Using panel data for partial identification of human immunodeficiency virus prevalence when infection status is missing not at random

Population‐based surveys are often considered the ‘gold standard’ to estimate the prevalence of human immunodeficiency virus (HIV) but typically suffer from serious missing data problems. This causes considerable uncertainty about HIV prevalence. Following the partial identification approach, we produce worst‐case bounds for HIV prevalence. We then exploit the availability of panel data and the absorbing nature of HIV infection to narrow the width of these bounds. Applied to panel data from rural Malawi, our approach considerably reduces the width of the worst‐case bounds. It also allows us to check the credibility of the additional assumptions that are imposed by methods that point‐identify HIV prevalence.