Multi-Method Tools for Data Security in Political Science Research

Session Submission Type: Full Paper Panel

Session Description

Political science research in recent decades has generated important new knowledge by collecting and analyzing qualitative and quantitative data that contain sensitive and personally-identifiable information. Sensitive data of various forms is used across subfields---scholars across the discipline rely on data from surveys, sensitive interviews, and non-anonymized administrative data to do work in comparative, American and international politics. While certain principles for working with sensitive, personally identifiable data have been set forth in the past (most often after high-profile failures to protect respondent privacy or maintain research transparency) creating a set of "best practices" suited to variety of 21st Century data sources and security challenges is still a work in progress. The papers on this panel provide new frameworks for thinking about data security in qualitative and quantitative research, and introduce or evaluate new tools for protecting respondent privacy and securing data.

The paper by Mir makes a novel contribution by introducing a framework for understanding and mitigating the challenges of research on ongoing international security crises, which he calls NatSec fieldwork. The author identifies the data sources and collection methods particular to "NatSec" fieldwork and describes context-specific challenges that threaten the security and practicality of "NatSec fieldwork." Mir's paper compels us to think more carefully about the problems (and solutions) unique to this kind of work, which is common in comparative politics and international security.


The paper by Evans, King, Schwenzfeier, and Thakurta introduces and validates a differentially private algorithm that facilitates safe analysis of data that contains personal information, while furnishing principled estimates of statistical uncertainty and correcting for statistical biases caused by common differential privacy tools like data censoring and the addition of random error. The authors' new algorithm provides a pathway for increasing the amount of data available to social science researchers by reducing the discretion of individual researchers over important privacy issues, and providing mathematical guarantees that researcher conduct cannot threaten individual privacy.

Finally, Milliff's paper provides a new conceptual assessment of researchers' data security obligations, the contemporary threats to privacy and data security, and the challenges particular to practitioner-academic collaborative research. Milliff also describes and tests three user-friendly tools for protecting against three threats to data security: legal expropriation of sensitive data, de-anonymization of technically non-identifiable shared data, and accidental revelation of identifying information. The paper highlights challenges particular to an increasingly common research arrangement, and introduces new workflows built around off-the-shelf computational tools to enhance data security and privacy protection.

The three papers in this panel contribute new concepts and new practical tools for enhancing data security and protecting respondent privacy in research across the methodological spectrum, from in-depth personal interviews, to analysis of massive administrative data. As privacy and security issues grow more and more concerning to the people whose data social scientists analyze to learn about the world, the development and implementation of credible tools to protect privacy and confidentiality will become central to the work of many political scientists, whether they collect data through their own fieldwork and extensive interviews, through large surveys and behavioral experiments, or through partnerships and sharing agreements. The research in this panel represents the leading edge of this work across the methodological and substantive spectrum in political science.

Sub Unit

• Division 8: Political Methodology

Individual Presentations

• NatSec Fieldwork - Asfandyar Ali Mir, Stanford University

• Statistically Valid Inferences from Privacy Protected Data - Georgina Evans, Harvard University; Gary King, Harvard University; Margaret Schwenzfeier, Harvard University; Abhradeep Guha Guha Thakurta, University of California Santa Cruz

• Protecting and Presenting Sensitive Information: New Tools for Data Security - Aidan Milliff, Massachusetts Institute of Technology

• Internet Connectivity Statistics: Reproducible and Privacy-Preserving Estimates - Suso B Baleato, Harvard University

Chair

• Margaret E Roberts University of California, San Diego

Discussant

• Jesse Driscoll University of California, San Diego