Chapter 12

Sampling on Successive Occasions

Many times, we are interested in measuring a characteristic of a population on several occasions to

estimate the trend in time of population means as a time series of the current value of population mean or
the value of population mean over several points of time.

When the same population is sampled repeatedly, the opportunities for a flexible sampling scheme are
greatly enhanced. For example, on the hth occasion we may have a part of sample that are matched with
(common to) the sample at (h  1)th occasion, parts matching with both (h  1)th and (h  2)th occasions,
etc.

Such a partial matching is termed as sampling on successive occasions with partial replacement of
units or rotation sampling or sampling for a time series.

Notations:
Let P be the fixed population with N units.
yt : value of certain dynamic character which changes with time t and can be measured for each unit on

a number of occasions, t  1, 2,.., n .

yij : value of y on j th unit in the population at the i th occasion, i  1, 2,..., h, j  1,..., N .

1
Yi 
N
y
j
ij : population mean for the i th occasion

1 N
Si2  
N  1 j 1
( yij  Yi ) 2 : population variance for the i th occasion.

Generally, we assume S12  S22  ...  S 2 .

1 N
 ii*   ( yij  Yi )( yi* j  Yi* ) .
N  1 j 1

is the population correlation coefficient between the observations at the occasions i and i*
(i  i*  1, 2,..., h) .
  12

si* : sample of size ni selected at the i th occasion

*
sim : part of si* which is common to (i.e. matched with) si*1 ,
*
sim  si*  si*1 , i  2,3,..., h ( s1m  s2 m )

Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur

Page 1
Note that s1m and s2 m are of the sizes n1* and n2* respectively.

siu* : set of units in si* not obtained by the selection in sim

*
.
Often
siu*  si*c1  si (i  2,..., h) ( s1*u  P  s1*m ) .

Note that siu* is of size ni** ( ni  ni* ) .

yi  sample mean of units in i th occasion.

yi*  sample mean of the units in sim

*
on the i th occasion.

yi**  sample mean of units in siu on the i th occasion.

yi***  sample mean of units in sim on the (i  1)th occasion, i  2,3,..., h

y ***
2  y1* , yi*** depends on yi 1 and yi**1 

Sampling on two occasions

Assume that ni  n
ni*  m
ni**  u ( n  m), i  1, 2

Suppose that the sample s1* is an SRSWOR from P. The sample

s2*  s2*m  s2*u
where s2*m is an SRSWOR sample of size m from s1* and
s2*u is an SRSWOR sample of size u from  P  s1*  .

Estimation of Population mean

Two types of estimators are available for the estimation of a population mean:
1. Type 1 estimators: They are obtained by taking a linear combination of estimators obtained from
s2*u and s2*m .

2. Type 2 estimators: They are obtained by considering the best linear combination of sample
means.

Type 1 estimators:
Two estimators are available for estimating Y2

(i) t2u  y2**

S 22 1
with Var ( y2** )   (say)
u Wu

Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur

Page 2
(ii) t2m = linear regression estimate of Y2 based on the regression of y2 j on y1 j

= y2*  b( y1  y1* )

 (y 1j  y1* )( y2 j  y2* )
is*2m
where b = is the sample regression coefficient.
 (y 1j  y1* ) 2
js1*m

Recall in case of double sampling, we had

1 1  1 1 
Var (Yˆregd )  S y2      2 s y2   * 
n N  n n 
S y2 1 1 
   2 S y2   *  .
n n n 
(1   2 ) 2  S y
2 2
1
 Sy  * (ignoring term of order ).
n n N
So in this case
S 22 (1   2 )  2 S 22
Var  t2 m   
m n
1
 (say).
Wm

If there are two uncorrelated unbiased estimators of a parameter, then the best linear unbiased estimator
of parameter can be obtained by combining them using a linear combination with suitably chosen
weights. Now we discuss how to choose weights in such a linear combination of estimators.

Let ˆ1 and ˆ2 be two uncorrelated and unbiased estimators of  , i.e., E (ˆ1 )  E (ˆ2 )   and

Var (ˆ1 )   12 , Var (ˆ2 )   22 , Cov(ˆ1 , ˆ2 )  0.

Consider ˆ  ˆ1  (1   )ˆ2 where 0    1 is the weight. Now choose  such that Var (ˆ) is
minimum.
Var (ˆ)   2 12  (1   ) 2  22
Var (ˆ)
0

 2 12  2(1   ) 22  0
 22
    * , say
1   2
2 2

 2Var (ˆ)
 0.
 2  *

Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur

Page 3
The minimum variance achieved by ˆ is

Var (ˆ) Min   *  12  (1   * ) 2  22

2

 24 12  14 22
 
   22     22 
2 2 2 2
1 1

 22 12 1
  .
1   2
2 2
1

1
 22  12
Now we implement this result in our case.

Consider the linear combination of t2u and t2 m as

Yˆ2  t2u  (1   )t2 m

where the weights  are obtained as

Wu

Wu  Wm

so that Yˆ2 is the best combined estimate.

The minimum variance with this choice of  is

1 S 2 (n  u  2 )
Var (Yˆ2 )   22 .
Wu  Wm (n  u 2  2 )

S2
For u  0 (complete matching), Var (Yˆ2 )  2 .
n
S2
For u  n (no matching), Var (Yˆ2 )  2 .
n

Type II estimators:
We now consider the minimum variance linear unbiased estimator of Y2 under the same sampling

scheme as under Type I estimator.

A best linear (linear in terms of observed means) unbiased estimator of Yˆ2 is of the form

Yˆ2*  ay1**  by1*  cy2*  dy2**

 m n 1
where constants a, b, c, d and matching fraction      are to be suitably chosen so as to
 n n 
minimize the variance.

Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur

Page 4
Assume S12  S22 .

Now E (Yˆ2* )  (a  b)Y1  (c  d )Y2 .

If Yˆ2* has to be an unbiased estimator of Y2 , i.e.

E (Yˆ2* )  Y2 ,

it requires
ab  0
c  d  1.

Since a minimum variance unbiased estimator would be uncorrelated with any unbiased estimator of
zero, we must have

Cov(Yˆ2* , y1**  y1* )  0 (1)

Cov(Yˆ * , y *  y ** )  0 .
2 2 2 (2)
Since
Cov( y2* , y1** )  0  Cov( y2* , y2** )
S2
Cov( y2* , y1* ) 
m
Cov( y , y )  Cov( y2** , y2* )  0
**
2
**
1

S2
Var ( y2* ) 
m
S2
Var ( y2 )  .
**

u
1
Now solving (1) and (2) by neglecting terms of order , we have
N

Cov(Yˆ2* , y1**  y1* )  Cov(ay1**  by1*  cy2*  dy2** , y1**  y1* )

 aVar ( y1** )  bC ov( y1* , y1** )  cCov( y2* , y1** )  dC ov( y2** , y1** )
 aC ov( y1** , y1* )  bVar ( y1* )  cC ov( y1* , y2* )  dCov( y2** , y1* )
or
S2 cS 2 (1  c) S 2
a   . (3)
m m u
Similarly, from (2), we have
Cov( y2* , y2*  y2** )  0
aS 2 c  S 2 aS 2
   . (4)
m m u

Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur

Page 5
Solving (3) and (4) gives
 
a , c
1  
2 2
1   2 2
u nu
where    1 ,  
n n
b   a, d  1  c.

Substituting a, b, c, d , the best linear unbiased estimator of Y2 is

  (1   2  ) y2** 
Yˆ2*    ( y1**  y1* )   y2*  .
 (1   
2 2
) 
For these values of a and c ,
 1   2 S 2 
Var (Yˆ2* )   2 2 
.
 1   n 

Alternatively, minimize Var (Yˆ2* ) with respect to a and c and find optimum values of a and c . Then

find the estimator and its variance.

Till now, we used SRSWOR for the two occasions. We now consider unequal probability sampling

schemes on two occasions for estimating Yˆ2 . We use the same notations as defined in varying probability

scheme.

Des Raj Scheme:

Let s1* be the sample selected by PPSWR from P using x as a size (auxiliary) variable.

xi
Then pi  is the size measure of i , where X tot is the population total of auxiliary variable.
X tot

s2*  s2*m  s2*u

where s2*m is an SRSWR( m ) from s1* and s2*u is an independent sample selected from P by PPSWR

using u draws ( m  u  n).

The estimator is
Yˆ2 des   t2 m  (1   )t2u ; 0    1

where

Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur

Page 6
  y2 j  y1 j   y1 j 
t2 m       
mp j  js1*  np j

js2* m  
y 
t2 u    2 j  .

js2* u  up j 

Assuming
2 2
N  Y1 j  N  Y2 j 
 Pj 
P
 Y1    Pj 
 P
 Y2   V0 (say) .

j 1  j  j 1  j 
For the optimum sampling fraction
m
 ,
n
V0 (1  2(1   )
Var (Y2 des ) 
2n
where
N
 Y1i   Y2i 
 P  P Y  P
i 1  Y2 

i 1  i  i 
 pps ( y1 ) pps ( y2 )
2
Z N

Varpps ( z )   Pi  i  Z    pps
2
( z)
i 1  Pi 
N
Z   Zi .
i 1

(ii) Chaudhuri-Arnab sampling scheme

Let s1* be a sample selected by Midzuno’s sampling scheme,

s2*  s2*m  s2*u

where s2*m = SRSWOR (m) sample from s1*

*
s2u = sample of size u from P by Midzuno’s sampling scheme.

Then an estimator of Y is

Yˆ2 ca   t2 m  (1   )t2u ; 0  1

where
 ( y2 j  y1 )n   y1 j 
t2 m    m j
    
js2* m   js1*   j 
y2 j
t2 u  
js2 u
*
j

 j  np j
 *j  up j .
Similarly, other schemes are also there.
Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur
Page 7
Sampling on more than two occasions
When there are more than two occasions, one has a large flexibility in using both sampling procedures
and estimating the character.
Thus on occasion i
 one may have parts of the sample that are matched with occasion (i  1)
 parts that are matched with occasion (i  2) j

 and so on.
One may consider a single multiple regression of all previous matchings on the current occasion.
However, it has been seen that the loss of efficiency incurred by using the information from the latest two
or three occasions only is fairly small in many occasions.

Consider the simple sampling design where

si*  sim
*
 siu* ,
*
where sim is a sample by SRSWOR of size mi from s(*i1) ,

siu* is a sample by SRSWOR of size ui ( n  mi ) from the units not already sampled.

Assume ni  n, S22  S 2 for all i .

On the i th occasion, we have therefore two estimators

S2 1
tiu  yi** with Var (tiu )  
ui Wiu

tim  yi*  b(i 1)i (Yˆ( i 1)  yi*** )

where b(i1),i is the regression of yij on y(i1) j

(y ( i 1) j  yi*** )( yij  yi* )

si**
b(i1),i 
(y ( i 1) j  yi*** ) 2
si**

and
S 2 (1   2 ) 1
Var (tim )    2Var (Yˆ(i 1) ) 
mi Wim

1
assuming that (i1),i   , i  2,3,...,. and terms of order are negligible.
N

Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur

Page 8
The expression of Var (tim ) has been obtained from the variance of regression estimator under the double

sampling
1 1  1 1 
Var ( yˆ regd )  Su2      2 S y2   * 
n N  n n 
S (1   )  S y
2 2 2 2

 y  *
n n
1
which is obtained after ignoring the terms of by using mi for n and replacing
N
 2 S y2
n *    V ( x *
2
by  2Var (Yˆ(i 1) ) since    and Si2 is constant. Using weights as the inverse of

variance, the best weighted estimator from tiu and tim is

Yˆiu  i  tiu  (1  i )  tim

where
Wiu
i  .
Wiu  Wim
Then
1 gi S 2
Var ( yˆi )   (say), i  1, 2,..., ( g1  1).
Wiu  Wim n

1 S2
Substituting 
Wiu ui

1 gi S 2
in  ,
Wiu  Wim n

we have
n 1
 ui  .
gi 1  2
 2 gi 1

mi n

n n
Now maximize with respect to mi so as to minimize Var ( yˆi ). So differentiate with respect to mi
gi gi
and substituting it to be zero, we get
2
(1   2 )  1   2  2 gi 1 
  
mi2  mi n 
n 1  2
 mˆ i  .
gi 1 (1  1   2 )

Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur

Page 9
 mˆ i
Now the optimum sampling fracture can be determined successively for i  2,3,... for given values of
n
n
 . Substituting this in the expression of , we have
gi

1 (1  1   2 ) 2
 1
gi gi1 2

or qi  1  aqi1

where

1 (1- 1-  2 )
qi  , q1  1, a  ; 0  a  1.
gi (1  1   2 )

Repeated use of this relation gives

qi 1  1  aqi  2
 qi  1  a(1  aqi 1 )
 1  a  a 2 qi 1
qi  2  1  aqi 3
 qi  1  a  a 2 (1  aqi  2 )

(1  a i ) 1
  as i  .
(1  a) 1  a

For sampling an infinite number of times, the limiting variance factor g  is

2 1  2
g  1  a  .
1 1  2

The limiting value of V (Yˆi ) as i   is

2S 2 1   2
lim Var (Yˆi )  Var (Yˆ )  .
i 

n 1 1  2

The limiting value of optimum sampling fraction as i   is

mˆ i mˆ  1  2 1
lim    .
i  n n

g 1  1   2 2

Thus for the estimation of current population mean by this procedure, one would not have to match more
than 50% of the sample drawn on the last occasion.

Unless  is very high, say more than 0.8, the reduction in variance (1  g h ) is only modest.

Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur

Page 10
Type II estimation
Consider

Yˆi  aiYˆi 1  bi yi**1  cyi***  di yi**  ei yi*

Now E (Yˆi )  (ai  bi  ci )Y( i 1)  (di  ei )Yi .

So for unbiasedness,
ci  (ai  bi )
.
di  1  ei. .
An unbiased estimator is of the form

Yˆi  ai Yˆ(i 1)  bi yi**  (ai  bi ) yi***  di yi**  (1  di ) yi* .

To find optimum weights, minimize

Var (Yˆi ) with respect to ai , bi , di .

Alternatively, one can consider that

Yˆi  ai Yˆi 1  bi yi**  (ai  bi ) yi***  di yi**  (1  ai ) yi* .

must be uncorrelated with all unbiased estimators of zero. Thus

Cov(Yˆi , yi**1  yi*** )  0

Cov(Yˆ , y **  y *** )  0
i 1 i 1 i

Cov(Yˆi , yi** 2  yi***

1 )  0

Using these restrictions, find the constants and get the estimator.

Sampling Theory| Chapter 12 | Sampling on Successive Occassions | Shalabh, IIT Kanpur

Page 11