Working Papers

  1. Dark-ish Money: What ‘Pop-Up PACs’ Can Teach Us About Donors and Disclosure.

    Abstract: Over the past decade, the amount of “dark money”–or independent expenditures from groups that don’t disclose their donors–has grown to encompass a significant slice of total spending in American elections. How does the absence of disclosure affect the behavior of both donors and electioneering organizations? While the secretive design of dark money makes data on them extremely limited, the 2018 election provided a useful stand-in to examine the role of disclosure in elections. ‘Pop-Up PACs’ are super PACs that form just after the last disclosure federal deadline before an election–allowing them to postpone revealing their donors until after voters have gone to the polls. In total, 63 super PACs employed this tactic during the 2018 midterms, spending a combined $21.6 million. Examining these late-disclosing super PACs, I find that they raise more money from other committees–indicating that they perceive that voters will penalize them for associations with certain donors. Further, I find that donors who gave to Pop-Up PACs supported a far less ideologically extreme set of candidate when giving directly and transparently to other candidates. These late-disclosing super PACs also supported more ideologically extreme candidates than did super PACs who disclosed their donors prior to an election. In sum, these results add credence to theories that donors alter their behavior in response to social pressures and that being able to hide the sources of one’s money correlates with an electoral organization eschewing moderates in favor of extremists.

  2. Extremism Grows in the Dark: Political Non-Profits, Disclosure and Ideological Interest Group Spending.

    Abstract: Following the Supreme Court’s 2010 decision in Citizens United v. FEC, independent expenditures have grown dramatically both in terms of raw dollars and as a percentage of spending in elections. A large and growing portion comes from political nonprofits—so called “dark money” groups—named because the terms of their incorporation allow them to partially obscure the sources of their income. I posit that the pathways for anonymous giving that emerged from the Citizens United decision allowed ideologically motivated interest groups to aggressively challenge more established factions of political parties in way previously unfeasible. Testing this theory, I find strong support that dark money groups back more extremist candidates—especially during primary elections—than either formal party organizations or access-oriented interest groups. These results indicate that the anonymous pathway to giving offered by dark money has created a new font for ideologically motivated interest groups to spend in American elections.

  3. The Noisy Neighbor Effect: How Negative Advertising in One State Influences Viewers Next Door.

    Abstract: The Supreme Court’s 2010 decision in Citizens United v. FEC unlocked a flood of new political advertising at all levels of government, leaving voters to decipher an unprecedented level of information before eventually casting their ballots. How competent are voters at sorting through all this information and shaping opinions based on relevant information? I exploit variations in state-election law and the timing of gubernatorial races to create a natural experiment on how seemingly irrelevant information can shape voter attitudes. I find that, all else equal, those living in multi-state media markets who were exposed to negative television ads for a gubernatorial race taking place in a neighboring state were more disapproving of their own governors than other residents of their state. This same exposure to out-of-state gubernatorial ads also led residents to rate their own governor’s as more ideologically distant from them. In a saturated era of political information–with a portion of it coming from active misinformation campaigns–the ability of voters to sift through and accurately process political messaging is vital to the function of democracy. These results call that ability into question and raise new implications for the regulation of political advertising in American elections.

  4. Buying the Ballot: Political Actors and Official Ballot Initiative Language with Mitch Downey, Economics, Stockholm IIES.

    Abstract: At the time of their inception in the Progressive Era, ballot initiatives were intended as a way for citizens to wrest control of the policy making process away from politicians corrupted by special interests. However, this measure of popular sovereignty may not be as immune to meddling these same special interests as either the founders of direct democracy or the literature suggests. Many voters learn about initiatives only from the titles and summaries appearing on ballots, which are themselves written by elected politicians. We study that special interest groups influence the wording of these official ballot initiative summaries, showing that summaries are more likely to use supporters’ preferred language as supporters contributed more to the summary writer. This relationship holds true even when the summary writer is running for a different office and does not exist in summaries written by an independent, apolitical non-profit group. We then show that donors are more likely to give to Attorney General candidates when relevant initiatives appear on the ballot. This result is not reflected in broader state-level political activity and exists only in states where the Attorney General writes the summary. Our results show the risks of tasking elected officials with fundamentally bureaucratic tasks and indicate that this process designed to reduce the power of special interests over the legislative process may have actually given them another avenue of control.

  5. The Supply and Demand of Fact v. Opinion in Presidential Tweets. with Thad Kousser, UCSD Political Science.

    Abstract: Do presidential candidates, and now President Trump, use social media to lay out the factual basis for their positions, or do they primarily communicate their values and opinions? We investigate this question by using human categorizations of over 8,000 tweets to train machine learning algorithms to code all of the tweets sent by all 23 candidates during and since the 2016 presidential election campaign. We argue that politicians should supply more opinions than facts, in order to match voter demand for opinion over fact and increase audience engagement with their tweets. In a descriptive analysis, we chart the flow of factual claims – many of which happen to be false, according to prominent fact-checking websites – versus opinion over the course of the campaign and across candidates. Donald Trump leaned heavily toward opinion during the campaign, but so did Hillary Clinton and many of the of the most successful candidates. Turning to the demand for tweets, we analyze the rate at which a candidate’s tweet is liked or retweeted, relative to her number of followers. We find that, for nearly every candidate, tweets that espouse an opinion generate significantly more engagement than factual claims. Negative opinions are especially engaging. These findings persist in multivariate models that hold constant the tone, subject matter, and ideology of the tweet, providing strong evidence that voters are more responsive to opinion than fact on social media.

  6. All the President’s Tweets: Studying the Big Data of Twitter Political Communication with a Small Data Approach. with Thad Kousser, UCSD Political Science.

    Abstract: This paper describes and demonstrates an approach to studying political communication on Twitter that combines the flexibility and subtlety of human coding on small datasets with the scalability of big data methods. We apply this method to study the 152,134 tweets sent by candidates and their SuperPACs during the 2016 American presidential election. We begin by outlining the importance of studying this increasingly prominent form of political communication, laying out both the opportunities and obstacles to analyzing the massive volume of social media political texts. We then explain our supervised learning method to classify the ideology and sentiment of tweets, what policy spheres they address, whether they make factual statements or espouse opinions, and other characteristics. To demonstrate what this approach to studying political communication can tell us about a campaign, we then focus on our measure of ideology. We map the candidates on a spatial spectrum based on their social media rhetoric, explore Donald Trump’s unique brand of populism by looking at where he placed himself on different issues, and asked whether he and Hillary Clinton followed the Downsian impulse to moderate their message from the primary to the general election.