A modified Monte Carlo EM algorithm for a three-parameter distribution

Statistical inference of a three-parameter distribution with closed-form mean residual life function is studied for complete and Type-II censored data. Because the log-likelihood function for this new distribution is not convex, the sequential quadratic programming method used for the maximum likelihood estimation easily diverges. Therefore, a modified Monte Carlo EM algorithm is proposed for parameter estimation. This modified algorithm is used to maintain the feasibility of the parameter vector during the evolutionary process, and is found to be robust in the sense that it is insensitive to the starting points.