Open research datasets and R packages from my research and the Social Analytics Data Lab.
Datasets
Research datasets
7 datasets · 3 peer-reviewed data papers · about 9,600 views and 2,100 downloads on Harvard Dataverse and Zenodo
SADLmarks datasets produced with the Social Analytics Data Lab.
Political Violence & Human Rights
Open · CC0
POLECAT Weekly Data
Global · 2018–2024 · News in 7 languages from 1,000+ sources
Machine-coded political event data: cooperative and hostile interactions between governments, groups and other actors, recording who did what to whom, when and where. The successor to ICEWS, built on the PLOVER event ontology, with millions of time-stamped, geolocated events.
7,438 views · 1,938 downloads on Harvard Dataverse
Scarborough, G.I., B.E. Bagozzi, A. Beger, J. Berrie, A. Halterman, P.A. Schrodt, and J. Spivey. 2023. "POLECAT Weekly Data." Harvard Dataverse. https://doi.org/10.7910/DVN/AJGVIT
Environmental Politics
SADL (produced with the Social Analytics Data Lab)
Open · CC0
Climate Change Talks Attendance Database
Every UN climate conference (COP) and pre-COP, 1991–2025 · 300,000+ attendees · 25,000+ delegations
Attendee-level records for every annual COP and pre-COP held under the UN Framework Convention on Climate Change and its precursor, extracted from the official participant lists with text-as-data methods. Includes original-language and English-translated information on attendees and their delegations.
Blinova, D., R. Emuru, and B.E. Bagozzi. 2024. "Individual Attendance Data for over 30 Years of International Climate Change Talks." Scientific Data 11: 1134.
Environmental Politics
SADL (produced with the Social Analytics Data Lab)
Open · CC0
Geo-Located CITES Conference Attendance Data
Every CITES Conference of the Parties to date · 20,297 attendee records · Geolocated
The first attendee-level dataset for meetings of the Convention on International Trade in Endangered Species, built by applying NLP to the meetings' PDF attendance rosters. Covers attendee locations, affiliations and gender, with details on each attendee's delegation.
United States and Canada · 1995–2007 · Event-level and grid-cell versions
Incident-level data on direct actions by the North American radical animal liberation movement, plus a version aggregated to 0.5° grid cells by year, with covariates for studying where and when these actions occur.
Almquist, Z.W., B.E. Bagozzi, and D. Blinova. 2025. "Historical Spatio-Temporal Data on North American Radical Environmental Direct-Action Events." Data in Brief 60: 111543.
Computational Methods
SADL (produced with the Social Analytics Data Lab)
Open · CC0
Linked Multi-Modal Data on Russian Domestic and Foreign Policy Speeches
Kremlin and Foreign Ministry speeches over several decades · Russian and English texts, images and captions · About 22 GB
Two linked corpora of official speeches by senior Kremlin and Ministry of Foreign Affairs officials. Each speech has aligned Russian and English texts, its images and captions where available, harmonized metadata (dates, speakers, locations, official tags), and validated topic annotations for both texts and images.
Blinova, D., G. Emuru, R. Emuru, K.S. Srivastava, M. Rulis, S. Chandrasekaran, and B.E. Bagozzi. 2026. "Linked Multi-Modal Data on Russian Domestic and Foreign Policy Speeches." Harvard Dataverse. https://doi.org/10.7910/DVN/SGI0VK
Misinformation & Media
Open IDs and labelsFull text restricted
U.S. 2024 Election X (Twitter) Harmful Content Dataset
97,696 posts · October 2024 to July 2025 · 5 harmful-content labels
Election-related posts labeled for speculation, sensationalism, conspiracy, hate speech and satire by an ensemble of six large language models, with a human-annotated subset for validation. The open release has post IDs, derived features and labels; the full text is available on application.
Wang, Q., P. Khatiwada, C. Coimbra Vieira, B.E. Bagozzi, K. Barner, and M.L. Mauriello. 2026. "A Multi-Label Dataset of LLM and Crowd-Annotated U.S. 2024 Election X (Twitter) Posts for Harmful Content Research." Zenodo. https://doi.org/10.5281/zenodo.22777827
30.2 million English posts, plus 4.4 million non-English · 2019 to January 2026 · 33 variables
Bluesky posts matching Trump-related search terms, with text, language, engagement, reply, quote, repost, hashtag and embed fields. Author identities are pseudonymized, so threads and per-author patterns can still be analyzed.
Wang, Q., A. Salloum, C. Coimbra Vieira, B.E. Bagozzi, M. Kivelä, K. Barner, and M.L. Mauriello. 2026. "Bluesky Data Trump Related Discourse 2019-2026." Zenodo. https://doi.org/10.5281/zenodo.22864223
Replication files
Replication files for 27 publications are linked from each paper on the Publications page.
Using the data
Most of these datasets are public-domain (CC0) releases on Harvard Dataverse; the Zenodo releases list their own terms. Please cite the data paper where there is one, or otherwise the dataset itself; each card’s Cite button shows the reference. For questions about access or collaboration, email bagozzib@udel.edu.
Software
R packages
Tools for detecting, modeling, and estimating count and ordered-outcome data structures that standard defaults overlook.
underdisp
v0.2.0
Diagnostics and models for underdispersed count data — conditional variance below the conditional mean, which Poisson and negative binomial defaults cannot represent. Includes a screening diagnostic, equidispersion tests and dispersion profiles; continuous parameter binomial and generalized event count regressions with zero-truncated, hurdle, zero-inflated and fixed-effects forms, alongside COM-Poisson, hyper-Poisson and other comparison families; and bootstrap and profile-likelihood inference, average marginal effects and prediction intervals. The likelihoods are implemented in C++.
Inflated ordered probit and logit models: the zero-inflated ordered probit of Harris and Zhao (2007) plus middle- and top-inflated extensions. Provides oprobit()/ologit() and iop()/iol(), panel versions with random intercepts and fixed effects, correlated-error variants (ZiOPC/MiOPC/TiOPC), and category-specific split equations.
772 downloads from CRAN
install.packages("iop")
Implements: Modeling Two Types of Peace: The Zero-Inflated Ordered Probit (ZiOP) Model in Conflict Research