Using Multiple Imputation Technique to Correct for Measurement Error and Statistical Disclosure Control in Sensitive Count Data in a National Survey

Measurement error in sensitive question is pervasive, therefore, biasing the estimation of most statistical models. The objective of this paper is to correct for measurement error in the number of life-time sexual partners by treating it as a missing data problem and using multiple imputation technique to synthesize this underlying “true” attribute. Bayesian Poisson model with diffuse Gaussian priors was fitted to the 1996 General Social Survey combining knowledge of data quality from the mode experiment conducted by Tourangeau and Smith (1996). Ignored in existing literature, the threat of augmented disclosure harm from releasing both imputed and original data to the public was recognized and tackled by statistical perturbation. Bias reduction and statistical integrity were evaluated. Markov Chain Monte Carlo algorithm was programmed using WinBUGS.