Brecracy

Analyzing Social Media Momentum
India’s 2011-12 Anticorruption Movement Prepared for U.S. Government Office of South Asia Policy

By Sasha Bong Kenneth Chung Karen Parkinson Andrew Peppard Justin Rabbach Nicole Thiher

Workshop in International Public Affairs Spring 2012

©2012 Board of Regents of the University of Wisconsin System All rights reserved. For additional copies: Publications Office La Follette School of Public Affairs 1225 Observatory Drive, Madison, WI 53706 www.lafollette.wisc.edu/publications/workshops.html [e-mail] The Robert M. La Follette School of Public Affairs is a teaching and research department of the University of Wisconsin–Madison. The school takes no stand on policy issues; opinions expressed in these pages reflect the views of the authors.

Table of Contents
List of Tables and Figures .................................................................................... v Foreword .............................................................................................................. vii Acknowledgments .............................................................................................. viii Executive Summary ............................................................................................. ix 1. Introduction ....................................................................................................... 1 2. Background ....................................................................................................... 1 2.A. Basic Theory of Social Networking ............................................................ 2 2.B. Social Media’s Emerging Role in Social Activism..................................... 2 2.C. Corruption in India and the Lokpal ............................................................. 3 2.D. India’s 2011 Anticorruption Movement and Anna Hazare......................... 4 2.E. The Role of Social Media in India’s Anticorruption Movement ................ 6 3. Facebook Data ................................................................................................... 8 3.A. Facebook Usage Terminology and Information Limitations ...................... 8 3.B. Descriptive Statistics of Selected Facebook Pages ..................................... 9 3.C. Social and News Media Volume Comparison .......................................... 10 4. Social and News Media Comparisons ........................................................... 11 4.A. March 25 – April 4: Why the Sub-Committee Meeting Failed ................ 12 4.B. December 22 –January 5: The Movement Changes Direction ................. 14 4.C. Implications for Further Analysis ............................................................. 16 5. Variable Identification and Development..................................................... 16 5.A. Determining Significant Facebook Events ............................................... 17 5.B. Determining Significant Phases ................................................................ 18 5.C. Real-World Event Variables ..................................................................... 19 5.D. Content Themes ........................................................................................ 20 5.E. Day of the Week ........................................................................................ 21 6. Model Specification......................................................................................... 22 6.A. Dependent Variables ................................................................................. 22 6.B. Independent Variables ............................................................................... 23 6.C. Regression Models .................................................................................... 23 7. Model Estimation and Inferences .................................................................. 24 7.A. Regression Model Inferences .................................................................... 27 7.B. Differences Between Natural Log of Likes and Natural Log of Comments Regression Results .................................................................................... 28 8. Summary and Recommendations for Further Work .................................. 29 Appendix A: Narrative Timeline ....................................................................... 33 Appendix B: Gathering Facebook Data............................................................ 36 Step-by-Step Instructions for Data Collection .................................................. 36 Summarizing Facebook Data in Excel with Visual Basic Macros ................... 37

Appendix C: Facebook Mechanics .................................................................... 39 Terms ................................................................................................................ 40 Appendix D: News Media Data Collection ....................................................... 43 Case Study of Problems with this Data Collection Method: December 5 – December 9 ............................................................................................... 43 Appendix E: Codebook....................................................................................... 46 Identifiers .......................................................................................................... 46 Activity Variables ............................................................................................. 47 Theme Variables ............................................................................................... 47 Content Variables ............................................................................................. 48 Phase Variables ................................................................................................. 48 Corruption/Phase Interaction Variables ............................................................ 49 Lokpal/Phase Interaction Variables .................................................................. 49 Hazare/Phase Interaction Variables .................................................................. 50 Hunger Strike/Phase Interaction Variables ....................................................... 51 Demonstration/Phase Interaction Variables...................................................... 51 Government/Phase Interaction Variables ......................................................... 52 Real-World Event Variables ............................................................................. 53 Appendix F: Summary Statistics ....................................................................... 54 Appendix G: Predictive Regression Model ...................................................... 57 Appendix H: Regression Results ....................................................................... 58 Appendix I: Diagnostic Tests and Limitations of the Models ......................... 61 Appendix J: Methodological Ideals and Realities ............................................ 62 Privacy Concerns .............................................................................................. 62 Cell Phones ....................................................................................................... 62 Twitter ............................................................................................................... 62 Computer System Requirements ...................................................................... 63 Demographic Information................................................................................. 63 Language ........................................................................................................... 63 Switch to Timeline ............................................................................................ 63 Appendix K: Alternative Means and Methodologies ...................................... 65 Ethical Considerations ...................................................................................... 65 Technical and Budgetary Limitations ............................................................... 66 Making Friends ................................................................................................. 66 Alternative Data Sources .................................................................................. 66 Analytical Toolkits ........................................................................................... 67 Works Cited ......................................................................................................... 73

List of Tables and Figures
Figure 1. Indian Facebook Users by Age Group .................................................... 7 Figure 2. Facebook Activity per Day: April 2011 – February 2012 ..................... 10 Figure 3. Social and News Media Volume, April 2011 – February 2012 ............ 11 Figure 4. Facebook Activity Threshold, 2011-2012 ............................................. 17 Table 1. Significant Events of the Anticorruption Movement, all 2011 ............... 18 Table 2. Phases of the Anticorruption Movement, 2011-2012 ............................. 18 Figure 5. Facebook Activity and Parliamentary Sessions .................................... 20 Figure 6. Change in Themes by Phase on AH and IAC Pages ............................. 21 Figure 7. Facebook AH, IAC Day-of-the-Week Activity..................................... 22 Table 3. Basic Model Summary Results, Statistically Significant Variables of Main Interest ...................................... 25 Table 4. Interaction Model Summary Results, Statistically Significant Variables of Main Interest ...................................... 26 Table D-1. Volume Comparison Between News Media and Facebook Activity .................................................................................. 44 Table D-2. Comparison of News Media Keywords.............................................. 45 Table F-1. Identifiers ............................................................................................ 54 Table F-2. Activity Variables ............................................................................... 54 Table F-3. Theme Variables ................................................................................. 54 Table F-4. Content Variables ................................................................................ 54 Table F-5. Phase Variables ................................................................................... 55 Table F-6. Corruption/Phase Interaction Variables .............................................. 55 Table F-7. Lokpal/Phase Interaction Variables .................................................... 55 Table F-8. Hazare/Phase Interaction Variables .................................................... 55 Table F-9. Hunger Strike/Phase Interaction Variables ......................................... 56 Table F-10. Demonstration/Phase Interaction Variables ...................................... 56 Table F-11. Government/Phase Interaction Variables .......................................... 56 Table F-12. Real-World Event Variables ............................................................. 56 Figure G-1. Idealized Predictive and Responsive Temporal Correlations in Ln Likes, Relative to Government Action ................................................ 57

v

vi

Foreword
The La Follette School of Public Affairs at the University of Wisconsin–Madison offers a two-year graduate program leading to a Master of Public Affairs or a Master of International Public Affairs degree. In both programs, students develop analytic tools with which to assess policy responses to issues, evaluate implications of policies for efficiency and equity, and interpret and present data relevant to policy considerations. Students in the Master of International Public Affairs program produced this report for the U.S. government’s Office of South Asia Policy. The students are enrolled in the Workshop in International Public Affairs, the capstone course in their graduate program. The workshop challenges the students to improve their analytical skills by applying them to an issue with a substantial international component and to contribute useful knowledge and recommendations to their client. It provides them with practical experience applying the tools of analysis acquired during three semesters of prior coursework to actual problems clients face in the public, non-governmental, and private sectors. Students work in teams to produce carefully crafted policy reports that meet high professional standards. The reports are research-based, analytical, evaluative, and (where relevant) prescriptive responses for real-world clients. This culminating experience is the ideal equivalent of the thesis for the La Follette School degrees in public affairs. While the acquisition of a set of analytical skills is important, it is no substitute for learning by doing. The opinions and judgments presented in the report do not represent the views, official or unofficial, of the La Follette School or of the client for which the report was prepared.

Melanie Frances Manion Professor of Public Affairs and Political Science May 2012

vii

Acknowledgments
Throughout this project, we benefited from the support of a few individuals. Dr. Melanie Manion, our advisor for this project, provided us with invaluable advice and guidance. The La Follette School of Public Affairs Publication Director, Karen Faster, assisted with meticulous editing of our work. We are grateful to Professors David Weimer and Jon Pevehouse for their help developing statistical models. Finally, we would like to thank Keith and the U.S. government Office of South Asia Policy for the opportunity to work on this project.

viii

Executive Summary
Social media sites such as Facebook are continuously expanding the number of connections among individuals and groups around the world. Organizers of social movements are taking advantage of these tools to spread their message and garner support. To look at the role of social media in social movements, this case study analyzes Facebook activity related to India’s anticorruption movement in 2011 and early 2012. We examine how the anticorruption movement used Facebook, how frequency of user activity correlated to the type of content being shared on Facebook, and ways in which real-world protest events and government actions affected user activity. We find strong correlations between important real-world protest events and substantial increases in user activity on Facebook. Using Facebook activity as a measure of social engagement, we offer conclusions and implications for relevant actors as they seek to monitor and manage the flow of social movements. We began our analysis by systematically gathering and electronically coding the content of 8,103 top-level posts created on Facebook pages for Anna Hazare and India Against Corruption from February 2011 through February 2012. Top-level Facebook posts provide access to direct messages from movement organizers to supporters. To evaluate the response to these individual posts, we collected information on the number of likes and comments on a given post as a measure of user activity. We observe large increases in user activity at the same time as significant protest events or government action related to the anticorruption movement. We believe this relationship allows us to treat the level of user activity as a measure of social engagement with the anticorruption movement at any given time in our sample period. We also believe this relationship reveals Facebook as a strong source of information that can provide insight into the themes that resonate most with supporters as the anticorruption movement looked to increase social engagement and on-the-ground participation. We analyzed thematic content of Facebook posts and used multivariate statistical models to determine how and how much the content influenced user activity. We developed variables and coded posts for significant themes present on Facebook. We ran regressions using these themes as our independent variables or main interest and user activity (measured by natural logs of likes and comments on Facebook posts) as our dependent variables. We ran several versions of these regressions, including additional independent variables for post content, day of the week of post creation, and occurrence of real-world protests or government action. We conclude that analysis of Facebook and other social media content can be useful to relevant actors to a social movement. People can use social media content to gauge the status of a movement and to identify the goals it seeks to attain. This content can assist actors outside the social movement, such as government officials, by offering insight into how a movement’s leaders might act or react to specific actions or events. Moreover, our analysis identifies specific ix

themes that resonated most with followers of the anticorruption movement. Increased use of these themes might have helped garner additional attention, support, and user activity, with implications for how movement organizers or outside actors could have increased social engagement with the movement. We believe a major benefit of this analysis is that it can serve as a starting point for future analyses of social movements’ use of social media. Continued refinement of the metrics we developed would allow them to be used to analyze similar social movements in real time. This implication is key, as relevant actors would want to be able to react quickly and effectively to steer, repress, or encourage certain aspects of the movement. This report lays the groundwork for use of the tools necessary for understanding social media content and changes in user activity. Applications of these metrics to future movements could provide further insight into the potential causal linkages between social media content and on-the-ground activity. Furthermore, as Facebook records and publishes the identity of each user who likes or comments on a post and links the activity to that user’s profile, researchers may be able to develop a tracking mechanism to record the frequency of likes and comments per individual, catalogue and correlate the temporal and thematic nature of those interactions, and explore the profiles of the most-engaged or most-influential individuals. In so doing, the researcher could determine the extent of an individual’s own friend circle. The researcher could learn about an individual’s stated attendance or role in protest activities and ability to mobilize friends to attend or participate in another way. Skilled computer programmers and library science personnel should be able extract this information through the native Facebook Application Programming Interface and organize it in an analytically useful way.

x

1. Introduction
Social media, a group of Internet-based applications facilitating the creation and sharing of messages, pictures, and videos by Internet users, continues to be a growing phenomenon. The popularity of online profile sites such as Friendster, Myspace, and SixDegrees has given way to a new wave of social media. Facebook, initially designed for access by users with college (.edu) e-mail addresses when launched in 2004, opened to the general public and, as of early December 2011, had more than 800 million users worldwide.1 Twitter, launched in March 2006, announced on its fifth anniversary that it hosted more than 140 million tweets a day and more than 1 billion a week.2 Through social media, people are connected to each other around the globe in a way never experienced before. More than 10 percent of the world’s 7 billion people are connected by Facebook alone.3 Many more non-Facebook account holders have profiles on Twitter, Google+, or one of many other global social media sites. This report analyzes Facebook activity during India’s 2011-12 anticorruption movement to determine how social media information can be used to better understand social engagement through social movements. After the background section, we begin our analysis by comparing the content of Facebook activity with the content of articles on web sites of two Indian newspapers in India. We find the two media types discuss the same events of the anticorruption movement but differ greatly in tone and language. Given these differences, we conclude that Facebook activity is a distinct information source to be analyzed. We systematically gathered and electronically coded 8,103 top-level posts on Facebook pages for Anna Hazare and India Against Corruption between February 2011 and February 2012. We theorize that particular features of Facebook posts, such as substantive thematic content, affected the volume of user activity on the pages during that time. We develop variables that measure these features and use multivariate statistical models to estimate their relationships with Facebook activity volume, our dependent variable. The results of these statistical analyses form the basis of our conclusions. We discuss the predictive ability of social media in terms of ebbs and flows of real-world action in social movements. We also discuss ways in which social media content provides information about the workings of social movements and the implications for individuals looking to manage, promote, or repress a similar movement.

2. Background
Social media have not only changed social networking, they provide a valuable tool for political and social organization and activism. When combined with a political problem, such as corruption among public officials, social media offer great potential for social movement participation, useful to the leaders who organize or emerge from such movements. This section explores the role of social media in India’s 2011 anticorruption movement. 1

2.A. Basic Theory of Social Networking
The study of social networking, whether online or in person, focuses on social capital. Social capital is the aggregation of actual and virtual resources an individual or group attains via networks built from meeting other people or groups.4 The underlying principle of social capital is that in extending one’s social network, a person or group then can draw on resources that people or groups in their network possess or can access. These resources can vary from information to relationships with other people; from a group’s perspective, resources increase capacity to organize members across organizations or causes.5 An important component of building an electronic social network, through Internet sites such as Facebook, is the ability to build two kinds of social capital: bonding and bridging. Bonding social capital involves a closer relationship among people, as typically found among family, friends, and close-knit communities. Bridging social capital is more heterogeneous and slightly more informal than bonding. Bridging social capital typically involves extending networks to more diverse people, in terms of personal characteristics or location; associations are based on one-time meetings or common causes, rather than close, personal relationships. As approximately 80 percent of social networking users join groups, opportunities for bridging social capital are likely to grow.6

2.B. Social Media’s Emerging Role in Social Activism
Groups with varying goals and missions have taken advantage of social media applications to attract and connect with members. Compared to news media, direct, widespread communications via social media can reach more people across larger geographic areas and can convey greater amounts of information.7 Online groups and forums also provide simpler means of organizing because they require significantly less physical effort to recruit members and distribute information. Petitions and meeting information can be distributed electronically, rather than making phone calls or distributing paper materials by mail.8 Groups and movements increasingly use Internet-based communications to sustain themselves through member recruitment and fundraising.9 An additional strength of social media, especially Twitter, is accessibility via cell phones and other handheld devices, for lightweight, portable communication.10 The types of people social media can reach are important for social movements. A barrier to social media sites can be Internet access, as it is not universal and it requires some basic technical skills and resources. These limitations may be less widespread for some key political groups, such as younger or middle class citizens. Social media sites are advantageous in this case, as a significant number of users are young, so these sites serve as conduits for organizations to encourage coveted youth interest and involvement. A drawback of social media is that false information and rumors can quickly spread. The content of posts, when not carefully considered and vetted, can

2

quickly lead to misinformation and controversy. Additionally, increased awareness and interaction via social media do not always result in real action. Social media sites allow for easy communication among members, but communication does not always lead to support and action.11 Finally, in places where Internet penetration rates are low, social media may not reach large segments of the population. Social media can be useful for organizing large numbers of individuals quickly, but what makes a social media campaign successful is unclear. Some governments have used censorship and regulation in an attempt to block activists from networking. In such an environment, social media can be used as organizing tools and open discussion forums about issues that may influence political change.12 Information posted on sites can also influence perceptions about another country. As in the case of the Mavi Marmara in Israel, online social media can be used to garner international support for policy decisions.13 Social media outlets can help measure public opinion of government behavior and help anticipate public uprising. In 2011, social media appear to have played significant roles in organizing and energizing social change movements, such as the Arab Spring, Occupy Wall Street Movement, and the 2011-12 anticorruption movement in India.

2.C. Corruption in India and the Lokpal
Political corruption is an ongoing problem in India, acknowledged domestically and internationally. Transparency International ranked India 95 out of 182 on its 2011 Corruption Perceptions Index, with a score of 3.1. India scored better than North Korea and Somalia, which had scores of 1. New Zealand ranked No. 1 with a score of 9.5. Additionally, Transparency International reported that more than 50 percent of Indian respondents disclosed they paid bribes to use basic public services, which indicates a relatively high level of actual corruption.14 Corruption can lead to decreased economic development, as theft diverts money for public services and infrastructure from its original purpose. It can also cause foreign investors to avoid new investments, as they may wish to avoid paying bribes levied by corrupt officials. India’s 2005 Right to Information Act called for increased transparency and required that the government disclose requested information to Indian citizens, allowing them to expose corrupt acts. However, this legislation does not directly address corruption, and complaints against corrupt officials often go unanswered.15 Several highly publicized scams involving the Commonwealth Games, the Indian Premier League, and the telecommunications industry helped spark more recent interest in curbing corruption.16 To combat corruption problems, India’s Parliament has debated establishing an independent commission with the authority to investigate and punish corrupt government officials. A lokpal (ombudsman) is a government official or organization that investigates individual complaints against public officials. However, differences between the government-drafted Lokpal bill and a version written by Indian activists caused political tension.

3

Legislators have proposed versions of the Lokpal bill since 1968 but it has not passed due to disagreements over jurisdiction and coverage of select government officials. The first Lokpal bill lacked enough support to survive a vote and failed not only in that introduction, but eight subsequent attempts as well, “none of which made it to a parliamentary vote.”17 A review in Economic and Political Weekly stated that the bill was “widely welcomed” at its 1985 introduction in Parliament. However, this “welcome” was short-lived. Controversy resulted, as the bill included the Indian prime minister as an investigable government official. Many critics thought the prime minister should be excluded.18 Continued calls by Indian citizens for a Lokpal bill have resulted in repeated introductions of legislation in Parliament. However, as in 1985, controversy in the 2010s over which public officials should be included under the provision of the new anticorruption body produced debates that ultimately blocked the bill’s passage. This controversy is easily understood: the individuals drafting the law are the people whom the lokpal would scrutinize. Thus the 44-year delay results from continual debate among ruling parties and individual reluctance to effectively self-police, given the significant potential gains from corrupt acts.

2.D. India’s 2011 Anticorruption Movement and Anna Hazare
In 2010, following major corruption scandals, the Indian government drafted a version of a Lokpal bill. Officials convened the Group of Ministers to consider this government measure for tackling corruption.19 Many citizens and social activists considered the proposed measure weak, as it did not cover the prime minister, members of Parliament, and cabinet ministers. Dissatisfaction gave rise to a national protest movement in 2011. We present a timeline of the 2011 Indian anticorruption movement’s major events in Appendix A. The face of the movement has been 74-year-old Anna Hazare of Talegan Siddhi, a small village in the state of Maharashtra. Hazare’s fame comes from his work as a community organizer, particularly his efforts to decrease alcoholism and ensure water access for individuals in rural areas.20 His tie to national issues came through his work in 2003 advocating for the Freedom of Information Act in India, which eventually passed in 2005.21 He has led anticorruption protests for two decades. The 2011 anticorruption movement, organized under the name India Against Corruption, has by far been the largest he has led. Hazare and others believed that the government’s version of the Lokpal bill was too weak because the ombudsman it would establish could not investigate actions of elected officials. The government argued that an ombudsman was a good idea, but that it would be too powerful if it was able to investigate elected leaders.22 Discussions among Anna Hazare, Prime Minister Manmohan Singh, and a selection of fellow ministers failed to result in an agreement on a Lokpal bill. In protest, Hazare began a hunger strike on April 5, 2011, to demand a stronger bill.23 Large protests erupted in support of Hazare. Protesters hoped to pressure the Indian government into forming a Joint Drafting Committee for a Lokpal bill with five ministers and five civil society members.24 This protest sparked extensive

4

discussions across news and social media about whether the bill should include the entire government. It also brought many people to the streets in support of Hazare’s movement.25 After four days of Hazare fasting, the government agreed to form the Joint Drafting Committee, and Hazare ended his fast. The committee met nine times over the course of the next two months. Midway through this series of meetings, significant differences arose between the ministers and civil society members. These differences were primarily over whom the Lokpal would have authority to investigate, the obstacle to earlier versions of the bill.26 Members were further divided after police arrested social activist Baba Ramdev and forcefully removed his supporters, who were peacefully protesting against corruption in Ramlila Maidan (a public square). In protest, the civil society members of the Joint Drafting Committee boycotted the sixth meeting.27 Differences about the strength of the Lokpal bill remained throughout the rest of the Joint Drafting Committee meetings, which ended on June 21.28 During a July 3 all-party meeting on the Lokpal bill, the political parties jointly agreed to bring a “strong and effective Lokpal Bill” to the next session of Parliament.29 At the end of July, the Lokpal committee completed a second draft, but it did not include top government officials. After approval from the Cabinet, legislators brought this version of the Lokpal bill before the Lok Sabha, the lower house of Parliament on August 4.30 Hazare rejected this version of the bill and announced a second hunger strike.31 Hazare announced that this second strike “to the death” would begin on August 16 and only end if the government agreed to pass a Lokpal bill that included the entire government. In response, the New Delhi police denied Hazare permission to stage an indefinite fast at the Jantar Mantar observatory. Police restrictions looked to limit the protest to one day and a maximum of 2,000 participants.32 At the same time, government officials began a smear campaign against Hazare that included accusations of corruption, which were quickly withdrawn.33 Prime Minister Manmohan Singh used his August 15 Independence Day speech to criticize Hazare and discuss the government’s efforts to pass a strong Lokpal bill.34 After refusing to agree to police instructions for his second hunger strike, Hazare and key members of India Against Corruption were arrested early in the morning of August 16, before Hazare started his hunger strike, and taken into “preventive” custody. Hazare’s supporters released a video that had been prepared in case of arrest. It called for fasts, peaceful protest, and a jail bharo, where protesters sought arrest so as to fill India’s prisons. Nearly 600 demonstrations and protests erupted across the country in response. Instead of Hazare’s arrest slowing the anticorruption movement, as politicians had hoped it would, the movement exploded. By the afternoon of August 16, large protests had occurred all over India, and 1,300 people had been arrested as part of the jail bharo.35 Hazare and his followers were officially released from jail later that day, but Hazare refused to leave until given permission to fast publicly without policeimposed restrictions.36 The government spent most of the next day negotiating

5

with him. After making a deal with police, Hazare left jail on August 18 and continued his hunger strike at Ramlila Maidan. Tens of thousands of people gathered in Ramlila Maidan to show their support and fast with Hazare, while thousands more supported Hazare online through social media. The way the government handled the situation angered many people, which left the ruling party humiliated.37 During the following days, the Indian government reduced its personal attacks on Hazare and instead suggested that a “foreign hand” was driving the protest movement. The government also repeatedly asked Hazare to end his fast.38 After further discussions with Hazare and debates in Parliament, the government agreed to debate all versions of the Lokpal bill.39 This agreement prompted Hazare to end his hunger strike on August 28.40 Despite strong support for the movement and its ability to gain several key concessions from the Indian government, the stronger version of the Lokpal bill developed in August has not passed. On December 27, 2011, the Lok Sabha, Parliament’s lower house, approved the government’s latest version of the bill, which Hazare thought was still too weak. He began another hunger strike in Mumbai to protest the government moving forward with the weaker Lokpal bill, and this strike drew a crowd of 10,000 supporters, a marked decrease from previous rallies. Hazare ended his third hunger strike, claiming health concerns, shortly after the Lok Sabha passed a weaker version of the Lokpal bill.41 In the Rajya Sabha, the upper house of India’s Parliament, both parties introduced more than 180 amendments to this version of the bill during a 13-hour session. Ultimately, they recessed without a vote.42 Promises were made to again take up the bill in March 2012 during the next legislative session. In the subsequent months, Hazare has been in and out of the hospital but has promised to continue the fight for a strong Lokpal bill when his health improves.43

2.E. The Role of Social Media in India’s Anticorruption Movement
Anticorruption movement organizers and supporters used social media to quickly broadcast information and organize protests. Indians also used social media to show support for India Against Corruption (IAC) and Anna Hazare, indicated on Facebook by “likes” on posts. In the first four days of its existence, IAC had 116,000 fans on its community Facebook page.44 People created many other Facebook pages, and individual social media users debated, posted statuses, and uploaded videos and photos throughout the movement. Social media analyst Gaurav Mishra estimates that the total online support for the movement was around 1.5 million people.45 Facebook hosts multiple Anna Hazare-related pages in English and Hindi, with tens of thousands of followers and supporters. The official IAC Facebook page had more than 500,000 followers as of February 7, 2012.46 Users can follow and access information about the anticorruption movement through applications for smart phones and other mobile devices. The IAC smart phone application has as many as 50,000 users.47 The organization used all these outlets to publish photos of Anna Hazare fasting, pro-Lokpal rallies, and examples of corruption. During

6

this social media onslaught, Hazare gained support from other prominent Indian activists, as well as the general populace. For a social movement to be able to use social media effectively to advance its cause, large parts of the population must have access to the Internet and people must be able to use the Internet and social media freely. The non-governmental research organization Freedom House scores the Internet in India as “mostly free” in its 2011 Freedom of the Net evaluation.48 India established the Internet Technology Act in 2000, and a 2008 amendment gave the government authority to block websites and Internet content, as well as outlaw offensive or inflammatory content.49 Comments about religion or caste can be particularly volatile in India, given its history of religion-based riots. Google gives government officials information about Internet protocol addresses and service providers when requested.50 Private blogs have had to remove posts upon threat of legal action.51 In the first six months of 2011, the Indian government requested 358 removals from Google, mostly from Orkut and YouTube, the majority for content criticizing the government.52 According to the World Bank, 5.3 percent of the total Indian population had Internet access in 2009.53 With a population of approximately 1.1 billion people, this usage rate means that more than 58 million Indians use the Internet.54 Fifty-six percent of Internet users use Facebook.55 Of those, 73 percent are men;56 50 percent are 18 to 24 years old,57 as pictured in Figure 1. Facebook posts and news reports show that the anticorruption movement centered on urban areas. The Facebook demographics suggest the movement engaged urban men but left large segments of the population out of the debate.
Figure 1. Indian Facebook Users by Age Group
50% Percentage of User Base 40% 30% 20% 10% 0% 13‐15 16‐17 18‐24 25‐34 35‐44 45‐54 55‐64 65+ Age Group
Source: Socialbakers.com

7

3. Facebook Data
The emergence of social media has changed how people engage with each other and with social movements. As more people access social media, the greater the opportunity to measure the extent to which Facebook activity reflects social engagement in a particular social movement. We determined the themes, events, and other variables that seemed to drive social media activity throughout the year. We then identified and analyzed relationships between Facebook activity and real-world events by using Facebook user activity as a reflection of social engagement with the movement. To analyze Facebook activity related to India’s 2011-12 anticorruption movement, we collected data on all top-level posts from February 2011 through February 2012 on Facebook pages for Anna Hazare and IAC. We detail how we collected the data in Appendix B: Gathering Facebook Data. Data on likes, comments, and shares indicate levels of social engagement in the movement. Top-level post content can be coded and used to identify the themes present throughout the movement and how Facebook discussions changed over the span of our analysis. We also analyzed articles about the movement published in two online English-language newspapers, The Hindu and The Times of India. We identified these articles through key word searches on the two news sites. We sought to determine whether Facebook activity mirrors news media activity in content and volume. A cross-referencing of Facebook top-level post content against traditional news stories helped us to distinguish points where Facebook content provides information different from what readers would encounter by following only the online newspaper accounts. Our reading of these sources determined that Facebook content provided a window into the intentions and workings of the movement, while news media accounts may have a very different message. The difference in content may be due to movement leaders distributing one message to news media as they try to win the support of a more neutral audience, while providing different or more detailed messages to Facebook users who would be assumed, on whole, to support the movement.

3.A. Facebook Usage Terminology and Information Limitations
The publisher of a Facebook page posts information on his or her page with a brief amount of text, photos, or links to websites. For our analysis, these are top-level posts. Responses to these top-level posts are what we describe as user activity. Three measures of user activity are available on Facebook: “likes,” comments, and “shares.” We only use likes and comments to measure user activity; we do not include shares in our definition. We describe these features briefly below and at length in Appendix C: Facebook Mechanics. Likes: The level of user interaction required to like a post is a single click, generating an increment of one additional like for that post. Facebook allows each user to like each unique post only once. Therefore, the total number of likes

8

for each post represents the total number of unique users who have viewed that post and clicked the like option. Comments: Comments involve a deeper level of user interaction and are independent of liking a top-level post. The act of adding a comment increases the comments count by one. A user may add an unlimited number of comments per post. Comments demonstrate that a unique post has been viewed and that the view the user wishes to share his or her thoughts on the post or on preceding comments with other users. Shares: While viewing a post, a user may click the share option that presents the ability to copy the post, along with an open comment box, to his or her personal wall, a friend’s wall, or into a private message. Sharing generates a cross-linked, independent thread in the sharing user’s profile page or the alternative location of his or her choosing. Facebook added shares as a feature late in our sample period. No data on shares are available for early in the sample period, and user familiarity was too low to have confidence in shares as reliable measures of social media momentum in the short time it was available during the sample period. Thus, we omitted shares from our analysis. A limitation of these measures is that users may like or comment on a post at any time after it has been posted to the page. Thus, our figures reflect the number of likes and comments on posts created on a specific date, not the number of likes and comments added on that date. However, as most likes and comments on a given post do occur soon after posting, they are strong indicators of social engagement in the movement on a given date.

3.B. Descriptive Statistics of Selected Facebook Pages
We selected two Facebook pages for data collection based on the volume of their followers: the IAC page (http://www.facebook.com/IndiACor), which boasts 550,000+ visitors who have liked the page, and the Anna Hazare page (AH page) (http://www.facebook.com/annahazare), which has 415,000+ liking visitors. Our search revealed no other Facebook pages related to the 2011-12 anticorruption movement that came close to this volume of followers. The two pages vary in some characteristics of author and visitor activity. The number of posts per day differs dramatically: the AH page averages two posts per day (range 0 – 20) and the IAC page averages 19 (range 3 – 157) during the time period under analysis. Also divergent are the numbers of likes per post and comments per post between the two pages. The AH page shows greater activity per post, averaging 1,201 likes per post (range 6 – 5,615) versus the IAC page’s 613 (range 16 – 1,880). The same pattern holds for comments per post. The AH and IAC pages often share content. Posts on one page are often crossposted on the other’s page. Post themes follow common patterns across pages. Theoretically, the IAC page encompasses a broader anticorruption movement,

9

while the AH page would focus more narrowly on the work of Anna Hazare. However, the period surveyed in this report is almost exclusively focused on Anna Hazare’s activities. For these reasons, we pooled AH page and IAC page data when analyzing social media user activity and relevant events. A high number of posts per day does not necessarily result in a high number of likes and comments, as seen in Figure 2. We attribute this disparity to “post overload”: the number of posts in a given day can exceed a user’s interest and time availability to continuously monitor and participate in the discussion. Prior to the August protest event, significantly more posts per day were required to elicit the same number of likes and comments. We believe this variation reflects a period of “audience building” during which more individual users became aware of the Facebook pages and began to respond to content. Once this audience had been built up, a dedicated group of AH and IAC page followers responded predictably to new page activity. Figure 2 demonstrates spikes in both the number of comments and likes that occurred at certain points. These spikes in social media activity occur at the same time as significant real-world events highlighted in our background section.
Figure 2. Facebook Activity per Day: April 2011 – February 2012
250 Likes, Comments per Day, Thousands 200 150 100 50 0 Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb AH, IAC Likes per Day AH, IAC Posts per Day
Source: Authors’ calculations

180 160 140 120 100 80 60 40 20 0 AH, IAC Comments per Day

3.C. Social and News Media Volume Comparison
The level of Facebook posts and activity provides one picture of the anticorruption movement and the type of content being posted to public Facebook pages. To develop alternative perspectives on the movement, we collected information about the volume of references to the movement in news media sources. We used two of the largest online English-language newspapers in India, The Times of India and The Hindu. We chose these sources, as other select media outlets do not have online archives that cover our entire sample period, and we were limited to sampling only English newspapers due to language restrictions.

10

Posts per Day

Further limitations and issues with the data are discussed in Appendix D: News Media Data Collection. Figure 3 demonstrates that the number of daily news stories referencing the movement mirrors the trend in social media activity throughout our sample period.
Figure 3. Social and News Media Volume, April 2011 – February 2012
Facebook Activity per Day, Thousands 250 200 150 100 50 0 Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb AH, IAC Likes per Day AH, IAC Comments per Day Times of India, The Hindu News Articles per Day
Source: Authors’ calculations

300 250 200 150 100 50 0 News Articles per Day

4. Social and News Media Comparisons
A relatively strong correlation exists between volume trends of Facebook and online versions of newspapers The Times of India and The Hindu, as demonstrated in Figure 3. Content, however, can be very different across sources. One source may give a different depiction of an event than the alternate source, or an event may be referenced in one source and not the other. The two examples that follow illustrate how an analysis of news media alone misses some information available through social media. Movement organizers are likely to recognize that information disseminated in news media is likely to go to a relatively more diverse audience. This understanding of different audiences may prompt organizers to put forth a different message or use a different tone across media. On the IAC and AH pages, the audience comprises mostly individuals who support the movement. The Facebook messages directed toward supporters may provide a better window into the true message of the movement, rather than polished statements tempered for news media. Additional information gained from social media can help to explain the real-world events of India’s 2011-12 anticorruption movement. Reading content on Facebook reveals insights into movement plans. Although we completed this analysis after significant social media activity on the movement subsided, we demonstrate how reading of the Facebook content in real time could have helped individuals predict a movement’s likely responses.

11

4.A. March 25 – April 4: Why the Sub-Committee Meeting Failed
As the April 5 start date of Anna Hazare’s threatened hunger strike approached, the government attempted to persuade IAC to reconsider. Organization members and a government sub-committee scheduled a meeting for March 28 to discuss an IAC-written draft of a Lokpal bill. The only way to get Hazare to cancel the fast and protest was if an agreement could be reached to move forward on passing IAC’s version of the bill. Specific to this meeting, progress would require the government agreeing to form a joint drafting committee, a body made up of IAC members and government officials, to draft official legislation mirroring the IAC version of the bill. IAC is quoted in the media after the meeting, stating that it had been “much anticipated.” Indo-Asian News, March 28 “The much anticipated sub-committee meeting on Lokpal Bill failed to yield any positive results with the government showing no intention of forming a joint committee,” IAC said in a statement. [sic]58 Despite IAC’s apparent anticipation of the meeting, the event received no coverage on the IAC or AH pages until a March 25 post announced that Hazare would not attend. This lack of promotion was a sharp contrast to other events created, supported, or attended by members of IAC that attracted multiple posts per day as advertisements and reminders. IAC Facebook post, March 25, 6:51 a.m. Breaking news from IAC Newswire!!!! Anna Hazare refuses to attend meeting with sub-committee. [sic]59 A post later that morning stated the movement’s negative feelings regarding further meetings with the prime minister or other government committees. IAC Facebook post, March 25, 12:11 p.m. After meeting with PM ended inconclusively, there is no point in meeting more ministers and committees – we demand Jan Lokpal and we will have it!!!! [sic]60 Despite declaring the pointlessness of further meetings, IAC sent a contingent of members to meet with the government on March 28. Afterward, IAC released a statement describing the lack of progress on inducing the government to form a joint drafting committee on which general public members held at least half of the seats. The release stressed that joint committees had been formed for other legislation. The government insisted that this type of action had no precedent and that the sub-committee did not have the power to establish a joint committee. Following the failed meeting on March 28, the prime minister expressed his “disappointment” that Hazare would continue with his fast, though he did state

12

that he respected “Hazare and his mission.” The government sought to depict IAC members as stubborn and unwilling to negotiate what would be included in the bill. Times of India, April 4 The sub-committee headed by defence minister A.K. Anthony met colleagues of Hazare “but the interactions proved fruitless as the activists were insisting on the government accepting their draft in full,” the PMO said.[sic]61 Conversely the AH page criticized the government. AH Facebook post, April 4, 12:17 p.m. Instead of welcoming Anna’s suggestions and try to bring systemic changes they are disappointed with this man who is willing to risk his life to fight against this disease – shame on them!!!! [sic]62 Though the March 28 meeting may have been an attempt by the government to appease IAC, Facebook content leads us to believe that IAC never actually considered entering the meeting with the intent to negotiate to an agreement that would have cancelled the fast and protest. The AH page was created on March 28, the day of the meeting. The first post was an invitation to the event page for the fast to be held in early April. Additionally, on the IAC page, information about Hazare’s fast consumed a large percentage of posts through March, leading up to the meeting. Posts do not indicate a tempering of plans in regard to Hazare’s fast in the event an agreement was reached. Language in the posts does not project a feeling of plans for “if,” but rather a sense of plans for “when.” IAC Facebook post, March 25, 11:09 a.m. Delhi / NCR Volunteer Meet on Sunday 27th March – to mobilise maximum people for Anna Hazare’s fast. Lets create a Tahrir Chowk at Jantar Mantar. All Delhi volunteers, please join us. [sic]63 This response suggests once the movement had raised enough support and reached a level of organization for this specific protest, it planned to go ahead regardless of government response or attempts at appeasement. The movement taking on this attitude toward this meeting appears indicative of a relative “tipping point” at which government discussions could not appease the movement. IAC decided that too much time and energy had been expended to organize and mobilize; cancelling the event could kill movement momentum. As support grew, so did the size of the concession needed by the government to cancel the fast. Monitoring the AH and IAC pages in the days leading up to the meeting would have revealed that the government was not taking strong enough action to lead to a cancellation of the protest. We see from Facebook a distinct choice by the movement to push ahead with preparation for the protest. If only the two newspapers, The Times of India and The Hindu, had been considered, the picture

13

would be one of the movement and the government each blaming the other for being inflexible and at fault for the lack of progress in negotiation.

4.B. December 22 –January 5: The Movement Changes Direction
In the days leading up to the debate and vote in Parliament on December 27, the AH and IAC pages called people to join protests planned for the time of the vote in response to the “weak version” of the Lokpal bill under consideration. Numerous posts on the AH and IAC pages asked people to join the jail bharo, and one post announced that more than 100,000 people had signed up to be arrested on December 29. The movement used strong words leading up to the protest: IAC Facebook post, December 25, 4:54 a.m. We are not afraid of being monitored. If a strong Lokpal is passed, we’ll go ahead with our next movement. If the government tries to pass a weak Lokpal, we’ll oppose it tooth and nail. We’ll oppose the party in power in center, until a strong Lokpal is passed. [sic] As Hazare began his fast, turnout was disappointing, though some credit this lack of participation to rumors that people who participated in the jail bharo would be penalized with job loss or loss of visas. IAC also worried that the government may have disrupted communication by limiting cell service. IAC Facebook post, December 27, 1:01 a.m. Serious communication issues in Mumbai – Airtel moblice phones and data cards at MMRDA not working – Govt’s dirty tricks, sabotage? [sic] As the bill passed the lower house, Hazare pledged to continue the fast. AH Facebook post, December 27, 11:49 a.m. Anna Hazare has fallen sick and is running 102 degrees temperature. He has refused to end his fast. Deep concern for his health continues as the team urges him to end his fast. Messages of immense worry are pouring in. IAC Facebook post, December 27, 11:37 p.m. Anna will continue on his resolve to fast for a strong Lokpal! IAC Facebook post, December 27, 11:45 p.m. People pressure made Parliament work, debate and pass Lokpal Bill after 42 years. This is no mean feat but the victory is incomplete. If Lokpal is to be functional CBI must be brought under Lokpal and pro-corruption sections must be removed. The battle is on, Jail Bharo from 30 December will ensure that India gets a strong anti-corruption law. Get ready!! 14

Despite these strongly worded pledges to continue, IAC posted quotes from Hazare calling off the fast and jail bharo less than six hours later. IAC Facebook post, December 28, 4:48 a.m. “Jail Bharo is suspended for now!” Anna IAC Facebook post, December 28, 4:49 a.m. “Due to what is happening in the Parliament, the Anshan [fast] will finish today and our aim for now will be voter awareness and the elections in the five states,” Anna. Organizers called off the fast due to Hazare’s failing health and the passage of a bill not supported by the movement. The fast drew disappointing numbers and failed to prevent the government from taking action; unlike its responses in April and August, the government did not move to appease the movement. This decline in interest led to an announcement regarding the new direction of the movement. IAC Facebook post, December 28, 4:28 a.m. “We have two years till the National elections. We will tour all over the country till then and spread awareness about the rights of people and ask them not to vote for the corrupt and criminals.” Anna These posts are a stark change in the direction of the movement. The decision seems to have been made quickly, given the abrupt change in tone and content of Facebook posts within a six-hour period. This change in direction may be why the movement seemed to lose momentum from this point forward. The movement threw away a month of buildup to the fast and jail bharo. It may have changed focus too quickly however, as organizers did not seem prepared to guide followers in the next step. IAC Facebook post, January 5, 2012, 9:21 p.m. The anti-corruption movement is at the crossroads today. Where do we go from here? …. If we do not go for the tour of election-bound states, what should we do? Should Anna go for another fast? But the government has already indicated that if people participating in the movement don’t translate into votes, they don’t care. Some have suggested we should form our own party. But we neither have the will nor the capacity. The excerpt above demonstrates that the movement broke down in some ways. It was not using Facebook to advertise specific events or call for volunteers and supporters to attend planned events. Rather, at this point, the movement looked to find direction itself. We believe this change in tone allowed for members to divide as they chose how to move forward. The dilution of support from various groups prevented the cohesion necessary to move forward. Without events to attend or a strong leader issuing statements about the way ahead, people lost faith and interest in the movement. Decrease in user activity on Facebook reflects this

15

disengagement. We do not see another spike in user activity during the remainder of our sample period. The previous April, IAC had refused to cancel the fast, given the amount of time and organization that had gone into it, for fear of killing momentum. In December, the movement chose the opposite path and canceled the fast and jail bharo. It was not able to recover momentum following this cancellation. It appears that a government action strong enough to force the movement to change its plans was able to kill momentum and quell support of the movement. This shift could be identified as the “tipping point” at which government action, rather than appeasement, affected the movement’s ability to move forward.

4.C. Implications for Further Analysis
Cross referencing Facebook posts with articles from The Hindu and The Times of India illustrates that gaps may exist between social and news media coverage of India’s 2011-12 anticorruption movement. Social media posts provide additional information as well as nearly instantaneous insight into changes in plans and directions of the anticorruption movement. Due to the availability of information, the number of users, and the amount of interaction these users have with the movement via social media, we used Facebook posts on the AH and IAC pages to measure movement action as we developed analytical models. Such use of posts makes sense given the increases in volume of user activity at the same time that real-world social engagement is occurring on the ground as a part of the movement. We used user activity and post content to measure social engagement and to identify the themes that keep individuals invested in the anticorruption movement at a given point in our sample period. Given massive amounts of content, it would be nearly impossible to monitor all Facebook posts and crosscheck content with newspaper articles in a reasonably short period of time through simple reading of posts and articles. This reality drives the need for a tool to more quickly analyze this content.

5. Variable Identification and Development
An initial analysis of user activity across our sample period demonstrated that a strong correlation existed among the level of user activity, the volume of news media coverage of the movement, and the occurrence of real-world protests and action by the movement. This correlation allowed us to use likes and comments on our selected Facebook pages as a measure the social engagement with the movement at a given point in time during our sample period. In our analysis we sought to determine the variables that affected the level of user activity with toplevel posts on Facebook and, based on the existing correlation, gain insight into how Facebook themes and content influenced social engagement with the movement. To measure effects on Facebook user activity we identified and developed variables for significant movement events, significant phases, content themes in individual posts, and days of the week to be used in analytical models

16

to determine the drivers of social engagement over the span of this movement. Subsequent parts of Section 5 discuss how we developed these variables for use in our quantitative analysis. Section 6 describes the regression models we estimated. Section 7 presents and interprets the results of our analysis.

5.A. Determining Significant Facebook Events
Graphical representation of the data, such as Figure 3, identifies dates during our sample period when Facebook user interaction on the AH and IAC pages spiked. Through comparison of the days of these Facebook interaction spikes and our timeline of real-world activity, we observed that most of these spikes seem to be closely correlated with real-world events. Therefore, we used spikes in Facebook activity to identify important events that may warrant further analysis. To distinguish between increases in daily Facebook interaction due to important events and more minor events (or noise), we developed a “signal-to-noise” threshold above which we define an event as “significant.” We set that threshold at one standard deviation above the mean of the natural log of likes and the natural log of comments. Additionally, to qualify as a significant event, the threshold must be crossed in terms of both likes and comments. Analysis by this metric yields nine significant Facebook events throughout the period of analysis, as seen in Figure 4.
Figure 4. Facebook Activity Threshold, 2011‐2012
12 11 Ln Likes per Day 2 10 9 8 7 Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Ln AH, IAC Likes per Day Mean + 1 s.d. Threshold
Source: Authors’ calculations

1

4 9

10 8 7 9 8 7 6 5 Ln AH, IAC Comments per Day

5

6

To establish effective date ranges around significant events, we reviewed the pooled likes and comments data and determined the date at which activity volume began an upward trend. This pre-peak data “valley” is our event start date. Where likes and comments differ in start date, we selected the earlier date of the pair. Event peaks are classified as the date of highest activity. Event end dates are 17

Ln Comments per Day

3

classified as the first day in which likes or comments fall below the level of its initial activity level, or valley, following the peak. The exception to these definitions is event 4, which was sustained for a longer date range and had multiple peaks and dips before dropping below threshold. We defined event 4’s start date as the first valley below threshold, its peak as the highest point during its date range as an event, and its end date as the first dip low enough to cross its initial starting valley. Identified events are listed in Table 1.
Table 1. Significant Events of the Anticorruption Movement, all 2011
Event 1 2 3 4 5 6 7 8 9 Month Apr Jun Aug Aug Sep Oct Dec Dec Dec Start Date 3 3 3 12 1 16 5 10 18 Peak Date 8 8 4 16 2 18 6 11 27 End Date 14 11 8 31 10 22 9 14 31

Source: Authors’ calculations

5.B. Determining Significant Phases
To incorporate the passage of time and to differentiate among distinct stages in the movement, we developed a phase variable based on real-world events. In general, this variable divides the movement by the three main protests that took place in April, August, and December and their interim periods. The periods of increased Facebook user activity around these three protests are phases 2, 4, and 6. The remaining phases, 1, 3, 5, and 7 are the lull periods, or periods of relatively lower user activity surrounding these protest events during the course of our sample period. The phases are listed in Table 2.
Table 2. Phases of the Anticorruption Movement, 2011‐2012
Phase 1 2 3 4 5 6 7 Start Date Jan 27 Apr 3 Apr 14 Aug 12 Sep 1 Dec 21 Dec 31 End Date Apr 2 Apr 13 Aug 11 Aug 31 Dec 20 Dec 30 Feb 19

Source: Authors’ calculations

18

5.C. Real-World Event Variables
We observed in Figure 3 that spikes in Facebook user activity and actions on the ground have a strong correlation. Due to this finding, we believed inclusion of real-world events in the regression models was important. The two types of realworld activities of interest are on-the-ground protests and government actions in response to (or related to) the movement. These variables help determine the importance of real-world events in the level of social media engagement. Organized protest events define the real-world protest event variable. We identified the dates of protests using our narrative timeline. As protests are usually planned events, a significant level of Facebook discussion and promotion about the protest happened on the day before it was to occur, and discussions about major protests likely continued afterward. As a result, we expanded our protest period to include a day before and after the organized protests. The result is a binomial variable coded as one for the dates of protests, including the date before and after, and zero otherwise. We divided the real-world government action variable into two categories, negative and positive government actions from the anticorruption movement’s perspective. We defined positive government actions as the government trying to appease the movement or to work toward meeting its demands. These actions include government officials meeting with protest leaders and with each other to discuss the Lokpal bill and the government making agreements with Hazare. Negative government actions are deliberate efforts to repress or unravel the movement and actions to which the movement strongly objects. These actions include the arrest of protestors, explicit attempts to discredit Hazare, or action to move forward with a version of the Lokpal bill that the movement believes is too weak. We assign each variable a one for each day of government action and the day after. Each code includes the day after an action to encompass subsequent Facebook discussion. The day preceding government action was not included because the dates for many government actions were not known to the public in advance, making it less likely that significant discussion in anticipation of these events occurred. All days not fitting this description of government action or following day are coded as zero. Last, we included a variable for whether Parliament was in session on a given day. The Parliament conducted the Budget, Monsoon, and Winter sessions during our sample period.64 This variable is coded as one on days Parliament was in session and as zero otherwise. Inclusion of this variable helped to control for the fact that protests were more likely to be scheduled to coincide with the times when government was more likely to take action on the Lokpal bill. The relevance of inclusion is seen in Figure 5, as two of the three largest spikes in user activity occurred while the parliament was in session.

19

Figure 5. Facebook Activity and Parliamentary Sessions
AH, IAC Likes and Comments per Day, Ten Thousands 2.5 2.0 1.5 1.0 0.5 0.0 Feb Mar Apr May Jun AH, IAC Likes Jul Aug Sep Oct Nov Dec Jan Feb AH, IAC Comments Parliamentary Session 0 Budget Monsoon Winter 1

Sources: PRSIndia.org and authors’ calculations

5.D. Content Themes
We developed variables to identify and code the themes in each Facebook post in our sample period. The purpose was to gain insight into the types of information that drove social media response. Using likes and comments as measures of social engagement, we believe this process helps to identify issues followers of the movement considered important and specific themes central to the message of the movement. We identified key words by studying the movement and reading a sample of Facebook posts. We organized these keywords into six major theme categories: Hazare, corruption, Lokpal, hunger strike, demonstration, and government. A full specification of the thematic keywords used to sort posts into these six theme categories (or no category) is presented in Appendix E. At the same time, we developed five post-content type categories: video, blog, news (traditional), web, and photo. The prevalence of the defined themes in the count of Facebook top-level posts differs across phases. Figure 6 shows the percentage of posts incorporating an individual theme in each phase. In phase 1, the themes of corruption and demonstration dominate; in phase 2, the majority of posts are about Hazare as he gains attention during his hunger strike. During phase 3, the Lokpal bill is the most common theme in Facebook posts, which is when the Joint Drafting Committee was meeting to discuss the bill. In phases 4 and 5, posts about Hazare are again the most common following his arrest and protest, and his continued rise in notoriety as the face of the movement. In phase 6, both the Lokpal bill and Hazare are important subjects in the posts, reflecting the bill’s passage in Parliament and Hazare’s ongoing fast. Phase 7 is primarily about Hazare, mainly discussing his hospitalization.

20

Parliamentary Session

Figure 6. Change in Themes by Phase on AH and IAC Pages
50% Percentage of Posts Containing Theme 40% 30% 20% 10% 0% 1 2 3 4 Phase Corruption Hunger Strike Lokpal Demonstration Hazare Government 5 6 7

Source: Authors’ calculations

5.E. Day of the Week
The average number of posts per weekday varies significantly, with the highest posting activity occurring on Thursday and the lowest on Monday. We depict this posting schedule in Figure 7. Also represented is the percentage of weekday posts with a significant number of likes and comments. This designation of “significant” reflects post engagement, indicating those posts that have accumulated a total number of likes and comments equal to at least the sample period average plus one standard deviation. Upon cursory review, we see that posts on Tuesday and Wednesday generate a greater percentage of posts with significant likes and comments than other days. Thursdays and Fridays reveal both a greater average number of posts per day and a lower percentage of posts that prove most engaging. This information is useful in presenting the possibility that a weekly rhythm of posting by the movement and user activity may exist.

21

Figure 7. Facebook AH, IAC Day‐of‐the‐Week Activity
Average Number of Posts per Weekday Percentage of Posts per Weekday that are Most Significant (Ln Likes, Comments Above Threshold) 23 9.5%

22

8.5%

21

7.5%

20

6.5%

19 Mon Tue Wed Thu Fri Sat Sun AvgPosts PctMostSigPost

5.5%

Source: Author’s calculations

6. Model Specification
To determine the effect of the variables described in section 5 on Facebook user activity during our sample period, we estimated multivariate statistical models. We ran regressions as ordinary least squares with panel-corrected standard errors accounting for first order autocorrelation. These models measure effects on Facebook user activity in terms of the percent change in likes or comments, utilizing various combinations of our independent variables and interaction terms. The results from these models describe the relationship between these independent variables and level of user activity. These relationships inform our conclusions on how varying social media content influences our measure of social engagement. This model assumes a post’s theme that increases the number of likes or comments is one that resonates with individuals involved with action on the ground during that time period.

6.A. Dependent Variables
Using the likes and comments to measure Facebook user activity, as described in section 3, we had four options to use as dependent variables: number of likes on a post, number of comments on a post, or the natural log of either of these variables. From these options, we selected the natural log of likes and natural log of comments as our dependent variables. We believe that taking the natural log of the variables for use as the dependent variable in our regression analysis makes sense for several reasons. Using this method allows us to realize a more normal distribution of our dependent variable. In a log-linear model, the presence of a given content variable is represented by a coefficient reflecting the percent, rather than unit, change in the dependent variable. Use of percent change helps us compare the effects of a specific theme or event on user activity across the

22

movement. Given that support grew for the movement over time and more users joined our sample Facebook pages, percent change is a more useful measure. For example, an increase of ten likes in phase 1 may be large, but only a drop in the bucket during phase 6, but if we find that the same theme caused a 5 percent increase in activity in those phases, we know that the variable has a similar effect, just at a different order of magnitude. Liking and commenting on a post are fundamentally different actions that a Facebook user can take. Although the number of likes and comments appears to be highly correlated over time, these two dependent variables could provide somewhat different results in our regression models. Commenting on a post represents a higher level of commitment than liking it, as the user must form and publicly share an opinion rather than make a single click. However, likes may be a stronger measure for several reasons. A Facebook user can only like a post once, so the number of likes is the number of people who liked the post. Facebook users can comment on a post an unlimited number of times, so a post could potentially have a high number of comments but a low number of commenters. We also know that a like is a demonstration of positive agreement or support for a post. Comments can be positive or negative about the post or not be related to the top-level post at all, which makes interpretation of results less clear.

6.B. Independent Variables
We developed a number of independent variables for use in our models. The variables of greatest interest are the theme variables reflecting the message being presented by Facebook content at any given point in the movement. We interacted these theme variables with phase variables, allowing us to analyze changes in theme effect on user activity. Other variables included in the models are: post characteristics, government actions, major protests, the day of the week the post was created, a categorical variable for the phase in which the posts were written, and interactions between each theme and phase. The interaction variables help us identify how the effects of these themes differ across the phases of the movement. The codebook describing these variables more fully is presented in Appendix E, and summary statistics are presented in Appendix F: Summary Statistics.

6.C. Regression Models
For this analysis, we developed two regression models using different combinations of our independent variables to analyze factors driving Facebook user activity. The basic versions of our model (models 1 and 2) include the variables for the theme of the post, the type of media the post used, real-world events that took place, the Facebook page where the post was made, the phase the post was written, and the day of the week it was posted. The interaction versions of our model (models 3 and 4) incorporate the interaction of theme and phase variables to analyze potential differences in the importance of different themes over time. We ran each of our

23

regression models twice, once with the natural log of likes as the dependent variable and once with the natural log of comments as the dependent variable. In addition to these models, we developed a basic model to test for temporal causality between Facebook activity and government action. The results of this model, elaborated on in Appendix G, suggested a strong Facebook response to negative government action. We see no persuasive link to suggest that Facebook activity alone compels government action—either positive or negative. As such, these models are not the focus of our work. They should be viewed as supplement to this analysis and as a base for further evaluation of the causal relationship between social media content and real-world action. Our four econometric models are: 1 2:

3

4:

∗

∗ ∗ ∗

∗

∗

7. Model Estimation and Inferences
This section presents the results and outlines general implications of our analysis. We discuss the importance of the identified themes as they relate to Facebook user activity and how the effect of specific themes changed over the course of our sample period. In our discussion of results in section 7.A., we focus on the models using the natural log of likes as the dependent variable (rather than the natural log of comments) because of the advantages in interpretation discussed in section 6.A. Still, as Facebook only offers users a like button, Facebook user reaction to posts that generate a negative response may be better measured by comments than likes. Table 3 and Table 4 summarize the statistically significant (p