The Existence Of MLE Of The Parameters In Censored Linear Regression Models With Unequal Variances

In experiment of designing,if we can observe all the complete datas,i.e,there are noe censored datas in experiment,we can analyze them by traditional means.But,it isn't suitble in censored linear regression models with censored observation datas.Likelihood methods can treat failure datas and censored datas easily.But it is a problem that MLE may be infinite ,namely,MLE is not existed, so it is important to consider the existence of MLE of unknown parameters.For the existence of maximum likelihood estimate of the parameters,the sufficient and necessary condition in linear models with censored data is introduced by Sivapulle and Burridge( 1986),and these conditions can transform to linear programming problem.Hamada and Tse(1988) gives a sufficient and necessary condition in constant variances too.But it is not mentioned in references to the existence of MLE in censored linear regression models with unequal variances and in dispersion effects models.Although Bihua(2004) decribe a iterative algorithm of identifing and estimating both location and dispersion effects in censored linear regression models with unequal variances,the existence of MLE isn't discussed too.In this paper,we give the sufficient condition and proof of MLE non-existence on the basis of Sivapulle and Burridge(1986) improving the sufficient and necessary condition in censored linear regression models with same varience. And these conditions can transform to linear programming problem,too.Then we give the condition and proof of MLE non-existence in dispersion model.