This validation study examined the joint effects of question threat and method of administration on response distortion using four interviewing techniques. The level of threat was varied by asking questions about library card ownership, voting, bankruptcy involvement, and having been charged with drunken driving. The results indicated that response distortion increased sharply as threat increased. None of the data methods was clearly superior to all other methods for all types of threatening questions. Randomized response gave the lowest distortion on questions about socially undesirable acts, but even with this procedure there was still a 35 percent understatement of drunken driving.

An Investigation of Interview Method, Threat and Response Distortion

Sensitive topics or highly personal questions are often being asked in medical, psychological and sociological surveys. This paper proposes two new models (namely, the triangular and crosswise models) for survey sampling with the sensitive characteristics. We derive the maximum likelihood estimates (MLEs) and large-sample confidence intervals for the proportion of persons with sensitive characteristic. The modified MLEs and their asymptotic properties are developed. Under certain optimality criteria, the designs for the cooperative parameter are provided and the sample size formulas are given. We compare the efficiency of the two models based on the variance criterion. The proposed models have four advantages: neither model requires randomizing device, the models are easy to be implemented for both interviewer and interviewee, the interviewee does not face any sensitive questions, and both models can be applied to both face-to-face personal interviews and mail questionnaires.

Two new models for survey sampling with sensitive characteristic: design and analysis

Three randomized response models for quantitative data are considered, the extension of the unrelated question model first considered by Greenberg, and two models which involve the addition or multiplication of a random number from a known distribution to the sensitive value. Estimation is considered in all three models and efficiencies compared. The addition and multiplication models appear to have potential but a practical comparison of the three models is needed.

A Comparison of Three Randomized Response Models for Quantitative Data

Randomized response methods for quantitative sensitive data are treated in an unified approach which includes the use of auxiliary information at the estimation stage. A class of estimators for the mean of a sensitive variable is proposed under a generic randomization model and the optimum estimator is obtained. Some special models are discussed in detail. To evaluate the degree of respondents’ confidentiality in models using auxiliary variables, a new measure of privacy protection is introduced. Different models are then compared both from the perspective of efficiency and privacy protection.

A class of estimators for quantitative sensitive data

Summary The primary purpose of this paper is to present a comprehensive review of a data-gathering device which was first introduced exactly ten years ago by Stanley L. Warner. This technique, known as randomized response, provides the researcher with an opportunity to reduce, if not eliminate, the response biases which usually occur when respondents in human population surveys are questioned about highly personal matters. Cooperation and honest answers to sensitive questions are expected with the randomized response technique since it does, in fact, protect the privacy of the respondent. In this paper the range and scope of the technique will be illustrated by discussing the highlights of various measurement designs that have been proposed. Finally, some applications will be presented.

Randomized Response: a Data-Gathering Device for Sensitive Questions

In sample surveys it is often difficult to get reliable information from the respondents. This is very often the case when the investigation deals with phenomena that are illegal or looked upon as morally condemnable by society. As, for example, illegal abortion and withholding incomes at self-assessment. Refusal to answer and consciously false information commonly causes large systematic errors when estimating parameters of interest. In order to diminish such errors when estimating the proportion of persons having a certain characteristic A (belong to group A) Warner [3] suggested an interview technique which provides true information with known probability but which does not reveal to which group the respondent belongs. Different techniques to get the information have been suggested by earlier authors. The randomizing instrument to be used here is a deck of cards. On each card there is one of the following two statements, which occur with relative frequency P and 1-P respectively: 1. I belong to group A 2. I belong to group A.

Sven A. Eriksson

Papers

A New Model for Randomized Response