A Unified Measure of Model Efficiency and Respondent privacy in Quantitative Randomized Response ModelsUntitled

Speaker: Sat Gupta, University of North Carolina

Abstract

Randomized response techniques (RRT) are important survey methods when dealing with sensitive topics where there is a significant risk of social desirability response bias unless respondent privacy is ensured. Two important measures to consider, when assessing the quality of a RRT model, are model efficiency and respondent privacy. Unfortunately, they move in opposite directions, so you can’t possibly control both. Statisticians deal with similar problem in reference to confidence level and precision associated with confidence intervals.

Focus of this talk will be on:

1.  Showing how seemingly very efficient looking RRT models might actually be very bad, and
2.   Presenting a unified measure of quality of quantitative RRT models which accounts for both model efficiency and respondent privacy.

We will also discuss why there is no extra loss of privacy with optional RRT models, contrary to such belief by some researchers.

Most of the contents of this talk are from a forthcoming paper Gupta, Mehta and Shabbir (2018, JSTP)

Light lunch in the same room at 12:30.

About the speaker:

Sat Gupta, PhD
Fellow of the American Statistical Association &
Professor of Statistics
Department of Mathematics and Statistics
The University of North Carolina at Greensboro
126 Petty Building, 317 College Ave,
Greensboro, NC 27412
https://sites.google.com/uncg.edu/sat-gupta/

Statistics REU at UNC Greensboro
https://www.uncg.edu/mat/undergraduate/2018asareu/

Chair: AISC 2018 Conference (October 6-8, 2018)
https://www.uncg.edu/mat/aisc/2018/

Editor-in-Chief: JSTP: Journal of Statistical Theory and Practice
http://www.tandfonline.com/loi/UJSP20

For submissions to JSTP, please use the Scholar One website
http://mc.manuscriptcentral.com/jrstp