Chapter 4

Stratified Sampling
An important objective in any estimation problem is to obtain an estimator of a population parameter
which can take care of the salient features of the population. If the population is homogeneous with
respect to the characteristic under study, then the method of simple random sampling will yield a
homogeneous sample and in turn, the sample mean will serve as a good estimator of population mean.
Thus, if the population is homogeneous with respect to the characteristic under study, then the sample
drawn through simple random sampling is expected to provide a representative sample. Moreover, the
variance of sample mean not only depends on the sample size and sampling fraction but also on the
population variance. In order to increase the precision of an estimator, we need to use a sampling
scheme which can reduces the heterogeneity in the population. If the population is heterogeneous with
respect to the characteristic under study, then one such sampling procedure is stratified sampling.

The basic idea behind the stratified sampling is to

 divide the whole heterogeneous population into smaller groups or subpopulations, such that the
sampling units are homogeneous with respect to the characteristic under study within the
subpopulation and
 heterogeneous with respect to the characteristic under study between/among the
subpopulations. Such subpopulations are termed as strata.
 Treat each subpopulation as separate population and draw a sample by SRS from each stratum.
[Note: ‘Stratum’ is singular and ‘strata’ is plural].

Example: In order to find the average height of the students in a school of class 1 to class 12, the
height varies a lot as the students in class 1 are of age around 6 years and students in class 10 are of
age around 16 years. So one can divide all the students into different subpopulations or strata such as

Students of class 1, 2 and 3: Stratum 1

Students of class 4, 5 and 6: Stratum 2
Students of class 7, 8 and 9: Stratum 3
Students of class 10, 11 and 12: Stratum 4

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 1

Now draw the samples by SRS from each of the strata 1, 2, 3 and 4. All the drawn samples combined
together will constitute the final stratified sample for further analysis.
Notations:
We use the following symbols and notations:
N : Population size
k : Number of strata
Ni : Number of sampling units in ith strata
k
N   Ni
i 1

ni : Number of sampling units to be drawn from ith stratum.

k
n   ni : Total sample size
i 1

Population (N units)

Stratum 1 Stratum 2 Stratum k

N1 units N2 units Nk units k
… … … N   Ni
i 1

Sample Sample Sample k

1 2 … … … n   ni
k i 1
n1 units n2 units nk
it

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 2

Procedure of stratified sampling
Divide the population of N units into k strata. Let the ith stratum has N1 , i  1, 2,..., k number of units.

 Strata are constructed such that they are non-overlapping and homogeneous with respect to the
k
characteristic under study such that N
i 1
i  N.

 Draw a sample of size ni from ith ( i  1, 2,..., k ) stratum using SRS (preferably WOR)
independently from each stratum.
 All the sampling units drawn from each stratum will constitute a stratified sample of size
k
n   ni .
i 1

Difference between stratified and cluster sampling schemes

In stratified sampling, the strata are constructed such that they are
 within homogeneous and
 among heterogeneous.

In cluster sampling, the clusters are constructed such that they are
 within heterogeneous and
 among homogeneous.
[Note: We discuss the cluster sampling later.]

Issue in the estimation of parameters in stratified sampling

Divide the population of N units in k strata. Let the i th stratum has Ni , i  1, 2,..., k number of units.

Note that there are k independent samples drawn through SRS of sizes n1 , n2 ,..., nk from each of the

strata. So, one can have k estimators of a parameter based on the sizes n1 , n2 ,..., nk respectively. Our
interest is not to have k different estimators of the parameters but the ultimate goal is to have a single
estimator. In this case, an important issue is how to combine the different sample information together
into one estimator which is good enough to provide the information about the parameter.

We now consider the estimation of population mean and population variance from a stratified sample.

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 3

Estimation of population mean and its variance
Let
Y : characteristic under study,
yij : value of jth unit in ith stratum j = 1,2,…,ni, i = 1,2,...,k,

1 Ni
Yi 
Ni
y
j 1
ij : population mean of ith stratum

1 ni
yi 
ni
y
j 1
ij : sample mean from ith stratum

1 k k

 
Ni
Y N Y
i i  i i : population mean where wi 
wY .
N i 1 i 1 N

Estimation of population mean:

First we discuss the estimation of population mean.
Note that the population mean is defined as the weighted arithmetic mean of stratum means in the case
of stratified sampling where the weights are provided in terms of strata sizes.
1 k
Based on the expression Y   NiYi , one may choose the sample mean
N i 1
1 k
y  ni yi
n i 1
as a possible estimator of Y .

Since the sample in each stratum is drawn by SRS, so

E ( yi )  Yi ,
thus
1 k
E( y )   ni E ( yi )
n i 1
1 k
  ni Yi
n i 1
Y

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 4

and y turns out to be a biased estimator of Y . Based on this, one can modify y so as to obtain an

unbiased estimator of Y . Consider the stratum mean which is defined as the weighted arithmetic mean
of strata sample means with strata sizes as weights given by
1 k
yst   Ni yi .
N i 1
Now
1 k
E ( yst )   Ni E( yi )
N i 1
1 k
  Ni Y i
N i 1
Y

Thus yst is an unbiased estimator of Y .

Variance of yst
k k ni
Var ( yst )   wi2 Var ( yi )    w w Cov( y , y ).
i j i j
i 1 i (  j ) 1 j 1

Since all the samples have been drawn independently from each of the strata by SRSWOR so
Cov( yi , y j )  0, i  j
Ni  ni 2
Var ( yi )  Si
Ni ni
where
1 Ni
Si2  
Ni  1 j 1
(Yij  Y i ) 2 .

Thus
k
Ni  ni 2
Var ( yst )   wi2 Si
i 1 Ni ni
k
 n  S2
  wi2 1  i  i .
i 1  Ni  ni
Observe that Var ( yst ) is small when Si2 is small. This observation suggests how to construct the

strata. If Si2 is small for all i = 1,2,...,k, then Var ( yst ) will also be small. That is why it was

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 5

mentioned earlier that the strata are to be constructed such that they are within homogeneous, i.e., Si2

is small and among heterogeneous.

For example, the units in geographical proximity will tend to be more closer. The consumption pattern
in the households will be similar within a lower income group housing society and within a higher
income group housing society whereas they will differ a lot between the two housing societies based
on income.

Estimate of Variance
Since the samples have been drawn by SRSWOR, so
E ( si2 )  Si2
1 ni
where si2   ( yij  yi )2
ni  1 j 1
 ( y )  N i  ni s 2
and Var i i
N i ni
k
so Var st  i  ( yi )
 ( y )  w2 Var
i 1
k
 N n  2
  wi2  i i  si
i 1  N i ni 

Note: If SRSWR is used instead of SRSWOR for drawing the samples from each stratum, then in this
case
k
yst   wi yi
i 1

E ( yst )  Y
k
 N 1  k
2
Var ( yst )   wi2  i  Si2   wi2 i
i 1  N i ni  i 1 ni
k 2 2
wi si
( y ) 
Var st 
i 1 ni
1 Ni
where  
ni
i
2
(y j 1
ij  yi ) 2 .

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 6

Advantages of stratified sampling
1. Data of known precision may be required for certain parts of the population.
This can be accomplished with a more careful investigation to few strata.
Example: In order to know the direct impact of hike in petrol prices, the population can be
divided into strata like lower income group, middle income group and higher income group.
Obviously, the higher income group is more affected than the lower income group. So more
careful investigation can be made in the higher income group strata.
2. Sampling problems may differ in different parts of the population.
Example: To study the consumption pattern of households, the people living in houses,
hotels, hospitals, prison etc. are to be treated differently.
3. Administrative convenience can be exercised in stratified sampling.
Example: In taking a sample of villages from a big state, it is more administratively
convenient to consider the districts as strata so that the administrative setup at district level may
be used for this purpose. Such administrative convenience and the convenience in organization
of field work are important aspects in national level surveys.
4. Full cross-section of population can be obtained through stratified sampling. It may be possible
in SRS that some large part of the population may remain unrepresented. Stratified sampling
enables one to draw a sample representing different segments of the population to any desired
extent. The desired degree of representation of some specified parts of population is also
possible.
5. Substantial gain in the efficiency is achieved if the strata are formed intelligently.
6. In case of skewed population, use of stratification is of importance since larger weight may
have to be given for the few extremely large units which in turn reduces the sampling
variability.
7. When estimates are required not only for the population but also for the subpopulations, then
the stratified sampling is helpful.
8. When the sampling frame for subpopulations is more easily available than the sampling frame
for whole population, then stratified sampling is helpful.
9. If population is large, then it is convenient to sample separately from the strata rather than the
entire population.
10. The population mean or population total can be estimated with higher precision by suitably
providing the weights to the estimates obtained from each stratum.

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 7

Allocation problem and choice of sample sizes is different strata
Question: How to choose the sample sizes n1 , n2 ,..., nk so that the available resources are used in an

effective way?
There are two aspects of choosing the sample sizes:
(i) Minimize the cost of survey for a specified precision.
(ii) Maximize the precision for a given cost.

Note: The sample size cannot be determined by minimizing both the cost and variability
simultaneously. The cost function is directly proportional to the sample size whereas variability is
inversely proportional to the sample size.

Based on different ideas, some allocation procedures are as follows:

1. Equal allocation
Choose the sample size ni to be the same for all the strata.
Draw samples of equal size from each strata.
Let n be the sample size and k be the number of strata, then
n
ni  for all i  1, 2,..., k .
k

2. Proportional allocation
For fixed k, select ni such that it is proportional to stratum size Ni , i.e.,

ni  Ni
or ni  CNi
where C is the constant of proportionality.
k k

 ni   CNi
i 1 i 1

or n  CN
n
C  .
N
n
Thus ni    Ni .
N
Such allocation arises from the considerations like operational convenience.

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 8

3. Neyman or optimum allocation
This allocation considers the size of strata as well as variability
ni  N i Si
ni  C * Ni Si
where C* is the constant of proportionality.
k k

 ni   C * N i Si
i 1 i 1
k
or n  C *  N i S i
i 1

n
or C *  k

N S
i 1
i i

nN i S i
Thus ni  k
.
N S
i 1
i i

k
This allocation arises when the Var  y st  is minimized subject to the constraint n
i 1
i (prespecified).

There are some limitations of the optimum allocation. The knowledge of Si (i  1, 2,..., k ) is needed to

know ni . If there are more than one characteristics, then they may lead to conflicting allocation.

Choice of sample size based on cost of survey and variability

The cost of survey depends upon the nature of survey. A simple choice of the cost function is
k
C  C0   Ci ni
i 1

where
C : total cost
C0 : overhead cost, e.g., setting up of office, training people etc

Ci : cost per unit in the ith stratum

k

 C n : total cost within sample.

i 1
i i

To find ni under this cost function, consider the Lagrangian function with Lagrangian

multiplier  as

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 9

  Var ( yst )   2 (C  C0 )
k
1 1  k
  wi2    Si2   2  Ci ni
i 1  ni N i  i 1
2 2
k
wS k k
w2 S 2
  i i   2  Ci ni   i i
i 1 ni i 1 i 1 Ni
2
kwS 
   i i   Ci ni   terms independent of ni .
i 1 
 ni 

Thus  is minimum when

wi Si
  Ci ni for all i
ni
1 wi Si
or ni  .
 Ci

How to determine  ?
There are two ways to determine  .
(i) Minimize variability for fixed cost .
(ii) Minimize cost for given variability.
We consider both the cases.

(i) Minimize variability for fixed cost

Let C  C0* be the pre-specified cost which is fixed.
k
So  Ci ni  C0*
i 1
k
wi Si
or C
i 1
i
 Ci
 C0*

 Ci wi Si
or   i 1
.
C0*

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 10

 1 wi Si
Substituting  in the expression for ni  , the optimum ni is obtained as
 Ci

 
wS  C0* 
ni*  i i  k .
Ci  
  Ci wi Si 
 i 1 

The required sample size to estimate Y such that the variance is minimum for given cost C  C0* is
k
n   ni* .
i 1

(ii) Minimize cost for given variability

Let V  V0 be the pre-specified variance. Now determine ni such that
k
1 1  2 2
 n   wi Si  V0
 i
i 1 Ni 
k
w2 S 2 k
wi2 Si2
or  i i  V0  
i 1 ni i 1 Ni
k
 Ci k
wi2 Si2
or 
i 1 wi Si
wi2 Si2  V0  
i 1 Ni
k 2 2
wS
V0   i i
Ni 1 wi Si
or   i 1
(after substituting ni  ).
k
 Ci
w S
i 1
i i Ci

Thus the optimum ni is

 k 
wi Si  
 wi Si Ci 
ni  i 1 .
Ci  k
wi2 Si2 
 0  N
V  
 i 1 i 
So the required sample size to estimate Y such that cost C is minimum for a
k
prespecified variance V0 is n   ni .
i 1

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 11

Sample size under proportional allocation for fixed cost and for fixed variance
k
(i) If cost C  C0 is fixed then C0  C n .
i 1
i i

n
Under proportional allocation, ni  Ni  nwi
N
k
So C0  n  wi Ci
i 1

C0
or n  k
.
 wiCi
i 1

Co wi
Thus ni  .
 wiCi
k
The required sample size to estimate Y in this case is n   ni .
i 1

(ii) If variance = V0 is fixed, then

k
1 1  2 2
 n
i 1  i

 wi Si  V0
Ni 
k
w2 S 2 k
w2 S 2
or  i i  V0   i i
i 1 ni i 1 Ni
k
wi2 Si2 k
wi2 Si2
or 
i 1 nwi
 V0  
i 1 Ni
(using ni  nwi )
k

w S 2 2
i i
or n  i 1
k
wi2 Si2
V0  
i 1 Ni
k

w S 2 2
i i
or ni  wi i 1
.
wi2 Si2k
V0  
i 1 Ni
This is known Bowley’s allocation.

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 12

Variances under different allocations
Now we derive the variance of yst under proportional and optimum allocations.

(i) Proportional allocation

Under proportional allocation
n
ni  Ni
N
and
k
 N n 
Var ( y ) st    i i wi2 Si2
i 1  N i ni 

 n 
k  Ni  Ni  2
 Ni  2
Varprop ( y ) st    N
   Si
Ni Ni   N 
i 1 
n
 N 
N n k
N i Si2
 
Nn i 1 N
N n k
  wi Si2 .
Nn i 1

(ii) Optimum allocation

Under optimum allocation
nN i Si
ni  k

N S
i 1
i i

k
1 1 
Vopt ( yst )      wi2 Si2
i 1  ni Ni 
k
w2 S 2 k w 2 S 2
  i i  i i
i 1 ni i 1 Ni
  k 
k    N i Si   k w2 S 2
   wi Si  i 1
2 2
   i i
i 1   nN i Si   i 1 N i
  
  
k
1 N S  k   k w2 S 2
   . i 2 i   N i Si     i i
i 1  n N  i 1   i 1 N i
2 2
1 k N S  k
w2 S 2 1  k  1 k
   i i    i i    wi Si   wS i i
2
.
n  i 1 N  i 1 N i n  i 1  N i 1

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 13

Comparison of variances of sample mean under SRS with stratified
mean under proportional and optimal allocation:
(a) Proportional allocation:
N n 2
VSRS ( y )  S
Nn
N  n k Ni Si2
Vp r op ( yst )   .
Nn i 1 N

In order to compare VSRS ( y ) and V prop ( yst ), first we attempt to express S 2 as a function of Si2 .

Consider
k Ni
( N  1) S 2    (Y ij  Y )2
i 1 j 1

k Ni 2

  (Yij  Yi )  (Yi  Y ) 
i 1 j 1
k Ni k Ni
  (Yij  Yi )2    (Y  Y ) i
2

i 1 j 1 i 1 j 1
k k
  ( N i  1) Si2   N (Y  Y ) i i
2

i 1 i 1

N 1 2 N 1 2 k k
Ni
S  i Si   (Yi  Y ) 2 .
N i 1 N i 1 N

For simplification, we assume that Ni is large enough to permit the approximation

Ni  1 N 1
 1 and 1.
Ni N
Thus
k
Ni 2 k Ni
S2   Si   (Yi  Y ) 2
i 1 N i 1 N
N n 2 N n k
Ni 2 N  n k
Ni N -n
or
Nn
S 
Nn

i 1 N
Si 
Nn

i 1 N
(Yi  Y ) 2 (Premultiply by
Nn
on both sides)

N n k
VarSRS (Y )  V prop ( y st ) 
Nn
 w (Y  Y )
i 1
i i
2

k
Since  w (Y  Y )
i 1
i i
2
 0,

 Varprop ( yst )  VarSRS ( y ).

Larger gain in the difference is achieved when Yi differs from Y more.

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 14

(b) Optimum allocation
2
1 k  1 k
Vopt ( yst )    wi Si   wS i i
2
.
n  i 1  N i 1

Consider

 N  n  k 2
1  k 
2
1 k 
V prop ( yst )  Vopt ( yst )     wi Si      wi Si    2
wi i 
S
 Nn  i 1   n  i 1  N i 1 
1 k  
2
 k
  i i   i i  

2
w S w S
n  i 1  i 1  
1 k 1
  wi Si2  S 2
n i 1 n
1 k
  wi ( Si  S )2
n i 1
where
k
S   wi Si
i 1

 Varprop ( yst )  Varopt ( yst )  0

or Varopt ( yst )  Varprop ( yst ).

Larger gain in efficiency is achieved when Si differ from S more.

Combining the results in (a) and (b), we have
Varopt ( yst )  Varprop ( yst )  VarSRS ( y )

Estimate of variance and confidence intervals

Under SRSWOR, an unbiased estimate of Si 2 for the ith stratum (i = 1,2,...,k) is

1 ni
si2  
ni  1 j 1
( yij  yi )2 .

In stratified sampling,
k
N i  ni 2
Var ( yst )   wi2 Si .
i 1 N i ni

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 15

So an unbiased estimate of Var ( yst ) is
k
N n
(y ) 
Var st 
i 1
wi2 i i si2
Ni ni
k
wi2 si2 k wi2 si2
 
i 1 ni

i 1 N i
k
wi2 si2 1 k
 
i 1 n
  wi si2
N i 1
i

The second term in this expression represents the reduction due to finite population correction.
The confidence limits of Y can be obtained as
( y )
yst  t Var st

assuming yst is normally distributed and ( y )

Var is well determined so that t can be read from
st

normal distribution tables. If only few degrees of freedom are provided by each stratum, then t values
are obtained from the table of student’s t-distribution.

The distribution of  ( y ) is generally complex. An approximate method of assigning an effective

Var st

 ( y ) is
number of degrees of freedom (ne ) to Var st

2
 k 2
  gi si 
ne   i k1 2 4
gi si

i 1 ni  1

Ni ( Ni  ni )
where gi 
ni
k
and Min(ni  1)  ne   (ni  1)
i 1

assuming yij are normally distributed.

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 16

Modification of optimal allocation
Sometimes in the optimal allocation, the size of subsample exceeds the stratum size. In such a case,
replace ni by Ni

and recompute the rest of ni ' s by the revised allocation.

For example, if n1  N1 , then take the revised ni ' s as

n1  N1
and
(n  N1 ) wi Si
ni  k
; i  2,3,..., k
wS
i2
i i

provided ni  Ni for all i = 2,3,…,k.

Suppose in revised allocation, we find that n2  N2 then the revised allocation would be

n1  N1
n2  N 2
( n  N1  N 2 ) wi Si
ni  k
; i  3, 4,..., k .
 wi Si
i 3

provided ni  Ni for all i  3, 4,..., k .

We continue this process until every ni  Ni .

In such cases, the formula for minimum variance of yst need to be modified as

( * wi Si )2  *
wi Si2
Min Var ( y st )  
n* N
where  *
denotes the summation over the strata in which ni  Ni and n* is the revised total sample

size in the strata.

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 17

Stratified sampling for proportions
If the characteristic under study is qualitative is nature, then its values will fall into one of the two
mutually exclusive complementary classes C and C’ . Ideally, only two strata are needed in which all
the units can be divided depending on whether they belong to C or its complement C’. Thus is difficult
to achieve in practice. So the strata are constructed such that the proportion in C varies as much as
possible among strata.
Let
Ai
Pi  : Proportion of units in C in the ith stratum
Ni
ai
pi  : Proportion of units in C in the sample from the ith stratum
ni
An estimate of population proportion based on the stratified sampling is
k
Ni pi
pst   .
i 1 N
which is based on the indicator variable
1 when jth unit belongs to the ith stratum is in C
Yij  
0 otherwise
and yst  pst .
Ni
Here Si2  Pi Qi
Ni 1

where Qi  1  Pi .
k
Ni  ni 2 2
Also Var ( yst )   wi Si .
i 1 Ni ni

1 k Ni2 ( Ni  ni ) PQ
So Var ( pst )   N  1 ni i .
N 2 i 1 i i

If the finite population correction can be ignored, then

k
PQ
Var ( pst )   wi2 i i
.
i 1 ni

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 18

If proportional allocation is used for ni , then the variance of pst is

N  n 1 k N i2 PQ
Varprop ( pst )   i i
N Nn i 1 N i  1
N n k
  wi PQ
Nn i 1
i i

and its estimate is

 prop ( p )  N  n w pi qi .
k
Var st  i
Nn i 1 ni  1

The best choice of ni such that it minimizes the variance for fixed total sample size is

N i PQ
ni  N i i i

Ni  1
 N i PQ
i i

N i PQ
Thus ni  n k
i i
.
N
i 1
i PQ
i i

k
Similarly, the best choice of ni such that the variance is minimum for fixed cost C  C0   Ci ni is
i 1

PQ
i i
nN i
Ci
ni  k
.
PQ
N
i 1
i
i i
Ci

Estimation of the gain in precision due to stratification

An obvious question crops up that what is the advantage of stratifying a population in the sense that
instead of using SRS, the population is divided into various strata? This is answered by estimating the
variance of estimators of population mean under SRS (without stratification) and stratified sampling by
evaluating
 SRS ( y )  Var
Var (y )
st
.

Var ( y ) st

This gives an idea about the gain in efficiency due to stratification.

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 19

 N n 2
Since VarSRS ( y )  S , so there is a need to express S 2 in terms of Si2 . How to estimate S 2
Nn
based on a stratified sample? Consider
k Ni
( N  1) S 2    (Yij  Y ) 2
i 1 j 1

k Ni 2

    (Yij  Yi )  (Yi  Y ) 
i 1 j 1
k Ni k
   (Yij  Y ) 2   N i (Yi  Y ) 2
i 1 j 1 i 1
k k
  ( N i  1) Si2   N i (Yi  Y ) 2
i 1 i 1
k
 k 
  ( N i  1) Si2  N   wiYi 2  Y 2 .
i 1  i 1 
In order to estimate S 2 , we need to estimates of Si2 , Yi 2 and Y 2 . We consider their estimation one by

one.

(I) For estimate of Si2 , we have

E ( si2 )  Si2

So Sˆi2  si2 .

(II) For estimate of Yi 2 , we know

Var ( yi )  E ( yi 2 )  [ E ( yi )]2
 E ( yi 2 )  Yi 2
or Yi 2  E ( yi 2 )  Var ( yi ).

An unbiased estimate of Yi 2 is

Yˆi 2  yi2  Var

( y )
i

 N n  2
 yi2   i i  si .
 N i ni 

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 20

(III) For the estimation of Y 2 , we know

Var ( yst )  E ( yst2 )  [ E ( yst )]2

 E ( yst2 )  Y 2
 Y 2  E ( yst2 )  Var ( yst )

So an estimate of Y 2 is

Yˆ 2  yst2  Var
( y )
st
k
 N n  2 2
 yst2    i i wi si .
i 1  N i ni 

Substituting these estimates in the expression (n  1) S 2 as follows, the estimate of S 2 is obtained as

k
 k 
( N  1) S 2   ( N i  1) Si2  N   wi Yi 2  Y 2 
i 1  i 1 
1 k
N  k

as Sˆ 2   ( N i  1) Sˆi2    w iYˆi 2  Yˆ 2 
N  1 i 1 N  1  i 1 
1  k 2 N  k  2  N i  ni  2    2 k N i  ni 2 2  
 
N  1  i 1
 N  1 s  
 N 1   w  yi    si     yst   wi si  
 i 1   N i ni    
i i i
 i 1 N i ni  
1  k 2 N  k k
N  ni 2 
  
N  1  i 1
 N i  1 si    
 N  1  i 1
wi ( yi  y st ) 2
  wi (1  wi ) i si  .
i 1 N i ni 
Thus

 SRS ( y )  N  n Sˆ 2
Var
Nn
N n  k 2 N ( N  n)  k k
N n 
  
N ( N  1)n  i 1
( N i  1) si    i i
 nN ( N  1)  i 1
w ( y  y st ) 2
  wi (1  wi ) i i si2 
i 1 N i ni 
and
N i  ni 2 2
k
( y ) 
Var st 
i 1 N i ni
wi si .

Substituting these expressions in

 SRS ( y )  Var
Var (y )
st
,

Var ( y ) st

the gain in efficiency due to stratification can be obtained.

If any other particular allocation is used, then substituting the appropriate ni under that allocation,

such gain can be estimated.

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 21

Interpenetrating subsampling
Suppose a sample consists of two or more subsamples which are drawn according to the same
sampling scheme. The samples are such that each subsample yields an estimate of parameter. Such
subsamples are called interpenetrating subsamples.

The subsamples need not necessarily be independent. The assumption of independent subsamples
helps in obtaining an unbiased estimate of the variance of the composite estimator. This is even
helpful if the sample design is complicated and the expression for variance of the composite estimator
is complex.

Let there be g independent interpenetrating subsamples and t1 , t2 ,..., tg be g unbiased estimators of

parameter  where t j ( j  1, 2,..., g ) is based on jth interpenetrating subsample.

Then an unbiased estimator of  is given by

1 g
ˆ   t j  t , say.
g j 1
Then
E (ˆ)  E ( t )  
and
1 g
 (ˆ)  Var
Var  (t ) 
 (t j  t )2 .
g ( g  1) j 1
Note that

1  g 
 (t ) 
E Var E   (t j   ) 2  g ( t   ) 2 
  g ( g  1)
 j 1 
1  g 
   Var (t j )  g Var ( t ) 
g ( g  1)  j 1 
1
 ( g 2  g )Var ( t )  Var ( t )
g ( g  1)
If the distribution of each estimator tj is symmetric about  , then the confidence interval of  can be
obtained by

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 22

 g 1
1
P  Min(t1 , t2 ,..., t g )    Max(t1 , t2 ,..., t g )   1    .
2

Implementation of interpenetrating subsamples in stratified sampling

Consider the set up of stratified sampling. Suppose that each stratum provides an independent
interpenetrating subsample. So based on each stratum, there are L independent interpenetrating
subsamples drawn according to the same sampling scheme.

Let Yˆij (tot ) be an unbiased estimator of the total of jth stratum based on the ith subsample ,

i = 1,2,...,L; j = 1,2,...,k.

An unbiased estimator of the jth stratum total is given by

1 J
Yˆj ( tot )   Yˆij (tot )
L i 1

and an unbiased estimator of the variance of Yˆj (tot ) is given by

1 L
 (Yˆ ) 
Var j ( tot ) 
L( L  1) i 1
(Yˆij (tot )  Yˆj (tot ) )2 .

Thus an unbiased estimator of population total Ytot is

k
1 L k
Yˆtot   Yˆj ( tot )   Yˆij (tot )
j 1 k i 1 j 1
And an unbiased estimator of its variance is given by
k
Var tot   (Yˆj (tot ) )
 (Yˆ )  Var
j 1

1 L k
 
L( L  1) i 1 j 1
(Yˆij ( tot )  Yˆj ( tot ) ) 2 .

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 23

Post Stratifications
Sometimes the stratum to which a unit belongs may be known after the field survey only. For
example, the age of persons, their educational qualifications etc. can not be known in advance. In such
cases, we adopt the post stratification procedure to increase the precision of the estimates.
Note: This topic is to be read after the next module on ratio method of estimation. Since it is related to
the startification, so it is given here.

In post stratification,
 draw a sample by simple random sampling from the population and carry out the survey.
 After the completion of survey, stratify the sampling units to increase the precision of the
estimates.
Assume that the stratum size Ni is fairly accurately known. Let

mi : number of sampling units from ith stratum, i = 1,2,...,k.

k

 m  n.
i 1
i

Note that mi is a random variable (and that is why we are not using the symbol ni as earlier).

Assume n is large enough or the stratification is such that the probability that some mi  0 is

negligibly small. In case, mi  0 for some strata, two or more strata can be combined to make the

sample size non-zero before evaluating the final estimates.

A post stratified estimator of population mean Y is

1 k
y post   Ni yi .
N i 1
Now

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 24

 1  k 
E   Ni E ( yi m1 , m2 ,..., mk ) 
E ( y post ) 
N  i 1 
1  k

 E   NiYi 
N  i 1 
Y

Var ( y post )  E Var ( y post m1 , m2 ,..., mk )   Var  E ( y post m1 , m2 ,..., mk ) 

 k  1 1  
 E   wi2    Si2   Var (Y )
 i 1  mi N i  
k   1   1 
  wi2  E       Si2 (Since Var (Y )  0).
i 1   mi   N i  

 1  1
To find E    , proceed as follows :
 mi  Ni
Consider the estimate of ratio based on ratio method of estimation as
n N

y
 yj Y
Y j

Rˆ  
j 1 j 1
n
, R  N
.
 xj Xj
x X
j 1 j 1

We know that
N  n RS X2  S XY
E ( Rˆ )  R  . .
Nn X2
1 if j th unit belongs to i th stratum
Let xj  
0 otherwise

and
y j  1 for all j = 1,2,...,N.

Then R, Rˆ and S x2 reduces to

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 25

 n

y j
n
Rˆ  j 1
n

x
ni
j
j 1
N

Y j 1
j
N
R N

X
Ni
j
j 1

1 N 2 2 1  N i2  1  Ni 2 
S    X j  NX    Ni  N 2    Ni 
2

N  1  j 1  N 1  N  N 1 
x
N 
1 N  1  N N
S xy    X jY j  NXY    N i  i 2   0.
N  1  j 1  N 1  N 

Using these values in E ( Rˆ )  R, we have

 n  N N ( N  n)( N  N i )
E ( Rˆ )  R  E     .
 ni  N i nN i2 ( N  1)
Thus
1  1 N N ( N  n)( N  N i ) 1
E    
 ni  N i nN i n 2 N i2 ( N  1) Ni
( N  n) N  N 1
 1   .
n( N  1) N i  N i n n 

Replacing mi in place of ni , we obtain

 1  1 ( N  n) N  N 1
E   1   
 mi  Ni n( N  1) Ni  Ni n n 
Now substitute this in the expression of Var ( y post ) as

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 26

 k   1  1 2
Var ( y post )   wi2  E     Si
i 1   mi  Ni 
k  N n N  N 1 
  wi2 Si2  . 1   
i 1  ( N  1)n Ni  nNi n  
N n k 2 2  1  1 1 
 
n( N  1) i 1
wi Si  1   
 wi  nwi n  
N n k  1
 
n ( N  1) i 1
2
wi Si2  n  1  
 wi 
N n k
 
n ( N  1) i 1
2
(nwi  1  wi ) Si2

N n k N n k
  i i n2 ( N  1) 
n( N  1) i 1
w S 2

i 1
(1  wi ) Si2 .

Assuming N  1  N .
N n n N n n
V ( y post ) 
Nn i 1
 i i n2 N 
w S 2

i 1
(1  wi ) Si2

N n n

2 
 V prop ( yst )  (1  wi ) Si2 .
Nn i 1
The second term is the contribution in the variance of y post due to mi ' s not being proportionately

distributed.

If Si2  S w2 , say for all i, then the last term in the expression is

N n k N n 2 k

2   w  1)
(1  wi ) Sw2  Sw (k  1) (Since
Nn2
i
Nn i 1 i 1

 k  1  N  n  2
   Sw
 n  Nn 
k 1
 Var ( yst ).
n
n
The increase in the variance over Varprop ( yst ) is small if the average sample size n  per stratum is
2
reasonably large.

Thus a post stratification with a large sample produces an estimator which is almost as precise as an
estimator in the stratified sampling with proportional allocation.

Sampling Theory| Chapter 4 | Stratified Sampling | Shalabh, IIT Kanpur Page 27