Levels of state repression and the frequency, severity, and targets of human rights abuses vary spatially within states. However, most previous studies on these topics have only considered repression typically aggregated into data at the country-year level. This is problematic because the high degree of spatial and temporal aggregation misses important subnational variation of violations, and implicitly assumes that states employ similar levels of abuse throughout their territory. Yet, even a cursory examination of the patterns of repressive events and behaviors suggests that repression is employed unevenly within the state. In order to capture subnational variation in repression --- and to ultimately aid in explaining this variation --- several coauthors and I have constructed a new dataset that provides new latent variable estimates of repression at the level of the sub-national unit for a global sample of states. In shot, scholars of human rights and repression have much to learn by systematically disaggregating and studying human rights content from annual human reports and other sources. 

 This research project is part of a collaborative National Science Foundation grant titled SNARP or The Sub-National Analysis of Repression Project. The SNARP project is a theoretically motivated data collection, categorization, and measurement project, that represents an ongoing collaboration with Rebecca Cordell (University of Texas at Dallas), Chad Clay (University of Georgia), Reed Wood (University of Essex), Thorin Wright (Arizona State University) and Christopher J. Fariss (University of Michigan). The project first identified each individual allegation of human rights abuse contained within a set of human rights documents. For each allegation, of which there are hundreds of thousands, the project team used hand coding, dictionary-based approaches, and supervised machine learning methods to categorize information about the perpetrator of the human rights abuse, the victim of the abuse, and information about the spatial and temporal context of the abuse. In some cases, these allegations are quite specific. In other cases, these allegations are rather general. To combine this information into comparable subnational and country- year estimates, we expand on several measurement models for repression using the allegation data classified in this project. To date, the SNARP collaborative research project has generated two published articles in Journal of Human Rights and International Studies Quarterly and several new working papers, and large-scale, publicly available dataset including 163,512 unique human rights abuse allegations in 196 countries between 1999 and 2016 (https://dataverse.harvard.edu/dataverse/SNARP). 

 Overall, the SNARP project sets a new standard for transparency and accessibility of human rights reporting content. It also will serve as the foundation for any new human rights coding projects that use the country year human rights reports as part of the document corpus because it is now the most easily accessible and searchable database for content from three annual human reports: the  State Department of the United States, Amnesty International, and Human Rights Watch. The SNARP project continues in several directions (1) new dictionary development, (2) updated new sets of sentences and allegations (pre-1998 and 2017-2020), (3) new machine learning classification modeling and validation of all categorized allegation features (e.g., scope, intensity, actor, location), (4) validation of subnational coding (capital city, region, geo-coding, location dictionaries), (5) validation of actor coding (UCDP, GED actor dictionaries), (6) SNARP web app and near-real time human coding, (7) allegation and sentence forecasting from new online information sources (e.g, social media data). All of this material is being added to the publicly available dataverse repository for the benefit of the human rights research community

Last Modified: 04/08/2022
Modified by: Christopher J Fariss