Date of Award

Spring 2022

Document Type


Degree Name

Doctor of Philosophy (PhD)



First Advisor

Christakis, Nicholas


Social Coordination is an essential feature of any social system. Coordination is a precondition for certain kinds of cooperation, as cooperation implies agents working together for a common cause, while coordination only implies agents synchronizing their behavior. For example, social contagion suggests a mechanism for groups to socially coordinate without necessarily cooperating. When behaviors spread across social ties the people may not be actively intending to spread the behavior, so they are not always cooperating. The necessity of coordination does not imply that it is easy to achieve. In Chapter 1, I survey two theoretical frameworks for operationalizing the challenges to coordination: The Prisoner's Dilemma Game and The Generals Problem. In the Prisoner's Dilemma game, the agents will both be better off if they cooperate, but have strong incentives to betray each other. In this classic case coordination is expected but it is not Pareto-optimal, highlighting the differences between agents simply coordinating (both playing the same strategy) and cooperation for each other's benefit. Here, the impediment to cooperation is that of misaligned incentives. The General's Problem presents a different barrier to coordination: unreliable communication channels. The Generals all have aligned incentives to coordinate and work together, but due to the chance that messages may fail to be transmitted, they are unable to effectively do so. The Byzantine Generals Problem in effect combines the challenges of coordinating with both misaligned incentives and faulty communication. In this setting, some of the agents are trying to prevent the rest of the group from reach consensus. Chapter 2 is a global consensus experiment in a "Byzantine" setting. The players have 10 rounds of communication to reach consensus among a set of arbitrary identifiers. The players are able to send full text messages to each other. This setting is Byzantine because some of the players disconnect from the game either through technical problems on their end or through failing to send their messages in enough time. Additionally, we found that a subset of the players did not perfectly understand the instructions of the game and made errors, thus demonstrating that players can engage in Byzantine behavior. The players were arranged into a Watts-Strogatz network and one of the interventions was altering the fraction of odd vertices in these graphs. The other intervention was to alter the instructions so as to change the story the players were told about why they were trying to reach consensus. We found that groups were more likely to reach consensus in groups with a lower fraction of odd vertices, but did not find the changing story about why players were coordinating had much impact. We found that the human consensus does exhibit some byzantine fault tolerance. For example, the effect of players dropping out of the game, a classic type of byzantine fault, had a negligible effect on the outcome. However, we also found that the fraction of players with misunderstandings or errors negatively impacted the consensus process. The players did demonstrate the ability to correct misunderstandings in others, but sometimes misunderstandings were contagious. Notably the effect of misunderstandings is comparable to effect of the fraction of players trying to vote for the first identifier in alphabetical order. This suggests that in this setting using a bad protocol is comparable in effect to a byzantine fault. Chapter 3 is a methodological exploration of creating preference-indifferent identifiers. Throughout the testing of identifiers to use in the consensus experiment in Chapter 1, we found that the players expressed preferences over random strings of letters and numbers. To remedy this, we generated nonsense words with an alternating vowel and consonant pattern to make them easily pronounceable to English speakers. We developed a software platform for users to evaluate these nonce words as forced-choice paired comparisons. We then used the Elo algorithm to generate scores for each of these words. We also developed techniques to find unobserved heterogeneity in ratings for this setting. We found that human raters do indeed have significant preferences even over these nonsense words, implying that even if the identifiers are randomly generated, they are not necessarily preference-equivalent. We also compared the preferences we observed with the predictions of Phonological Cue Theory and found that our results were not entirely consistent. While this was not initially devised as a Phonology experiment, the platform we develop may have benefits for conducting Phonology experiments. Chapter 4 is an agent-based model assessing the impact of capacity constraints in a threshold contagion model. Many sorts of contagious phenomenon, such as music, do not exist in isolation but as part of a competitive marketplace. In these settings there are often superstars with out-sized popularity along with a large number of flops with little popularity. I suggest that capacity constraints may be a structural factor that influences these disparities. In this model, there are multiple potentially cascading states that the agent can potentially occupy. The agents have a certain capacity of states that they can occupy at once. For example, suppose someone has a workout playlist that lasts 1 hour. As they discover new music to add to the playlist, they have to remove songs currently in the playlist to keep the playlist 1 hour. Thus, in this setting, the states indirectly trade off with each other by virtue of the capacity constraint. Increasing the number of states in excess of capacity increased the unpredictability of which states become popular as well as increased the disparities between popular and unpopular states. This suggests that capacity constraints may play a role in explaining market concentration and superstar phenomenon.