ARO MURI: Evolution of Cultural Norms and Dynamics of Sociopolitical Change Annual Report, August 21, 2014

PIs: Ali Jadbabaie*, Michael Kearns (Penn) Daron Acemoglu, Fotini Christia, Asuman Ozdaglar, and Munzer Dahleh (MIT)

Larry Blume, Jon Kleinberg (Cornell) Jure Lescovek, Matt Jackson (Stanford)

Jeff Shamma** (Georgia Tech) 1. Introduction Many of the central questions related to evolution of social and cultural norms and dynamics of sociopolitical change involve interactions among individuals and groups with different identities. The understanding of these interactions is crucial for creating a new interdisciplinary science that on the one hand utilizes domain-specific expertise from Economics, Political Economy, Political science and Sociology, and on the other hand leverages the advances in the science of networks, systems theory, and computation and our understanding of la1rge scale networked systems. Our goal is to create new theories, analyses, experiments, methodologies, and predictive models that provide insight and understanding for policymakers. To this end, we need to have an understanding of how endogeneity of behavior relates to patterns of communication, dissemination of information, and expectations of others' behaviors. The goal of our research program is to address the following needs: –Need for more quantitative approaches, beyond descriptive analysis –Need for theory, principled modeling, data analysis, lab experiments, and field surveys –Need to educate a new breed of computational social scientists and engineers. In what follows, we summarize the significant advances to address these needs, along the thrusts of the project outlined below Thrust 1. Social change, social aggregation, and evolution of social norms [S1]Evolution of social norms [S2] Aggregation of beliefs and social learning [S3]The emergence of generalized trust and economic outcomes Thrust 2. Political change and societal stability: [P1]Emergence of democracies and challenges to law and order [P2]Field experiments: role of post-conflict development [P3]Empirical study of social interaction on political change [P4]Behavioral experiments in political and societal influence Thrust 3. Modeling and analysis of social interaction, cascades, and contagion [M1]Modeling of social interaction: *on leave at MIT **no longer on MURI Team as of June 2014

[M2]Modeling and analysis of cascades in networks [M3]Idiosyncratic shocks, systemic risk Thrust 4. Control of sociopolitical change [C1] Robust dynamic mechanisms and sequential networked global games [C2] Influencing social evolutionary dynamics Synergies between Thrusts We investigate how patterns of communication, interaction, and learning in networked societies (Thrust 1) affect the evolution of behavior (Thrusts 2 and 3), how this understanding (the endogeneity of behavior) affects the optimal design of policies and interventions designed to influence beliefs and social norms (Thrust 4). In Thrust 3, we investigate the micro-foundations of local interaction models, and using domain-dependent knowledge, models, and data, we investigate how this leads us to understand evolution of social norms and social learning (Thrust 1), as well as social and political change, and role of development in both (Thrust 2), and finally how to control these phenomena using principled interventions (thrust 4). We have coordinated the thrusts so that our modeling, analysis and theory work on models of local interaction which lead to dynamics of contagion, diffusion and cascade both inform our work on information aggregation and social learning as well as the thrust on political change.

Team, Research Thrusts, and Connections

2. Progress Report Our efforts over the past year have lead to significant progress on all 4 Thrusts. Below we will present some highlights and emphasize collaborations. These have also lead to joint publications in highly visible and selective venues across many disciplines, ranging from FOCS (the premiere computer science conference) to Science, Journal of Political Economy, Review of Economic Studies, and American Economic Review (last three are most selective journals in the Economics discipline). Below each task, we have highlighted one project and then list the relevant publications. For each task, we provide a brief highlight of one project and then list all publications. The last section includes the list of awards and notable achievements of PIs. The code in bracket represents the task number. S stands for social norms (Thrust 1), P stands for political Change (Thrust 2). M stands for Thrust 3 (Modeling, identification and analysis of contagion). C stands for Control of sociopolitical change (thrust 4) S1, S2, S3] Evolution of Social Norms, Social Learning, and Dynamics of Generalized Trust We [Jadbabaie, Acemoglu, Jackson, and Blume] have extensively studied the question of emergence and evolution of social norms, as well as evolution of cooperation or lack of cooperation in a dynamic environment. Acemoglu has been investigating the interaction

between institutions and social norms in joint work with Matthew Jackson. Their first paper, History, Expectations and Leadership in the Evolution of Social Norm, constructs a tractable model of evolution of social norms, where individuals follow social norms in their behavior when they are uncertain about the exact actions taken by previous generations and interpret the signals they receive through social norms. We show how history-dependent social norms can emerge, and also change as a result of shocks. Prominent agents, whose actions are more broadly observed, and then anchor others' expectations and endogenously emerge as potential leaders. This paper is in the final revision for the Review of Economic Studies. In a new work, Norms and the Enforcement of Laws, Acemoglu and Jackson study the interaction between enforcement of laws and social norms. If laws are highly inconsistent with prevailing social norms (patterns of behavior in the population), they will be ineffective because the information necessary for their enforcement will not be provided by the agents. They show how social norms can change endogenously in response to laws as well as being a constraint on their effectiveness. Furthermore they show how, in the presence of multiple laws regulating multiple types of behavior, one law that is strongly out of line with norms can create contagious spread of lawlessness into other types of behavior. This is work now completed and submitted. Following this work, Jackson and Xing examined how social norms and culture shape different populations' play in coordination games in online experiments with over a thousand subjects. Subjects played a two-player coordination game that had multiple equilibria: two equilibria with highly asymmetric payoffs and another equilibrium with symmetric payoffs but a slightly lower total payoff. Subjects were predominantly from India and the United States. Overall, each population fared better when matched with players from itself than when matched across populations, both in terms of frequency of coordination and payoffs. Subjects residing in India played the strategies leading to asymmetric payoffs significantly more frequently than subjects residing in the U.S. who showed a greater play of the strategy leading to the symmetric payoffs. In addition, when prompted to play asymmetrically, the population from India responded even more significantly than those from the U.S. Subjects' did not expect substantial differences in play across populations, and their predictions about play were more accurate about their own population. Only slightly more than half of the subjects were more confident in playing with their own population than in playing of the other population. In a related work by Jadbabaie and his former student Molavi (who is now working with Acemoglu) study emergence of coordination on an unknown yet profitable outcome for agents in a dynamic, networked environment has been investigated. They show that a network of rational yet myopic agents who want to coordinate on a profitable yet unknown state, by repeatedly observing private signals and the actions of their neighbors, can do so, even though they do not observe the action of every other agent in the network. Under very mild connectivity assumptions, they show that coordination and learning can happen in this beauty-contest-like scenario, except for a set of measure zero of priors, for which coordination will be suboptimal due to lack of global access to all observations. This work also connects to the results of Dahleh, Zmpulis, Tahbaz-Salehi,

and Tsitsiklis, on coordination in global games. Also, studying the limits of coordination, Blume has shown that in resource constraints situations were experimenting is costly, failure to coordinate can occur as a contagion phenomena, essentially showing when the above result breaks down. In another related effort, Jadbabaie and coauthors have shown how the quality of information available to agents, and their location in the network can affect the speed of social learning, and have developed an axiomatic approach for creating tractable social learning models from ideal Bayesian updates. This is in agreement with Jackson’s work, which has theoretically and experimentally addressed the importance of quality of information and location of agents for spreading of information [Thrust 3]. Jadbabaie and coauthors have shown that the same phenomena are present in social learning. In particular, they show that when the informativeness of different agents’ signal structures are comparable in the sense of Blackwell (1953), then a positive assortative matching of signal qualities and eigenvector centralities maximizes the rate of learning. On the other hand, if information structures are such that each individual possesses some information crucial for learning, then the rate of learning is higher when agents with the best signals are located at the periphery of the network. Finally, we show that the extent of asymmetry in the structure of the social network plays a key role in the long-run dynamics of the beliefs. This theory work essentially agrees with the empirical observations of the work of Jackson and coauthors. In that work, Banerjee, Chandrasekhar, Duflo and Jackson study network models of behavior and diffusion in rural Indian villages. In particular, they examined how to easily identify the members of a community best placed to diffuse information by seeing whether community members were able to name highly central individuals? The researchers defined a measure of diffusion centrality (which nests other standard centrality measures) and theoretically demonstrate that boundedly-rational individuals can, simply by tracking sources of gossip, identify those who are diffusion central. Finally, they compare this prediction with the data. Using data from 35 Indian villages, they find that respondents accurately nominate those who are diffusion central, and moreover these nominees are more central than ``village leaders'' and geographically central individuals. Another related but new area of research is the economic analysis of the dynamics of conflict and trust. In new work joint with Alexander Wolitzky, Cycles of Trust: An Economic Model, Acemoglu develops a simple framework for the study of spirals of conflict and distrust. The idea is simple and has been suggested in several key contributions and international relations: aggressive behavior by one party triggers aggressive behavior by the other, and now conflict spirals because the first group starts thinking that the second is inherently aggressive. We show that this reasoning requires finite memory, but then finite memory implies that spirals don't go on forever but contain the seeds of their own dissolution. We provide a simple analysis of this sort of cyclic equilibria, and explain why these insights are robust and can be applied to a variety of circumstances, on the way proposing several new ideas on dynamics of communication and political polarization as well as civil war. This paper is now published in the American Economic Review.

The proposed model is a two period model: Each agent lives for two periods and interacts with agents from the previous and next generations via a coordination game. Social norms emerge as patterns of behavior that are stable in part due to agents' interpretations of private information about the past, influenced by occasional commonly-observed past behaviors. For sufficiently backward-looking societies, history completely drives equilibrium play, leading to a social norm of high or low cooperation. In more forward-looking societies, there is a pattern of reversion whereby play starting with high (low) cooperation reverts toward lower (higher) cooperation. The impact of history can be countered by occasional prominent agents, whose actions are visible by all future agents, and who can leverage their greater visibility to influence expectations of future agents and overturn social norms of low cooperation. Other Publications D. Acemoglu, and M. Jackson, “Norms and the Enforcement of Laws,” NBER Working paper, 2014 D. Acemoglu, and M. Jackson “History, Expectations and Leadership in the Evolution of Social Norm,” Review of Economic Studies, forthcoming D. Acemoglu, and A. Wolitzky, “Cycles of Trust: An Economic Model,” American Economic Review, 2014 A. Banerjee, A. G. Chandrasekhar, E. Duflo, and M. O. Jackson. “The Diffusion of Microfinance,” Science  341,  1236498  (2013) A. Jadbabaie, P. Molavi, and A. Tahbaz-Salehi. “Information Heterogeneity and the Speed of Learning in Social Networks,” Revise and Resubmit at Review of Economic Studies, 2014 P. Molavi, C. Eksin, A. Ribeiro, and A. Jadbabaie," Learning to coordinate," Operations Research, Final revision, 2014 P. Molavi, C. Eksin, A Ribeiro, and A. Jadbabaie, " Learning to Coordinate in a Beauty Contest Game ," Proceedings of the IEEE Conference on Decision and Control, December 2013, Florence, Italy M. A. Dahleh, A. Tahbaz-Salehi, J. N. Tsitsiklis, and S. I. Zoumpoulis, “Coordination with Local Information, “ Revise and Resubmit at Operations Research, 2014 M.A Rahimian, P. Molavi, and A. Jadbabaie, “(Non-)Bayesian Learning without Recall,”Proceedings of the IEEE Conference on Decision and Control, forthcoming,2014 Abhijit Banerjee, Arun Chandrasekhar, Esther Duflo, Matthew O. Jackson, “Gossip: Identify Central Individuals in a Social Network,'' , under review.

Daron Acemoglu and Matthew O. Jackson, ``Networks of Military Alliances, Wars, and International Trade, '' ``Social Norms and the Enforcement of Laws,'' Philippe Aghion and Matthew O. Jackson, ,``Inducing Leaders to Take Risky Decisions: Dismissal, Tenure, and Term Limits,'' under review.

Roland Fryer, Philipp Harms, Matthew O. Jackson ``Updating Beliefs with Ambiguous Evidence: Implications for Polarization,'' - doing experiments, working paper

Matthew O. Jackson, ``Networks and the Identification of Economic Behaviors'' forthcoming in the Journal of Economic Perspectives. Matthew O. Jackson and Yiqing Xing,``Culture-dependent strategies in coordination games,'' Proceedings of the National Academy of Sciences (July 14, 2014 edition). Matthew L. Elliott, Benjamin Golub, Matthew O. Jackson,``Financial Networks and Contagion'' Forthcoming in the American Economic Review. Arun Chandrasekhar and Matthew O. Jackson, ``Tractable and Consistent Exponential Random Graph Models'' under review. Matthew O. Jackson and Yves Zenou ,``Games on Networks'' , in press: Handbook of Game Theory Vol. 4, edited by Peyton Young and Shmuel Zamir, Elsevier Science. [M1-M3]Modeling, identification, and analysis of social Interaction and Cascades Our team has made significant progress in this area by following up on last year’s work studying various aspects of local interaction models that focus on how one can solve an inverse problem related to contagion and determine how should one seed the network to get the desired contagion result. In particular, We (Michael Kearns and Hoda Heidari who is a PhD student jointly advised with Jadbabaie) have developed a game-theoretic framework for the study of competition between firms who have budgets to ``seed'' the initial adoption of their products by consumers located in a social network. The payoffs to the firms are the eventual number of adoptions of their product through a competitive stochastic diffusion process in the network. This framework yields a rich class of competitive strategies, which depend in subtle ways on the stochastic dynamics of adoption, the relative budgets of the players, and the underlying structure of the social network. Also, Co-PI Kearns, along with Moez Draief (Imperial College London) and doctoral student Hoda Heidari (jointly advised with Jadbabaie), developed natural extensions of the competitive contagion model originally introduced and studied by Kearns and Goyal. This work established that although the Kearns-Goyal bounds on the Price of Anarchy and Budget Multiplier are largely preserved in settings where payoffs depend not just on the number of infections but on

their interconnections, these quantities become unbounded if player budgets are endogenous. In a related line of work that was motivated by the work of Kearns, MURI Postdoc Amir Ajorlou with Jadbabaie and student Arastoo Fazeli have developed a model on competition and contagion in a duopoly that studies the balance of investing in the quality of a product to make the product more contagious, or investing in seeding more nodes in the network. The result shows that based on the network structure, and node centralities, how should the nodes decide whether to invest in seeding or on improving quality to increase the spreading of the product. Furthermore, Jadbabaie with postdocs Ajorlou and Kakhbod (who is now working with Ozdaglar and Acemoglu at MIT), have developed and analyzed a model of pricing a durable product, when the product is spread by Word of Mouth diffusion. This is a notoriously difficult problem, as the network formed is endogenous to the process: individuals will know about the product only if they are already friends with someone who has bought the product. Given that individual’s valuation of the product is unknown: this leads generally to an intractable model for analysis. By a clever choice of modeling, Jadbabaie and colleagues were able to show that the monopolist can achieve maximum penetration in the network by setting the price to zero infinitely often. The culmination of our efforts on the study of local interaction models has resulted in exciting development son the study of binary threshold models that are notoriously hard to analyze. In a joint work between student Elie Adam and PIs Ozdaglar and Dahleh, the starting point is a networked coordination game where each agent’s payoff is the sum of the payoffs coming from pairwise interaction with each of the neighbors. We first establish that the best response dynamics in this networked game is equivalent to the linear threshold dynamics with heterogeneous thresholds over the agents. While the previous literature has studied such linear threshold models under the assumption that each agent may change actions at most once, a study of best response dynamics in such networked games necessitates an analysis that allows for multiple switches in actions. We have developed such an analysis and established that agent behavior cycles among different actions in the limit, and characterized the length of such limit cycles, revealing bounds on the time steps required to reach them. We have also proposed a measure of network resilience that captures the nature of the involved dynamics. We have also studied a stochastic linear threshold model of diffusion of innovation in networks in which each agent has a threshold drawn from some distribution. Starting from a seed set of agents, each agent adopts the innovation with a probability p if at least t fraction of his neighbors adopted the innovation. Previous works focused on deterministic versions of this model where agent thresholds are fixed, which makes the analysis intractable. In recent work, we provide explicit analytical characterizations for expected number of adopters when thresholds are random. This characterization allows us to characterize networks that maximize overall diffusion when the seed set is random or targeted. We are currently working on generalizing this to the case of a duopoloy competition

More recently, and following up on our data-driven efforts, Co-PI Kearns, along with doctoral students Kareem Amin and Hoda Heidari, conducted an extensive theoretical and experimental investigation of algorithms for learning an unknown social network structure from multiple observed contagion processes. Unlike prior works in this area, the approach developed does not require full “timestamp” data (providing the exact order of infections throughout the network), but only initial and subsequent snapshots of the state. This weaker data is more natural in many real applications. The approach was proved correct and efficient in theoretical settings, and validated with extensive experiments on both simulated and real network structures. CoPI Blume and coauthors have developed a systematic analysis of identification in linear social interactions models. This is a theoretical and econometric exercise as the analysis is linked to a rigorously delineated model of interdependent decisions. They develop an incomplete information game that describes individual choices in the presence of social interactions. The equilibrium strategy profiles are linear. Standard models in the empirical social interactions literature are shown to be exact or approximate special cases of our general framework, which provides a basis for understanding the microeconomic foundations of those models. The Key innovation is that they consider identification of both endogenous (peer) and contextual social effects under alternative assumptions regarding the analyst’s a priori knowledge of social structure or access to individual-level or aggregate data. Publications Sanjeev Goyal, Hoda Heidari, and Michael Kearns, “Competitive Contagion in Networks,” Game and Economic Behavior, Revise and Resubmit, 2014 A Ajorlou, A. Jadbabie, and A. Kakhbod, “Strategic Information Diffusion: Spread vs. Exploit”, Proceedings of the 2014 Workshop on Network Economics: NetEcon 2014, Austin, Texas, June 2014 A Ajorlou, A. Jadbabie, and A. Kakhbod, “Dynamic pricing in social networks: the Word of Mouth effect”, Working paper J. Cheng, L. A. Adamic, P. A. Dow, J. Kleinberg, J. Leskovec, “Can Cascades be predicted,” Proceedings of the 2014 World Wide Web Conference K. Amin, H. Heidari, M. Kearns. "Learning from Contagion (Without Timestamps),” Proceedings of the International Conference on Machine Learning, 2014. M. Draief, H. Heidari, M. Kearns, "New Models for Competitive Contagion”, Proceedings of AAAI, 2014.

L. Blume, W. Brock, S. Durlauf and R. Jayaraman,” Linear local interaction models,” Journal of Political Economy, Forthcoming, 2014 E. Adam, A. Ozdaglar, M. Dahleh, “On Threshold Models over Finite Networks,” working paper

Y. Lim, A. Ozdaglar, A. Teytelboym, “A simple model of cascades in networks,” working paper, 2014.

Y. Lim, A. Ozdaglar, A. Teytelboym, “Platform competition in social networks,” working paper, 2014.

K. Bimpikis, A. Ozdaglar, E. Yildiz, “Competing over networks,” revise-resubmit in Operations Research, 2013.

E. M. Adam, M. A. Dahleh, and A. Ozdaglar, “On threshold models over finite networks,”arXiv preprint arXiv:1211.0654, submitted for publication, 2014.

[P2] Empirical Study of post-conflict Development: from Sierra Leone to Egypt, Afghanistan Yemen, and Columbia Our data-driven modeling effort on role of development in sociopolitical change has made significant advances over the past years, with rigorous studies across multiple countries. Below we will highlight this progress Sierra Leone We (Acemoglu) have investigated the political economy role of paramount chiefs in Sierra Leone. Chiefs play a central role in the politics of sub-Saharan Africa. Nevertheless, there is relatively little systematic that's known about their role. Some view them as despots, while others argue that they are popular and responsive to local needs. We exploit the unique history of Sierra Leone, where British colonialists created or managed ruling families from which local chiefs must originate. Based on the original field research, we trace the histories of these chiefs and ruling families, and argue that the number of ruling families is a crucial determinant of the power of chiefs. This is both because rivals from other families can compete and also because governance also involves bargaining across ruling families in the tribal council. Using this idea, we show that places with fewer ruling families, where presumably chiefs have more power, have robustly worse development outcomes, particularly education and occupational composition. Interestingly, chiefs in such places are generally respected and individuals seem take part in activities that can be recognized as bridging and bonding social capital. We argue that this constellation is the result of the peculiar political economy of sub-Saharan Africa, where social capital is built not to control politicians or chiefs, but managed by chiefs. In places where chiefs are more powerful, they are able to structure various local organizations, inducing individuals to invest more in various forms of social capital, but this doesn't enable better political control for better development outcomes. This paper has been published in Journal of Political Economy Columbia

In another related paper, we (Acemoglu) study the relationship between local state capacity and economic development within Columbia. Differently from the small literature that has investigated these issues within economics (and despite a larger literature within political science), we distinguish the direct and spillover effects of local state capacity. This very much relates and connects to Blume’s work on identification of linear interaction models on Thrust 3. We model the indirect effects using the network of Colombian municipalities. Also differently from the previous literature, we anchor our empirical work to a microeconomic model of local state capacity choices, based on a setup in which local and national state capacity is determined as part of a network game in which each municipality, anticipating the choices and spillovers created by other municipalities and the decisions of the national government, invests in local state capacity and the national government chooses the presence of the national state across municipalities to maximize its own payoff. This model not only disciplines our empirical work but also shows why reduced-form empirical work will face fundamental identification problems. Based on this model, we estimate the parameters of this model using reduced-form instrumental variables techniques and structurally (using GMM, simulated GMM or maximum likelihood). To do so we exploit both the structure of the network of municipalities, which determines which municipalities create spillovers on others, and the historical roots of local state capacity as the source of exogenous variation. These historical instruments are related to the presence of colonial royal roads and local presence of the colonial state in the 18th century, factors which we argue are unrelated to current provision of public goods and prosperity except through their impact on their own and neighbors' local state capacity. Our estimates of the effects of state presence on prosperity are large and also indicate that state capacity decisions are strategic complements across municipalities. As a result, we find that bringing all municipalities below median state capacity to the median, without taking into account equilibrium responses of other municipalities, would increase the median fraction of the population above poverty from 57% to 60%. Approximately 57% of this is due to direct effects and 43% to spillovers. However, and crucially illustrating the importance of the equilibrium modeling and estimation in this paper, when the equilibrium response of other municipalities is taken into account, the median would instead increase to 68%, a sizable change driven by equilibrium network effects. This paper is being revised for the American Economic Review. Egypt In ongoing work, we have investigated the impact of various events and protests during the Egyptian Arab spring on the stock market returns up with politically-connected firms. We find that the market anticipated the strengthening of the military-connected firms at the very early stages of the protests, and in general larger protests destroy the value of firms connected to incumbent powers. The evidence suggests that this was not simply a reallocation of rents from one group to another, though especially in the early stages there was some anticipation of such reallocations.

Afghanistan We (Christia) have had a unique opportunity to perform rigorous field experiments in Afghanistan and Yemen. Our yearlong effort on data gathering in Afghanistan has finally produced significant results, which we now report here. In a paper coauthored with collaborators entitled “Electoral Rules and the Quality of Politicians: Theory and Evidence from Afghanistan” that was issued as a working paper in the series of the National Bureau for Economic Research (NBER, see here: http://www.nber.org/papers/w20082), we use our unique field experiment, which induced randomized variation in the method of council elections in 250 villages in Afghanistan to examine the effect of electoral rules on the quality of elected officials. In particular, we compare at-large elections, with a single multi-member district, to district elections, with multiple single member districts. We propose a theoretical model where the difference in the quality of elected officials between the two electoral systems occurs because elected legislators have to bargain over policy, which induces citizens in district elections to vote strategically for candidates with more polarized policy positions even at the expense of candidates' competence. Consistent with the predictions of the model, we find that elected officials in at-large elections are more educated than those in district elections and that this effect is stronger in more heterogeneous villages. We also find evidence that elected officials in district elections have more biased preferences. We are a week away from submitting this paper for review at the Journal of Political Economy, one of the top-five journals in Economics. Three more papers on the Afghanistan project that we had referenced and described in detail in previous reports, coauthored with Andrew Beath and Ruben Enikolopov, remain at different stages of the publication process. We are two weeks away from re-submitting our paper entitled “Winning Hearts and Minds through Development: Evidence from a Field Experiment in Afghanistan” which has an R&R at the Quarterly Journal of Economics, one of the top-five journals in Economics. We are also revising two more working papers available on SSRN, “Direct Democracy and Resource Allocation: Experimental Evidence from Afghanistan” and “Do Elected Councils Improve Governance? Experimental Evidence on Local Institutions in Afghanistan” to be submitted for review within the next couple of months. Yemen CoPI Christia has secured three years worth of cell-phone call records metadata from 2010-2013, which we are using to address the overarching question of how contexts of conflict and violence shape social networks. This project involves PI Jadbabaie, CoPIs Dahleh, and Leskovec. Christia formulates the social science question, Leskovec creates the software infrastructure for the big data analysis, and Dahleh and Jadbabaie, joint with students and postdocs study the network effects. We have done this collaborative effort by looking at two types of violent events that transpired during this three-year time period: drone strikes that are exogenous violent shocks; and Arab Spring protests that are indigenous to the Yemeni society. (Please note

that Google Ideas is very interested in having us over to present and in assisting us with the visualization of our findings once we have things to share.) In terms of drone strikes, we use data on these strikes from the New America Foundation to look at their effect on Yemen’s social network as elicited through patterns of communication captured by cell phone metadata (calls and SMS texts). We are directing our study around three main questions of interest: detecting the effect of drone strikes on cell phone activity; classifying users based on their communication patterns as related to drone strikes; and predicting drone strikes from cell phone activity. Together with David Hallac, coPIs Leskovec and Christia’s research investigated the disruptions of the social network connectivity structure that correlate with the violent events. In particular, they developed a method that allows for identification and localization of violent events through the traces collected from cellphone calls. We are presently focusing on the first thrust of questions by looking at volume of communications; direction of calls (incoming or outgoing); who people call at such instances (whether they call close or more distant friends as determined by call patterns in a meaningful time window preceding the drone strike); duration of calls; type of communication (if there is a deviation from the average ratio of voice over text usage); geographic range of effect (whether effect is picked up only by most proximate or more distant antennas); extent of a “retweeting effect”, i.e., events of chain communications following the strike; extent of displacement of users after the strike, i.e., people moving to different areas after the strike; the resilience of the network to link removal as captured by turn-around in phone numbers and more specifically the extent to which nodes with relatively high activity disappear, and new nodes appear after the strike, or other users assume the call patterns of the disappearing nodes (which can be attributed to key individuals either hiding, or getting killed, or changing SIM cards because of the strike, and other individuals possibly replacing them). As measurable predictors, which interact with the effect of a drone strike, we use: distance of strike location and subnetworks/antennas of interest; demographics of affected population, i.e., percentage of civilians versus militants killed; lethality of the strike, i.e., total number of deaths (potentially injuries as well); day and time of the strike (e.g., regular day versus religious holiday, day versus night). Baselines against which we will measure the drone strike effects will be: 1) the same day of the week a month before and after the strike; 2) other events that are bound to instigate changes in call patterns such as religious celebrations to see how the deviation in calling patterns during those meaningful events differs from calling patterns after drone strikes. Christia also spent this past year traveling to Yemen to secure primary data that allowed for the coding of over 1,500 Arab Spring related demonstrations with information on the date and location of event as well as its character (violent or non-violent). We are specifically interested in looking at how cellphones were used for opposition mobilization and government counter-mobilization during the Arab Spring in Yemen.

The richness of our data allows us to look at volume of calls, frequency, direction, duration, dispersion, and mobility within the network so as to get a sense of an individual’s network connectedness and centrality. We also exploit the location of the event and the callers, as well as the direction of the call, to separate out “mobilization” activity (ie activity that brings people to the protest) from pure “information” activity (where people get information but remain outside the primary area of events). Apart from the communications in the year of the Arab spring, we also have a baseline year of communication activity that reveals the network structure of interest before the protests, as well as a year’s worth of calls after the Arab spring, that allows us to look at the potential persistence or decay or the network effects identified during the protests. In addition to dynamics, this data also offers great cross-sectional variation as it covers the whole country, which exhibited variant intensities of protest activity. In addition, the timing and location of the actual protest events, and the sequential entering in the protests of different groups as discussed below (students first, then opposing tribes, then counter-mobilization activities from the government etc) allows us to also get at differential network effects for the distinct groups participating in the revolution. We will report on the progress on these fronts in the next year. Publications: Daron Acemoglu, Tristan Reed and James A. Robinson, “Chiefs - Economic Development and Elite Control of Civil Society in Sierra Leone,” forthcoming, Journal of Political Economy. Daron Acemoglu, Camilo Garcia-Jimeno, James Robinson, “State Capacity and Economic Development: A Network Approach”, under revision, American Economic Review Fotini Christia, Andrew Beath, Georgy Egorov and Ruben Enikolopov “Electoral Rules and the Quality of Politicians: Theory and Evidence from Afghanistan”, Journal of Political Economy, final revision, 2014 Fotini Christia, Andrew Beath, Georgy Egorov and Ruben Enikolopov “Winning Hearts and Minds through Development: Evidence from a Field Experiment in Afghanistan” R&R at the Quarterly Journal of Economics, Fotini Christia, Andrew Beath, Georgy Egorov and Ruben Enikolopov “Do Elected Councils Improve Governance? Experimental Evidence on Local Institutions in Afghanistan”, Working paper, to be submitted

Fotini Christia, Andrew Beath, Georgy Egorov and Ruben Enikolopov “Direct Democracy and Resource Allocation: Experimental Evidence from Afghanistan”, working paper Daron Acemoglu, Tarek Hassan and Ahmed Tahoun “Did the Arab Spring Disrupt Rent-Seeking Network in Egypt?” Working paper [P1] Emergence of political change, challenges to law and order We (Acemoglu and coauthors) have followed up on our theoretical studies from last year and developed and analyzed models of dynamic collective decision-making in non-cooperative environments with arbitrary discounting (so that individuals care about transition path rather than long-term outcomes) and a rich set of stochastic shocks. We show how equilibria can be characterized in such environments, and how the results are different from the more specialized environments we focused on previous work. Nevertheless, we also show that despite the much greater generality of this set of environments, equilibria can be characterized fairly tightly and there are "monotone" comparative statics of political equilibria, whereby changes in parameters or payoffs affect the equilibrium path in a monotone fashion. We consider several applications of this general setup, most originally, to the relationship between radical politics and repression. Our framework shows how radicalism and repression of radicalism feed to each other, and provide sharp results clarifying the role of institutions, preferences and their interaction on the dynamics of political rights and repression. Publications Daron Acemoglu, Georgy Egorov and Konstantin Sonin, “Political Economy in a Changing Environment,” Journal of Political Economy, forthcoming [P3,P4] Empirical and Behavioral studies on social and political change MURI-supported student Myers and coPI Leskovec studied a problem where in online social media systems users are not only posting, consuming, and resharing content, but also creating new and destroying existing connections in the underlying social network. While each of these two types of dynamics has individually been studied in the past, much less is known about the connection between the two. How does user information posting and seeking behavior interact with the evolution of the underlying social network structure? Myers and Leskovec studied ways in which network structure reacts to users posting and sharing content. They examined the complete dynamics of the Twitter information network, where users post and reshare information while they also create and destroy connections. They found that the dynamics of network structure can be characterized by steady rates of change, interrupted by sudden bursts. Information diffusion in the form of cascades of post re-sharing often creates such sudden bursts of new connections, which significantly change users' local network structure. These bursts transform users' networks of followers to become structurally more cohesive as well as more homogenous in terms of follower interests. They also explored the effect of the

information content on the dynamics of the network and find evidence that the appearance of new topics and real-world events can lead to significant changes in edge creations and deletions. Lastly, they developed a model that quantifies the dynamics of the network and the occurrence of these bursts as a function of the information spreading through the network. The model can successfully predict which information diffusion events will lead to bursts in network dynamics. Most recently, co-PI Kearns and Penn doctoral student Lili Dworkin have embarked on a significant extension of the long series of behavioral studies in networked interactions conducted in the Kearns group. The new experiments will examine the extent to which individuals can solve complex problems in small networks, and compare to group performance. Several test studies have been conducted already, and plans are in place to use Kearns’ Coursera MOOC “Networked Life” to execute large-scale studies. A preliminary version of the platform can be tried at http://upenn-aaai.herokuapp.com. Another related effort in this area has been a data-driven and modeling effort on whether cascades can be predicted. In a joint work between coPIs Leskovec and Kleinberg, The PIs develop a framework for addressing cascade prediction problems. On a large sample of photo reshare cascades on Facebook, they find strong performance in predicting whether a cascade will continue to grow in the future. The PIs find that the relative growth of a cascade becomes more predictable as we observe more of its reshares, that temporal and structural features are key predictors of cascade size, and that initially, breadth, rather than depth in a cascade is a better indicator of larger cascades. This prediction performance is robust in the sense that multiple distinct classes of features all achieve similar performance. They also discover that temporal features are predictive of a cascade’s eventual shape. Observing independent cascades of the same content, it is determined that while these cascades differ greatly in size, they are still able to predict which ends up the largest. Publications S. Myers, J. Leskovec, “The Bursty Dynamics of the Twitter Information Network.” ACM International Conference on World Wide Web (WWW), 2014. J. Cheng, L. Adamic, A. Dow, J. Kleinberg, J. Leskovec, “ Can Cascades be Predicted. “ACM International Conference on World Wide Web (WWW), 2014. J. Cheng, C. Danescu-Niculescu-Mizil, J. Leskovec.How Community Feedback Shapes User Behavior AAAI International Conference on Weblogs and Social Media (ICWSM), 2014. D. Shahaf, J. Yang, C. Suen, J. Jacobs, H. Wang, J. Leskovec , “Information Cartography: Creating Zoomable, Large-Scale Maps of Information.” ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2013. [C1,C2] Control of Sociopolitical Change

In this thrust we have studied several variations of sequential/networked global games and Extended analysis of social influence over networks: robustness to dynamics, network uncertainty, global game setting. We have also developed models and analysis for social coordination over networks: Variants of "language game" as a framework for evolution of convention and studied Cascading failure in networks by coupling of network centrality and immunity. This thrust will be further developed in thee coming year. In particular, PI Jadbabaie, joint with students Zargham and Eniyoha and other coauthors, studied dynamics of epidemic models and control of epidemic outbreaks in networks. The work addresses how to optimally allocate vaccines in a network to control an epidemic outbreak. Since vaccination resources are typically limited, optimally allocating vaccines to control the outbreak is vital. Existing literature on this problem considers homogeneous networks, limited the discussion to undirected networks, and largely proposed network centrality-based vaccination strategies as heuristics. In practice, networks are directed and comprise heterogeneous nodes. In this work, the PIs studied the problem of minimum-cost allocation of resources to control the spread of an epidemic outbreak on weighted, directed networks comprising heterogeneous nodes. The epidemic control problem was formulated as a Geometric Program (GP), for which a convex characterization was achieved leading to an optimal solution, outperforming network centrality-based strategies previously proposed in the literature. Motivated by this work, coPI Ozdaglar and collaborators studied a dynamic policy for the rapid containment of a contagion process modeled as an SIS epidemic on a bounded degree undirected graph with n nodes. They showed that if the budget of curing resources available at each time is lower bounded by the CutWidth of the graph, and also of order Ω(logn), then the expected time until the extinction of the epidemic is within a constant factor from optimal, as well as sublinear in the number of nodes. Furthermore, if the CutWidth increases only sublinearly with n, a sublinear expected time to extinction is possible with a sublinearly increasing budget. Both of these results have implications on utilizing the network structure to achieve efficient curing policies for epidemics and contagions Publicatons G. Kotsalis and J.S. Shamma, "Dynamic mechanism design in correlated environments", IEEE Conference on Decision and Control, December 2013. VM Preciado, M Zargham, C Enyioha, GJ Pappas, A. Jadbabaie, “Optimal Resource Allocation for Network Protection Against Spreading Processes,” IEEE Transactions on Control of Network Systems, Vol. 1, No. 1, pp. 99-108 VM Preciado, M. Zargham, C. Eniyoha, A. Jadbabaie, and G. Pappas, " Optimal Vaccine Allocation to Control Epidemic Outbreaks in Arbitrary Networks," Proceedings of the IEEE Conference on Decision and Control, December 2013, Florence, Italy

K Drakopoulos, AE Ozdaglar and JN Tsitsiklis, “An efficient curing policy for epidemics on graphs,” IEEE Transactions on network Science and Engineering, Rrvise and Resubmit 3. List of awards and Achievements (2013-2014)

• Ali Jadbabaie received the Alfred Fitler Moore Endowed Chair in Network Science from Penn’s School of Engineering and Applied Sciences, September 2013

• Munzer Dahleh received the William Coolidge Endowed Chair and Professorship from MIT’s department of Electrical Engineering and Computer Science, July 2014

• Munzer Dahleh became the director of a new Entity to coordinate MIT’s activities

in Socio-techncial systems, network science, statistics and Information and Decision Systems, May 2013

• Ali Jadbabaie was elected as an IEEE Fellow, Jan 2014

• Ali Jadbabaie was elected as the inaugural Editor-in-Chief of the new IEEE Transactions on Network Science and Engineering, March 2014

• Fotini Christia was promoted to Associate Professor with Tenure at the department of Political Science at MIT, July 2014

• Daron Acemoglu received an Honorary doctorate from the University of Athens.

• Daron Acemoglu was elected to the National Academy of Science

• Asu Ozdaglar received the Spira Teaching Award from MIT EECS Department in

2014

• Matt Jackson received Honorary Doctorate (Doctorat Honoris Causa), from Aix-Marseille Universit ́e, 2013

• Matt Jackson received the 2013 Dean’s Award for Distinguished Teaching in Humanities and Sciences at Stanford.