TY - JOUR
T1 - A Bayesian approach for parameter estimation in multi-stage models
AU - Pham, Hoa
AU - Nur, Darfiana
AU - Pham, Huong T.T.
AU - Branford, Alan
PY - 2019/5/19
Y1 - 2019/5/19
N2 - Multi-stage time evolving models are common statistical models for biological systems, especially insect populations. In stage-duration distribution models, parameter estimation for the models use the Laplace transform method. This method involves assumptions such as known constant shapes, known constant rates or the same overall hazard rate for all stages. These assumptions are strong and restrictive. The main aim of this paper is to weaken these assumptions by using a Bayesian approach. In particular, a Metropolis-Hastings algorithm based on deterministic transformations is used to estimate parameters. We will use two models, one which has no hazard rates, and the other has stage-wise constant hazard rates. These methods are validated in simulation studies followed by a case study of cattle parasites. The results show that the proposed methods are able to estimate the parameters comparably well, as opposed to using the Laplace transform methods.
AB - Multi-stage time evolving models are common statistical models for biological systems, especially insect populations. In stage-duration distribution models, parameter estimation for the models use the Laplace transform method. This method involves assumptions such as known constant shapes, known constant rates or the same overall hazard rate for all stages. These assumptions are strong and restrictive. The main aim of this paper is to weaken these assumptions by using a Bayesian approach. In particular, a Metropolis-Hastings algorithm based on deterministic transformations is used to estimate parameters. We will use two models, one which has no hazard rates, and the other has stage-wise constant hazard rates. These methods are validated in simulation studies followed by a case study of cattle parasites. The results show that the proposed methods are able to estimate the parameters comparably well, as opposed to using the Laplace transform methods.
KW - Bayesian analysis
KW - destructive samples
KW - multi-stage models
KW - stage duration
KW - stage frequency data
UR - http://www.scopus.com/inward/record.url?scp=85057301684&partnerID=8YFLogxK
U2 - 10.1080/03610926.2018.1465090
DO - 10.1080/03610926.2018.1465090
M3 - Article
AN - SCOPUS:85057301684
VL - 48
SP - 2459
EP - 2482
JO - COMMUNICATIONS IN STATISTICS-THEORY AND METHODS
JF - COMMUNICATIONS IN STATISTICS-THEORY AND METHODS
SN - 0361-0926
IS - 10
ER -