Research
Publications
“Reform and Community Level Participation: The Overturn of SQF in New York City” Urban Affairs Review (2024) https://doi.org/10.1177/10780874241241266
Abstract
Though there has been a wealth of work showing that negative experiences with government lead to less political participation, there is little understanding of how to address this problem. That is, it is unclear whether reform to negative government behavior can also address the negative consequences of that behavior. This paper uses the overturn of Stop, Question, and Frisk in New York City to show that despite a significant change to police behavior, having experienced a high level of policing continues to depress voter turnout. Furthermore, it uses residents who have moved within New York City after the change in policy to show that residing in a community that was heavily policed continues to influence voter behavior, even if an individual never directly experienced high levels of policing. This is in line with communities both being changed by police policies and community transmission of norms around participation.
“Large Language Models Can Argue in Convincing and Novel Ways About Politics” Political Science (2024) https://doi.org/10.1080/00323187.2024.2335471 With Arthur Spirling. Github
Abstract
All politics relies on rhetorical appeals, and the ability to make arguments is considered perhaps uniquely human. But as recent times have seen successful large language model (LLM) applications to similar endeavours, we explore whether these approaches can out-compete humans in making appeals for/against various positions in US politics. We curate responses from crowdsourced workers and an LLM and place them in competition with one another. Human (crowd) judges make decisions about the relative strength of their (human v machine) efforts. We have several empirical ‘possibility’ results. First, LLMs can produce novel arguments that convince independent judges at least on a par with human efforts. Yet when informed about an orator’s true identity, judges show a preference for human over LLM arguments. This may suggest voters view such models as potentially dangerous; we think politicians should be aware of related ‘liar’s dividend’ concerns.
Working Papers
“Stories of Success: Narrative Propaganda in Kenya”
Abstract
As many countries note declining trust in government institutions, the academic and public discussion has largely focused on the connection between trust and government performance. However, many governments and institutions use other strategies to build trust, including storytelling. Rather than improving government performance or procedures, telling a story (following a rising action-climax-falling action format) can increase empathy towards government representatives and a more positive a ect towards institutions. I use the case of the main investigative police service in Kenya, the Directorate of Criminal Investigations, which has been sharing stories of their cases on Twitter as a way to build trust in the police. Using sentiment classi cation, I show that people who follow these stories (rather than straightforward reports of similar police actions) write more positively about the police. I also use data from a survey experiment in Kenya to demonstrate that structuring information as a story increases audience belief that the relevant actors can be trusted. Finally, I compare the e ects of a story with elements of procedural justice to establish this strategy is at least as e ective as shifting opinions about performance at building trust. This project contributes to our understanding of how trust develops and the e ect of structure as compared to content.
“Why Are Citizens So Cynical? And About What?” With Gwyneth McClendon
Abstract
Observers have remarked that cynicism about politics—the belief that political actors have bad intentions and act in bad faith—is currently very high around the world. Is this cynicism mostly about political elites, or also about a wider range of actors? What are some of the drivers of political cynicism, and what are some of its consequences? Using original survey data from the United States (with planned data collection in Kenya), we argue that there are at least two dimensions of politically-relevant cynicism that should be parsed: (1) cynicism about human nature generally and (2) cynicism specifically about political elites. These two dimensions are conceptually distinct from each other: human nature cynicism views all people as badly intentioned, whereas elite cynicism allows that most people are well-intentioned but does not see this goodness manifested in current elite circles. These two dimensions of cynicism have divergent consequences. Individuals who are highly cynical about human nature are less motivated to participate in politics, are more tolerant of leaders’ bad behavior, even willing to support leaders who strategically lie and cheat in order to get things done, and are not particularly keen to support leaders who diverge from existing elite norms. By contrast, individuals who are highly cynical about elites but not so cynical about human nature tend to be highly motivated to participate in politics, express outrage about current elites’ lying and cheating behavior, and look for explicit signals that the candidates they support are fundamentally different from current elites. We also hypothesize that these two types of cynicism have different drivers. Using survey experiments, we find that perceiving that the current government is against your in-group increases elite cynicism; by contrast, perceiving that governments (not just the current one) have for a long time been against your in-group, increases human nature cynicism. We discuss the implications for understanding contemporary politics.
“Books vs Bullets: Measuring Rebel-Provided Services”
Abstract
Despite the growing literature on rebel governance, the difficulty in obtaining reliable data during active conflicts has limited our understanding of internal variation in governance and service provision. This paper uses a within household cohort design drawn from DHS reporting of educational attainment to look at service provision by the Fources Nouvelles in Cote d’Ivoire. This measure if validated by existing surveys on service provision. I test several existing theories around service provision which have contrasting predictions in the current literature, and show that the patterns of provision in this case on consistent with a group focused on building long term, inclusive civilian support rather than short term resource gathering.
Works in Progress
“How to Use Masked Language Models to Classify Narrative Structure in Political Texts”
Abstract
This paper develops an original method for classifying the latent structure of a piece of text using a Masked Language Model to classify the similarity of component parts and then scoring documents by the overall ordering of each category. Though there is a wealth of work categorizing text based on its topic and tone, there is little methodology for sorting text by the way it is structured. This is despite the fact that anyone who speaks or writes thinks carefully about the best order in which to convey information and the subsequent impact on an audience. For instance, relating information as a story rather than in a straightforward man ner has been shown to be more persuasive and increase empathy. However, in contexts where time is limited or an audience is more skeptical, it may be best to lead with the main point. Being able to classify the structure of text will allow us to both understand how politicians convey information and the contexts in which this varies. The benefits of the method include that it can be used on short form documents, which are often more difficult to break into discrete parts, and each stage can be verified by an analyst rather than black-boxing what is being classified.
“Replication for Language Models: Problems, Principles, and Best Practice for Political Science.” with Arthur Spirling and Christopher Barrie
Abstract
Excitement about Language Models (LMs) abounds: these generative tools require minimal researcher input and yet make it possible to annotate and generate large quantities of data. But while LMs promise to replace conventional approaches to our efforts, there has been almost no systematic research into the reproducibility of research using these methods. This is a problem: the status quo for their use lacks the scientific integrity we expect in our field. We give a new theoretical framework for replication in the discipline and show that much LM work is uniquely wanting. We then demonstrate the problem empirically using a rolling iterated replication design in which we compare crowdsourcing and LMs on repeated, multiple tasks, over a long period of time (six months). We find that while LMs can match or exceed the accuracy of crowdworkers, the observed variance in LM performance is unexpectedly and unacceptably high. Indeed, in many cases the LM findings cannot be re-run, let alone replicated. We conclude with preliminary recommendations for best practice.
“Cynicism and Corruption in Kenya.” with Gwyneth McClendon
Abstract
A robust literature outlines a vicious cycle between political corruption and a lack of electoral accountability in many countries. Some studies characterize particular countries as so steeped in corruption that it becomes ``normal" and acceptable to voters. Through an original, nationally representative survey in Kenya conducted in 2024, we argue for additional nuance in these claims. We consider cynical beliefs---beliefs that others are badly intentioned and/or deeply selfish---and their possible role in perpetuating voter support for corrupt politicians. But we underscore an important distinction between two distinct types of cynical beliefs---(1) cynical beliefs about political elites alone versus (2) cynical beliefs about human nature more generally---and argue that these two types of cynical beliefs have different implications for how citizens respond to purportedly ``clean" politicians. Citizens who hold cynical beliefs about politicians alone are more open to the possibility that there are ``clean" candidates out there and thus are more willing to divert their votes to candidates running clean campaigns and away from those who are openly corrupt. By contrast, citizens who hold cynical beliefs about human nature in general (about the motivations both of politicians and of the more general population) are inclined to believe that all possible candidates are corrupt. They are thus more susceptible to campaign claims that seemingly ``clean" candidates are actually corrupt and are more likely to vote for openly corrupt politicians than those running an ostensibly clean campaign. We discuss implications for the literature on accountability and on populism.
Other Publications
“Using proprietary language models in academic research requires explicit justification.” Natural Computational Science (2023) https://doi.org/10.1038/s43588-023-00585-1 With Arthur Spirling and Noah Smith.