Randomized Response Techniques and Stratified Sampling Designs: Methods for Dealing with Social Desirability and Gender Response Bias
by
Sat Gupta
University of Southern Maine
Coauthors: Bill Thornton, Javid Shabbir, Sunil Singhal (University of Southern Maine)

Randomized response techniques have been presented as a means to overcome social desirability response bias on sensitive questions in personal surveys or interviews. In the present research, the relative effectiveness of a partial and a full randomized response technique were compared using a survey of health-related attitudes and behaviors (e.g., alcohol/drug use, sexual practices, diet, exercise). Compared to benchmark results from an anonymous/confidential survey, and results using a "bogus pipeline technique" in which respondents were made to believe that deceptive responses on their part could be detected on the basis of their physiological responses, the partial (vs. full) randomization technique proved more accurate in estimating the "true" responses on personally sensitive items (e.g., driving while intoxicated); no difference was apparent on nonsensitive items (e.g., eating breakfast). In addition to social desirability response bias, a potential gender response bias was apparent on certain items (e.g., number of sexual partners). The present research also was able to demonstrate how the use of a stratified sampling design may be used effectively in overcoming this gender bias.

Date received: September 7, 2003