Comparative Analysis of Stratified Randomized Response Models for HIV Seroprevalence Surveys

Abstract

Seroprevalence surveys of HIV pandemic are highly sensitive especially in Africa. The objective of this study is to reach research frontier to devise a two-way randomized response model (RRM) in stratification and use same to estimate HIV seroprevalence rates in a given population and compare results with the existing seroprevalence rates. The randomized response techniques (RRT) guarantees the anonymity of respondents in surveys aimed at determining the frequency of stigmatic, embarrassing or criminal behaviour where direct techniques for data collection may induce respondents to refuse to answer or give false responses. The motivation was to improve upon the existing RRMs as well as to apply them to estimate HIV seroprevalence rates. Warner proposed the pioneering RRM for estimating the proportion of persons bearing a socially disapproved character. Quatember produced unified criteria for all RRTs, Kim and Warde proposed a stratified RRM and so many others. The proposed two-way RRM in stratification for HIV seroprevalence surveys was relatively more efficient than the Kim and Warde stratified estimator for a fixed sample size. The chosen design parameter was 0.7, using the criteria of Quatember who derived the statistical properties of the standardized estimator for general probability sampling and privacy protection. Furthermore, the model was used to estimate the HIV seroprevalence rate in a sampled population of adults 3,740 people aged 18 years and above attending a clinic in Kaduna, Nigeria using a sample size of 550. The findings revealed that HIV seroprevalence rate, as estimated by Model I, stood at 6.1% with a standard error of 0.0082 and a 95% confidence interval of [4.5%, 7.7%]. These results are consistent with that of Nigerian sentinel survey (2003) conducted by NACA, USAID and CDC which estimated the HIV seroprevalence in Kaduna State as 6.0%. Hence, the RRTs herein can serve as new viable methods for HIV seroprevalence surveys. Similarly, the result for model II show that, using the survey data, the model estimated the HIV seroprevalence rate as 8.74% with a standard error of 0.0134 and 95% confidence bands of [6.1%, 11.4%]. Accordingly, the sentinel projected seroprevalence rate, using the EPP Package, for the next ten years (2013) was 9.7%; very consistent with the 95% confidence interval. Hence, the RRTs herein can also serve as new viable methods for HIV seroprevalence surveys. Model II has a better chance of estimating HIV seroprevalence because it has higher privacy preservation.