Evaluating HIV Estimates from National Population-Based Surveys for Bias due to Non-Response

Vinod Mishra, ORC Macro
Ties Boerma, World Health Organization (WHO)
Ann Way, ORC Macro
Bernard Barrere, ORC Macro
Fred Arnold, ORC Macro
Anne R. Cross, ORC Macro
Rathavuth Hong, ORC Macro
Kiersten Johnson, ORC Macro
Shane Khan, ORC Macro

Recently, HIV seroprevalence data have been collected in several national population-based surveys, such as Demographic and Health Surveys and AIDS Indicators Surveys. Such surveys enable direct estimation of population HIV prevalence. A major challenge for surveys is bias due non-response, both from refusal and absence. We evaluate survey estimates from eight sub-Saharan African countries for non-response bias. For each country, we predict HIV prevalence among non-responding adults based on multivariate statistical models of HIV for those who were interviewed and tested, using a common set of predictor variables. Predictions are made separately for adult males and females who were absent or refused testing. Results from Burkina Faso, Cameroon, Ghana, Kenya, and Tanzania (Lesotho, Malawi, and Uganda to be added) indicate that although non-tested males and females tend to have higher predicted HIV prevalence than those tested, overall effects of non-response on the observed national HIV prevalence estimates tend to be small.