Social Media

Pernicious Personalization. A Political Audit of the Twitter Recommender System
(with Simon Roth and Fabio Votta)



Although social media only recently emerged, the accumulation of evidence undermining the ‘echo chamber’ hypothesis is striking. While self-selective exposure to congruent content - the echo chamber - is not as salient as expected, the ideological bias induced primarily by algorithmic selection - the filter bubble - has been less scrutinized in the literature. In this study, we propose a new experimental research design to investigate recommender systems. To avoid any behavioral confounder, we rely on automated agents, which 'treat' the algorithm with ideological and behavioral cues. For each agent, we compare the ideological slant of the recommended timeline with the ideological slant of the chronological timeline and, hence, isolate the ideological bias of the recommender system. This allows us to investigate two main questions : (1) how much bias is induced by the recommender system? (2) what role is played by implicit and explicit cues, when triggering ideological recommendations?

The experiment has been pre-registrated and features 170 automated agents, which were active for three weeks before and three weeks after the 2020 American presidential election. We find that, after three weeks of delivering ideological cues (following accounts and liking content), the average algorithmic bias is about 5%. In other words, the timeline as structured by the algorithm entails 5% less cross-cutting content than it does when it is structured chronologically. While the algorithm relies on both implicit and explicit cues to formulate recommendations, the effect of implicit cues is significantly stronger. This study is, up to our knowledge, the first experimental assessment of the ideological bias induced by the recommender system of a major social media platform. Recommendations are biased and rely above all on behavioral cues unwarily and passively shared by the user. As affective polarization becomes a greater contemporary challenge, our results raise important normative questions about the possibility of opting-out from the ideological bias of recommender systems. In addition, it points out that more transparency is urgently needed around the recommendation questions: How are algorithms trained? What cues or features do they use? Against which biases have they been tested? In parallel, the results demonstrate the failure of ‘in-house bias correction’ and calls for an external auditing framework, that would facilitate this kind of research and crowd-sources the scrutiny of recommender systems.

Fifteen Seconds of Fame: TikTok and the Supply Side of Social Video
(with Fabio Votta and Kevin Munger)



TikTok has rapidly developed from a punchline for jokes about ``kids these days" into a formidable force in American politics. The speed of this development is unprecedented, even in the rapidly-changing world of digital politics. Through a combination of hashtag and snowball sampling, we identify 11,546 TikTok accounts who primarily post about politics, allowing us to analyze trends in the posting, viewing and commenting behavior on 1,998,642 tiktoks they have uploaded. We test a number of theories about how the unique combination of affordances on TikTok shapes how it is used for political communication. Compared to the dominant platform for political videos (YouTube) we find that a higher percentage of TikTok users upload videos, TikTok view counts are more dominated by virality, and viewership of videos are less dependent on a given accounts' number of followers/subscribers. We discuss how these findings affect the production of content that ultimately determines the experience of TikTok consumers.

Measuring Ideology

Measuring MPs' Ideological Position: Expert Pairwise-Comparisons
(with Christian Breunig and Simon Roth)


The ideological position of individual legislators is a central concept in comparative politics. A particularly thorny problem is estimating these positions in legislatures with strong partisan discipline and are rare opportunities of unconstrained voting. We provide a novel approach for estimating legislators' ideological positions that comprise of three core features. First, we survey members of parties' youth organisation as they possess detailed knowledge about elected representatives. Second, these experts are asked to compare pairs of legislators along a left-right dimensions. Finally, we utilize these data for estimating a Bradley-Terry model that generates an ideological position of each legislator. As an empirical illustration, we estimate the ideological position of the 709 members of the German Bundestag. Several validity tests show that our model captures variation within and across political parties well.

Partisan Semantic Overlaps: Floor-speeches and Ideological Position
(with Simon Roth)

Paper Slides


Estimating ideological position has always been challenging for political scientists. The technical progress of the last decades -digitalization, computational improvements- opened new opportunities to measure ideological position. While classical approaches relied on surveys, new strategies relying on other sources of data such as social networks or text have been explored. This paper develops a strategy to estimate the ideological position of representatives on the basis of their floor speeches.

One big challenge when scaling from text is to ensure that the captured dimension matches the left-right axis. Existing models deal differently with this issue: Wordscores requires the identification of two documents meant to represent the extreme points of the dimension, Wordfish assumes the principal component maps the left-right and Wordshoal require to filter the document to build a homogeneous corpus structured around the dimension to measure. We propose another approach, which relies on a two-stage model. First, we train a neural network to predict the party of each speaker. Thus, we obtain for each document a vector of probabilities, which informs on the affiliation to each party. In other words, we represent the speeches in a low-dimensional space, where each dimension corresponds to one party. In the second stage, we further reduce the dimensions to obtain one linear scale, expected to map the left-right dimension.

To test the measure, we use each floor speech pronounced in the German, British, Spanish and French Parliaments in the last two decades (~2000000 Speeches). The speech-level ideological point estimation allows representing each representative as a distribution of ideological position, which offers more flexibility than single-point estimation. After validating the measure using more classical measure of political ideology, we use it to substantially investigate how the ideological position of representatives responds to electoral incentives.

Scaling Models in Political Science
(with Daniel Braby and Marius Sältzer)


Models of political competition, questions of representation and voting behaviour alike require an understanding of what policies political actors prefer, offer and promote. To make positive assessments of these processes, political scientists have developed and refined tools to measure the relative positions of political actors on dimensions of politics.

Where do parties or voters stand on the left-right dimension? What policies define it? Is left-right really the most salient dimension in a political system? To answer these questions, so-called scaling models have been developed to estimate the positions of actors on the basis of their observable behaviours. To this end, many data sources have been explored: from voting behaviour in parliament (Poole and Rosenthal 1985), over party manifestos (Budge 2001) and parliamentary speech (Lauderdale and Herzog 2016), to positions in social media networks (Barbera 2015). This contribution gives an intuition on how this can be achieved, an introduction to the math behind them and provides an overview of their applications, both technical and substantive. We close with criticisms of these methods, which speaks to the development of further validation methods.


Elections Sets the Agenda: Mandate Responsiveness in Germany
(with Isabelle Guinaudeau)



In democracies, elections are meant to shape public policy. But how much leeway do elected representatives actually have to implement their mandate? Influential scholars think that constraints linked to regional integration, budget restrictions, and counter-majoritarian institutions dilute mandate responsiveness. However, empirical evidence for this important claim remains scarce. This article provides an empirical account of the extent to which different types of constraints limit governing parties’ ability to set their electoral priorities on the agenda. Using panel negative binomial regression of German electoral and legislative priorities over a period of over three decades (1983-2016), we conclude that – even when controlling for most confounders – electoral priorities affect policies to a greater extent than scholarship has acknowledged so far. We confirm, however, the constraining effect of Europeanization, shrinking budget leeway, and intra-coalition disagreement. We elaborate on the implications for theories of public policy, democratic representation, and comparative politics.

Do Parliaments Legislate? A Text Comparison Approach



Legislative studies have extensively benetted from the recent improvement of textual analysis

and scholars have learned to take advantage of the large amount of text produced by legislatures.

Among others, they have used text to estimate the ideological position of individual actors;

they have expanded the scope of their previous analyses by automating the coding processes of

textual documents and they also have improved our understanding of the strategic behaviours of

legislative actors, by looking in detail at their verbal interactions.

Following those developments, this paper proposes to take advantage of the textual modications

adopted during the legislative review, to estimate the amount of inuence, that a parliament exerts.

In doing so, it addresses an old -but widely accepted- idea, that, in parliamentary democracies,

most policies are written by the government all alone and the parliament is restricted to a mere

adoption role. Despite the large consensus supporting this expectation, there is only limited

empirical evidence.

Using data from the British House of Common and the German Bundestag and following an idea

rst proposed by Martin and Vanberg (2011), I compare the introduced and the nal versions of

each bill adopted during the last decade. Then, I compute the number of adopted modications

and obtain a Parliamentary Inuence Score (PIS). This PIS is expected to capture the extent to

which a bill is inuenced by members of Parliaments. The rest of the paper is dedicated to the

validation of the measure.