Estimating a Finite Population Mean under Random Non-Response in ...

Open Journal of Statistics, 2017, 7, 834-848 http://www.scirp.org/journal/ojs ISSN Online: 2161-7198 ISSN Print: 2161-718X

Estimating a Finite Population Mean under Random Non-Response in Two Stage Cluster Sampling with Replacement Nelson Kiprono Bii1, Christopher Ouma Onyango2, John Odhiambo1 Institute of Mathematical Sciences, Strathmore University, Nairobi, Kenya Department of Statistics, Kenyatta University, Nairobi, Kenya

1 2

How to cite this paper: Bii, N.K., Onyango, C.O. and Odhiambo, J. (2017) Estimating a Finite Population Mean under Random Non-Response in Two Stage Cluster Sampling with Replacement. Open Journal of Statistics, 7, 834-848. https://doi.org/10.4236/ojs.2017.75059 Received: September 1, 2017 Accepted: October 24, 2017 Published: October 27, 2017 Copyright © 2017 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY 4.0). http://creativecommons.org/licenses/by/4.0/

Abstract Non-response is a regular occurrence in Sample Surveys. Developing estimators when non-response exists may result in large biases when estimating population parameters. In this paper, a finite population mean is estimated when non-response exists randomly under two stage cluster sampling with replacement. It is assumed that non-response arises in the survey variable in the second stage of cluster sampling. Weighting method of compensating for non-response is applied. Asymptotic properties of the proposed estimator of the population mean are derived. Under mild assumptions, the estimator is shown to be asymptotically consistent.

Keywords Non-Response, Nadaraya-Watson Estimation, Two Stage Cluster Sampling

Open Access

1. Introduction In survey sampling, non-response is one source of errors in data analysis. Nonresponse introduces bias into the estimation of population characteristics. It also causes samples to fail to follow the distributions determined by the original sampling design. This paper seeks to reduce the non-response bias in the estimation of a finite population mean in two stage cluster sampling. Use of regression models is recognized as one of the procedures for reducing bias due to non-response using auxiliary information. In practice, information on the variables of interest is not available for non-respondents but information on auxiliary variables may be available for non-respondents. It is therefore desirable to model the response behavior and incorporate the auxiliary data into DOI: 10.4236/ojs.2017.75059 Oct. 27, 2017

834

Open Journal of Statistics

N. K. Bii et al.

the estimation so that the bias arising from non-response can be reduced. If the auxiliary variables are correlated with the response behavior, then the regression estimators would be more precise in estimation of population parameters, given the auxiliary information is known. Many authors have developed estimators of population mean where non-response exists in the study and auxiliary variables. But there exist cases that do not exhibit non-response in the auxiliary variables, such as: number of people in a family, duration one takes to go through education. Imputation techniques have been used to account for non-response in the study variable. For instance, [1] applied compromised method of imputation to estimate a finite population mean under two stage cluster sampling, this method however produced a large bias. In this study, the Nadaraya-Watson regression technique is applied in deriving the estimator for the finite population mean. Kernel weights are used to compensate for non-response.

Reweighting Method Non-response causes loss of observations and therefore reweighting means that the weights are increased for all or almost all of the elements that fail to respond in a survey. The population mean, Y , is estimated by selecting a sample of size n at random with replacement. If responding units to item y are independent so that i is pij ( i 1,= the probability of unit j responding in cluster = 2, , n; j 1, 2, , m ) then an imputed estimator, yI , for Y , is given by

= yI where wij =

  ∗  ∑ wij yij + ∑ wij yij  i , j∈sm ∑ i, j∈swij i, j∈sr  1

(1.0)

1 gives sample survey weight tied to unit j in cluster i and π ij

= π ij p [i, j ∈ s ] is its second order probability of inclusion, sr , is the set of r units responding to item y and sm is the set of m units that failed to respond to item y so that r + m = n and yij∗ is the imputed value generated so that the

missing value yij is compensated for, [2].

2. The Proposed Estimator of Finite Population Mean Consider a finite population of size M consisting of N clusters with N i elements in the ith cluster. A sample of n clusters is selected so that n1 units respond and n2 units fail to respond. Let yij denote the value of the survey variable Y for unit j in cluster i, for i = 1, 2, , N , j = 1, 2, , N i and let population mean be given by

Y=

1 N Mi ∑∑ Yij MN i =i 1 =j 1

(2.1)

ˆ Let an estimator of the finite population mean be defined by Y as follows: ˆ = Y

DOI: 10.4236/ojs.2017.75059

1 M 835

 Yij 1 1  ∑∑ δ ij + n2   n1 i∈s j∈s π ij



1 ˆ    Yijδ ij   ij   

∑∑ 1 − π i∈s j∉s



(2.2)


N. K. Bii et al.

where δ ij is an indicator variable defined by 1, if j th unit in the i th cluster responds 0, elsewhere

δ ij = 

and n1 and n2 are the number of units that respond and those that fail to respond respectively.

π ij is the probability of selecting the jth unit in the ith cluster into the sample.

( )

Let w xij =

1

π ij

to be the inverse of the second order inclusion probabilities

and that xij is the ith auxiliary random variable from the jth cluster. It follows that Equation (2.2) becomes ˆ Y =

1 M

 1 1  ∑∑ w ( xij ) Yijδ ij + n2  n1 i∈s j∈s



∑∑ (1 − w ( xij ) ) Yîjδ ij 

(2.3)



i∈s j∉s

Suppose δ ij is known to be Bernoulli random variables with probability of δ ij δ ij∗ 1 − δ ij∗ , [3]. Thus, success δ ij∗ , then, E δ ij = pr δ ij= 1= δ ij∗ and =

( )

(

)

(

( )

)

the expected value of the estimator of population mean is given by 1 1  ˆ 1  = E Y   ∑∑E w ( xij ) Yij δ ij + M n n    2  1 i∈s j∈s

(

)

 

∑∑E ( (1 − w ( xij ) ) Yîj ) δ ij∗ 

 

i∈s j∉s

(2.4)

Assuming non-response in the second stage of sampling, the problem is therefore to estimate the values of Yîj . To do this, a linear regression model applied by [4] and [5] given below is used;

= Yîj m ( xîj ) + eîj

(2.5)

where m (.) is a smooth function of the auxiliary variables and eîj is the residual term with mean zero and variance which is strictly positive, Substituting Equation (2.5) in Equation (2.4) the following result is obtained:

 ˆ  1  1 E Y  =  ∑∑E m ( xîj ) + eîj w ( xij ) δ ij   M  n1 i∈s j∈s 1  + ∑∑E 1 − w ( xij ) m ( xîj ) + eîj δ ij∗  n2 i∈s j∉s 

((

)

)

(

)(

)

(2.6)

Assuming that n= n= n , and simplifying Equation (2.6) we obtain the fol1 2 lowing

1   ˆ E Y  = ∑∑E m ( xîj ) + eîj w ( xij ) δ ij Mn    i∈s j∈s

((

(

+ ∑∑E 1 − w ( xij ) i∈s j∉s

)

)

)(

 m ( xîj ) + eîj δ  

)

(2.7)

∗ ij

( )

A detailed work done by [5] proved that E eîj = 0 . Therefore Equation (2.7) reduces to

1   ˆ E Y  = ∑∑E m ( xîj ) E w ( xij ) δ ij   Mn  i∈s j∈s

(

(

+ ∑∑E 1 − w ( xij ) i∈s j∉s

DOI: 10.4236/ojs.2017.75059

836

) (

)

 E m ( xîj ) + eîj δ  

) (

)

(2.8)

∗ ij


N. K. Bii et al.

The second term in Equation (2.8) is simplified as follows:

 1  * ∑∑E 1 − w ( xij ) E m ( xîj ) + eîj δ ij  Mn  i∉s j∉s 

(

) (

)

 1  ∑∑E 1 − w ( xij ) m ( xîj ) δ ij  Mn  i∉s j∉s 

(

=

+

)

(2.9)

 1  ∑∑E 1 − w ( xij ) eijδ ij  Mn  i∉s j∉s 

(

)

)

(

But E m= ( xîj ) m ( xij ) , [6]. Thus we get the following: ( xij ) m=

 1  ∗ ∑∑E 1 − w ( xij ) E m ( xîj ) + eîj δ ij  Mn  i∉s j∉s  M N   1  =  ∑ ∑ δ ij m ( xij ) − w ( xij ) δ ij m ( xij )  Mn i =m +1 j =n +1   1  M N +  ∑ ∑ E ( eij δ ij ) − E w ( xij )( eij δ ij )  Mn i =m +1 j =n +1 

(

) (

)

(

(2.10)

)

1   ∗ ∑∑E 1 − w ( xij ) E m ( xîj ) + eîj δ ij  Mn  i∉s j∉s  1 = ( M − ( m + 1) ) ( N − ( n + 1) ) (δ ij ) m ( xij ) − w ( xij ) δ ij m ( xij ) Mn 1 + ( M − ( m + 1) ) ( N − ( n + 1) ) δ ij E ( eij ) − E ( eij ) δ ij w ( xij ) Mn

(

) (

)

{

} }

{

(2.11)

( )

But E eij = 0 , for details see [5]. On simplification, Equation (2.11) reduces to

1   ∗ ∑∑E 1 − w ( xij ) E m ( xîj ) + eîj δ ij  Mn  i∉s j∉s  1 1 M m N n − + − + ( ) ( ) ( )( ) δ m x 1− w x = ( ij ) ( ij ) ij Mn

(

) (

)

{

( )

Recall w xij =

(

)}

(2.12)

1

π ij

so that Equation (2.12) may be re-written as follows:

1   ∗ ∑∑E 1 − w ( xij ) E m ( xîj ) + eîj δ ij  Mn  i∉s j∉s    1 1 M m N n − + − + ( ( ) ) ( ( ) ) δ m x π ij − 1  =    ij ( ij )  Mn   π ij  

(

) (

)

(2.13)

Assume the sample sizes are large i.e. as n → N and m → M , Equation (2.13) simplifies to

1   ∗ ∑∑E 1 − w ( xij ) E m ( xîj ) + eîj δ ij  Mn  i∉s j∉s     π ij − 1   1  =   δ ij m ( xij )  Mn   π ij  

(

DOI: 10.4236/ojs.2017.75059

837

) (

)

(2.14)


N. K. Bii et al.

Combining Equation (2.14) with the first term in Equation (2.08) becomes;  δ ij 1    ˆ E Y  = ∑∑E m ( xij ) E    Mn   π ij  i∈s j∈s

(

)

  + ∑∑δ ij m ( xîj )  i∈s j∉s 

(

 π ij − 1      π   ij  

) 

(2.15)

Since the first term represents the response units, their values are all known. The problem is to estimate the non-response units in the second term. Let the indicator variable δ ij = 1 , the problem now reduces to that of estimating the function m xîj , which is a function of the auxiliary variables, xij . Hence the

( )

expected value of the estimator of the finite population mean under non-response is given as;  ˆ  1  = E Y  ∑∑Yij + ∑∑δ ij m ( xîj )   Mn  i∈s j∈s i∈s j∉s

(

 π ij − 1      π ij  

) 

(2.16)

In order to derive the asymptotic properties of the expected value of the proposed estimator in 2.16, first a review of Nadaraya-Watson estimator is given below.

3. Review of Nadaraya-Watson Estimator Given a random sample of bivariate data

( xi , yi ) , , ( xn , yn )

having a joint pdf

g ( x, y ) with the regression model given by = Yij m ( xij ) + eij as in Equation (2.5), where m (.) is unknown. Let the error term satisfy the following conditions:

E ei , e j ) 0 for i ≠ j = ( eij ) 0, Var= ( eij ) σ ij2 , cov (=

(3.0)

Furthermore, let K (.) denote a symmetric kernel density function which is twice continuously differentiable with:   ∞  ∫−∞ wk ( w ) dw = 0  ∞  2 ∫−∞ k ( w ) dw < ∞   ∞ 2 ∫−∞ w k ( w ) dw = d k  k ( w= ) k ( −w)   ∞

∫−∞ k ( w ) dw = 1

(3.1)

In addition, let the smoothing weights be defined by

 x − X ij  K   b =  , i 1,= 2, , n; j 1, 2, , m = w ( xij )  x − X ij  K ∑ i∈s ∑ i∈s  b   

(3.2)

where b is a smoothing parameter, normally referred to as the bandwidth such that,

∑ i ∑ j w ( xij ) = 1 .

( )

Using Equation (3.2), the Nadaraya-Watson estimator of m xij DOI: 10.4236/ojs.2017.75059

838

is given by:


N. K. Bii et al.

= m ( xîj )

w ( xij ) Yij ∑∑= i∈s j∈s

 x − X ij   Yij  b=  , i 1,= 2, , n; j 1, 2, , m (3.3) x − X  ij  ∑ i∈s ∑ j∈sK  b   

∑ i∈s ∑ j∈sK 

( )

= Yîj m xîj + eîj and the conditions of the error term as exGiven the model plained in 3.0 above, the expression for the survey variable Yij relative to the auxiliary variable X ij can be given as a joint pdf of g xij , yij as follows:

(

m ( x= E (Yij X= x= ij ) ij ij ) where

∫ g ( x , y ) dy

)

∫ yg ( x, y ) dy dy ∫ yg [ y x ] = ∫ g ( x , y ) dy

(3.4)

is the marginal density of X ij . The numerator and the de-

nominator of Equation (3.4) can be estimated separately using kernel functions as follows:

g ( x, y ) is estimated by;

gˆ ( x, y ) =

 1  x − X ij  1  y − Yij 1  K  K ∑∑ mn i j  b  b  b  b

   

(3.5)

and

1

1

 x − X ij  b

∫ ygˆ ( x, y ) dy = mn ∑∑ ∫  b K  i j 

 1  y − Yij  K b  b

   ydy 

(3.6)

Using change of variables technique; let

y − Yij   b  = y wb + Yij   dy = bdw  w=

(3.7)

So that 1 1  x − X ij K ∑∑ ∫ mn i j b  b

= ∫ ygˆ ( x, y ) dy =

 x − X ij 1 K ∑∑ mnb i j  b

1  ( bw + Yij ) K ( w ) bdw b

 1    ∫wK ( w ) bdw + Yij ∫K ( w ) bdw b  

(3.8)

(3.9)

From the conditions specified in Equation (3.1), the following (3.9) simplifies to = ∫ ygˆ ( x, y ) dy

 x − X ij 1 K ∑∑ mnb i j  b

  0 + Yij  

(3.10)

which reduces to: 1

 x − X ij  b

K ∫ ygˆ ( x, y ) dy = mnb ∑∑ i j

  Yij 

(3.11)

Following the same procedure, the denominator can be obtained as follows: DOI: 10.4236/ojs.2017.75059

839


N. K. Bii et al.

1

1

 x − X ij  1  y − Yij  K  b b  b

∫gˆ ( x, y ) dy = mn ∑∑ ∫  b K  i j 

   dy 

(3.12)

 x − X ij  1  y − Yij  1 = K  K  dy ∑∑ mnb =i 1 =j 1  b  ∫ b  b  n

m

Using change of variable technique as in Equation (3.7), Equation (3.12) can be re-written as follows: 1

n

m

 x − X ij  1  ∫ K ( w ) bdw  b  b

K ∫gˆ ( x, y ) dy = mnb =∑∑ i 1 =j 1

(3.13)

which yields

∫gˆ ( x, y ) dy = Since

1

∫ b K ( w ) bdw

1 n m  x − X ij   ∑∑K  mnb =i 1 =j 1  b 

(3.14)

is a pdf and therefore integrates to 1.

( )

It follows from Equations ((3.11) and (3.14)) that the estimator m xîj is as given in Equation (3.3). Thus the estimator of m xij is a linear smoother

( )

since it is a linear function of the observations, Yij . Given a sample and a specified kernel function, then for a given auxiliary value xij , the corresponding y-estimate is obtained by the estimator outlined in Equation (3.3), which can be written as:

ˆ = yîj m= NW ( xij )

∑∑Wij ( xij ) Yij i

( )

where mNW xîj

(3.15)

j

is the Nadaraya-Watson estimator for estimating the un-

known function m (.) , for details see [7] [8]. This provides a way of estimating for instance the non-response values of the

survey variable Yij , given the auxiliary values xij , for a specified kernel function.

ˆ 4. Asymptotic Bias of the Mean Estimator Y Equation (2.16) may be written as N M  1  n m  ˆ = E Y  ∑∑Yij + ∑ ∑ mNW ( yîj )    MN  i=1 j =1  i =n +1 j =m+1

(4.1)

Replacing x by xij and re-writing Equation (3.15) using the property of symmetry associated with Nadaraya-Watson estimator, then

= mNW ( xîj )

=

DOI: 10.4236/ojs.2017.75059

 X ij − xij   Yij  b=  , i 1,= 2, , n; j 1, 2, , m  X ij − xij  ∑ i∈s ∑ j∈sK  b   

∑ i∈s ∑ j∈sK 

 1  X ij − xij   K   Yij  ∑∑ gˆ ( xij )  mnb i j  b  

(4.2)

1

(4.3)

840


N. K. Bii et al.

( )

where gˆ xij

is the estimated marginal density of auxiliary variables X ij .

But for a finite population mean, the expected value of the estimator is given in Equation (4.1). The bias is given by

 ˆˆ   ˆ Bias = Y  E Y − Y       

(4.4)

N M  1  n m   ˆ Bias  Y  E  =  ∑∑Yij + ∑ ∑ m ( xîj )    i =n +1 j =m+1  MN  i=1 j =1  n m N M   1  −  ∑∑Yij + ∑ ∑ Yij   MN  i=1 j =1 i =n +1 j =m+1   

(4.5)

Which reduces to N M  N M  1    ˆ = Bias  Y   ∑ ∑ m ( xîj ) − ∑ ∑ Yij    MN  i=n+1 j =m+1  i =n +1 j =m+1 

=

N M  N M  1    ∑ ∑ m ( xîj ) − ∑ ∑ m ( xij )  MN  i=n+1 j =m+1  i =n +1 j =m+1 

( ) =m ( x ) +  m ( X ) − m ( x )  + e

(4.6) (4.7)

= Yij m X ij + eij as Re-writing the regression model given by Yij

ij

ij

ij

(4.8)

ij

So that from Equation (4.3) the first term in Equation (4.7) before taking the expectation is given as:

 1  X ij − xij   N M K  ∑  Yij  i= n +1∑ j = m+1  b 1  mnb      MN  gˆ ( xij )      1  1  N M  X ij − xij  =  ∑ ∑ K  m ( xij ) MN  gˆ ( xij ) i=n+1 j =m+1  b   1 N M  X ij − xij  + ∑ ∑ K  b   m ( X ij ) − m ( xij ) mnb i=n+1 j =m+1   N M  X ij − xij   1 + K  eij  ∑ ∑ mnb i=n+1 j =m+1  b  

(4.9)

Simplifying Equation (4.9) the following is thus obtained: N M   X ij − xij   1 K   Yij  ∑ ∑ b    mnbgˆ ( xij ) i=n+1 j =m+1  N M   gˆ x m ( xij ) + mˆ 1 ( xij ) + mˆ 2 ( xij )   1  ∑ i= n +1∑ j = m+1  ( ij )  =   MN  mnbgˆ ( xij )  

1 MN

(4.10)

where mˆ 1 ( xij ) =

DOI: 10.4236/ojs.2017.75059

841

N

M

 X ij − xij b 

∑ ∑ K

i =n +1 j =m+1

   m ( X ij ) − m ( xij )   Open Journal of Statistics

N. K. Bii et al.

mˆ 2 ( xij ) =

N

M

 X ij − xij   eij b  

∑ ∑ K

i =n +1 j =m+1

Taking conditional expectation of Equation (4.10) we get

 ∑ N ∑ M M ( xîj )  i= n +1 j= m+1  E   xij    1 N M  mˆ 1 ( xij ) mˆ 2 ( xij )   1  m ( xij ) +  E = + ∑ ∑ MN  mnb i=n+1 j =m+1  gˆ ( xij ) gˆ ( xij )     

(4.11)

To obtain the relationship between the conditional mean and the selected bandwidth, the following theorem due to [6] is applied; Theorem: (Dorfman, 1992)

Let k ( w ) be a symmetric density function with

∫wk ( w) dw = 0

and

n → π with N 0 < π < 1 . Besides, assume the sampled and non-sampled values of x are in the interval [ c, d ] and are generated by densities d s and d p−s respectively both bounded away from zero on [ c, d ] and assumed to have continuous second derivatives. If for any variable  , E (  U= u= ) A ( u1) + O ( B ) and   , then Var (  U= u= O C = A u O +  ) ( ) ( ) p  B + C 2  .   Applying this theorem, we have

∫w k ( w) dw = k2 . 2

Assume n and N increase together such that

 ˆ Y  MSE  xij  +

 = 1  ( MN )2 

( MN − mn ) 4m 2 n 2

2

( )

2 2   ( MN − mn ) ∫k w dw  mnbg ( xij ) 

 2 g ′′ ( xij ) m′ ( xij )   b k ( k )  m′′ ( xij ) + g ( xij )   4

2

2 2

(4.12)

 ( MN − mn )2 1   + O b4 + O  +  mnb mnb    

( )

This theorem is stated without proof. To prove it, we partition it into the bias and variance terms and separately prove them as follows:

(

)

From Equation (3.0) it follows that E eij X ij = 0 . Therefore, E  mˆ 2 ( xij )  = 0 . Thus E  mˆ 1 ( xij )  can be obtained as follows: N

E∑

M

∑

i =n +1 j =m+1

1 = MN

 mˆ 1 ( xij )   

 1  N M  X ij − xij E∑ ∑ K   mnb i=n+1 j =m+1  b

     m ( X ij ) − m ( xij )    

(4.13)

Using substitution and change of variable technique below V − xij w =so that V = xij + bw and dV = bdw b

(4.14)

Equation (4.13) can simplify to: DOI: 10.4236/ojs.2017.75059

842


N. K. Bii et al. N

M

E∑

∑

i =n +1 j =m+1

 mˆ 1 ( xij )   

1  MN − mn  = k ( w )  m ( xij + bw ) − m ( xij )  ∫g ( xij + bw ) bdw  MN  mnb ∫  =

1  MN − mn  k ( w )  m ( xij + bw ) − m ( xij )  g ( xij + bw ) dw  ∫ MN  mn 

(4.15)

(4.16)

Using the Taylor’s series expansion about the point xij , the kth order kernel can be derived as follows:

g ( xij + bw= ) g ( xij ) + g ′ ( xij ) bw +

( )

(4.17)

( )

(4.18)

1 1 g ′′ ( xij ) b 2 w2 +  + g k ( xij ) b k wk + O b 2 2 k!

Similarly,

m ( xij + bw = ) m ( xij ) + m′ ( xij ) bw + 12 m′′ ( xij ) b2 w2 +  + k1! mk ( xij ) bk wk + O b2 Expanding up to the 3rd order kernels, Equation (4.18) becomes

1 1  m ( xij + bw ) − m= ( xij  m′ ( xij ) bw + m′′ ( xij ) b 2 w2 + m′′′ ( xij ) b3 w3  2 3!

(4.19)

( )

In a similar manner, the expansion of Equation (4.16) up to order O b 2

is

given by: N

E∑

M

∑

i =n +1 j =m+1

 mˆ 1 ( xij )   

 1  MN − mn 1   = k ( w )  m′ ( xij ) bw + m′′ ( xij ) b 2 w2  g ( xij ) + g ′ ( xij ) bw dw  ∫ MN  mn 2   

(

)

(4.20)

Simplifying Equation (4.20) gives; 1  MN − mn   mˆ 1 ( xij )  =   g ( xij ) m′ ( xij ) b ∫wk ( w ) dw   MN  mn  i =n +1 j =m +1  MN − mn  2 2 (4.21) +  g ′ ( xij ) m′ ( xij ) b ∫w k ( w ) dw  mn   MN − mn  1 2 2 2  +  g ( xij ) m′′ ( xij ) b ∫w k ( w ) dw + O b   mn  2  N

E

M

∑ ∑

( )

Using the conditions stated in Equation (3.1), the derivation in (4.21) can further be simplified to obtain: N

M

E∑

∑

i =n +1 j =m+1

 mˆ 1 ( xij )   

1  MN − mn   1  2 2  =    g ′ ( xij ) m′ ( xij ) + g ( xij ) m′′ ( xij )  b d k + O b  MN  mn   2  

( )

(4.22)

Hence the expected value of the second term in Equation (4.11) then becomes: N

E

M

∑ ∑

i =n +1 j =m +1

1 = MN

DOI: 10.4236/ojs.2017.75059

 mˆ 1 ( xij )   

  g ′ ( xij ) m′ ( xij )  2  MN − mn   1   b dk + O b2     m′′ ( xij ) + g ( xij )  mn   2     

( )

843

(4.23)


N. K. Bii et al.

1 = MN

=

  −1  MN − mn   m′′ ( xij )  2 2   ′ ′   g x g x m x b d O b + + (4.24) ( ) ( ) ( )   ij ij ij k    mn   2    

( )

1  MN − mn  2 2    b dk C ( x ) + O b  MN  mn  

( )

(4.25)

where = C ( x)

−1 1 m′′ ( xij ) +  g ( xij )  g ′ ( xij ) m′ ( xij ) 2

(4.26)

and d k is as stated in Equation (3.1) Using equation of the bias given in (4.4) and the conditional expectation in Equation (4.11), we obtain the following equation for the bias of the estimator:

1  MN − mn  2  ˆ 2  = Bias  Y    b dk C ( x ) + O b    MN  mn   1  MN − mn  2 2  =   b dk C ( x ) + O b  MN  mn  

( )

(4.27)

( )

ˆ 5. Asymptotic Variance of the Estimator, Y From Equations ((4.9) and (4.11)), m2′ ( xij ) =

1 n m  X ij − xij  ∑∑K  b  eij mnb =i 1 =j 1  

(5.0)

Hence

1  MN − mn  ∑  mˆ 2 ( xij ) =  2  i =n +1 j =m+1 ( MN )  mnb  N

Var ∑

M

2 n

m

∑∑Var ( Dx )

(5.1)

i =1 j =1

where  X ij − xij  Dx = K   eij b  

Expressing Equation (5.1) in terms of expectation we obtain: = Var ∑ ∑  mˆ 2 ( xij )  i =n +1 j =m+1 N

M

1

( MN )

2

 ( MN − mn )2  2 2   E [ Dx ] −  E ( Dx )  2 mnb  

{

}

(5.2)

Using the fact that the conditional expectation

E ( eij X ij ) = 0 , the second term in Equation (4.13) reduces to zero. Therefore, 1  mˆ 2 ( xij )  = 2   i =n +1 j =m+1 ( MN ) N

Var ∑

M

∑

 ( MN − mn )2  2  σ(x ) mnb 2   ij

(5.3)

where

E ( eij X ij ) = σ 2x ( ij ) 2

Let X = X ij , and x = xij , and making the following substitutions DOI: 10.4236/ojs.2017.75059

844


N. K. Bii et al.

X −x  b  X −x= bw dX = bdw  w=

(5.4)

( MN − mn )  X − x  2 K ∑  mˆ 2 ( xij ) = 2  σ g ( X ) dX 2 ∫   b  ( xij ) mnb ( MN ) i =n +1 j =m+1 N

2

M

Var ∑

2

( MN − mn ) 2 K w σ 2x g ( x + bw ) bdw 2 ∫ ( ) 2 ( ) mnb ( MN )

(5.5)

2

=

ij

(5.6)

which can be simplified to get:

( MN − mn ) 2  1  Var ∑ ∑  mˆ 2 ( xij )  K w g ( x ) σ 2x dw + O  =  2 ∫ ( ) ( ij )  mnb  mnb ( MN ) i =n +1 j =m+1 N

2

M

N

Var ∑

M

∑

i =n +1 j =m+1

( MN )

 mˆ 1 ( xij )     1 n m  X ij − xij   ∑  mnb ∑∑K  b    M ( X ij ) − m ( xij ) i =n+1 j =m+1  i =1 j =1     N

1

=

2

(5.7)

M

Var ∑

N M  X ij − xij ( MN − mn ) = Var ∑ ∑  mˆ 1 ( xij )  Var K  2 2 mnb ( MN ) i =n +1 j =m+1  b 2

   M ( X ij ) − m ( xij )  

(5.8)

(5.9)

Hence N

Var ∑

M

∑

i =n +1 j =m+1

 mˆ 1 ( xij )   

2 ( MN − mn )   X − x 2 E  ∫K    M ( X ) − m ( x )   g ( X ) dX 2 2 mnb ( MN )   b   2

=

(5.10)

= bw + x so that dX = bdw . where X Changing variables and applying Taylor’s series expansion then N

Var ∑

M

∑

i =n+1 j =m+1

 mˆ 1 ( xij )    (5.11)

2 ( MN − mn ) 2 K w  m x + bw ) − m ( x )  g ( x + bw ) dw 2 ∫ ( )  ( 2 mnb ( MN ) 2

=

2 ( MN − mn ) 2 K w  m x + m′ ( x ) bw +  − m ( x )  ( g ( x ) + g ′ ( x ) bw ) dw 2 ∫ ( )  ( ) 2 mnb ( MN ) 2

=

(5.12)

which simplifies to  ( MN − mn )2 b 2    ˆ = m x O  ∑  1 ( ij )  mnb i =n +1 j =m+1   N

Var ∑

M

(5.13)

For large samples, as n → N , m → M and for b → 0 , then mnb → ∞ . Hence the variance in Equation (5.12) asymptotically tends to zero, that is, N

Var ∑

M

∑

i =n +1 j =m+1

DOI: 10.4236/ojs.2017.75059

845

 mˆ 1 ( xij )  → 0  


N. K. Bii et al.

 ˆ  ( MN − mn ) Var  Y  = 2   mnb ( MN )

2



∑ ∑ Var  m ( xij ) + N

M

i =n +1 j =m+1



m1′ ( xij ) + m2′′ ( xij )   gˆ ( xij ) 

(5.14)

On simplification,

 N M ( MN − mn )   ˆ Var  Y  = Var  ∑ ∑  mˆ 2 ( xij )   2 2   mnb ( MN )  gˆ ( xij )  i=n+1 j =m+1    2

(5.15)

Substituting Equations ((5.7) into (5.15)) yields the following:

 ˆ = Var  Y   

=

1

( MN )

2

1

( MN )

2

 ( MN − mn )2 K ( w )2 σ 2 dw   ( MN − mn )2 ∫ 1  ( xij )  + + O     (5.16) mnb mnb   mnb g ( xij )    

(

)

 ( MN − mn )2 H ( w ) σ 2   ( MN − mn )2 1  ( xij )  + + O    mnb mnb   mnb g ( xij )    

(

)

(5.17)

where, H ( w ) = ∫K ( w ) dw 2

It is notable that the variance term still depends on the marginal density func-

( )

tion, g xij

of the auxiliary variables X ij . It can also be observed that the va-

riance is inversely related to the smoothing parameter b. This implies that an increase in b results in a smaller variance. However, increasing the bandwidth would give a larger bias. Therefore there is a trade-off between the bias and variance of the estimated population mean. A bandwidth that provides a compromise between the two measures would therefore be desirable.

6. Mean Squared Error (MSE) of the Finite Population Mean ˆ Estimator Y ˆ The MSE of Y combines the bias and the variance terms of this estimator that is,

 ˆ ˆ  MSE = Y  E Y − Y     

2

(6.0)

ˆ   ˆ  ˆ  ˆ MSE  Y  =E  Y − E Y  + E Y  − Y         

2

(6.1)

Expanding Equation (6.1) gives:

  ˆ   ˆ ˆ  MSE  Y  = E  Y − E Y   + E  E Y  − Y               ˆ ˆ ˆ        + 2 E  Y − E Y   E Y  − Y         2

 ˆ = Var  Y  + Bias 2 + 0  

2

(6.2)

(6.3)

Combining the bias in Equation (4.27) and the variance in Equation (5.17) and conditioning on the auxiliary values xij of the auxiliary variables X ij then DOI: 10.4236/ojs.2017.75059

846


N. K. Bii et al.

ˆ  MSE  Y X ij = xij     MN − mn )2 H ( w ) σ 2  1 ( xij ) (  = + O  2  MN mnb g ( xij ) ( MN )    1  MN − mn  2 2  +   b dk C ( x ) + O b  MN  mn   1

(

)

  ( MN − mn )2 1    +    mnb mnb     

(6.4)

( )

ˆ  MSE  Y X ij = xij    2 2  1  ( MN − mn ) H ( w ) σ ( xij ) = 2  mnb g ( xij ) ( MN )   2 2   2 g ′ ( xij ) m′ ( xij )  MN mn − ( ) 4 2  b d k  m′′ ( xij ) + +  4 ( mn )2 ( MN )2 g ( xij )   

(

( )

+ O b4 +

)

(6.5)

1   MN − mn  1         O  + MN   mnb  mnb    

where H ( w ) = ∫K ( w ) dw , d k = ∫w2 K ( w ) dw , 2

−1 1 m′′ ( xij ) +  g ( xij )  g ′ ( xij ) m′ ( xij ) as used earlier in the rest of the de2 rivations.

= C ( x)

7. Conclusion If the sample size is large enough, that is as n → N and m → M the 𝑀𝑀𝑀𝑀𝑀𝑀 of  ˆ  Y  in Equation (6.5) due to the kernel tends to zero for sufficiently a small    ˆ bandwidth b. The estimator  Y  is therefore asymptotically consistent since   its MSE converges to zero.

References

DOI: 10.4236/ojs.2017.75059

[1]

Singh, S. and Horn, S. (2000) Compromised Imputation in Survey Sampling. Metrika, 51, 267-276. https://doi.org/10.1007/s001840000054

[2]

Lee, H., Rancourt, E. and Sarndal, C. (2002) Variance Estimation from Survey Data under Single Imputation. Survey Nonresponse, 315-328.

[3]

Bethlehem, J.G. (2012) Using Response Probabilities for Assessing Representativity. Statistics Netherlands, International Statistical Review, 80, 382-399.

[4]

Ouma, C. and Wafula, C. (2005) Bootstrap Confidence Intervals for Model-Based Surveys. East African Journal of Statistics, 1, 84-90.

[5]

Onyango, C.O., Otieno, R.O. and Orwa, G.O. (2010) Generalized Model Based Confidence Intervals in Two Stage Cluster Sampling. Pakistan Journal of Statistics and Operation Research, 6. https://doi.org/10.18187/pjsor.v6i2.128

[6]

Dorfman, A.H. (1992) Nonparametric Regression for Estimating Totals in Finite Populations. In: Proceedings of the Section on Survey Research Methods, American 847


N. K. Bii et al. Statistical Association Alexandria, VA, 622-625.

DOI: 10.4236/ojs.2017.75059

[7]

Nadaraya, E.A. (1964) On Estimating Regression. Theory of Probability & Its Applications, 9, 141-142. https://doi.org/10.1137/1109020

[8]

Watson, G.S. (1964) Smooth Regression Analysis. Sankhya: The Indian Journal of Statistics, Series A, 359-372.

848
