Data and Replication Files
Publications
|
Disaggregating Repression: Identifying Physical Integrity Rights Allegations in Human Rights Reports
Rebecca Cordell, K Chad Clay, Christopher J Fariss, Reed M Wood, Thorin M Wright [VIEW ARTICLE] [Replication Files] Most cross-national human rights datasets rely on human coding to produce yearly, country-level indicators of state human rights practices. Hand-coding the documents that contain the information on which these scores are based is tedious and time-consuming, but has been viewed as necessary given the complexity and detail of the information contained in the text. However, advances in automated text analysis have the potential to streamline this process without sacrificing accuracy. In this research note, we take the first step in creating this streamlined process by employing a supervised machine learning automated coding method that extracts specific allegations of physical integrity rights violations from the original text of country reports on human rights. This method produces a dataset including 163,512 unique abuse allegations in 196 countries between 1999 and 2016. This dataset and method will assist researchers of physical integrity rights abuse because it will allow them to produce allegation-level human rights measures that have previously not existed and provide a jumping-off point for future projects aimed at using supervised machine learning to create global human rights metrics. |
Cordell, R., Clay, K. C., Fariss, C. J., Wood, R. M., & Wright, T. M. (2022). Disaggregating repression: Identifying physical integrity rights allegations in human rights reports. International Studies Quarterly, 66(2), 1-11. doi.org/10.1093/isq/sqac016.
|
|
Changing Standards or Political Whim? Evaluating Changes in the Content of US State Department Human Rights Reports Following Presidential Transitions
Rebecca Cordell, K. Chad Clay, Christopher J. Fariss, Reed M. Wood, Thorin M. Wright [VIEW ARTICLE] [Replication Files] The annual US State Department Country Reports on Human Rights Practices represents one of the principal data sources used to generate multiple commonly used human rights measures. Despite the frequency with which these indicators are used in quantitative studies of human rights, scholars have rarely considered how the qualitative information in the source has varied over time. We contribute to this area of research by investigating the general changes in the amount of information included in the reports as well as the administration-specific changes in this information. Using automated text analysis techniques, we find that the amount of information in the reports generally increases over time. However, our analysis also reveals that the rate (and direction) of change varies across different human rights topics and across presidential administrations. Consequently, we find evidence to support a changing standard of accountability as well as evidence that political considerations shape human rights reporting. |
Cordell, R., Clay, K. C., Fariss, C. J., Wood, R. M., & Wright, T. M. (2020). Changing standards or political whim? Evaluating changes in the content of US State Department human rights reports following presidential transitions. Journal of Human Rights, 19(1), 3–18. doi.org/10.1080/14754835.2019.1671175.
|
Works in Progress
|
Repressive Actions Dataset (RAD--SNARP): Classifying the Content of Physical Integrity Rights Abuse Allegations using Machine Learning Methods
Rebecca Cordell, K. Chad Clay, Christopher J. Fariss, Reed M. Wood, Thorin M. Wright Theories of state repression now routinely recognize that agents select from a menu of physically coercive actions to accomplish their objectives, such as quelling dissent or (re)asserting control over populations or territory. These agents likewise exercise discretion over the intensity with which repression is deployed and the specific targets and locations of these actions. Despite the rapid expansion of quantitative scholarship in these areas in recent decades, the high costs of collecting and coding sufficiently granular data on state repression impedes scholars’ ability to test hypotheses on these topics. In an effort to reduce these costs, in this research note we introduce and evaluate an automated coding process that produces categorical information about physical integrity rights violations contained in the annual human rights reports. Allegation sentences present information about specific types of physical integrity violation (disappearances, torture, killing and political imprisonment) and whether the state perpetrated. Empirical assessments showcase the relative validity improvements of the new allegations data. |
Cordell, R., Clay, K. C., Fariss, C. J., Wood, R. M., & Wright, T. M. Repressive Actions Dataset (RAD--SNARP): Classifying the Content of Physical Integrity Rights Abuse Allegations using Machine Learning Methods. Working Paper.
|
|
Measuring Sub-national Intensity Levels of Repression Using A Large Language Model and Latent Variable Model
Rebecca Cordell, K. Chad Clay, Christopher J. Fariss, Reed M. Wood, Thorin M. Wright We present a large language measurement model (LLM) that aggregates human rights information from country-year allegation data drawn directly from annual human rights reports published by Amnesty International, Human Rights Watch, and the US State Department. We begin by reviewing categorized information about four specific types of physical integrity violations (disappearances, torture, killing and political imprisonment) and their intensity levels. In addition, each allegation is coded for content that includes actor type and an aggregate physical integrity indicator (any of the four violations). This dataset, generated through a LLM coding process, contains 3.6 million sentences covering all countries from 1980–2023. By disaggregating the reports at the sentence level, our measurement models can combine any single piece of information—or any combination of information—into estimates for higher-level units: report, allegation type, country, year, and subnational units. We present preliminary model performance results for each of our new LLM-produced variables. To illustrate the value of our approach, we visualize patterns and trends in the intensity levels of torture allegations perpetrated by state actors at the country-year level for reports published in 2020. We then use the LLM-coded sentence-level data to develop latent variable models that aggregate this information to the first-administrative-level-year and country-year. |
Cordell, R., Clay, K. C., Fariss, C. J., Wood, R. M., & Wright, T. M. Measuring Sub-national Intensity Levels of Repression Using A Large Language Model and Latent Variable Model. Working Paper.
|
All Repression Is Local: The Sub-national Determinants of Physical Integrity Rights Abuse
Rebecca Cordell, K. Chad Clay, Christopher J. Fariss, Reed M. Wood, Thorin M. Wright
Why do we observe different levels of respect for human rights in different regions of the same country? Why are citizens’ human rights generally uniformly protected (or abused) within the borders of some countries while within other countries these rights are upheld in some locations and severely restricted in others? Prior research investigating patterns of human rights protection and violation has typically treated states as centralized decision-makers and examined state respect for human rights as a single, countrywide phenomenon. This approach masks important variations in the actors perpetrating abuses, motives for the abuse, targets of the abuse, and severity of abuse. We propose that cross-national human rights researchers must break their focus on the country as the unit of analysis and look at the sub-national characteristics of repressive behaviors. We argue that antigovernment dissent encourages government agents to respond with repression. However, this response is particularly likely when government power is highly decentralized, when the dissent takes place far from the national capital, and when the local government is largely incapable of controlling its repressive agents. We test this theory using an original dataset that represents the first attempt to document the level of repression at the subnational level for a global sample of countries over space and time.
Rebecca Cordell, K. Chad Clay, Christopher J. Fariss, Reed M. Wood, Thorin M. Wright
Why do we observe different levels of respect for human rights in different regions of the same country? Why are citizens’ human rights generally uniformly protected (or abused) within the borders of some countries while within other countries these rights are upheld in some locations and severely restricted in others? Prior research investigating patterns of human rights protection and violation has typically treated states as centralized decision-makers and examined state respect for human rights as a single, countrywide phenomenon. This approach masks important variations in the actors perpetrating abuses, motives for the abuse, targets of the abuse, and severity of abuse. We propose that cross-national human rights researchers must break their focus on the country as the unit of analysis and look at the sub-national characteristics of repressive behaviors. We argue that antigovernment dissent encourages government agents to respond with repression. However, this response is particularly likely when government power is highly decentralized, when the dissent takes place far from the national capital, and when the local government is largely incapable of controlling its repressive agents. We test this theory using an original dataset that represents the first attempt to document the level of repression at the subnational level for a global sample of countries over space and time.