orange publication summary banner small

Document Type
Working Papers
Publication Topic(s)
HIV Prevalence
Mapeta et al and ICF Macro, Calverton, Maryland, USA
Publication Date
February 2010
Publication ID

Order a Hard Copy: Please use electronic copies of DHS publications whenever possible. Hardcopies of publications are intended primarily for those in developing countries where internet connections are limited or unavailable.


Objectives: This study evaluates the extent of potential bias due to non-response in the 2005-06 Zimbabwe Demographic and Health Survey (ZDHS) and assesses the impact of such bias on the survey estimates of HIV prevalence. The study also examines the extent to which exclusion of non-household population groups from the ZDHS may have biased the HIV prevalence estimates. Methods: Analysis in this study is based on 8,342 eligible men and 9,870 eligible women age 15-49. For the analysis of non-response bias, we divided the eligible survey population into three categories—interviewed and tested, interviewed but not tested, and neither interviewed nor tested. HIV prevalence for respondents in the two non-tested groups was predicted based on statistical models for those who were interviewed and tested, using a common set of predictor variables. For the analysis of non-household bias, we used the 2002 Population Census estimate of the size of the non-household population and simulated total HIV prevalence among all adults (household and non-household) under scenarios of whether HIV prevalence among the non-household population was double, triple, or quadruple that of the household population. Results: Of eligible adults age 15-49 in the 2005-06 ZDHS, 36% of men and 24% of women did not have a valid HIV test result. Refusal was a more important reason for non-response than absence from the household. Non-tested men (but not women) had a slightly higher predicted prevalence of HIV than those tested. The overall effects of non-response bias on observed HIV prevalence estimates were small and not statistically significant, among both men and women. The estimated effects of exclusion of non-household population from the ZDHS were also negligible. Even under the unlikely scenario that 72.4% of adults in the non-household population were HIV-positive, the survey-based estimate of HIV prevalence of 18.1% would only increase to 18.8%. Conclusions: The findings that potential biases in the HIV prevalence estimates, whether due to non-response in the ZDHS or to exclusion of the non-household population, are negligible suggest that Zimbabwe can confidently continue using household-based national surveys to obtain reliable estimates of HIV prevalence. The anonymously linked socioeconomic and behavioural information with the HIV serostatus of individuals in such surveys is useful for identifying higher-risk and vulnerable populations and for informing prevention, care, and treatment programs. Nonetheless, it is recommended that similar evaluations of potential bias in survey-based estimates of HIV prevalence be conducted after each survey to ensure that the observed estimates of HIV prevalence from household surveys are reliable. Key words: Non-response; bias; HIV estimate; HIV prevalence; survey; Zimbabwe