A parametric mixture model of three different distributions: An approach to analyse heterogeneous survival data
Yusuf Abbakar Mohammed, Bidin Yatim, Suzilah bt Ismail · AIP conference proceedings · 2014
A parametric mixture model of three different distributions is proposed to analyse heterogeneous survival data. The maximum likelihood estimators of the postulated parametric mixture model are estimated by applying an Expectation Maximization Algorithm (EM) scheme. The simulations are performed by generating data, sampled from a population of three component parametric mixture of three different distributions. The parameters estimated by the proposed EM Algorithm scheme are close to the parameters of the postulated model. To investigate the consistency and stability of the EM scheme, the simulations are repeated several times. The repetitions of the simulation gave parameters closer to the values of postulated models, with relatively small standard errors. Log likelihood, AIC and BIC are computed to compare the proposed mixture model with parametric mixture models of one distribution. The calculated values of Log likelihood, AIC and BIC are all in favour of the proposed parametric mixture model of different distributions.