Protecting and Presenting Sensitive Information: New Tools for Data Security

• In Event: Multi-Method Tools for Data Security in Political Science Research

Abstract

An large and growing segment of contemporary political science research involves collecting, storing, and analyzing personal information gathered from research subjects or from administrative data, often with assistance or facilitation from practitioner partners. As research using personally identifiable information and research involving practitioner partners becomes more common, however, political scientists still lack a systematic framework for a) assessing data security obligations of researchers and b) implementing best practices to meet those obligations in light of relevant threats. This paper introduces a new framework of obligations/imperatives and threats to data security (both general to political science research with personally identifiable information, and particular to collaborative efforts with practitioners), and outlines three easy-to-implement practices for meeting obligations to privacy and transparency in the collection, storage, and presentation of work based on personally identifiable information. The framework focuses on researchers' obligations to protect respondent privacy while producing transparent and reproducible researh findings. Threats to these obligations include legal expropriation of sensitive information, theft of information, and de-anonymization or re-identification of shared or published data. Work with practitioner partners is noted as particularly exposed to this final threat because practitioner partners have excellent access to contextual information to re-identify respondents. The paper then introduces three practices that contribute to mitigating these threats, including a PGP-based workflow for protecting data while in the "field," a data segregation strategy to hold-out personally identifiable information while analyzing data, and a new implementation of Topic Modeling for presenting trends in sensitive qualitative data.

Author

• Aidan Milliff, Massachusetts Institute of Technology