Page 45 - Contributed Paper Session (CPS) - Volume 2
P. 45

CPS1416 Jungtaek O. et al.
            should be used after identifying the characteristics of the data. Eventually the
            accuracy and the precision of the estimation is dependent on the parameter 
            which is included in the variance. Therefore the better results are expected if
                                                                2
            the estimated value of  is used when () =  σ , 0 < ν and  ≠ 1. The
                                                              
            simple way to estimate  is using MLE method. However it is necessary to know
            the  population  distribution,  so  there    is  a  limitation  in  general  use.
            Consequently, if there is an estimator that maintains the good properties of
            the regression estimator and also has a simple expression like the ratio-type
            estimator, this estimator may be very useful in the actual data analysis.
                In this paper, we suggest a generalized ratio-type estimator which has the
            advantages of the regression estimator, also has a good features of the ratio
            estimator. So this estimator has good properties such as relatively accuracy,
            robustness and convenience of calculation, ease of use. In addition, a method
            of  estimating    using  MLE  is  studied  under  the  assumption  that  the  error
            follows the normal distribution.

            2.  Generalized regression estimator
                Consider the regression model with p auxiliary variables following as
                                       =   +    +     + ⋯ +   +      (1)
                                                       2 2
                                                1 1
                                           0
                                      
                                                                     
                                                                           
                                    2
            Where  ~(0, ∏     ) and 0 ≤  ,  = 1, … , .
                    
                                                 
                                  
                             =1
            Then from equation (1), if  = 0,  = 1, ⋯ ,  the usual multiple regression
            estimator is obtained by
                                    ̂
                                            ̂
                                                 ̂
                                                        ̂
                                                                    ̂
                                                             Y ̅   =  +  X ̅ +  X ̅ + ⋯ +  X ̅                (2)
                                                         2 2
                                                  1 1
                                             0
                                                                     
            where   =   −   1 −  2 − ⋯ −    and  ,  = 1, ⋯ ,  are sample means
                                             ̂
                                   ̂
                            ̂
                   ̂
                    0
                             1
                                              
            and  ,  = 1, ⋯ ,  are known population means of auxiliary variables. Also if
                                                                                      2
            νk,  = 1, ⋯ ,  are unknown, we can estimate the 0, ,,  = 1, ⋯ ,  and σ
            using MLE with the following log-likelihood function defined by
                                   (| ,  , … ,  ,  , … ,  ,  ) =
                                                             2
                                                
                                                   1
                                                         
                                           1
                                        0
                                                                     2
                               1      ( −  −   − ⋯ −   )          (3)
                                                                     
                                                        1 1
                                                   0
                                              
                               2
            − ∑ log (∏        ) −     ∑ {                             },
                         =1    2 2             ∏      
              =1                      =1             =1  

            Eventually the following regression estimator is obtained by
                              ̂
                                                                   X ̅
                                                Y ̅   =  ̂ 0  +  ̂ 1 1  ̂  ,     (4)
                                                    X ̅ + ⋯ + 
                                                      
            using  ̂   = ( ̂ 0  +  ̂ 1  + ⋯ +  ̂  ) obtained  from  equation  (3).
            Therefore, several auxiliary variables can be used for the mean estimation in
            order for improving the precision and the accuracy of parameter estimation.

                                                                34 | I S I   W S C   2 0 1 9
   40   41   42   43   44   45   46   47   48   49   50