In this plaper,a random effect Poisson regression model is considered for prediction of the failiure rate which would follow a lognormal distribution.A two stage procedure is used to obtain the regression estimator of the failiure rate as well as the shrinkage estimator.These estimators are compared to both the raw estimator which entirely depends on the historical failiure records and a shrinkage estimator in whilch a gamma distribution is used mistakenly in place of the lognormal plrior distribution.Results of Monte-Carlo simulation indicate the following in terms of the MSE:(1)overall,the shrinkage estimator based on the lognormal prior distribution performs best;(2)with the failure rates (0-2.5),the performance of the shrinkage estimator based on the gamma distribution is not significantly different from that of the shrinkage estimator based on the lognormal distribution;(3)when there exists considerable variability in the failure rates(0-10),the raw estimator appears to replace shrinkage estimations.In terms of the Bias,the raw estimator performs better than the others.
展开▼