I am a PhD Candidate at the LSE.

I am interested in Text as Data and other Data Science-based approaches in social sciences.

News about Ukraine on Russian Domestic Television, 2009-2019

Solo-authored. Under review

This article is the first study to expose large-scale evidence of attempts by Russian state-controlled domestic television to amplify the importance of the Ukrainian agenda for its audience in the years prior to the full-scale war. Drawing on a data set composed of 2.4 million transcripts of news stories transmitted in Russia between 2009–2019, I offer the evidence of mass media agenda-setting and distraction. First, applying the difference in differences technique, I show that state-enforced change in news management in 2016 induced previously independent television network RBC to increase its reporting about Ukraine by more than half. Second, I demonstrate the that on nine domestic media outlets, less intense reporting about domestic events was attributed to more reporting about the events in Ukraine rather than in other countries. The data set employed in this article covers most of popular national domestic television news flow during the period.

Key words: autocratic resilience, Russia, Ukraine, television, news

How Russia Solved Dictator's Dilemma in 2019

Solo-authored. Work in progress

How do news manipulations in Russia vary across different types of mass media? I conjecture that state-controlled news published online, for consumption by a digitally literate audience, exhibits less pro-regime bias than programming created for those who rely on national television to learn about daily affairs. I test the conjecture using two data sets: headlines displayed on the Yandex News Aggregator and transcripts of evening news on Channel One, examining how daily reporting varied in topics across in two media outlets. Using crowdsourcing, supervised machine learning, and dictionary techniques, I analyse 70,000 news reports from 2018-2020. Results indicate that the state-controlled online platform, in contrast to national television channels, demonstrates less tendency to censor stories about political opposition and bad economic news, but promotes stronger measurable bias in the coverage of foreign affairs, specifically news that involves Ukraine and the US.

Key words: autocracy, media bias, news aggregators, Russia

Are domestic war crimes trials biased? Evidence from Serbia

with Ivor Sokolić (LSE), Denisa Kostovicova (LSE), and Sanja Vico (LSE). Work in progress

How nations prosecute war crimes committed by their members after inter-group violence is an unassailable indicator of their commitment to peace and reconciliation. Because nations are challenged to come to terms with their wrongdoing after an identity conflict, domestic war crimes trials are thought to be inherently biased towards one’s ethnic group. Scholars have argued that nations are not willing to prosecute in-group members, and, if they do, they are more lenient towards them than towards members of the out-group(s). On the one hand, these claims have not been put to rigorous scrutiny. On the other, scholarly preoccupation with ethnic bias because of salience of identity issues in perpetration of violence, has resulted in their overlooking other conflict dynamics and their possible impact on war crimes prosecutions. We ask: who is punished for war crimes? We test the in-group bias argument, along with the pattern of punishment of different types of conflict actors, distinguishing paramilitaries from state security forces. We apply statistical modelling and quantitative text analysis to two original datasets based on 559 verdicts delivered to defendants in domestic war crimes trials in Serbia from 2000 until 2019. While we do not find evidence of ethnic bias, we find evidence of bias against paramilitaries belonging to an in-group disproportionately who are punished disproportionately in comparison with state security forces. The study advances the emerging scholarship, overwhelmingly focused on international criminal justice, between war crimes trials and peace-building. We show how the strategy of states shifting responsibility for wrongdoing in which they are implicated to non-state actors obstructs peace-building and reconciliation even when sentencing is not ethnically-biased.

Key words: transitional justice, content analysis

It Depends on Who Is Asking and Who Is Being Asked: Gendering Speaking Behavior in Parliamentary Interactions

with Denisa Kostovicova (LSE), Tolga Sinmazdemir (SOAS), and Vesna Popovski (LSE). Work in progress

Women have a distinctive voice in parliaments. Analysis of female legislators’ speaking behavior, including floor apportionment and policy issues they address, informs inferences about their effectiveness. In this vein, less adversarial tenor of female MPs’ speeches has been taken as an indicator of their less robust questioning of the members of the executive. To advance our understanding of the gendered patterns of legislative behaviour and their implications, we draw on the scholarship on gender and language in the field of social psychology and scrutinize the legislators’ speaking style. We analyze the text corpus comprised of parliamentary questions and answer sequences in the Croatian parliament from 2004 and 2020, and conduct quantitative content analysis that combines computer-assisted techniques and human coding of the discourse. A novel assessment of how female legislators’ interrogate ministers, rooted in Conversation Analysis―an approach to the study of interactive features of talk―systematically captures linguistic, deliberative and semantic features of discourse. The results show that female parliamentarians are no less robust questioners than their male counterparts. They put pressure on the answerer to account for government policy using different linguistic forms, rather than open aggression evident in semantic features of a speaker’s language. This research furthers the study of gender and politics by demonstrating that the masculine discourse in parliaments is used not only to ensure male dominance in parliaments, as scholars of feminist institutionalism have argued, but also to obscure the agency of female MPs and their effective oversight of the executive.

Key words: gender, legislature, content analysis

How is Russia studied? Meta analysis of publications in leading political science and economic journals

Work in progress

A working paper for a special issue of PSA

Grand-Standing Instead of Policy-Making: Legislators, Parliamentary Questions and Transitional Justice in the Croatian Parliament

with Denisa Kostovicova (LSE)

We know little about how legislators engage with post-conflict justice in contrast to our good grasp of the role of domestic and transnational civil society groups in policy deliberation. We study how politicians ask questions about transitional justice. Parliamentary questions, which can be oral and written, are an important tool used by legislators to hold government to account. We argue that publicness of parliamentary questions matters. When a politician asks a question publicly in parliament, they are concerned about reputational costs to themselves and to their party. In post-conflict societies, the costs are determined by the degree of dissent from dominant nationalist norms. We analyze 738 parliamentary questions about transitional justice in the Croatian Parliament (2004-2018). We find differences between oral and written questions. Legislators belonging to nationalist parties use oral oral questions more than liberals, and more than written questions. Overall, the share of oral question about war veterans, the most privileged stakeholder, is larger than their share in written questions. Lastly, oral questions are used to demonstrate partisanship, while written questions are used for policy deliberation across party lines. We identify the limitations of public policy deliberation on post-conflict justice, where parliamentary questions are used for nationalist grand-standing.

Entropy-based Newsmap

a very early stage project with Kohei Watanabe

A methods paper about an improved version of Newsmap

Newsmap is a semi-supervised Naive Bayes-based model for geographical document classification. While fully supervised models are trained on manually classified data, Newsmap learns from “seed words” in toponyms dictionaries which helps classify large data sets without relying on human coders. We offer a new, entropy-based approach which helps achieve higher classification accuracy and precision.

Key words: news, semi-supervised learning, text classifiers

Blog post about Telegram and its role in protests at Media@LSE

Protests in Russia raise questions about the role of new media in democratisation

Gender bias in news

frozen project started at SICSS London

with Anastasiia Menshikova (LiU SE) and Katharina Lawall (LSE)

we are training word embeddings models (mostly BERT) on a set of large English language news corpora

Key words: gender, news