A Study of the Effects of Correlation and Prevalent Rate Distribution in the Item List of Non-Random Response Models: ICT, DICT, SSC

• Activity Number: 553
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Jay Schaffer* and Caroline Emsermann
• Companies: University of Northern Colorado and University of Northern Colorado
• Keywords: item count technique ; double item count technique ; single sample count technique

Abstract

Non-random response (NRR) techniques are meant to reduce bias in estimating sensitive behavior rates by embedding sensitive questions in an item list of innocuous questions. This introduces variation and depending on the prevalent rate distribution of the innocuous questions "ceiling" or "floor" effects occur resulting in bias. Manipulation of the prevalent rates and correlation of the item list is thought to reduce variability and the ceiling and floor effects. A large simulation study including a wide range of sensitive behavior rates and sample sizes was performed to determine optimal model parameters of the item count technique (ICT), double item count technique (DICT) and the single sample count technique (SSC). Simulation parameters included item list size, correlations within and between lists, and distribution of prevalent rates of the item list. Truthful reporting was assumed and variances of each simulated technique were compared to the variance of the direct questioning technique. Results indicated correlating between sensitive questions within a list or between lists reduce variation. Distribution of prevalent rates was more effective in reducing variation for the ICT.