Adult Mortality Determinants Controlling for Migration Biases – A Two-Stage Competing Risks Model applied to Nairobi HDSS Data

Event history analyses make the explicit assumption of independence between censoring and event. Under this hypothesis right censoring due to survey time does not create a selection bias. However, when censoring is not independent from the event of interest (e.g. migration in relation to death) then results suffer from potential bias. This paper presents a model to deal with non-independent right- as well as left-censoring, when the same determinants may cause in-migration, out-migration and mortality. The model follows the rationale of two-stage regression models controlling for selection biases to control for both observed and unobserved heterogeneity in migration. The method is applied on longitudinal adult mortality data collected by the APHRC Nairobi Urban Health and Demographic Surveillance System (NUHDSS), situated in two Nairobi slums – Korogocho and Viwandani – where circular migration is high. Results confirm selection for both out- and in-migration. The method produces higher adult mortality rates than raw estimates. After controlling for migration effect, median age at death from age 15 would be 46 year old (against gross estimate of 70) for females, and 52 (against 70) for males. Migration is most likely an important strategy adopted by slum-dwellers to avoid health risks attached to slum environment.