![]() | Only 14 pages are availabe for public view |
Abstract In this thesis, three resampling techniques are considered, namely, bootstrap, jackknife and jackknife after bootstrap. The main objective is to study the performance of the resampling techniques in the maximum likelihood estimation of the parameters using EM, ECM, ECME algorithms for Grubbs model, when the latent response follows normal distribution, skew normal distribution, scale mixture of normal (such as student-t (T), slash normal (SL) and contaminated normal (CN)) and scale mixture of skew normal distributions (such as skew-t (ST), skew slash (SSL), skew contaminated normal (SCN) ). The performance of these techniques is discussed in the detection of the influential observations using local influence method for assessing the robustness of these parameter estimators under different perturbation schemes for Grubbs model. The performance is illustrated through an application using real data sets under different bootstrap replications. Our results provide resampling techniques offer better fit, protect against outlying observations and more precise inferences than traditional techniques |