CONCEPT OF STRATIFICATION

There may often be factors which divide the population into several sub-populations or groups and
we may expect the measurements of interest to vary among the different sub-populations. This has
to be accounted for when we select a sample from the population in order that we obtain a sample
that is representative of the population. This is achieved by stratified sampling. The sub-
populations making up the whole population are called strata. The strata are non-overlapping and
homogeneous within themselves. Sampling frames are constructed for each of these strata and the
sampling is performed independently within each stratum. If a random sampling strategy is
followed to select the sample within each stratum, the whole procedure is called stratified random
sampling.

DEFINITION: Stratified random sampling is a sampling plan in which we divided the population
into several non-overlapping strata and select a random sample from each stratum in such a way
that units within the strata are homogeneous but between strata they are heterogeneous.
Strata are generally formed on the basis of some known characteristics of the population, which is
believed to be related to the variable of interest. Such stratification is usually done in such a way as
to reduce the variability of the strata estimates. This is generally achieved by forming the strata
such that they are homogeneous within themselves with respect to a suitably chosen auxiliary
variable called stratification variable or stratification factor.

Sampling procedure within stratum

Sampling within each stratum may be made proportionately or disproportionately. A
proportionate stratified sample is achieved if the sampling fraction is the same for every stratum.
Under this design, the sample size in a stratum is proportional to the size of the population in the
stratum. If, on the other hand, there are variable sampling fractions, the sample is a
disproportionate stratified sample. We will discuss these in turn.

The justification of adopting a stratified sampling is that if we know nothing about the structure of
the population apart from its size, we cannot do better than take a simple random sample. However,
it is an extreme situation that we know nothing about the population other than its size. Most often,
we know that a population consists of individuals who can be classified by their characteristics such
as religion, socio- economic status, income, expenditure, occupation, race and level of education. It
the researcher, for instance, studying the living and working condition of the people, had wished to
ensure that different types of area (e. g. City corporation, Municipal area, urban, semi- urban, rural
etc.) were adequately represented in the sample, he could have stratified the population by area
type. Here area type is the stratification variable, which is clearly related to the study variables. In
studying the television viewing habit among the university student, academic performance score
(high, medium, low) or the place of residence (urban rural) may serve as stratification variables,
since each of these variables is believed to be related to the television viewing habit of the students.

Features of stratification process

Stratification process, as we will see, does not imply any departure from the principal of
randomness. The process possesses the following salient features:
▪ The entire population is divided into several distinct sub-populations, called strata.
▪ Within each stratum, a separate and independent sample is selected.
1
 ▪ For each individual stratum, stratum mean, proportion, variance and other statistics are
computed.
▪ These estimates are the properly weighted to form a combined estimate for the entire
population.

Principles of stratification
The process of stratification involves dividing the population into several sub- population, which
we call strata. In forming such strata, few principles should be followed to take full advantage of
stratified sampling.
These are:
▪ The strata should be non- overlapping and exhaustive so that they together comprise the
whole population. Thus several administrative divisions of the country, ecological zones.
rural-urban residence and the like may be thought of as different strata. The strata should
be made as homogeneous as possible, ensuring greater similarity within the strata than
between the strata.
▪ Strata are to be formed on the basis of some known characteristics of the population,
which are believed to have some relationship with the subject of inquiry and variables of
interest.
▪ When stratification with respect to the characteristics under study becomes difficult for
practical reasons, administrative convenience may be considered as the basis for forming
the strata.
▪ With a view to improve the sampling design, strata should be formed on the basis of
natural characteristics as far as possible.
▪ Past data, intuition, expert judgment or preliminary findings from pilot surveys may also
be used to set-up the strata. This, however, requires that we have prior knowledge of the
nature of the population from which we are sampling.

Examples of stratified sampling

We cite a few examples below where stratified sampling procedure is warranted:
EXAMPLE 4.1: In all national surveys of Bangladesh, the entire country is divided in to two broad
areas: urban and rural. The 1999-2000 Bangladesh Demographic and Health Survey considered
these two areas as two separate strata in collection their health and demographic data. The major
objectives of the survey were to assess the overall demographic situations in Bangladesh in terms of
fertility, mortality, awareness, approval and use of family planning, nutrition level, maternal and
child health. The survey adopted a stratified sampling design. What led them to adopt this design?
We note down the following points in deciding the stated sample design:
1. The population of households falls into two natural groupings: an urban area and a rural area. We
may thus treat these areas as two separate strata simply for administrative convenience in selecting
the samples and carrying out the fieldwork.
2. It is also expected that each of the two groups of households should have similar behavior
patterns among residents within the group. And as a result, we expect to see relatively small
variability in the measurements of the survey variables among households within a group.
3. The survey had a wish to generate estimates on every aspect of the objectives (say of the
proportion of children malnourished, percentage of women wanting no more children) for each area
separately.
2
The above observations are precisely the arguments that dictate the use of stratified random
sampling for the problems in hand.
EXAMPLE 4.2: Suppose we wish to estimate the number of students in secondary school in
Dhaka city. It is known that majority of the schools are small and middle sized in terms of the
number of students and that there are only a few large schools. It is also learnt that these large
schools account for a substantial proportion of the total number of students. If a sample random
sample technique is employed to estimate the total number of students in the city, there is a good
chance that our sample may contain either too many or too few of the very large schools. If this is
the case, the sample may not adequately represent the population.
A convenient solution to the problem is to stratify the sampling units (schools here) prior to
sampling into three group on the basis of school size (i.e. small, medium, large), and then select,
using simple random sampling technique, certain number of schools from each of the three groups.
An estimate of the total number of students can then be obtained from the combined results of the
three strata.
EXAMPLE 4.3: An iodine deficiency disorder (IDD) survey was conducted in Bangladesh in 1999
by Dhaka University with financial assistance from UNICEF with the primary objective of
estimating the Prevalence rate of goiter in the country. Ecologically, the country is divided into
three different zones: Hilly, Flood-prone and the plain. It was intended to conduct the survey on the
basis of these zones considering these zones as three different natural strata. This division was in
consonance with the general subject knowledge that iodine deficiency disorders are more prevalent
in hilly and flood -prone areas than elsewhere because of leaching away of iodine from the soil of
these areas by the heavy rain and flood waters.
Note that the example 4.3 is an ideal one that demonstrates the rationale behind the use of stratified
sampling technique. First the three zones were naturally created. Second, expert judgment and
preliminary findings suggest that goiter prevalence are different in these zones implying that the
zones are internally homogeneous but externally heterogeneous. Third, different sampling plan
could be employed in different zones because of accessibility problem. Fourth one of the main
objectives of the survey was to provide the funding agency with the zonal estimates. And finally,
supervision and monitoring of the fieldwork locally, was considered to be of immense convenience
from administrative point of view.

Steps involved in stratified sampling

In earring out stratified sampling, some important points need to be carefully considered
(Murthy,1977). These include, among others,
(a) Choice of stratification variable,
(b) Formation of strata,
(c) Number of strata,
(d) Sampling within the strata and
(e) Allocating sample to strata

Choice of stratification variable: The stratification variable provides supplementary information

on the basis of which a study population is partitioned into several sub- populations or strata. This
information should be clearly related to the objective of the survey. There is no gain in precision
from stratifying by a factor unrelated to the survey. In forming the strata care must be taken to
make the strata as much homogeneous as possible within themselves with respect to the
3
stratification variable. The choice of the stratification variable will however, depend on the
sampling design used within the strata.

Formation of strata: Stratification with uniform sampling fraction, almost always increases
precision. In general, the effect in the gain in precision from stratified sampling with uniform
sampling fraction depends on the variation between stratum averages. The greater the variation
between the stratum averages, the greater the gain in precision. In extreme cases, when the stratum
averages are exactly equal, there may actually be a loss in precision due to the stratification. But
this can only happen when the sampling fraction is appreciable (Stuart, 1984:44). It is thus
important to note that our aim when forming the strata for sampling with a uniform sampling
fraction must be to make the strata averages differ as much as possible. What about the case when
we have variable sampling fractions? Use of variable sampling fractions may increase or decrease
precision as compared with a uniform sampling fraction. Except in extreme cases, precision will be
increased if larger sampling fraction are used in the more variable strata, and decreased if they are
used in the less variable strata. A mathematical rule seems to work well in this regard: to maximize
the precision in each stratum sampling fraction should be made proportional to the square root of
the variance in that stratum (Stuart, 1984:41). Even if it becomes difficult to follow the above rule,
nevertheless, it is easy to realize that if we divide a population into strata with the aim of
maximizing strata variations with respect to their averages, we automatically minimize the total
variation within the strata, since the population contains a fixed number of variables. This also
takes into account the case of variable sampling fraction to maximize the precision.

Number of strata: The decision on the number of strata to be formed from any single stratification
variable is of crucial importance. The choice however is to be made keeping in view that it takes
into account such factors as the (i) effect of increasing the number of strata (ii) in the reduction of
variance and (ii) cost of the survey.
While deciding on this, it should be borne in mind that very small strata contribute little to the gains
from stratification. The formation of only a few large strata will typically yield most of the possible
gains from a variable. Further subdivisions of these would result in only small additional gains.
Sampling within the strata: The different sub-groups of a population may markedly vary in terms
of its physical distribution and peculiarity. This calls for adopting different sampling designs in
different strata to account for the degree of variability in the population. The sample selection
within the stratum may follow simple random sampling, sampling with probability proportional to
size (PPS), or systematic sampling especially at the primary and other stages down to the
penultimate stage.
Allocating sample to strata: The essence of stratification is the classification of the population
into sub-populations or strata and then the selection of the separate samples from each of the strata.
This should not, however be done entirely arbitrarily if precision of the estimates is of concern. We
must therefore frame some rule or strategy so as to allocate the sample of a fixed size to different
strata. These are what we call principles of allocation. It is however of importance to note that if
one that if one has to balance the requirements of overall precision on the one side and stratum
level precision on the other, a variety of compromise solution are possible.

Advantages of stratified sampling

4
  Adequate representation of minority subgroups of interest can be ensured by stratification
and varying sampling fraction between strata as required.
 Stratification decreases the variance of estimates; hence increase the precision of estimates.
 Since each stratum is treated as an independent population, different sampling approaches
can be applied to different strata

Disadvantages of stratified sampling

 Sampling frame of entire population has to be prepared separately for each stratum
 Stratification is often difficult job. When examining multiple criteria, stratifying variables
may be related to some, but not to others, further complicating the design, and potentially
reducing the utility of the strata.
 in some cases (such as designs with a large number of strata, or those with a specified
minimum sample size per group), stratified sampling can potentially require a larger sample
than would other methods

ESTIMATORS AND THEIR PROPERTIES

Consider a population consisting of N units. For stratified sampling, the population is first divided
into L sub-populations. These sub-populations are non-overlapping and together they comprise the
whole of the population so that N1+N2+…..+NL=N. These sub-population are treated as strata.
When the strata have been identified, we proceed to draw sample of predetermined size from each
stratum. If n1, n2, ….., nL denote to the sizes of the samples to be drawn from strata of sizes N1, N2,
...., NL respectively, then n1+n2+…..+nL =n . If now a random sampling procedure is followed in
selecting samples from each stratum, the whole procedure is described as stratified random
sampling.
The notation used for the stratified sampling may be kept exactly analogous to the simple random
sampling. We use the symbol “h” to denote the stratum and “i” the unit within the stratum. Thus,
yhi is used to denote the value of the ith unit in the hth stratum.

Notations and formulae

Let the population of N units be divided into a number of 𝐿 strata with strata sizes 𝑁ℎ (ℎ =
𝐿
1, 2, ……𝐿) where 𝑁 = ∑ℎ=1 𝑁ℎ .
Suppose a simple random sample of size 𝑛ℎ be drawn from hth stratum.
Let 𝑦ℎℎ denotes ith observation in hth stratum.
Following table presents useful notations for hth stratum.
Population Sample
Size 𝑁ℎ 𝑛ℎ
𝑛ℎ
Total 𝑁 ℎ

𝑌ℎ = 𝑦ℎℎ 𝑦ℎ = 𝑦ℎℎ
𝑖=1 𝑖=1
Mean 𝑌ℎ 𝑦ℎ
𝑌ℎ = 𝑦ℎ =
𝑁ℎ 𝑛ℎ

5
 𝑛
Variance 𝑁
∑𝑖=1
ℎ
𝑦ℎℎ ― 𝑌ℎ
2
∑𝑖=1
ℎ
(𝑦ℎℎ ― 𝑦ℎ )2
𝑆2ℎ = 𝑠2ℎ =
𝑁ℎ ― 1 𝑛ℎ ― 1
Stratum weight 𝑁ℎ
𝑊ℎ =
𝑁
Sampling fraction 𝑛ℎ
𝑓ℎ =
𝑁ℎ

Population mean in stratified sampling

In stratified sampling, the population mean 𝑌 can be expressed as
𝐿

𝑌= 𝑊ℎ 𝑌ℎ ,
ℎ=1
where ,
𝑁ℎ
𝑊ℎ =
𝑁
𝑁ℎ
𝑌ℎ ∑𝑖=1 𝑦ℎℎ
𝑌ℎ = = .
𝑁ℎ 𝑁ℎ

Proof:
Let the population of N units be divided into a number of 𝐿 strata with sizes 𝑁ℎ (ℎ = 1, 2, ……𝐿)
𝐿
where 𝑁 = ∑ℎ=1 𝑁ℎ .
Let a random sample of size 𝑛ℎ be drawn from hth stratum using simple random sampling without
replacement.
Let
𝑌=population mean,
𝑌ℎ =population mean for hth stratum,
𝑦ℎ =sample mean for hth straum and
𝑌 =population total.

𝐿 𝑁
Then population total 𝑌 = ∑ℎ=1 𝑌ℎ where 𝑌ℎ = ∑𝑖=1
ℎ
𝑦ℎℎ .
𝑌
So, population mean 𝑌 =
𝑁
𝐿
∑ℎ=1 𝑌ℎ
=
𝑁
𝐿
1 𝑌ℎ 𝑁ℎ
=
𝑁 𝑁ℎ
ℎ=1
𝐿
𝑁ℎ
= 𝑌
𝑁 ℎ
ℎ=1
𝐿

= 𝑊ℎ 𝑌ℎ
ℎ=1
where
𝑁ℎ
𝑊ℎ = 𝑁
is called the stratum weight or population proportion for hth stratum.

6
Stratified sample mean (𝒚𝒔𝒔 )
For the population mean per unit, the estimator used in stratified sampling is 𝑦𝑠𝑠 , known as
stratified sample mean and is given by
𝐿
∑ℎ=1 𝑁ℎ 𝑦ℎ
𝑦𝑠𝑠 =
𝑁
𝐿

= 𝑊ℎ 𝑦ℎ
ℎ=1
where
𝑁ℎ
𝑊ℎ =
𝑁
and
𝑁1 + 𝑁2 + … 𝑁𝐿
That is, stratified sample mean 𝑦𝑠𝑠 is the weighted average of stratum means.

Note:
Stratified sample mean 𝑦𝑠𝑠 is not in general the same as the ordinary sample mean 𝑦
where
𝐿 𝐿
∑ℎ=1 𝑛ℎ 𝑦ℎ
𝑦= = 𝑤ℎ 𝑦ℎ
𝑛
ℎ=1
where
𝑛ℎ
𝑤ℎ = .
𝑛

It is evident that 𝑦 coincides with 𝑦𝑠𝑠 provided that in every stratum

𝑤ℎ = 𝑊ℎ
𝑛ℎ 𝑁ℎ
=> =
𝑛 𝑁
𝑛ℎ 𝑛
=> =
𝑁ℎ 𝑁
=> 𝑓ℎ = 𝑓.

That is 𝑦 coincides with 𝑦𝑠𝑠 if the sampling fraction is the same in all strata. This stratification is
described as stratification with proportional allocation of the 𝑛ℎ .

Estimation in stratified sampling

Theorem:
Stratified sample mean 𝑦𝑠𝑠 is an unbiased estimator of population mean.

Proof:
Let the population of N units be divided into a number of 𝐿 strata with sizes 𝑁ℎ (ℎ = 1, 2, ……𝐿)
𝐿
where 𝑁 = ∑ℎ=1 𝑁ℎ .

7
Let a random sample of size 𝑛ℎ be drawn from hth stratum using simple random sampling without
replacement.
Let
𝑌 =population mean,
𝑌ℎ =population mean for hth stratum,
𝑦ℎ =sample mean for hthstraum and
𝑌 =population total.

Stratified sample mean 𝑦𝑠𝑠 is given by

𝑦𝑠𝑠 = 𝑊ℎ 𝑦ℎ
ℎ=1
where
𝑁ℎ
𝑊ℎ =
𝑁
Now,
𝐿

𝐸(𝑦𝑠𝑠 ) = 𝐸 𝑊ℎ 𝑦ℎ
ℎ=1
𝐿

= 𝑊ℎ 𝐸(𝑦ℎ )
ℎ=1
𝐿

= 𝑊ℎ 𝑌ℎ [ ∵ 𝐸(𝑦ℎ ) = 𝑌ℎ , 𝑠𝑖𝑛𝑐𝑒 𝑡ℎ𝑒 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑜𝑟 𝑦ℎ 𝑖𝑠 𝑢𝑛𝑏𝑖𝑠𝑒𝑑 𝑓𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑠𝑡𝑟𝑎𝑡𝑢𝑚]

ℎ=1
𝐿
𝑁ℎ
= 𝑌
𝑁 ℎ
ℎ=1
𝐿
1
= 𝑁ℎ 𝑌ℎ
𝑁
ℎ=1
𝐿
1 𝑌ℎ
= 𝑁ℎ
𝑁 𝑁ℎ
ℎ=1
𝐿
1
= 𝑌ℎ
𝑁
ℎ=1
𝑌
=
𝑁
= 𝑌. (proved)

Theorem:
In stratified sampling, the ordinary sample mean 𝑦 is not an unbiased estimator of population mean
𝑌.

Proof:

8
 𝐿
𝑛ℎ 𝑦ℎ
𝐸(𝑦) = 𝐸
𝑛
ℎ=1
𝐿
𝑛ℎ
= 𝐸(𝑦ℎ )
𝑛
ℎ=1

𝐿
𝑛ℎ
= 𝑌
𝑛 ℎ
ℎ=1
≠ 𝑌. (proved)
𝑛ℎ 𝑁ℎ
So, 𝑦 is not an unbiased estimator of population mean𝑌 unless 𝑛
= 𝑁
for all h=1, 2, ….L.

Theorem:
Variance of stratified sample mean is given in one of three following forms:

𝐿
𝑆2ℎ 𝑛ℎ
𝑎. 𝑉(𝑦𝑠𝑠 ) = 𝑊2ℎ (1 ― 𝑓ℎ ) , 𝑤ℎ𝑒𝑟𝑒 𝑓ℎ =
𝑛ℎ 𝑁ℎ
ℎ=1
𝐿
1 𝑆2ℎ
𝑏.𝑉(𝑦𝑠𝑠 ) = 2 𝑁ℎ (𝑁ℎ ― 𝑛ℎ )
𝑁 𝑛ℎ
ℎ=1
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ
𝑐.𝑉(𝑦𝑠𝑠 ) = ―
𝑛ℎ 𝑁ℎ
ℎ=1 ℎ=1
Proof:
Let the population of N units be divided into a number of 𝐿 strata with sizes 𝑁ℎ (ℎ = 1, 2, ……𝐿)
𝐿
where 𝑁 = ∑ℎ=1 𝑁ℎ .
Let a random sample of size 𝑛ℎ be drawn from hth stratum using simple random sampling without
replacement.
Let
𝑌 =population mean,
𝑌ℎ =population mean for hth stratum,
𝑦ℎ =sample mean for hthstraum and
𝑌 =population total.

a.
𝐿

𝑉(𝑦𝑠𝑠 ) = 𝑉 𝑊ℎ 𝑦ℎ
ℎ=1
𝐿 𝐿

= 𝑊2ℎ 𝑉(𝑦ℎ ) +2 𝑊ℎ 𝑊𝑗 𝐶𝑜𝑣(𝑦ℎ , 𝑦𝑗 )

ℎ=1 ℎ=1 𝑗>ℎ
But since samples are drawn independently in different strata, all covariance terms vanish.
Therefore,

9
 𝐿

𝑉(𝑦𝑠𝑠 ) = 𝑊2ℎ 𝑉(𝑦ℎ )

ℎ=1
𝐿
𝑆2ℎ
= 𝑊2ℎ (1 ― 𝑓ℎ )
𝑛ℎ
ℎ=1

b.
We have
𝐿
𝑆2ℎ
𝑉(𝑦𝑠𝑠 ) = 𝑊2ℎ (1 ― 𝑓ℎ )
𝑛ℎ
ℎ=1

𝐿
𝑁2ℎ 𝑁ℎ ― 𝑛ℎ 𝑆2ℎ
=
𝑁2 𝑁ℎ 𝑛ℎ
ℎ=1
𝐿
1 𝑆2ℎ
= 2 𝑁ℎ (𝑁ℎ ― 𝑛ℎ )
𝑁 𝑛ℎ
ℎ=1
c.
We have
𝐿
𝑆2ℎ
𝑉(𝑦𝑠𝑠 ) = 𝑊2ℎ (1 ― 𝑓ℎ )
𝑛ℎ
ℎ=1
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ 𝑓ℎ
= ―
𝑛ℎ 𝑛ℎ
ℎ=1 ℎ=1
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ 𝑛ℎ
= ― ∵ 𝑓ℎ =
𝑛ℎ 𝑁ℎ 𝑁ℎ
ℎ=1 ℎ=1
(proved)

Theorem:
𝑌𝑠𝑠 = 𝑁𝑦𝑠𝑠 is an unbiased estimator of population total 𝑌.

Proof:
Let the population of N units be divided into a number of 𝐿 strata with sizes 𝑁ℎ (ℎ = 1, 2, ……𝐿)
𝐿
where 𝑁 = ∑ℎ=1 𝑁ℎ .
Let a random sample of size 𝑛ℎ be drawn from hth stratum using simple random sampling without
replacement.
Let
𝑌 =population mean,
𝑌ℎ =population mean for hth stratum,
𝑦ℎ =sample mean for hthstraum and
𝑌 =population total.

First, prove that 𝐸(𝑦𝑠𝑠 ) = 𝑌.

10
𝐸(𝑌𝑠𝑠 ) = 𝐸(𝑁𝑦𝑠𝑠 )
= 𝑁𝐸(𝑦𝑠𝑠 )
= 𝑁𝑌[ ∵ 𝐸(𝑦𝑠𝑠 ) = 𝑌]
𝑌
=𝑁
𝑁
= 𝑌.
(proved)

Theorem:
Variance of stratified sample total is given as:

𝐿
𝑆2ℎ
𝑉(𝑌𝑠𝑠 ) = 𝑁ℎ (𝑁ℎ ― 𝑛ℎ )
𝑛ℎ
ℎ=1

Proof:
Let the population of N units be divided into a number of 𝐿 strata with sizes 𝑁ℎ (ℎ = 1, 2, ……𝐿)
𝐿
where 𝑁 = ∑ℎ=1 𝑁ℎ .
Let a random sample of size 𝑛ℎ be drawn from hth stratum using simple random sampling without
replacement.
Let
𝑌 =population mean,
𝑌ℎ =population mean for hth stratum,
𝑦ℎ =sample mean for hth straum and
𝑌 =population total.

First prove that

𝐿
1 𝑆2ℎ
𝑉(𝑦𝑠𝑠 ) = 2 𝑁ℎ (𝑁ℎ ― 𝑛ℎ )
𝑁 𝑛ℎ
ℎ=1
We have

𝑉(𝑌𝑠𝑠 ) = 𝑉(𝑁𝑦𝑠𝑠 )
= 𝑁 2 𝑉(𝑦𝑠𝑠 )
𝐿
1 𝑆2ℎ
= 𝑁2 2 𝑁ℎ (𝑁ℎ ― 𝑛ℎ )
𝑁 𝑛ℎ
ℎ=1
𝐿
𝑆2ℎ
= 𝑁ℎ (𝑁ℎ ― 𝑛ℎ )
𝑛ℎ
ℎ=1
(proved)

11
Estimator of 𝑽(𝒚𝒔𝒔 )

Theorem:
If SRS is done without replacement within each stratum, an unbiased estimator of 𝑆2ℎ is
𝑛ℎ
1
𝑠2ℎ = (𝑦ℎℎ ― 𝑦ℎ )2 .
(𝑛ℎ ― 1)
𝑖=1

Now replacing 𝑆2ℎ by its estimator 𝑠2ℎ , an unbiased estimator of 𝑉(𝑦𝑠𝑠 ) is given in one of three
following forms:

𝐿
𝑠2ℎ 𝑛ℎ
𝑎. 𝑣(𝑦𝑠𝑠 ) = 𝑠 2 (𝑦𝑠𝑠 ) = 𝑊2ℎ (1 ― 𝑓ℎ ) , 𝑤ℎ𝑒𝑟𝑒 𝑓ℎ =
𝑛ℎ 𝑁ℎ
ℎ=1
𝐿
1 𝑠2ℎ
𝑏.𝑣(𝑦𝑠𝑠 ) = 𝑠 2 (𝑦𝑠𝑠 ) = 2 𝑁ℎ (𝑁ℎ ― 𝑛ℎ )
𝑁 𝑛ℎ
ℎ=1
𝐿 𝐿
𝑊2ℎ 𝑠2ℎ 𝑊2ℎ 𝑠2ℎ
𝑐.𝑣(𝑦𝑠𝑠 ) = 𝑠 2 (𝑦𝑠𝑠 ) = ―
𝑛ℎ 𝑁ℎ
ℎ=1 ℎ=1

Note: The last formula i.e.

𝐿 𝑊ℎ 𝑠ℎ 2 2
𝐿 𝑊ℎ2 𝑠ℎ2
𝑠 2 (𝑦𝑠𝑠 ) = ∑ℎ=1 ― ∑ℎ=1 𝑁ℎ
is usually used for computing purpose.
𝑛ℎ
Hence the estimator of standard error of 𝑦𝑠𝑠 is given by
2 2 2 2
𝐿 𝑊ℎ 𝑠ℎ 𝐿 𝑊ℎ 𝑠ℎ
𝑠(𝑦𝑠𝑠 ) = 𝑣(𝑦𝑠𝑠 ) = ∑ℎ=1 ― ∑ℎ=1 which is used for computing purpose.
𝑛ℎ 𝑁ℎ

Estimator of 𝑽(𝒀𝒔𝒔 )
Similarly, replacing 𝑆2ℎ by its estimator 𝑠2ℎ , an unbiased estimator of 𝑉(𝑌𝑠𝑠 ) is given by
𝐿
𝑠2ℎ
𝑣(𝑌𝑠𝑠 ) = 𝑠 2 (𝑌𝑠𝑠 ) = 𝑁ℎ (𝑁ℎ ― 𝑛ℎ )
𝑛ℎ
ℎ=1
Hence the estimator of standard error of 𝑉(𝑌𝑠𝑠 ) is given by
𝐿
𝑠2ℎ
𝑠(𝑌𝑠𝑠 ) = 𝑣(𝑌𝑠𝑠 ) = 𝑁ℎ (𝑁ℎ ― 𝑛ℎ ) .
𝑛ℎ
ℎ=1

12
Allocation of total sample to different strata
Suppose that we want to draw a total sample of size n from a population which is divided into a
number of strata. Then the total sample of size n is to be distributed among different strata. This
distribution of sample over different strata is called allocation of total sample or simply allocation.
Types of allocation
There are four types of allocation.
a) Equal allocation
b) Proportional allocation
c) Neyman allocation
d) Optimum allocation

Factors considered for allocation

The allocation of the sample to different strata is done by considering the following three factors.
a) The number of units in the stratum i.e. stratum size.
b) The variability within the stratum.
c) The cost of taking observations per sampling unit in the stratum.

Now we describe all types of allocations through following steps.

• First, we provide definition a specific allocation.
• Then, we describe how to obtain that allocation i.e. we describe how to obtain 𝑛ℎ under that
allocation.
• Also, we find 𝑉(𝑦𝑠𝑠 ) under that allocation.
• Besides, we describe some other issues related to certain allocation.

Equal allocation

Definition:
If the total sample is distributed equally among all the strata, the allocation is called equal
allocation.

How to obtain Equal allocation

We have
𝐿

𝑛ℎ = 𝑛
ℎ=1
⇒𝐿𝑛ℎ = 𝑛 [since under equal allocation 𝑛ℎ is constant for all h]
𝑛
⇒𝑛ℎ =
𝐿
That is under equal allocation, we obtain 𝑛ℎ by dividing the sample size by the number of strata.

𝒚𝒔𝒔 under equal allocation

For equal allocation, the stratified sample mean is
𝐿

𝑦𝑠𝑠 = 𝑊ℎ 𝑦ℎ
ℎ=1

13
 𝐿 𝑛
𝑁ℎ ∑𝑖=1 𝑦ℎℎ
ℎ

=
𝑁 𝑛ℎ
ℎ=1

𝐿 𝑛ℎ
𝑁ℎ 𝐿 𝑛
= 𝑦ℎℎ ∵ 𝑛ℎ =
𝑁 𝑛 𝐿
ℎ=1 𝑖=1
𝐿 𝑛ℎ
𝐿
= 𝑁ℎ 𝑦ℎ ∵ 𝑦ℎ = 𝑦ℎℎ
𝑛𝑁
ℎ=1 𝑖=1
𝑽(𝒚𝒔𝒔 ) under equal allocation
We have
𝐿
𝑆2ℎ
𝑉(𝑦𝑠𝑠 ) = 𝑊2ℎ (1 ― 𝑓ℎ )
𝑛ℎ
ℎ=1

𝐿
𝑁2ℎ 𝑁ℎ ― 𝑛ℎ 𝑆2ℎ
=
𝑁2 𝑁ℎ 𝑛ℎ
ℎ=1
𝐿
1 𝑆2ℎ
= 2 𝑁ℎ (𝑁ℎ ― 𝑛ℎ )
𝑁 𝑛ℎ
ℎ=1
𝐿
1 𝑛 𝑆2ℎ
= 2 𝑁ℎ 𝑁ℎ ―
𝑁 𝐿 𝑛
ℎ=1 𝐿
𝐿
1 𝐿𝑁ℎ ― 𝑛 𝑆2ℎ
= 𝑁ℎ 𝑛
𝑁2 𝐿
ℎ=1 𝐿
𝐿
1 𝐿𝑁ℎ ― 𝑛 2
= 𝑁ℎ 𝑆ℎ
𝑁2 𝑛
ℎ=1
𝐿
1
= 𝑁ℎ (𝐿𝑁ℎ ― 𝑛)𝑆2ℎ
𝑛𝑁 2
ℎ=1

Proportional allocation

Definition:
If the total sample is distributed over different strata in such a way that the size of sample in any
stratum is proportional to the stratum size, then this allocation is called proportional allocation.

How to obtain Proportional allocation

By definition of proportional allocation we have for hth stratum
𝑛ℎ ∞ 𝑁ℎ where 𝑛ℎ =hth stratum sample size and 𝑛ℎ =hth stratum size
⇒𝑛ℎ = 𝑘𝑁ℎ ………..(1)

14
 𝐿 𝐿

⇒ 𝑛ℎ = 𝑘 𝑁ℎ [summing over all strata]

ℎ=1 ℎ=1
⇒𝑛 = 𝑘𝑁
𝑛
⇒𝑘 =
𝑁
𝑛
Putting 𝑘 = 𝑁 in (1),
𝑛
𝑛ℎ = 𝑁ℎ .
𝑁
𝑛𝑁ℎ
⇒𝑛ℎ = 𝐿 .
∑ℎ=1 𝑁ℎ

Sampling fraction for proportional allocation

Since under proportional allocation,
𝑛
𝑛ℎ = 𝑁ℎ
𝑁
𝑛ℎ 𝑛
⇒ =
𝑁ℎ 𝑁
⇒ 𝑓ℎ = 𝑓
Therefore, under this allocation, sampling fraction is constant for all strata.

𝒚𝒔𝒔 under proportional allocation

For proportional allocation, the stratified sample mean is
𝐿

𝑦𝑠𝑠 = 𝑊ℎ 𝑦ℎ
ℎ=1
𝐿
𝑁ℎ
= 𝑦
𝑁 ℎ
ℎ=1

𝐿
𝑛ℎ 𝑛 𝑛ℎ 𝑁ℎ
= 𝑦ℎ Since under proportional allocation,𝑛ℎ = 𝑁ℎ ⇒ =
𝑛 𝑁 𝑛 𝑁
ℎ=1
𝐿 𝑛ℎ
𝑛ℎ 𝑦ℎ
= where𝑦ℎ = 𝑦ℎℎ
𝑛 𝑛ℎ
ℎ=1 𝑖=1
𝐿
𝑦ℎ
=
𝑛
ℎ=1
𝐿
∑ℎ=1 𝑦ℎ
=
𝑛
=𝑦
So, under proportional allocation, 𝑦𝑠𝑠 = 𝑦, i.e. stratified sample mean is equal to ordinary sample
mean.
Therefore, under proportional allocation, ordinary sample mean 𝑦 is an unbiased estimator of
population mean 𝑌 .

15
𝑽(𝒚𝒔𝒔 ) under proportional allocation
We have
𝐿
𝑆2ℎ
𝑉(𝑦𝑠𝑠 ) = 𝑊2ℎ (1 ― 𝑓ℎ )
𝑛ℎ
ℎ=1

𝐿
𝑆2ℎ
= 𝑊2ℎ (1 ― 𝑓)
𝑛ℎ
ℎ=1
𝐿
𝑁2ℎ 𝑆2ℎ
= (1 ― 𝑓)
𝑁 2 𝑛ℎ
ℎ=1
𝐿
𝑁2ℎ 𝑆2ℎ
= (1 ― 𝑓)
𝑁 2 𝑛𝑁ℎ
ℎ=1
𝑁
𝐿
(1 ― 𝑓) 𝑁ℎ 2
= 𝑆
𝑛 𝑁 ℎ
ℎ=1

𝐿
1 1
= ― 𝑊ℎ 𝑆2ℎ
𝑛 𝑁
ℎ=1
𝐿 𝐿
∑ℎ=1 𝑊ℎ 𝑆2ℎ ∑ℎ=1 𝑊ℎ 𝑆2ℎ
= ―
𝑛 𝑁

Neyman allocation (optimum allocation without cost)

Definition:
If the total sample is distributed over different strata in such a way that the size of sample in any
stratum is proportional to i) stratum size as well as to ii) variance of the stratum, then this allocation
is called Neyman allocation.

How to obtain Neyman allocation (how to obtain 𝒏𝒉 under this allocation)

𝐿
Neyman allocation is obtained by minimizing V(𝑦𝑠𝑠 ) subject to the condition ∑ℎ=1 𝑛ℎ = 𝑛.
We have
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ
V(𝑦𝑠𝑠 ) = ―
𝑛ℎ 𝑁ℎ
ℎ=1 ℎ=1

Now second part of right hand side is constant (with respect to 𝑛ℎ ). So we shall minimize only first
𝐿 𝑊ℎ2 𝑆ℎ2
part i.e. ∑ℎ=1 𝑛ℎ
= 𝑉 ′ (say).
Now minimizing 𝑉 ′ for fixed sample size n or minimizing n for fixed 𝑉 ′ is equivalent to
minimizing the product
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ
𝑉′𝑛 = 𝑛ℎ
𝑛ℎ
ℎ=1 ℎ=1
Let 𝑎ℎ and 𝑏ℎ are two sets of L positive numbers, then by Cauchy–Schwarz inequality, we have

16
 2
𝑎2ℎ 𝑏2ℎ ≥ 𝑎 ℎ 𝑏ℎ
𝑏
Where equality holds if 𝑎ℎ is constant for all h.
ℎ
𝑊ℎ 𝑆ℎ
Let 𝑎ℎ = 𝑛ℎ
and 𝑏ℎ = 𝑛ℎ ∴ 𝑎ℎ 𝑏ℎ=𝑊ℎ 𝑆ℎ
Hence by Cauchy–Schwarz inequality
𝐿 𝐿 𝐿 2
𝑊2ℎ 𝑆2ℎ
𝑉′𝑛 = 𝑛ℎ ≥ 𝑊ℎ 𝑆ℎ
𝑛ℎ
ℎ=1 ℎ=1 ℎ=1
𝐿 2
Therefore, the smallest value of 𝑉 ′ 𝑛 is ∑ℎ=1 𝑊ℎ 𝑆ℎ .
This minimum value occurs when
𝑏ℎ
= 𝑘 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡
𝑎ℎ
𝑛ℎ
⇒ =𝑘
𝑊ℎ 𝑆ℎ
𝑛ℎ
𝑛ℎ
⇒ =𝑘
𝑊ℎ 𝑆ℎ
⇒𝑛ℎ = 𝑘𝑊ℎ 𝑆ℎ
𝑁ℎ
⇒𝑛ℎ = 𝑘 𝑆
𝑁 ℎ
𝑘
⇒𝑛ℎ = 𝑘 ′ 𝑁ℎ 𝑆ℎ ……………(1) [𝑝𝑢𝑡𝑡𝑖𝑛𝑔 = 𝑘′]
𝑁
𝐿 𝐿

⇒ 𝑛ℎ = 𝑘 ′ 𝑁ℎ 𝑆ℎ [𝑠𝑢𝑚𝑚𝑖𝑛𝑔 𝑜𝑣𝑒𝑟 𝑎𝑙𝑙 𝑠𝑡𝑟𝑎𝑡𝑎]

ℎ=1 ℎ=1
𝐿

⇒𝑛 = 𝑘 ′ 𝑁ℎ 𝑆ℎ
ℎ=1
𝑛
⇒𝑘 ′ = 𝐿
∑ℎ=1 𝑁ℎ 𝑆ℎ
Putting the value of 𝑘 ′ in (1)
𝑛𝑁ℎ 𝑆ℎ
⇒𝑛ℎ = 𝐿 ……………(2)
∑ℎ=1 𝑁ℎ 𝑆ℎ
Equation (2) gives the formula for 𝑛ℎ in Neyman allocation.

Note: Equation (2) gives 𝑛ℎ ∞𝑁ℎ 𝑆ℎ which provides the definition of this allocation where
𝑛
proportionality constant 𝑘 ′ = 𝐿 .
∑ℎ=1 𝑁ℎ 𝑆ℎ

Minimum variance in Neyman allocation

We have
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ
V(𝑦𝑠𝑠 ) = ―
𝑛ℎ 𝑁ℎ
ℎ=1 ℎ=1

17
 𝐿 𝐿
𝑊2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ
⇒𝑉𝑁𝑁𝑁 = ― [𝑝𝑢𝑡𝑡𝑖𝑛𝑔 𝑡ℎ𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑛ℎ 𝑢𝑛𝑑𝑒𝑟 𝑁𝑒𝑦𝑚𝑎𝑛 𝑎𝑙𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛]
𝑛𝑁ℎ 𝑆ℎ 𝑁ℎ
ℎ=1 ℎ=1
∑ 𝑁ℎ 𝑆ℎ
𝐿 𝐿
1 𝑁2ℎ 𝑆2ℎ 1 𝑁2ℎ 𝑆2ℎ
= ― 2
𝑛𝑁 2 𝑁ℎ 𝑆ℎ 𝑁 𝑁ℎ
ℎ=1 ℎ=1
∑ 𝑁ℎ 𝑆ℎ
𝐿 2 𝐿
1 1
= 𝑁ℎ 𝑆ℎ ― 2 𝑁ℎ 𝑆2ℎ ……………(3)
𝑛𝑁 2 𝑁
ℎ=1 ℎ=1
2
𝐿 𝐿
∑ℎ=1 𝑊ℎ 𝑆ℎ ∑ℎ=1 𝑊ℎ 𝑆2ℎ
= ―
𝑛 𝑁2
Minimum value of n for given variance V
From (3) we have
𝐿 2 𝐿
1 1
𝑉= 𝑁ℎ 𝑆ℎ ― 2 𝑁ℎ 𝑆2ℎ
𝑛𝑁 2 𝑁
ℎ=1 ℎ=1
𝐿 2 𝐿
1 1
⇒ 𝑁ℎ 𝑆ℎ =𝑉+ 2 𝑁ℎ 𝑆2ℎ
𝑛𝑁 2 𝑁
ℎ=1 ℎ=1

2
1 𝐿
2 ∑ℎ=1 𝑁ℎ 𝑆ℎ
∴𝑛= 𝑁
1 𝐿
𝑉 + 𝑁 2 ∑ℎ=1 𝑁ℎ 𝑆2ℎ
2
𝐿
∑ℎ=1 𝑁ℎ 𝑆ℎ
⇒𝑛 = 𝐿 .
𝑁 2 𝑉 + ∑ℎ=1 𝑁ℎ 𝑆2ℎ

Optimum allocation
Definition:
If the total sample is distributed over different strata in such a way that the sample size in hth
𝑁ℎ 𝑆ℎ
stratum is proportional to the quantity 𝑐ℎ
where
𝑁ℎ = hth stratum size
𝑆2ℎ =variance of hth stratum
𝑐ℎ =cost per unit for sampling in hth stratum,
then this allocation called optimum allocation.
That is in this allocation,
𝑁ℎ 𝑆ℎ
𝑛ℎ ∞ .
𝑐ℎ

How to obtain optimum allocation (how to obtain 𝒏𝒉 under this allocation)

For this allocation, we consider a linear cost function

18
 𝐿

𝑐 = 𝑐0 + 𝑐ℎ 𝑛ℎ ………(1)
ℎ=1
𝐿
⇒𝑐 = 𝑐0 + 𝑐 ′ …………(2) where 𝑐 ′ = ∑ℎ=1 𝑐ℎ 𝑛ℎ

In equation (1)
𝑐 = total cost
𝑐0 = overhead cost (e.g. cost of setting up and maintaining an office, recruiting survey personnel,
capital expenses etc.)
𝑐ℎ = cost of sampling per unit from hth stratum
…………………………………………
Note: For example, in a household survey, in stratum 1, survey-cost per unit per household is c1
and n1 households are taken from stratum 1. So the total cost for n1 households is c1n1. Similarly
𝐿
for stratum 2, the cost is c2n2 and so on. Therefore, the total cost for survey only is ∑ℎ=1 𝑐ℎ 𝑛ℎ and
𝐿
hence the total cost for whole activities of the survey is 𝑐0 + ∑ℎ=1 𝑐ℎ 𝑛ℎ = 𝑐 (𝑠𝑎𝑦).
………………………………………
Optimum allocation is obtained by minimizing V(𝑦𝑠𝑠 ) for a given cost c or minimizing c for given
value of V(𝑦𝑠𝑠 ).
Now,
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ
V(𝑦𝑠𝑠 ) = ―
𝑛ℎ 𝑁ℎ
ℎ=1 ℎ=1
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ
= 𝑉′ ― ………(3) 𝑤ℎ𝑒𝑟𝑒 𝑉 ′ =
𝑁ℎ 𝑛ℎ
ℎ=1 ℎ=1
In V(𝑦𝑠𝑠 ), we consider only 1st term i.e. 𝑉 ′ of equation (3) and
in the expression for c, we consider 2nd term i.e. 𝑐 ′ of equation (2), because other parts are constants
(as the remaining parts in V and c are independent of 𝑛ℎ ).

Now minimizing 𝑉 ′ for 𝑐 ′ or minimizing 𝑐 ′ for given 𝑉 ′ is equivalent to minimizing the product
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ
𝑉′𝑐′ = 𝑐ℎ 𝑛ℎ
𝑛ℎ
ℎ=1 ℎ=1
Let 𝑎ℎ and 𝑏ℎ are two sets of L positive numbers, then by Cauchy–Schwarz inequality, we have
2
𝑎2ℎ 𝑏2ℎ ≥ 𝑎 ℎ 𝑏ℎ
𝑏ℎ
where equality holds if 𝑎ℎ
is constant for all h.
𝑊ℎ 𝑆ℎ
Let 𝑎ℎ = 𝑛ℎ
and 𝑏ℎ = 𝑐ℎ 𝑛ℎ
𝑊ℎ 𝑆ℎ
∴ 𝑎 ℎ 𝑏ℎ = 𝑐ℎ 𝑛ℎ = 𝑊ℎ 𝑆ℎ 𝑐ℎ
𝑛ℎ
Hence by Cauchy–Schwarz inequality
𝐿 𝐿 𝐿 2
𝑊2ℎ 𝑆2ℎ
𝑉′𝑐′ = 𝑐ℎ 𝑛ℎ ≥ 𝑊ℎ 𝑆ℎ 𝑐ℎ
𝑛ℎ
ℎ=1 ℎ=1 ℎ=1
𝐿 2
Therefore, the smallest value of 𝑉 ′ 𝑐 ′ is ∑ℎ=1 𝑊ℎ 𝑆ℎ 𝑐ℎ .
19
This minimum value occurs when
𝑏ℎ
= 𝑘 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡
𝑎ℎ
𝑐ℎ 𝑛ℎ
⇒ =𝑘
𝑊ℎ 𝑆ℎ
𝑛ℎ
𝑛ℎ 𝑐ℎ
⇒ =𝑘
𝑊ℎ 𝑆ℎ
𝑊ℎ 𝑆ℎ
⇒𝑛ℎ = 𝑘
𝑐ℎ
𝑘 𝑁ℎ 𝑆ℎ
⇒𝑛ℎ =
𝑁 𝑐ℎ
𝑁ℎ 𝑆ℎ 𝑘
⇒𝑛ℎ = 𝑘 ′ ……………(4) [𝑝𝑢𝑡𝑡𝑖𝑛𝑔 = 𝑘 ′ ]
𝑐ℎ 𝑁
𝐿 𝐿
𝑁ℎ 𝑆ℎ
⇒ 𝑛ℎ = 𝑘 ′ [𝑠𝑢𝑚𝑚𝑖𝑛𝑔 𝑜𝑣𝑒𝑟 𝑎𝑙𝑙 𝑠𝑡𝑟𝑎𝑡𝑎]
ℎ=1 ℎ=1
𝑐ℎ
𝐿
𝑁ℎ 𝑆ℎ
⇒𝑛 = 𝑘 ′
ℎ=1
𝑐ℎ
𝑛
⇒𝑘 ′ = 𝐿
∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ
Putting the value of 𝑘 ′ in (4)
𝑛𝑁ℎ 𝑆ℎ / 𝑐ℎ
⇒𝑛ℎ = 𝐿 ……………(5)
∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ
Equation (5) gives the formula for 𝑛ℎ in optimum allocation.
………………………………………
𝑁ℎ 𝑆ℎ
Note: Equation (5) gives 𝑛ℎ ∞ 𝑐ℎ
which provides the definition of this allocation where
𝑛
proportionality constant 𝑘 ′ = .
∑𝐿ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ
………………………………………

Minimum variance in optimum allocation

We have
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ
V(𝑦𝑠𝑠 ) = ―
𝑛ℎ 𝑁ℎ
ℎ=1 ℎ=1
𝐿 𝐿
𝑊2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ
⇒𝑉𝑜𝑜𝑜 = ― [𝑝𝑢𝑡𝑡𝑖𝑛𝑔 𝑡ℎ𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑛ℎ 𝑢𝑛𝑑𝑒𝑟 𝑜𝑝𝑡𝑖𝑚𝑢𝑚 𝑎𝑙𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛]
𝑛𝑁 𝑆 / 𝑐ℎ 𝑁ℎ
ℎ=1 𝐿 ℎ ℎ ℎ=1
∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ

𝐿 𝐿
1 𝑁2ℎ 𝑆2ℎ 𝑊2ℎ 𝑆2ℎ
⇒𝑉𝑜𝑜𝑜 = 2 ―
𝑁 𝑛𝑁 𝑆 / 𝑐ℎ 𝑁ℎ
ℎ=1 𝐿 ℎ ℎ ℎ=1
∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ

20
 𝐿 𝐿 𝐿
1 1
⇒𝑉𝑜𝑜𝑜 = 𝑁 ℎ 𝑆 ℎ 𝑐ℎ 𝑁 ℎ 𝑆 ℎ / 𝑐ℎ ― 2 𝑁ℎ 𝑆2ℎ ………(6)
𝑛𝑁 2 𝑁
ℎ=1 ℎ=1 ℎ=1

Value of n for given cost c

From (1), we have
𝐿

𝑐 = 𝑐0 + 𝑐ℎ 𝑛ℎ
ℎ=1
𝐿
𝑛𝑁ℎ 𝑆ℎ / 𝑐ℎ
⇒𝑐 = 𝑐0 + 𝑐ℎ 𝐿
ℎ=1
∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ
𝐿
𝑛 ∑ℎ=1 𝑁ℎ 𝑆ℎ 𝑐ℎ
⇒𝑐 ― 𝑐0 = 𝐿
∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ
𝐿
(𝑐 ― 𝑐0 ) ∑ℎ=1 𝑁ℎ 𝑆ℎ 𝑐ℎ
⇒𝑛 = 𝐿 ………(7)
∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ

Minimum variance for given cost c

Putting the value of n from (7) in (6), we have
𝐿 𝐿 𝐿 𝐿
1 ∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ 1
𝑉𝑜𝑜𝑜 = 2 𝑁 ℎ 𝑆 ℎ 𝑐ℎ 𝑁ℎ 𝑆ℎ / 𝑐ℎ ― 2 𝑁ℎ 𝑆2ℎ
𝑁 (𝑐 ― 𝑐 ) ∑𝐿 𝑁 𝑆 𝑐 𝑁
0 ℎ=1 ℎ ℎ ℎ ℎ=1 ℎ=1 ℎ=1
2 𝐿
𝐿
1 ∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ 1
= 2 ― 2 𝑁ℎ 𝑆2ℎ
𝑁 (𝑐 ― 𝑐0 ) 𝑁
ℎ=1
2 𝐿
𝐿
∑ℎ=1 𝑊ℎ 𝑆ℎ / 𝑐ℎ 𝑊ℎ 𝑆2ℎ
= ―
(𝑐 ― 𝑐0 ) 𝑁
ℎ=1

Value of n for given variance V

From (6), we have
𝐿 𝐿 𝐿
1 1
𝑉= 𝑁ℎ 𝑆ℎ 𝑐ℎ 𝑁 ℎ 𝑆 ℎ / 𝑐ℎ ― 2 𝑁ℎ 𝑆2ℎ
𝑛𝑁 2 𝑁
ℎ=1 ℎ=1 ℎ=1
𝐿 𝐿 𝐿
1 1
⇒ 𝑁ℎ 𝑆ℎ 𝑐ℎ 𝑁 ℎ 𝑆 ℎ / 𝑐ℎ =𝑉+ 2 𝑁ℎ 𝑆2ℎ
𝑛𝑁 2 𝑁
ℎ=1 ℎ=1 ℎ=1
𝐿 𝐿 𝐿
1 1
⇒ 𝑁ℎ 𝑆ℎ 𝑐ℎ 𝑁 ℎ 𝑆 ℎ / 𝑐ℎ = 2 𝑁2𝑉 + 𝑁ℎ 𝑆2ℎ
𝑛𝑁 2 𝑁
ℎ=1 ℎ=1 ℎ=1
𝐿 𝐿 𝐿

⇒𝑛 𝑁 2 𝑉 + 𝑁ℎ 𝑆2ℎ = 𝑁 ℎ 𝑆 ℎ 𝑐ℎ 𝑁ℎ 𝑆ℎ / 𝑐ℎ
ℎ=1 ℎ=1 ℎ=1
𝐿 𝐿
∑ℎ=1 𝑁ℎ 𝑆ℎ 𝑐ℎ ∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ
⇒𝑛 = 𝐿
𝑁 2 𝑉 + ∑ℎ=1 𝑁ℎ 𝑆2ℎ

21
Comparison of precision in different allocation with SRS
1
Theorem: If terms 𝑁ℎ ′s are so large that 𝑁 ≃ 0 (ℎ = 1, 2, …𝐿) then
ℎ

𝑉𝑟𝑟𝑟 ≥ 𝑉𝑝𝑝𝑝𝑝 ≥ 𝑉𝑁𝑁𝑁

or
𝑉𝑁𝑁𝑁 ≤ 𝑉𝑝𝑝𝑝𝑝 ≤ 𝑉𝑟𝑟𝑟

Proof:
We know
𝑆2
𝑉𝑟𝑟𝑟 = (1 ― 𝑓)
𝑛
𝑛 𝑆2
⇒𝑉𝑟𝑟𝑟 = (1 ― )
𝑁 𝑛
1 1
⇒𝑉𝑟𝑟𝑟 = ― 𝑆2
𝑛 𝑁
𝑆2 1
𝑉𝑟𝑟𝑟 = …………(1) since𝑁ℎ ′s are large 𝑁 is also large so that ≃ 0
𝑛 𝑁

𝐿
(1 ― 𝑓)
𝑉𝑝𝑝𝑝𝑝 = 𝑊ℎ 𝑆2ℎ
𝑛
ℎ=1
𝐿
1 1
⇒𝑉𝑝𝑝𝑝𝑝 = ― 𝑊ℎ 𝑆2ℎ
𝑛 𝑁
ℎ=1
𝐿
1 1
⇒𝑉𝑝𝑝𝑝𝑝 = 𝑊ℎ 𝑆2ℎ …………(2) since ≃0
𝑛 𝑁
ℎ=1
Now
𝐿 𝑁
∑ℎ=1 ∑𝑖=1
ℎ
(𝑦ℎℎ ― 𝑌) 2
𝑆2 =
𝑁―1
𝐿 𝑁ℎ

⇒(𝑁 ― 1)𝑆 2 = (𝑦ℎℎ ― 𝑌) 2

ℎ=1 𝑖=1
𝐿 𝑁ℎ
2
= [ 𝑦ℎℎ ― 𝑌ℎ ) + (𝑌ℎ ― 𝑌 ]
ℎ=1 𝑖=1
𝐿 𝑁ℎ 𝐿 𝑁ℎ
2 2
= 𝑦ℎℎ ― 𝑌ℎ + 𝑌ℎ ― 𝑌 [𝑐𝑟𝑜𝑠𝑠 𝑝𝑟𝑜𝑑𝑢𝑐𝑡 𝑡𝑒𝑟𝑚𝑠 𝑣𝑎𝑛𝑖𝑠ℎ]
ℎ=1 𝑖=1 ℎ=1 𝑖=1

𝐿 𝐿
2
= (𝑁ℎ ― 1)𝑆2ℎ + 𝑁ℎ 𝑌ℎ ― 𝑌
ℎ=1 ℎ=1
𝐿 𝐿
2
∴ 𝑁𝑆 2 = 𝑁ℎ 𝑆2ℎ + 𝑁ℎ 𝑌ℎ ― 𝑌 [∵ 𝑁ℎ ― 1 ≃ 𝑁ℎ 𝑎𝑛𝑑 𝑁 ― 1 ≃ 𝑁]
ℎ=1 ℎ=1

22
 𝐿 𝐿
2
⇒𝑆 2 = 𝑊ℎ 𝑆2ℎ + 𝑊ℎ 𝑌ℎ ― 𝑌
ℎ=1 ℎ=1
𝐿 𝐿
𝑆2 1 1 2
⇒ = 𝑊ℎ 𝑆2ℎ + 𝑊ℎ 𝑌ℎ ― 𝑌
𝑛 𝑛 𝑛
ℎ=1 ℎ=1
𝐿
1 2
⇒𝑉𝑟𝑟𝑟 = 𝑉𝑝𝑝𝑝𝑝 + 𝑊ℎ 𝑌ℎ ― 𝑌 [𝑢𝑠𝑖𝑛𝑔 (1)𝑎𝑛𝑑 (2)]
𝑛
ℎ=1
𝐿
1 2
∴ 𝑉𝑟𝑟𝑟 ≥ 𝑉𝑝𝑝𝑝𝑝 …………(𝐴) ∵ 𝑊ℎ 𝑌ℎ ― 𝑌 ≥0
𝑛
ℎ=1
Again we have

2
𝐿 𝐿
∑ℎ=1 𝑊ℎ 𝑆ℎ ∑ℎ=1 𝑊ℎ 𝑆2ℎ
⇒𝑉𝑁𝑁𝑁 = ―
𝑛 𝑁

2 𝐿
𝐿
∑ℎ=1 𝑊ℎ 𝑆ℎ ∑ℎ=1 𝑊ℎ 𝑆2ℎ
⇒𝑉𝑁𝑁𝑁 = …………(3) ∵ ≃ 0 𝑓𝑜𝑟 𝑙𝑎𝑟𝑔𝑒 𝑁
𝑛 𝑁

Now
𝐿 𝐿 2
1 1
𝑉𝑝𝑝𝑝𝑝 ― 𝑉𝑁𝑁𝑁 = 𝑊ℎ 𝑆2ℎ ― 𝑊ℎ 𝑆ℎ [𝑢𝑠𝑖𝑛𝑔 (2)𝑎𝑛𝑑 (3)]
𝑛 𝑛
ℎ=1 ℎ=1
𝐿 𝐿 2
1
= 𝑊ℎ 𝑆2ℎ ― 𝑊ℎ 𝑆ℎ
𝑛
ℎ=1 ℎ=1

𝐿 𝐿
1 2
= 𝑊ℎ 𝑆ℎ ― 𝑆 𝑤ℎ𝑒𝑟𝑒 𝑆 = 𝑊ℎ 𝑆ℎ
𝑛
ℎ=1 ℎ=1
𝐿
1 2
∴ 𝑉𝑝𝑝𝑝𝑝 ― 𝑉𝑁𝑁𝑁 = 𝑊ℎ 𝑆ℎ ― 𝑆 ≥0
𝑛
ℎ=1
∴ 𝑉𝑝𝑝𝑝𝑝 ≥ 𝑉𝑁𝑁𝑁 …………(𝐵)

It follows from (A) and (B),

𝑉𝑟𝑟𝑟 ≥ 𝑉𝑝𝑝𝑝𝑝 ≥ 𝑉𝑁𝑁𝑁
or
𝑉𝑁𝑁𝑁 ≤ 𝑉𝑝𝑝𝑝𝑝 ≤ 𝑉𝑟𝑟𝑟
(proved)

Note 1: 𝑉𝑁𝑁𝑁 is sometimes referred to as 𝑉𝑜𝑜𝑜 , e.g. in Cochran (p. 100).

Note 2:
Cross product term

23
 𝐿 𝑁ℎ
2
= 𝑦ℎℎ ― 𝑌ℎ )(𝑌ℎ ― 𝑌
ℎ=1 𝑖=1
𝐿 𝑁ℎ
2
= 𝑦ℎℎ ― 𝑌ℎ )(𝑌ℎ ― 𝑌
ℎ=1 𝑖=1
𝐿 𝑁ℎ
2
= (𝑌ℎ ― 𝑌) 𝑦ℎℎ ― 𝑌ℎ
ℎ=1 𝑖=1
𝑁ℎ
2
=0 ∵ 𝑦ℎℎ ― 𝑌ℎ =0
𝑖=1
Note 3:

Let
𝑊ℎ = 𝑓𝑖 , 𝑆ℎ = 𝑥𝑖 , 𝑆 = 𝑥 𝑎𝑛𝑑 𝐿 = 𝑛
Then
𝐿 𝐿 2

𝑊ℎ 𝑆2ℎ ― 𝑊ℎ 𝑆ℎ
ℎ=1 ℎ=1
𝑛 𝑛 2

= 𝑓𝑖 𝑥𝑖2 ― 𝑓𝑖 𝑥𝑖
𝑖=1 𝑖=1

𝑛 2 𝑛 𝐿
𝑛
∑𝑖=1 𝑓𝑖 𝑥𝑖
= 𝑓𝑖 𝑥𝑖2 ― 𝑛 ∵ 𝑓𝑖 = 𝑊ℎ = 1
𝑖=1
∑𝑖=1 𝑓𝑖 𝑖=1 ℎ=1

= 𝑓𝑖 (𝑥𝑖 ― 𝑥) 2
𝑖=1
𝐿
2
= 𝑊ℎ 𝑆ℎ ― 𝑆
ℎ=1

Theorem: Given the result of stratified random sample, an unbiased estimator of variance of SRS
is given by
𝐿 𝑛ℎ
𝑁―𝑛 1 𝑁ℎ
𝑣𝑟𝑟𝑟 (𝑦) = 𝑦2ℎℎ ― 𝑦2𝑠𝑠 + 𝑣(𝑦𝑠𝑠 )
𝑛(𝑁 ― 1) 𝑁 𝑛ℎ
ℎ=1 𝑖=1
𝑤ℎ𝑒𝑟𝑒 𝑣(𝑦𝑠𝑠 )𝑖𝑠 𝑎𝑛 𝑢𝑛𝑏𝑖𝑎𝑠𝑒𝑑 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑜𝑟 𝑜𝑓 𝑉(𝑦𝑠𝑠)

Proof:
We have
𝑁―𝑛 2
𝑉𝑟𝑟𝑟 (𝑦) = 𝑆
𝑛𝑁

24
 𝐿 𝑁
𝑁 ― 𝑛 ∑ℎ=1 ∑𝑖=1 (𝑦ℎℎ ― 𝑌) 2
ℎ

⇒𝑉𝑟𝑟𝑟 (𝑦) =
𝑛𝑁 𝑁―1

2 𝐿 𝑁
𝑁 ― 𝑛 ∑ℎ=1 ∑𝑖=1 𝑦ℎℎ ― 𝑁𝑌2
ℎ

⇒𝑉𝑟𝑟𝑟 (𝑦) =
𝑛𝑁 𝑁―1

𝐿 𝑁ℎ
𝑁―𝑛 1
⇒𝑉𝑟𝑟𝑟 (𝑦) = 𝑦2ℎℎ ― 𝑌2 ………(1)
𝑛(𝑁 ― 1) 𝑁
ℎ=1 𝑖=1

Now,
𝐿 𝑛ℎ
1 𝑁ℎ
𝐸 𝑦2ℎℎ
𝑁 𝑛ℎ
ℎ=1 𝑖=1

𝐿 𝑛ℎ
1 𝑁ℎ
= 𝐸(𝑦2ℎℎ )
𝑁 𝑛ℎ
ℎ=1 𝑖=1

𝐿
1 𝑁ℎ
= 𝑛 𝐸(𝑦2ℎℎ )
𝑁 𝑛ℎ ℎ
ℎ=1

𝐿 𝑁ℎ
1 1
= 𝑁ℎ 𝑦2ℎℎ
𝑁 𝑁ℎ
ℎ=1 𝑖=1
𝐿 𝑛ℎ 𝐿 𝑁ℎ
1 𝑁ℎ 1
∴𝐸 𝑦2ℎℎ = 𝑦2ℎℎ ………(2)
𝑁 𝑛ℎ 𝑁
ℎ=1 𝑖=1 ℎ=1 𝑖=1

Again
2
𝑉(𝑦𝑠𝑠 ) = 𝐸 𝑦2𝑠𝑠 ― 𝐸(𝑦𝑠𝑠 )
⇒𝐸 𝑣(𝑦𝑠𝑠 ) = 𝐸 𝑦2𝑠𝑠 ― 𝑌2 ∵ 𝐸 𝑣(𝑦𝑠𝑠 ) = 𝑉(𝑦𝑠𝑠 ) 𝑎𝑛𝑑 𝐸(𝑦𝑠𝑠 ) = 𝑌

⇒ 𝐸 𝑦2𝑠𝑠 ― 𝐸 𝑣(𝑦𝑠𝑠 ) = 𝑌2

⇒ 𝐸 𝑦2𝑠𝑠 ― 𝑣(𝑦𝑠𝑠 ) = 𝑌2 ………(3)

Hence 𝑦2𝑠𝑠 ― 𝑣(𝑦𝑠𝑠 ) is an unbiased estimator of 𝑌2 .

Using the results of (2) and (3) in (1), we get an unbiased estimator of 𝑉𝑟𝑟𝑟 (𝑦) as:
𝐿 𝑛ℎ
𝑁―𝑛 1 𝑁ℎ
𝑣𝑟𝑟𝑟 (𝑦) = 𝑦2ℎℎ ― 𝑦2𝑠𝑠 + 𝑣(𝑦𝑠𝑠 ) .
𝑛(𝑁 ― 1) 𝑁 𝑛ℎ
ℎ=1 𝑖=1

25
Use of this theorem
Using the above theorem, we get the estimate of gain in precision due to stratification as
1 1
― .
𝑣(𝑦𝑠𝑠 ) 𝑣𝑟𝑟𝑟 (𝑦)

Transformation of one allocation to another

For proportional allocation
𝑛𝑁ℎ
𝑛ℎ = 𝐿 …………………(1)
∑ℎ=1 𝑁ℎ
For Neyman allocation
𝑛𝑁ℎ 𝑆ℎ
𝑛ℎ = 𝐿 ……………….(2)
∑ℎ=1 𝑁ℎ 𝑆ℎ

For optimum allocation

𝑛𝑁ℎ 𝑆ℎ / 𝑐ℎ
𝑛ℎ = 𝐿 ……………(3)
∑ℎ=1 𝑁ℎ 𝑆ℎ / 𝑐ℎ
Now,
If 𝑆ℎ is constant for all ℎ, (2) turned into (1).
If 𝑐ℎ is constant for all ℎ, (3) turned into (2).
If 𝑆ℎ and 𝑐ℎ are constant for all ℎ, (3) turned into (1).

26