in this case, the state-sponsored violence). Second, there is the coordination effect, which takes place due to strategic complementarities. These would lead individuals to engage in violence when mass media induces other individuals to participate. The author uses the Rwandan topology as an instrument to access differences in radio coverage, and finds that radio increases the participation in the genocide in nearly 51k individuals. The mechanisms proposed by Yanagizawa Drott are useful for analyzing the role of social media in the June protests in Brazil, in which individuals were invited to demonstrations through Facebook events, where information about the demands and complains of protesters were largely disseminated. Finally, on addressing the impact of the public services quality in the June protests, our work is a contribution to a still incipient literature that seeks to understand the causes and consequences of the June protests in Brazil. Moseley, M., & Layton, M. (2013) estimate a predictive model of protests in Latin America, and show that Brazil has extremely low levels of support for the political system and satisfaction with public services. Soares, E. L. (2013) emphasizes the importance of the large decrease in inequality Brazil experienced over the last 20 year in the arising of a new contingent of individuals demanding new political and social achievements. Winters and Weitz- Shapiro (2013) investigate how the protests influence the feelings of anti-partisanship and the rejection to the governing Partido dos Trabalhadores (Worker s Party; PT). Institutional Background During the first half of June 2013, the Movimento Passe Livre (Free Fare Movement; MPL) organized a series of protests in São Paulo, claiming for the revocation of a R$ 0,20 increase (nearly 8 cents) in the bus fares. What seemed to be just another series of acts among several others organized by the MPL over the decade, turned into a wave of social unrest that spread over the country (The Economist, June 2013). In the period between June 17 and 30, more than 775 protests in 433 cities led at least 2.8 million individuals to the streets. Protesters held posters claiming that It is not just about 20 Cents and demanding FIFA standard public services. These individuals were as diverse as their demands, that ranged from dissatisfaction with public services to the overwhelming costs of the 2014 World Cup, from corruption inside public institutions to police brutality while repressing demonstrations. When satisfaction with the government headed by President Dilma Rousseff and the Partido dos Trabalhadores (workers Party; PT) droped 27 percentage points in just two weeks, Mrs Dilma Rousseff announced that she was willing to hear the voice of the streets. The raise in the bus fares was revoked or postponed in several municipalities, and the government presented a wide package of policies contemplating themes like public services, mobility, fiscal discipline and a political reform (BBC News, June 2013). The peak of protests eventually ceased, but since then Brazil has experienced a much more intense routine of strikes and demonstrations. The demonstrations were largely organized, covered and broadcasted through social media (The Guardian, July 2013). Protests were called in Facebook events, that rapidly reached hundreds of thousands of guests as individuals invited their peers organically, in a diffusion proccess where social ties may have played a prominent role. Individuals

7 7 collaboratively recorded and broadcasted the acts, so that an even larger contingent could watch everything that was happening in each demonstration. Furthermore, the live broadcasting of the acts allowed protesters to register violent repression. The collected footage helped not only to hold responsibles accountable, but also to tactically react to coertion. This sort of organization was possible due to the widespread usage of comunication devices and social media in Brazil. According to the PNAD 2011, about 46.5% of the overall population, and 71.8% among those between years old use the internet. In addition, Brazil has the third larger population of Facebook user in the world, with 76 million users (June 2013). In figure 3, we show that more than 30% of Brazilians use social media. Over the last 20 years, Brazilians improved their wellfare considerably. Public services like health and primary education were virtually universalized, and income grew up and became better distributed, with the Gini coefficient reaching its lower level since What could justify the eruption of such a massive wave of protests, under this conjuncture? We believe that the universalization of public services like education or transport without a propper increase in the supply of those services may have impacted quality drastically, leaving both the old and the new costumers unsatisfied. In other words, quality may have been compressed, as a massive contingent of new users showed up, which left both the old and new costumers unsatisfied. Our data from Latino Barometro support the idea that Brazilians are very unhappy with their public services. In Figures 4 and 5, we show how Brazilians are sistematically less satisfied with the quality of their public services, as well as reporting a larger corruption perception than the rest of Latin America. Moreover, due to the recent improvements in wellfare, Brazilians may feel more entitled to public services of a better quality, and hence start to engage in protests to demand from the government the level of quality that they feel entitled to. In figure 6 we can see how brazilians have a much larger perception of protest effetiveness than the rest of the region. As points Soares (2013), improvements have combined in such a way that certain situations, which in the past would have been tolerated passively, have become unnaceptable. Data Protests We use data compiled from G1, a major news website in Brazil, to identify whether a municipality had protests during the period between June 17 and 30. While many studies cross-country use the Banks el al. (2012) database to analyze protests, data about demonstrations within country as a whole, and particularly in Brazil, is scarce. Thus, we use a similar approach to Yanagizawa Drott. et al (2013), where authors compiled information about tax day rallies attendance over U.S. counties at New York Times, as well as other newspapers. G1 s extensive coverage of the June demonstrations allows us to estimate whether a protest happened, how many people attended according to police estimates, and how many days with protests a municipality had during our period of interest. Our database accounts for 775 protests in 433 municipalities over the country (Table 1A), where more than 2.8 million people attended to demonstrations.

8 8 Two concerns arise from our approach. A first limitation, as pointed by Yanagizawa Drott. et al (2013) would be an eventual correlation between media coverage and the size of the municipalities: if media coverage of protests were more likely in big cities than in small cities, our results would be biased. Second, as protest attendance was measured by different sources, usually local police departments, we have limited access to the methodologies used to compute how many people showed up in each protest, and hence, to its precision. The first concern would be a problem if media coverage (through population) was correlated to protest incidence. If, however, small cities that had protests or that did not were equally ignored by media coverage, which seems to be the case, our estimates would not be biased. The second concern requires more attention, and is harder to mitigate when analyzing protests intensity. Thus, we focus our analysis on protest incidence, and leave protest intensity for descriptive and anecdotal exercises. Public Opinion: We use data from Latino Barometro s 2005, 2006, 2009 and 2010 survey waves in order to measure public opinion dimensions, like satisfaction with public services, perception of corruption, and preferences over institutional features. Moreover, Latino Barometro provides us information about social networks and media usage, like Facebook, Orkut and Youtube. Latino Barometro interviews more than 1200 people over 90 Brazilian municipalities, which allows us to collapse the database at the municipal level so that we can estimate the impact of several public opinion features on protests. Public Services Quality While Latino Barometro provides us data about satisfaction with public services, its limited size (90 municipalities) may threat the power of our estimations, and prevent us to analyze the determinants of protests in Brazil as a whole. Hence, we lie in IDEB (Education) and IDSUS (Health) indexes in order to measure the impact of public service s quality on protest incidence. In addition, since these indexes measure actual quality instead of satisfaction, we can extend our analysis into a broader evaluation of the impact of public services as a determinant of the demonstrations, as quality and satisfaction, though correlated, are not the same thing. The IDEB of a school is constructed through the multiplication between a performance indicator, the average grades in Math and Reading in the Prova Brasil, and the average school s pass rate, collected from Censo Escolar and collapsed in a 0-10 scale. The index is measured each two years, since We use the public school s municipal average in 2011 to access public school quality. The IDSUS uses administrative data, combining 24 indicators - such as the fraction of the population covered by the primary attention system, the cure rate in new cases of syphilis and leprosy and proportion of deaths in ICU admissions covering access and efficacy conditions of the public health system. Unfortunately, as the index started to be measured in 2011, we just have access to one year of data. In our estimations, we use the municipal average in the IDSUS as a proxy of public health quality. Census

9 9 We use data from the 2000 and 2010 Brazilian Census, organized and collapsed at the municipal level by the PNUD at the Atlas of Human Development in Brazil (ADH) to control for municipal demographical and socioeconomic characteristics. In addition, we build our first measure of computer usage measuring the proportion of households that own a computer in each municipality. After describing our datasets, we present in Table 1B our descriptive statistics. Empirical Strategy In this work we estimate whether public services quality influence negatively the incidence of the June Protests in Brazil, and how this influence depends on the availability of information and communication technologies. In order to identify a causal impact of public services quality on protests, we would need public services quality to be orthogonal to the incidence of protests: that would be the case if, for example, we chose through a lottery 200 municipalities to replace their best teachers and doctors by professionals of a lower quality, which for several reasons would be neither feasible, nor socially desirable. Thus, in the absence of an experimental intervention, our empirical strategy seeks to mitigate bias by controlling our estimates for observable municipal characteristics that, according to the social unrest literature, influence protest incidence, where X is a vector that include all these characteristics, which also implies that we can interpret our results as causal effects. We include in X the following variables: The logarithm of the municipal per capita GPD (log_gpd), in order to control for the state of the economy, as in Passarelli and Tabellini (2013), or the endowment of economic resources available in society; the municipal Gini coefficient, to control for redistributive conflicts (Passarelli, F., & Tabellini, G. (2013)) and heterogeneity on income; the logarithm of the municipal population (log_population), which is important to a) mitigate eventual biases caused by correlations between media coverage and the size of the municipalities, b) due to the fact that larger cities usually host not only more political institutions, but also more investments related to the 2014 World Cup, two remarkable targets of the June protests, and c) due to the coordination effects (Yanagizawa Drott, 2013) that lower the costs of attending demonstrations as the potential attendance to the acts becomes larger; the proportion of individuals between 15 and 24 years old (Young), as youngsters are the age group which attend more frequently to protests, as well as to control for the municipal age profile; the proportion of women (Women), to control for gender differences over municipalities; the municipal proportion of urban households (Urban), as social unrest is deeply correlated to urban density (DiPasquale, D., & Glaeser, E. L., 1998); the municipal average of years of schooling (Years_Schooling) as a proxy of human capital, which, as shown by (Campante, F. R., & Chor, D. 2012), happens to be a strong predictor of social unrest, and finally, to the municipal proportion of households that own at least a computer (Computer), which is an important organizational tool for collective action. We use data from the Brazilian Census 2010 to construct all the variables included in X. Our first step to understand whether public services quality affects protests is the regression of the municipal averages in the IDEB 2011 (IDEB) and IDSUS 2011 (IDSUS), our proxies for quality of public schools and hospitals, respectively, in the

10 10 dependent variable manifest, equal to 1 if a city had a protest between June 17 and 30, and 0 otherwise. Hence, in the next session we estimate using an OLS regression the following models: Where is a set of dummies for each of the Brazilian states, in order to control for regional differences, is a vector of municipal characteristics of municipality i, and is an error term. If both bad quality of public hospitals and schools increase the likelihood of a protest happens, we expect that < 0 and, respectively. The models we present in equations (1a) and (1b) estimate the impact of the quality of public services, but they fail to capture the dynamics between the availability of information and communication technologies and social unrest caused by poor public services. In the case of the June protests, information and communication technologies may have played a crucial role in the organization and dissemination of the demonstrations. In this sense, disparities in the availability of those technologies may affect the magnitude of the effects caused by poor public services quality. Thus, in our next models we interact IDSUS and IDEB with Computer, in order to account for those heterogeneities: Where and are interactions between the proportion of households that own computers, and our measures of public services quality, is an extension of the vector used in (1a) and (1b), with the difference that we take Computer from X and include it in the main equation, is a set of dummies for each of the Brazilian states and is an error term. We estimate the model using an OLS regression. We test whether the impact of quality increases on the availability of computers, thus, whether < 0 and. Our next models use data from Latino Barometro 2010 to estimate how satisfaction with public services affect protests. The following models differ from the equations presented until now in several aspects: first, they test whether satisfaction with public services, instead of quality, as in equations (1a) and (1b), affects protests. Second, although in the models presented above we test whether the effects of quality change according to computers availability, these measures do not tell us nothing about how people use computers. As Latino Barometro provides us data on Facebook, Twitter, Orkut and Youtube usage, which were the platforms massively used to organize, broadcast and disseminate the June Protests, we can test how social media usage in a given municipality affects the indignation caused by dissatisfaction with public services. Third, instead of education and health, we address a broader set of public services. We test the impacts of satisfaction with public schools, public hospitals, police, and public

11 11 transport. Fourth, our vector of controls, X, presents a few differences from X, as we include Social Media usage instead of the proportion of households that own computers, and use data from Latino Barometro to estimate the municipal proportions of women, youngsters and the municipal average of years of schooling. Finally, due to the reduced number of municipalities in our Latino Barometro sample, 88, we cannot use state dummies. We present our equations for satisfaction with public services using satisfaction with public schools as an example, as the estimation for satisfaction with public hospitals, police and transportation are analogous. We estimate the models using OLS regressions: Where is the interaction between satisfaction with public services and social media usage, is a vector of municipal characteristics, is an extension of without social media, and is an error term. In equation (3), we test whether low levels of satisfaction with public schools increase the likelihood of a protest to happen, thus, if. In equation (4), we test whether the effects of low levels of satisfaction with public schools on the likelihood of a protest to happen become larger as social media usage is more widespread in a given municipality, that is, if. Our estimates for corruption are analogous to equations (3) and (4), with the difference that we expect large levels of corruption perception to affect positively the chance of a protest to happen, hence, that and in equations (5) and (6), respectively. We estimate (5) and (6) using two different measures of corruption perception, Known a Corruption Act, which is the proportion of the individuals in a given municipality that said yes to the question Have you or someone in your family known of a corruption act?, and No Progress on Reducing corruption, which is the proportion of the individuals in a given municipality that answered little, or no progress at all in the question How much progress do you think has been made on reducing corruption in the State institutions during the last 2 years? Finally, we test the robustness of our findings by estimating equations (2a), (2b), (4) and (6) as we control for interactions between or measures of quality, satisfaction or corruption perception, and Young and Years of Schooling. This is important to rule out the possibility that the effects we find in the standard specifications might come instead from covariates deeply correlated both to protest incidence and social media usage, as Years of schooling and Young. In equation (7), we show how this robustness check applies for equation 4, as the estimation is analogous for equations (2a), (2b) and (6):

12 12 Results Public Services Quality In Table (2), we present our estimations of the equations (1a) and (1b), where we test whether low public services quality increases the chances of a protest to happen, that is, if < 0 and < 0, respectively. We show that while school s quality affects protest incidence negatively, we cannot reject that the quality of public hospitals have null effects on protests, once we control for municipal characteristics. In columns 1-3, we estimate the impact of public hospitals quality on protests. We find in column 1 a negative coefficient of IDSUS 2011, statistically significant at a 1% level. Nevertheless, the effects disappear as we include in columns (2) and (3) demographic controls and state dummies, respectively. We investigate in columns (4)-(6) the effect of school quality on protest incidence, using the IDEB 2011 as a proxy of school quality. We find a positive and statistically significant coefficient of IDEB in column (4). As we include demographic controls and state dummies in columns (5)-(6), the coefficient of IDEB becomes negative ( and ), and statistically significant at a 1% level. The magnitude of the effect of a standard deviation (.754) increase in the IDEB decreases the chance of a protest to happen in 1.71 percentage points, or a decrease of 21.3%, compared to the mean of protest incidence, which is 8.01%. We account for computer usage in our estimates of table (3) in order to test whether the impact of public services poor quality increase as a larger proportion of the population have access to computers (equations 2a and 2b), hence, if < 0 and. We find that increases in both IDEB and IDSUS affect protest incidence more negatively, as larger the computer access in a given municipality is. Columns (1) and (4) present our naïve specifications, with no control variables. We add controls in columns (2) and (5), and state dummies in columns (3) and (6). In columns (1)-(3), we show that the interaction between IDSUS 2011 and computer is negative and statistically significant at a 1 % level in all of our specifications, with the coefficient decreasing slightly in magnitude as we include controls in column 2 (-.202) and state dummies in column 3 (-.176). We present our estimates for school quality in the columns (4)-(6). The coefficient of IDEB 2011 X Computer - columns (4) and (5) - are negative and statistically significant at a 1% level, and correspond to the values of -.2 and -.125, respectively. The effect decreases in magnitude to , statistically significant at a 5% level, as we include state dummies in column (6). We verify the magnitude of our results estimating the effects of a standard deviation increase in the IDSUS 2011 (.837) in distinct levels of computer usage. In column (3), which is our preferred specification due to the complete set of controls and state dummies, a standard deviation increase in the IDSUS 2011 raises the likelihood of protest incidence in 3.02 percentage points [F( 1, 5526) = 22.16; P-Value =.000], or 38.8%, compared to the mean of protest incidence (.078), in municipalities in the 10 th centile of the computer usage distribution (.057), and decreases the likelihood of protest

13 13 incidence in 2.17 percentage points [F( 1, 5526) = 13.16; P-Value =.00003], or 28% compared to the mean of protest incidence in Brazilian municipalities. A standard deviation increase in the IDEB 2011 in column (6) decreases the likelihood of a protest happens in.57 percentage points [F( 1, 5320) =.69; P-Value =.4078], or -7.14% compared to the mean of protest incidence, for municipalities in the 10 th centile of computer usage. However, due to the low level of the F-Statistic, we cannot reject that IDEB helps to explain protests at that level of computer usage. Nevertheless, for municipalities in the 90 th centile of the computer usage distribution, a standard deviation increase in the IDEB 2011 decreases the likelihood of a protest happens in 3.19 percentage points [F( 1, 5320) = 16.04; P-Value =.0001], or % compared to the mean of protest incidence. Public Services Satisfaction In Table (4) we present our estimations of equation (3). We show that satisfaction with public services has no effects on protest incidence. In columns (1) (2), we estimate the effects of satisfaction with public hospitals where we find no effects on protest likelihood. In columns (3) (4), we estimate the effects of satisfaction with public schools. We find, in column (3), that a percentage point decrease in the proportion of satisfaction with public schools increase the likelihood of a protest happens in.59 percentage points, statistically significant at a 1% level, although this effect disappears as we include demographic controls in column (4). Finally, in columns (5)-(6) and columns (7)-(8), we estimate, respectively, the effects of satisfaction with the police and satisfaction with transport on protest incidence. We find no effects in any of these estimates. In table (5), we present our estimation of equation (4). Here, we test whether the effects of satisfaction with public services are more negative when social media usage is larger, that is, < 0. We show that, when accounting for social media usage, the satisfaction with a variety of public services affects protest incidence. In addition, as social media usage spreads over the population, the effects of low satisfaction with public services on increasing protest likelihood intensify. In columns (1)-(2), we estimate the effects of public hospitals satisfaction using the dummy manifest as a dependent variable. Our estimate in column (1) shows that the coefficient of interaction between satisfaction with public hospitals and social media usage is negative (-2.84) and statistically significant at a level of 1%. The second column adds demographic controls, and even though the interaction coefficient decreases in magnitude (-1.64), the direction and the significance of the coefficient remains unchanged. In order to enlighten the magnitude of our finds from columns (1)-(2), we evaluate the effects of satisfaction with public hospitals over different centiles of the social media usage distribution. In the column (2), our favorite specification, a standard deviation increase in the quality of public hospitals (.272) increases the likelihood of a protest happens in 8.94 percentage points [F( 1, 77) = 3.76; P-Value =.056] for municipalities in the 10 th centile of the social media usage distribution (.071) and decrease in 13.3 percentage points [F( 1, 77) = 4.82; P-Value =.031] for municipalities in the 90 th centile of the social media usage distribution (.57). Given the sample means of protest incidence, those effects

14 14 correspond to, respectively, an 18.7% increase and a 27.8% decrease in the protest likelihood. We present our estimates for satisfaction with public schools in the columns (3)-(4). We find a negative and statistically significant at 1% level coefficient (-2.632) in the interaction between satisfaction with public schools and social media usage in column (4). The direction of the effect ( ) holds for the controls introduced in column (4), now statistically significant at a 5% level. In terms of magnitude, a standard deviation increase in satisfaction with public schools (.273) increases protests in 4.9 percentage points, in the 10 th centile of the social media usage distribution, and decreases protests in 18 percentage points in the 90 th centile of social media distribution in column (4), which implies, compared to the mean of protest incidence, respectively an increase of 10.3%, and a decrease of 37.7%. Finally, our results for satisfaction with the police and satisfaction with transport lie in columns (5)-(8). The coefficient of the interaction between satisfaction with the police and social media is negative (-3.835) and statistically significant at a 1% level in (5), nevertheless, losing its significance in column (6) with the inclusion of controls. In the columns (7)-(8), we examine the coefficients of the interaction between satisfaction with transport and social media. The coefficient is negative (-2.825) and statistically significant at a 1% level in the column (7), and it holds negative (-1.311) in column (8), despite relevant differences both in magnitude and significance, now, statistically significant at a 10% level. The effects of a standard deviation increase in satisfaction with transport on protest incidence (column 11) are of an increase of 5.18 percentage points [F( 1, 77) =.95; P-Value =.033] in the 10 th centile of the social media usage distribution, or 10.8% compared to the mean of protest incidence, and percentage points [F( 1, 77) = 3.36; P-Value =.071] in the 90 th centile of the social media usage distribution, or -26.8% compared to the mean of protest incidence. Corruption In table 6, we present our estimations of equation (5), where we test whether increases in our two proxies of corruption perception, No Progress on Reducing Corruption (columns 1 and 2) and Known of a Corruption Act (columns 3 and 4) affect positively the likelihood of a protest to happen. In column (1), the coefficient of No Progress on Reducing Corruption is positive (.374) and statistically significant at a 10% level, but loses its significance as we include controls in column (2). The coefficients of Known of a Corruption Act are positive in both of our estimates (.748 and.223), statistically significant at a 1% and at a 10% level in column (3) and column (4), respectively. The magnitude of the effects of a standard deviation (.274) increase in Known of a Corruption Act are of an increase in 6.1 percentage points in the likelihood of a protest happens, or 12.79% compared to the mean of protest incidence. Next, we evaluate whether our conclusions from Table 6 change as we interact our measures of corruption perception with social media usage. We present our results in table 7. We show, in opposition to our results from table 6, a positive significant effect in the interaction between No Progress on Reducing Corruption and Social Media Usage (Columns 1 and 2), while we cannot reject that the interaction between Known of a Corruption Act and Social Media Usage has no effect on protest incidence (Columns

15 15 3 and 4). In column (1), No Progress on Reducing Corruption X Social Media Usage is positive (1.804) and statistically significant at a 10% level. The coefficient decreases slightly in magnitude (1.408), statistically significant at a 5% level, if we add controls in column (2). A standard deviation increase in No Progress on Reducing Corruption (.2702) decreases the likelihood of protest incidence in municipalities at the 10 th centile of the social media usage distribution in 4.98 percentage points [F( 1, 77) =.71; P- Value =.409], or % compared to the mean of the dependent variable - although the low F-Statistic indicate the limitations of the interpretation of this estimate. Nevertheless, for municipalities in the 90 th centile of the social media usage distribution, a standard deviation increase in No Progress on Reducing Corruption increases protest likelihood in percentage points [F( 1, 77) = 6.43 ; P-Value =.0132], or 29.42%. Robustness Checks In order to access the robustness of our findings, we estimate whether our results hold as we include interactions between our measures of public services, and variables Years of Schooling and Young (equation 7). This is important because the additional effects that arise from high levels of social media usage may come from other sources, as both social media usage and protests may have a strong correlation with Years of Schooling and Young. In other words, if younger and more educated individuals are more willing to protest as well as to use social media, as strongly suggested by evidence Campante, F. R., & Chor, D. (2012), our results could be capturing this effect, rather than the effects of social media usage per se, which is our hypothesis. In table (8) we present our results for public services quality, measured through IDSUS and IDEB, as in Table (3). We can see that our results remain mostly unchanged, compared to our findings from Table (3). The coefficient of the interaction IDSUS 2011 X Computer in column (1) decreases slightly (-.167), but remains statistically significant at a 1% level. In column (2) the coefficient of IDEB 2011 X Computer decrease as well, though not in an expressive magnitude, now statistically significant at a 10% level, which is less than the 5% level found in the original model (Table 3, column 6). Thus, our results hold the inclusion of the controls of potentially cofound variables. Next, we evaluate how our estimates of the impact of public services satisfaction found in Table 5 change with the inclusion of our new controls. We present our results in Table 9, where we find, in general, negative coefficients of a slightly larger magnitude, with larger statistical significance in some cases. Furthermore, the coefficient of Satisfaction with Police X Social Media Usage, which had lost its significance with the inclusion of controls in the column 6 of Table 5, now becomes negative (-1.862) and statistically significant at a 5% level. Lastly, in Table 10 we present our robustness checks for our results on corruption perception from Table 7. Again, our conclusions do not change at all. The magnitude of the coefficient of No Progress on Reducing Corruption X Social Media Usage increases to 1.949, still statistically significant at a

16 16 1% level (Column 1). Moreover, as it happened in our model from Table (7), we cannot reject that Known of a Corruption Act has null effect on protest incidence. Conclusion Our findings provide several insights about the June Protests. First, we find that municipalities that have poor public services quality, and which are less satisfied with public services, have, on average, more protests. We show that this is particularly true for education and public health, while less consistent for police and public transport, for example. Second, we find that the dynamics between public services quality and protests is very heterogeneous, depending crucially of the local levels of social media usage and hence, of the availability of information and communication technologies. This is a very straightforward conclusion, as social media may lower the costs of collective action in several ways, as in the information and coordination effects proposed by Yanagizawa-Drott, D. (2010). For municipalities that have a high penetration of ICTs, like São Paulo or Rio de Janeiro, decreases in both satisfaction and public services quality increase drastically the likelihood of a protest to happen. Nevertheless, in municipalities where these technologies are barely available, a decrease in quality or satisfaction may have no effects on protests, or on increasing the likelihood of a protest to happen, as we find in a few of our models. A plausible mechanism for the positive effects of public services quality on protests is that better public services may push political activism, which makes sense in the case of school quality, for example, given evidence found by Campante, F. R., & Chor, D. (2012) of the impacts of schooling in the Arab Spring. Third, we find that municipalities that face high levels of corruption perception have on average more protests, although these findings are less robust, as the results for our two proxies of corruption, Known of a Corruption Act and No Progress on Reducing Corruption, differ in all our estimations, for no trivial reasons. As happened with public services, we show that the impact of corruption on protests varies a lot given the levels of social media usage. Our findings make sense given a vast literature that shows how the responsiveness of political action to corruption depends considerably of the availability of communication technologies that allow the diffusion of the corruption episodes Ferraz, C., & Finan, F. (2008). In municipalities in the 90 th centile of the social media usage distribution, a standard deviation increase in No Progress on Reducing Corruption may increase the likelihood of a protest happening in 29.4% (Table 7, Column 2), compared to the mean of protest incidence. However, in municipalities in the 10 th centile of the social media distribution, the same standard deviation increase in No Progress on Reducing Corruption decreases in 10.44% the likelihood of a protest happens. Nonetheless, our trust in this last estimate is very limited as we can reject the null hypothesis in a F-Test. The main limitation of our work is that we lack a good and testable identification strategy, which is a source of concern, once we may have endogenous relations between public services and protests. For example, politicians from cities that protest more frequently may be more responsive to this form of social unrest, and hence, systematically improve the quality of public services over the time. Thus, we are unable

17 17 to interpret our findings as actual causal relations. However, our results do provide the first description of the correlations between plausible causes and the incidence of the June protests toward Brazilian cities. Another limitation is that due to the expressive amount of missing data, as well as our limited access to the precision and the methodologies used to estimate protest attendance, our data prevents us to analyze the intensity of the June protests. We did try to estimate the effects on the amount of protesters and the number of days on which protests occurred, but with no success. We intend to extend this work in several directions. The first main challenge would be to find exogenous variations in public services quality in order to access the causal impact of quality in the June Protests. Second, it would be nice to expand the data available for the June Protests into a broader panel of protests and other forms of informal political action, like strikes, over the time in Brazil, for since the June projects took place, this kind of phenomenon has been remarkably more frequent. Finally, to enlighten the effects of the June protests in political, institutional, and public services outcomes would teach us to what extent public policy is responsive to forms of political action other than voting. We leave all these questions for future research. References Campante, F. R., & Chor, D. (2012). Schooling, political participation, and the economy. Review of Economics and Statistics, 94(4), Campante, F. R., Durante, R., & Sobbrio, F. (2013). Politics 2.0: The Multifaceted Effect of Broadband Internet on Political Participation (No. w19029). National Bureau of Economic Research. CASTELLS, Manuel. Networks of outrage and hope: Social movements in the internet age. John Wiley & Sons, De Mesquita, E. B. (2010). Regime change and revolutionary entrepreneurs. American Political Science Review, 104(3), DellaVigna, S., List, N. J. A., Malmendier, N. U., & Rao, N. G. Voting to Tell Others. version Feb 2013 DiPasquale, D., & Glaeser, E. L. (1998). The Los Angeles riot and the economics of urban unrest. Journal of Urban Economics, 43(1), Esteban, J., Mayoral, L., & Ray, D. (2012). Ethnicity and conflict: An empirical study. The American Economic Review, 102(4), Ferraz, C., & Finan, F. (2008). Exposing corrupt politicians: The effects of Brazil's publicly released audits on electoral outcomes. The Quarterly Journal of Economics, 123(2), Gentzkow, M. (2006). Television and voter turnout. The Quarterly Journal of Economics, 121(3), Gerber, A. S., Green, D. P., & Larimer, C. W. (2008). Social pressure and vote turnout: Evidence from a large-scale field experiment. American Political Science Review, 102(1), 33. Gladwell, M. (2010). Small change. Why the revolution won t be tweeted. The New Yorker, 4.

18 18 Glaeser, E. L., Sacerdote, B. I., & Scheinkman, J. A. (2003). The social multiplier. Journal of the European Economic Association, 1(2-3), Granovetter, M. S. (1973). The strength of weak ties. American journal of sociology, Harvey, D., Maricato, E., Zizek, S., Davis, M., Maior, J. S., Iasi, M.,... & de Oliveira, P. R. (2013). Cidades rebeldes: Passe livre e as manifestações que tomaram as ruas do Brasil. Boitempo Editorial. Hirschman, A. O. (1970). Exit, voice, and loyalty: Responses to decline in firms, organizations, and states (Vol. 25). Harvard university press. Lohmann, S. (1994). Information aggregation through costly political action.the American Economic Review, Machado, F., Scartascini, C., & Tommasi, M. (2011). Political institutions and street protests in Latin America. Journal of Conflict Resolution, 55(3), Madestam, A., Shoag, D., Veuger, S., & Yanagizawa-Drott, D. (2013). Do Political Protests Matter? Evidence from the Tea Party Movement*. The Quarterly Journal of Economics, 128(4), Moseley, M., & Layton, M. (2013). AmericasBarometer Insights: Neri, M. C. (2008). A nova classe média. Rio de Janeiro: FGV/IBRE, CPS. Olson, M. (1965). logic of collective action; public goods and the theory of groups Passarelli, F., & Tabellini, G. (2013). Emotions and Political Unrest. Ponticelli, J., & Voth, H. J. (2011). Austerity and anarchy: Budget cuts and social unrest in Europe, Available at SSRN Romer, P. M. (1996). Preferences, promises, and the politics of entitlement. In Individual and social responsibility: Child care, education, medical care, and long-term care in America (pp ). University of Chicago Press. Romero, D. M., Meeder, B., & Kleinberg, J. (2011, March). Differences in the mechanics of information diffusion across topics: idioms, political hashtags, and complex contagion on twitter. In Proceedings of the 20th international conference on World wide web (pp ). ACM. Soares, E. L. (2013) The Ground Shakes in the Country of Inequalities and Paradoxes. Available in Tullock, G. (1971). The paradox of revolution. Public Choice, 11(1), Voth, H. J. (2012). Tightening Tensions: Fiscal Policy and Civil Unrest in Eleven South American Countries, Available at SSRN WINTERS, Matthew S.; WEITZ-SHAPIRO, Rebecca. Partisan Protesters and Nonpartisan Protests in Brazil. Journal of Politics in Latin America, v. 6, n. 1, 2014.

19 19 Woo, J. (2003). Economic, political, and institutional determinants of public deficits. Journal of Public Economics, 87(3), Yanagizawa-Drott, D. (2010). Propaganda and conflict: Theory and evidence from the rwandan genocide. Working Paper, Harvard University. The data we compiled from G1 is available in the following link:

20 Figures and Tables 20

21 21 Figure 3: Social Media Usage in Latin America (2010) Figure 4: Satisfaction with Public Services in Latin America Figure 5: Perception of Corruption in Latin America (2010)