Refusal bias in HIV prevalence estimates from nationally representative seroprevalence surveys
Jeff Eaton, Imperial College
Nationally-representative seroprevalence surveys are increasingly used for HIV prevalence estimates. We explore non-response bias in these estimates because of refusals for testing. The few studies on this topic have failed to identify any substantial bias, but they typically ignore that refusals may be informed by respondents’ prior knowledge of their HIV status. Using longitudinal data for Malawi in a sample where respondents know their HIV status, we find that HIV positives are 4.62 (95%-CI: 2.60-8.21) times more likely to refuse a repeat test than HIV negatives. We use this parameter to develop a heuristic model of refusal bias and apply it to empirical data from six Demographic and Health Surveys. Our model suggests that a downward bias of 13.3% (95%-CI:7.2%-19.6%) in national HIV prevalence estimates is possible. In some urban populations, bias exceeds 20%. Because refusal rates are higher in men, seroprevalence surveys tend to overestimate the female to male ratio of infections as well.