An Analysis on Generalized Models For Capture–Recapture Studies With Individual Covariates |

ABSTRACT:

Capture-recapture methods are used to estimate theincidence of a disease, using a multiple-source registry. Usually, log-linearmethods are used to estimate population size, assuming that not all sources ofnotification are dependent. Where there are categorical covariates, astratified analysis can be performed. The multinomial logit model hasoccasionally been used. In this paper, the authors compare log-linear and logitmodels with and without covariates, and use simulated data to compare estimatesfrom different models. The crude estimate of population size is biased when thesources are not independent. Analyses adjusting for covariates produce lessbiased estimates. In the absence of covariates, or where all covariates arecategorical, the log-linear model and the logit model are equivalent. A 4exible method for modelling capture–recapture datawith continuous covariates that describe heterogeneous catch ability isdeveloped. The well-established generalized additive modelling framework isused. An estimator of population size is developed using this method. Theperformance of the method is demonstrated using neural tube defectcapture–recapture data from the Netherlands,with the birth weight of a child as a covariate. The parametric bootstrap isused for variance estimation. Registrations in epidemiological studies suffer fromincompleteness, thus a general consensus is to use capture-recapture models.Inclusion of covariates which relate to the capture probabilities has beenshown to improve the estimate of population size. The covariates used have tobe measured by all the registrations. In this article, we show how multipleimputation can be used in the capture-recapture problem when some lists do notmeasure some of the covariates or alternatively if some covariates areunobserved for some individuals.