Summary: False information about the legitimacy of recent American elections has prompted a barrage of harsh rhetoric against the officials who administer them. We measure how this negativity has played out on Twitter: the extent of this negativity, how it has trended over time, which state administrations are targeted by it most, and what sorts of accounts are sending it. We show that the usage of keywords related to election fraud has spiked in recent years, while the sentiments of the replies have grown almost universally harsher.
American election results at the precinct level. Nature: Scientific Data, 2022. With Alexander Agadjanian, Declan Chin, John Curiel, Kevin DeLuca, James Dunham, Jennifer Miranda, Connor Halloran Phillips, Annabel Uhlman, Cameron Wimpy, Marcos Zárate, and Charles Stewart III.
Summary: We describe the process of collecting, standardizing, and quality assuring precinct-level results of elections in the United States. The MIT Election Data and Science Lab collected the vote totals for nearly every candidate in nearly every office above the very local level in the 2016, 2018, and 2020 American elections, and we show that the resulting 40 million row dataset is extremely accurate.
Summary: We identify an unexpectedly strong consensus about which countries have been democratic in which years. While searching for the cutpoint that dichotomizes many-valued measures of democracy so that they match different binary measures of democracy as closely as possible, we found that, out of a thousand possible values, several cutpoints are exactly the same.
Summary: We re-examine the effects of state capacity, executive regime type, and party systems on democratic stability and quality. The precision and coverage of Varieties of Democracy data allows us to consider how institutional arrangements affect the probability of a democratic step-up or a democratic step-down when democracies are defined in hundreds of different ways.
Summary: One in five political scientists with biographies on Wikipedia are women, while nearly half are American. Biases on Wikipedia can cause real harm, so I created or expanded a political science-related Wikipedia article every day for a year, focusing on writing new pages about political scientists from underrepresented groups.
Working papers
Simulating electoral system changes and how voters might respond
Abstract: A crucial determinant of electoral system choice is how many seats a specific political party would win under some alternative electoral system. For decades this quantity has been estimated using a simple approach that is known to have multiple major flaws. I address these problems by developing a computational formal model that estimates how the seat totals in a particular single-member district plurality election might have been different if the election had been held under another electoral system, explicitly simulating how strategic voters might have voted differently. The model is falsifiable, and outperforms the widespread simple approach when applied to New Zealand’s electoral system change. I simulate recent Canadian and British elections under ranked-choice voting and proportional representation. The results suggest that several parties opposed election reforms that, when voters’ preferences are imputed from survey data and strategic voting is accounted for, could actually have benefited those parties.
How strategic are most voters? Evidence from simulations
Abstract: Election results do not just represent peoples’ true preferences, but is that because most voters are slightly strategic, or because a few voters are highly strategic? This study examines the prevalence of four types of voters: sincere voters, partially strategic voters, precisely strategic voters, and those who vote based on party identification. A new method for comparing the results of a computational formal model to real election results suggests that many voters are slightly strategic, some voters are highly strategic, and a smaller number of voters are not strategic at all. A novel partially strategic heuristic is shown to be highly accurate, and models with only sincere voters actually produce less accurate results than models where every voter maximizes their utility. The results suggest that sincere voting, partially strategic voting, and precisely strategic voting were all prevalent in recent Canadian and British elections.
The probability of casting a pivotal vote in Instant Runoff Voting
Abstract: I derive the probability that a vote cast in an Instant Runoff Voting election will change the election winner. I show that there can be two types of pivotal event: direct pivotality, in which a voter causes a candidate to win by ranking them, and indirect pivotality, in which a voter causes one candidate to win by ranking some other candidate. This suggests a reason that voters should be allowed to rank at most four candidates. I identify all pivotal events in terms of the ballots that a voter expects to be cast, and then I compute those probabilities in a common framework for voting games. I provide pseudocode, and work through an example of calculating pivotal probabilities. I then compare the probability of casting a pivotal vote in Instant Runoff Voting to single-vote plurality, and show that the incentives to vote strategically are similar in these two systems.
Abstract: Election forensics identifies fraudulent activity from empirical distributions of election turnout and vote choice, but strategic behavior can affect these distributions in ways that might resemble fraud. Many types of election forensics interpret a multiplicity of modes in election data as indicators of frauds, but strategic behavior induces correlations among electors' behavior that can produce multimodalities. We design simulations to match equilibria derived in models of wasted vote logic and of strategic abstention, and we simulate elections in which we know that strategic behaviour is present while fraud is not. The results suggest that fraud detection models might misidentify legitimate strategic coordination as a signal of election fraud.
Abstract: In the period before a large democratic election, thousands of electors, with diverse preferences and very different ways of making decisions, update their ideas about which candidates are competitive and how they should vote. Iterative computer models allow us to study this process without abstracting away the notion of time, in a framework that avoids some of the less realistic assumptions of pencil-and-paper models. But how do we know when the model has converged to an election result? I demonstrate that certain broad categories of iterative election models produce cyclic behaviours in the electors' strategic intentions, and that those cycles obey simple rules. This opens up a new modeling framework that captures some of the dynamics of elections better than existing formal models can.