Algorithmic Versus Human Taxonomies in Torture Research


How you define your research variables has a profound impact on research outcomes. This is particularly true with social, political-science and psychological research, where taxonomies are critical to the categorization and analysis of data: Psychological research depends on the DSM V definitions, for example, while political and social sciences research is frequently dependent on concepts/constructs such as ethnicity, nationality, religiosity and sexual orientation.

This project seeks to determine whether a common machine learning algorithm will identify cohorts similar to those used by PSOT researchers in their research projects, when presented with the data-sets used by PSOT researchers.

This project is a side investigation of a much bigger research project, which is an analysis of whether identity (religion, sexual orientation and/or ethnicity), contingency (wrong place / wrong time) or the political activity of victims of torture has any predictive power as to the intensity of survivor’s levels of psychological trauma (specifically PTSD and depression) and / or the arc of their recovery.

The machine learning project is a way of testing the assumption that the categories of identity, contingency and activity have any relevance at all to the clinical history of our clients.