HIV estimation using population-based surveys with non-response: A partial identification approach

• Main Authors: ADEGBOYE, Oyelola A., FUJII, Tomoki, LEUNG, Denis H. Y., LI, Siyu
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2024
• Subjects: demographic and health surveys; HIV; instrumental variable; non-response; partial identification; Econometrics; Health Economics
• Online Access: https://ink.library.smu.edu.sg/soe_research/2749; https://ink.library.smu.edu.sg/context/soe_research/article/3748/viewcontent/Statistics_in_Medicine___2024___Adegboye___HIV_estimation_using_population_based_surveys_with_non_response__A_partial.pdf
• Institution: Singapore Management University

HIV estimation using data from the demographic and health surveys (DHS) islimited by the presence of non-response and test refusals. Conventional adjust-ments such as imputation require the data to be missing at random. Methodsthat use instrumental variables allow the possibility that prevalence is differentbetween the respondents and non-respondents, but their performance dependscritically on the validity of the instrument. Using Manski’s partial identifica-tion approach, we form instrumental variable bounds for HIV prevalence from apool of candidate instruments. Our method does not require all candidate instruments to be valid. We use a simulation study to evaluate and compare ourmethod against its competitors. We illustrate the proposed method using DHSdata from Zambia, Malawi and Kenya. Our simulations show that imputationleads to seriously biased results even under mild violations of non-random miss-ingness. Using worst case identification bounds that do not make assumptionsabout the non-response mechanism is robust but not informative. By takingthe union of instrumental variable bounds balances informativeness of thebounds and robustness to inclusion of some invalid instruments. Non-responseand refusals are ubiquitous in population based HIV data such as those col-lected under the DHS. Partial identification bounds provide a robust solutionto HIV prevalence estimation without strong assumptions. Union bounds aresignificantly more informative than the worst case bounds without sacrificingcredibility.