Research

Political Image Analysis with Deep Neural Networks

Abstract

Politicians and political organizations routinely interact with voters and the public at large using images, yet until recently, computational limitations have precluded efforts to gain systematic knowledge about how images function as a medium of political communication. New developments in machine learning, however, are bringing the systematic study of images within reach. In this paper, we provide a framework for political image analysis with deep neural networks. In addition to introducing neural networks and deep learning methods, we discuss some of the promises and pitfalls of these techniques for political image analysis. Using a database of 296,460 photos from the Facebook pages of members of the U.S. House and Senate, we provide two illustrative examples of how these techniques can be used to study home style

Measuring and Modeling Legislative Accomplishment with Directed Graphs (with Anthony Madonna)

Measuring legislative accomplishment and the productivity of political institutions is fundamental to understanding the political economies and trajectories of democratic nations. Recent research measuring legislative accomplishment has enabled scholars to assess the importance of legislation across a wide range of time but entails cumbersome and expensive methods which do not allow for continuous updating of new pieces of legislation past 1994. In this paper, we develop an algorithm which allows us to constantly update and track the relative importance of legislation over time as enacted legislation becomes available. This is accomplished by first modeling enacted legislation across time as a directed network of citations using bill text and changes to the United States Code (amendments, repeals and additions). The importance of each piece of legislation in this network is then measured as a function of its pagerank centrality. Using this new measure, we reassess the importance of several pieces of key legislation and the productivity of Congress from 1926, the year that the first edition of the United States Code was published, to 2017.

Understanding delegation in the European Union through machine learning (with Anthony M. Bertelli)

The delegation of powers by legislators is essential to the functioning of modern government, and presents an interesting tradeoff in multi-level states such as the European Union (EU). More authority for member states mitigates ideological drift by the European Commission, but less authority reduces the credibility of commitments to centralized policies. Extant empirical studies of this problem have relied on labor-intensive content analysis that ultimately restricts our knowledge of how delegation responded to legislative and executive power changes in recent years. We present a machine--learning approach to replicating the content analysis of 158 laws between 1958--2000 by \citet{franchino2001delegation,franchino2007powers} that will train'' classifiers to examine EU laws enacted since 2000 in a similar way. Using the trained classifier with the highest overall performance, we introduce probabilistic delegation ratios (PDR) as an alternative to the delegation ratio first introduced by~\cite{epstein1999delegating} and also demonstrate that our trained classifier is able to automatically estimate delegation ratios in legislation as well. While our principal interest is in the European Union, the method we employ can be used to understand delegation in a variety of contexts.

A Formal Model of Segregation and Political Polarization

Racially segregated cities tend to be politically polarized cities, leading to inequalities in public goods provision, political and social isolation, concentrated poverty and the perpetuation of a sense of hopelessness among many living in America's urban centers. While the links between racial segregation and political polarization are well established, it is less clear why, or through what mechanism, both arise simultaneously. In this article, we derive a formal model which we demonstrate can partially account for this puzzle. This model allows us to derive ideological tipping points'': changes in neighborhood demographics at which all members of one or more groups along the ideological spectrum (liberal, conservative, moderate) relocate. We then validate the model and demonstrate that racial segregation and political polarization consistently emerge in equilibrium under a wide variety of conditions by simulating movement of individuals between Census tracts in the largest 10 cities in the United States.

Visible Homestyles in Congress (with Dhruvil Badani, Shiry Ginosar, Crystal Lee and Jake Williams)

While members of Congress routinely communicate with constituents using images, there is little systematic knowledge about how images are used as a means of strategic communication due to computational limitations. New developments in computer vision, however, are bringing the systematic study of images within reach. We develop a framework for understanding visual political communication by extending Fenno’s analysis of home style to images and then apply this framework to study racial representation in ∼19,000 images collected from MCs Facebook profiles using a machine learning technique known as convolutional neural networks. We demonstrate that Democrats and Republicans in the House of Representatives strategically use the race of individuals that they pose with for political ends. When compared with their district demographics, Democrats tend to “over-represent” African-Americans in photos that they post on social media while Republican House members tend to “under-represent” African-Americans in their photos.

The Logic of Modern Collective Action (with Jake Williams)

The ubiquity of social media has given rise to new forms of collective action which are not well understood. Here, we introduce a theory of social action which we demonstrate can be applied to systematically identifying these new forms of collective action. To accomplish this, we used this theory to design a series of Bayesian machine learning classifiers along with a subsampled database of over 600 million geo–located social media posts collected between April 1st, 2014 and April 30th, 2015. We demonstrate that these algorithms can be used to accurately detect peaceful and violent collective action in social media at fine–grained levels of time and geography.