My principles are twofold. First, we will not be saved by big data; we will be saved by the analytical techniques that allow us to draw actionable, clear insights from them. Second, a method is only as useful as the substantive conclusions you are able to draw because of it.
Data science and “big data” are increasingly used to conduct empirical research in politics and economics. For example, scholars frequently use natural language processing to predict political ideology from text, or to scale news organizations against voting outcomes. The validity of these methods are questionable, however, because we lack a theoretical framework for what data are permissible to link together, and when one ought to use certain data science tools to model them. In fact, the issue of how to preserve inferential power when linking disparate data is one of the least understood aspects of model specification. Such considerations are perhaps obfuscated further in big data analyses because the sheer scale of the data seems to make classical inference irrelevant.
My research demonstrates how big data methods, particularly analyses involving machine analysis of text, can introduce non-ignorable bias and inefficiency into results and undermine conclusory exercises. I then develop theoretically supported frameworks to overcome this bias and inefficiency, and derive statistical methods to operationalize these frameworks. The subject matter is part of the emerging field of computational social science, which combines data science and political science to conduct research at scale.
Substantively, I focus on political behavior. My greatest interest is in the present phenomenon of affective polarization, in which voters express extreme dislike for each other without basis in issue disagreement. One manifestation of this phenomenon is in the way we socially sort through relationships: believe it or not, marriages between members of opposite parties have decreased dramatically over the past three decades! What explains this phenomenon, and what are its implications? How can we mitigate it?
One of my favorite findings, which Jon Rogowski and I published in the journal Political Behavior, is that we may be able to overcome affective polarization by presenting individuals with “humanizing” information about the other side. The RCT suggested that when voters were treated with small vignettes about the out-group’s day-to-day life, even the most extreme haters came around and moderated their evaluations of the out-group. Due to partisan sorting, issue consistency, and economic trends in technology and inequality, perhaps we just don’t see enough of the other side these days.
Another one of my areas of interest that I consider to be extremely important, is trends in levels of self-censorship by Americans. A person “self-censors” when they feel the cost of expressing their true opinion is so punitively high that they would rather keep their mouth shut, even in the face of potentially harmful dominant opinions. As my research with Jim Gibson shows, fully 4 in 10 Americans today do not feel free to speak their minds (yes, today in 2020!). Levels of self-censorship today are 4 times as high as they were during the Red Scare of the 1960’s—a time when simply appearing to sympathize with the communist persuasion could render you ruined and in jail.
What is the root cause of this finding? It’s unrelated to macro trends in political intolerance and affective polarization, but it is related to factors that increase socialization, such as education and metropolitanism. People with higher levels of education self-censor more; so do people who live in cities. We hypothesize that a micro-level mechanism called the “spiral of silence” has potential as an explanation—we self-censor because negative feedback from those around us causes us to recoil from honest public discourse. The problem is, of course, that these micro-level phenomena are impossible to study with survey data (…enter the next design proposal!).
In a time of increasing partisanship and polarization, it is easy to overlook the fundamental role new technologies play in defining the relationship between parties and voters. New technologies, including depersonalized communication at scale, human–machine interaction, and enhanced prediction have precipitously reduced the costs of voter engagement. Are voters starting to catch on? For example, initial evidence in my research with Don Green and the DNC suggests that as the costs of engagement have decreased, the value of costly signalling has increased. In a world where contact no longer suggests care, voters only respond to interventions where the cost of engagement is perceived to be high
At the core of my research is an obsession with understanding how the digital transformation of our society – first niche television, then the internet, now cell phones and predictive personalization at scale – has changed the distribution of power in our society. These topics in political behavior are intimately related with, and thread through, technology and the new digital economy. Together, they may provide a powerful engine of insight into the ways society will resolve collective action problems in the future.