Structurally missing capture–recapture data

Registrations in epidemiological studies suffer from incompleteness, thus a general consensus is to use capture–recapture models. Log–linear models are typically used when the registrations measure the same population and the covariates are measured by all registrations. This article shows how data can be analysed if some covariates are unobserved in some registrations and the registrations do not all measure the whole population.