Advances and Applications in Statistics
Volume 34, Issue 1, Pages 65 - 84
(May 2013)
|
|
A LATENT CLASS APPROACH WITH COVARIATES AND LOCAL DEPENDENCE IN CAPTURE-RECAPTURE MODELS
Joanne Thandrayen and Yan Wang
|
Abstract: Traditional capture-recapture methods assume that lists operate independently (local independence) and that capture probabilities are homogeneous. In studies involving human populations, these assumptions are often violated. This paper presents an approach where dependence between the lists and the effects due to the observable covariates are modelled directly in the capture probability. For this purpose, we employ a multinomial latent class model. Estimation of the model parameters is based on the maximum likelihood method via the EM algorithm. An approximation for the variance of the unknown population size is also formulated. |
Keywords and phrases: conditional likelihood, covariate, EM algorithm, heterogeneity, latent class model, multinomial logit. |
|
Number of Downloads: 375 | Number of Views: 1111 |
|