HIV prevalence estimates and their use in regression models: cautionary evidence from Zimbabwe and studies of the relationship between armed conflict and HIV

Includes abstract.Includes bibliographical references.This dissertation makes two central arguments. The first is that regressions on country-level HIV prevalence are compromised by the fact that the HIV data used are estimates and not empirical data points. The HIV prevalence rates published by UNAIDS are estimates derived from epidemiological modelling (using EPP and Spectrum) in which data from antenatal clinics (sometimes supplemented by population survey data) are translated into adult HIV prevalence estimates