Measuring Human Rights Abuse from Access to Information Requests

Existing measures of human rights abuses are often only available at the country-year level. Several more fine-grained exhibit spatio-temporal inaccuracies or reporting biases due to primary sources upon which they rely. To address these challenges, and increase diversity generally, this study provides first quantitative effort measure from textual records citizen-government interactions. Using a dataset encompassing over 1.5 million access-to-information (ATI) requests made Mexican federal government June 2003 onward, supervised classification is used identify subset that pertain various types. The results machine learning exercise validated against (i) gold standard ATI pertaining past in Mexico (ii) several accepted external sub-national sub-annual abuses. In doing so, we demonstrate measurement citizen-submitted request texts can provide abuse both high validity notable specificity, relative existent datasets variables.