Research
Political violence, online extremism, and computational social science
Why do movements embrace violence that undercuts their own aims, and what do the communities around that violence get out of it?
I treat digital platforms not as communication channels but as constitutive environments, where political identities form, grievances are framed, violence is celebrated and remixed, and the next act is collectively shaped. Answering that question at scale requires measurement strategies that surveys and interviews cannot supply, so my empirical work combines transformer-based text models, network analysis, and time-series causal inference with formal theory and archival work on the movements themselves.
A longer version of this agenda is on the research statement page. Code, models, and datasets are described under code and data.
Job market paper
“The Happening Is Coming”: Apocalyptic Rhetoric on 4chan Following Mass-Casualty Attacks
I train a transformer classifier to measure apocalyptic rhetoric across eleven years of 4chan /pol/ posts, then use intervention analysis to test how that measure responds to mass-casualty terrorist attacks. Apocalypticism rises sharply in the immediate aftermath of attacks — evidence that online communities participate in violence after the fact, and the first empirical piece of the dissertation’s account of how celebration feeds future attacks. Presented at APSA 2025; draft available on request.
Publications
Peer-reviewed articles
Newhouse, A. and Kowert, R. (2025). “Extremist Identity Creation Through Performative Infighting on Steam.” Frontiers in Psychology.
Extremist identity crystallizes through performative displays of ideological purity and intra-group conflict, showing how gaming communities sustain cohesion and radicalization without central leadership.
Gaming Extremism Identity formation
Kowert, R., Kilmer, E., and Newhouse, A. (2024). “Taking it to the Extreme: Prevalence and Nature of Extremist Sentiment in Games.” Frontiers in Psychology, 15:1410620.
A survey of 423 game players measuring the prevalence, location, and impact of extremist sentiment in games. More than half of players report encountering hate, harassment, or abuse in gaming spaces. Funded by the Department of Homeland Security (EMW-2022-GR-00036).
Gaming Survey research Online harm
Kowert, R., Kilmer, E., and Newhouse, A. (2024). “Culturally Justified Hate: Prevalence and Mental Health Impact of Dark Participation in Games.” Proceedings of the 57th Hawaii International Conference on System Sciences (HICSS).
Gaming Online harm Mental health
Newhouse, A. (2021). “The Threat is the Network: The Multi-Node Structure of Neo-Fascist Accelerationism.” CTC Sentinel, 14(5).
Atomwaffen Division was not the apex of a hierarchy but one node in a larger network of violent accelerationists held together by membership fluidity, frequent communication, and a shared goal of social collapse — with the implication that enforcement against individual groups is necessary but not sufficient.
Networks Accelerationism Political violence
Under review
Benton, A., Newhouse, A., and Philips, A. “Mind the Gap! Policy Dissonance and Financial Markets.” Under review.
Newhouse, A., Kaur, K., and Philips, A. “Partisan Polarization Shapes Visual Framing of COVID-19 in U.S. Media.” Under review.
Book chapters
Newhouse, A. and Kowert, R. (2025). “Recruitment and mobilization in digital gaming spaces.” In L. Schlegel (ed.), Handbuch Gaming & Rechtsextremismus.
Kowert, R. and Newhouse, A. (2025). “Digital games as cultural assets of influence.” In S. Lakhani and A. Amarasingam (eds.), The Sociology of Violent Extremism.
Newhouse, A. and Kowert, R. (2024). “Digital games as vehicles for extremist recruitment and mobilization.” In L. Schlegel and R. Kowert (eds.), Gaming and Extremism: The Radicalization of Digital Playgrounds.
Preprints and research reports
McGuffie, K. and Newhouse, A. (2020). “The Radicalization Risks of GPT-3 and Advanced Neural Language Models.” arXiv preprint 2009.06807.
Among the first assessments of generative language model misuse risk, showing that GPT-3 could produce interactive and influential content usable for far-right radicalization.
AI safety LLM misuse Extremism
Kriner, M., Conroy, M., Newhouse, A., and Lewis, J. (2022). “Understanding Accelerationist Narratives: The Great Replacement Theory.” Global Network on Extremism and Technology.
Shadnia, D., Newhouse, A., Kriner, M., and Bradley, A. (2022). “Militant Accelerationism Coalitions: A Case Study in Neo-Fascist Accelerationist Coalition-Building Online.” CTEC.
Kowert, R., Botelho, A., and Newhouse, A. (2022). “Breaking the Building Blocks of Hate: A Case Study of Minecraft Servers.” Anti-Defamation League.
A full list, including public writing and earlier reports, is in the CV and on the writing page.
Current projects
Dissertation — The Role of Participatory Violence in Contentious Politics
Why social movements embrace disproportionate, theatrical violence even when it makes their stated goals less likely. I introduce participatory violence — violence whose production and meaning are distributed across a large population of contributors who accrue benefits decoupled from movement success — and test the argument with a formal model, case studies of the Years of Lead, the Arab Spring, and QAnon, and transformer NLP applied to a multi-platform database of 4chan, Gab, and Telegram. Prospectus defended April 2026; expected defense May 2028. Full overview.
Committee: Andrew Q. Philips and Jennifer Fitzgerald (co-chairs), Alexandra Siegel, with Brian Keegan (Information Science) as external advisor.
Partisan framing of the pandemic
With Andrew Q. Philips and Komal P. Kaur. Partisan differences in the visual and textual framing of COVID-19 coverage across U.S. news outlets, using computer vision alongside text analysis. Presented at APSA 2025 and MPSA 2025; under review.
Policy dissonance and financial markets
With Allyson L. Benton and Andrew Q. Philips. How gaps between political rhetoric and policy action move financial markets. Under review.
The American humanities workforce
A large-scale computational study of humanities career trajectories using anonymized LinkedIn data, conducted at the National Humanities Alliance. The measurement strategy combines ensembles of local LLMs with conventional NLP for free-text career histories.
Methods
Text as data. Transformer fine-tuning for political concepts (PyTorch, Hugging Face), classical supervised text classification, and validation strategies for contested constructs. See the worked example of a DistilBERT classifier for political text.
Networks. Diffusion and contagion across forum and cross-platform networks, using igraph and statnet.
Causal inference with time series. Interrupted time-series and intervention analysis for online–offline effects, plus regression with novel observational datasets.
Data collection. Large-scale archival social media collection, survey design, and qualitative coding under sensitive-data constraints.
Data, code, and collaboration
Replication materials, classifiers, and dataset documentation are described on the code and data page. For collaboration, data access, or media requests, email alex.newhouse@colorado.edu.
Google Scholar · ORCID 0009-0003-6346-9527 · GitHub
Last updated: September 2026