Questions based on Stat 513 (Sampling Techniques)

QUESTIONS AND ANSWERS OF SECTION A

(BOLD option shall provide the correct answer of the questions)

Section A

Objectives on Sampling (Stat 513)

1. A sample consists of
(a) All units of the population
(b) 5% units of the population
(c) 10% units of the population
(d) Any fraction of the population

2. Sampling is used in the situations

(a) Blood test of the patients
(b) Cooking rice in an utensil
(c) Purchase of food commodity from shopkeeper
(d) All the above
3. The number of possible samples of size n out of N population size in SRSWOR
is equal to
(a) Ncn answer is (a)
(b) N n

(c) (N-n)/N
(d) n/N
4. The number of possible samples of size n out of N population size in SRSWR is
equal to

(a) Ncn

(b)Nn

(c) (N-n)/N

(e) n/N
5. The number of possible samples of size 2 out of 5 population size in SRSWOR is
equal to
(a) 10
(b) 4
(c) 2
 (d) 12
6. The number of possible samples of size 2 out of 5 population size in SRSWR is
equal to
(a) 25

(b) 20

(c) 2

(d) 12

7. Probability of a drawing unit at each subsequent draw remains same in

(a) SRSWOR
(b) SRSWR
(c) Both (a) &(b)
(d) None
8. The sampling fraction in usual notation is expressed as
(a) n/N
(b) N/n
(c) 1-n/N
(d) None.
9. The finite population correction in usual notation is expressed as
(a) (N-n)/N
(b) 1-(n/N)
(c) Both(a)&(b)
(d) None
10. A selection procedure of sampling having no involvement of probability is known
as
(a) SRSWOR
(b) Purposive sampling
(c) SRSWR
(d) None
11. For gathering information on rare events, sampling is used
(a) SRSWOR
(b) Stratified random sampling
(c) Inverse sampling
(d) None
12. If a larger units have more probability of their inclusion in the sample, the
sampling is known as
(a) SRSWOR
(b) PPS sampling
 (c) Stratified random sampling
(d) None
13. Simple random samples can be drawn with of help of
(a) Random numbers table
(b) Chit Method
(c) Roulette wheel
(d) All the above
14. Sampling frame is a list of
(a) A list of units of a population
(b) A list of random numbers
(c) A list of natural numbers
(d) None
15. In SRSWR, the same sampling unit may be included in the sample
(a) Only once
(b) Two times
(c) More than once
(d) None
16. The discrepancies between the estimate and the population parameter is known
as
(a) Sampling error
(b) Non-sampling error
(c) Formula error
(d) None
17. The error in a survey other than sampling error is known as
(a) Sampling error
(b) Non-sampling error
(c) Formula error
(d) None
18. A function of sample observations is known as
(a) Statistic
(b) Estimator
(c) Both (a)&(b)
(d) None
19. If the sample sizes are large from the population, then which error will contribute
more errors
(a) Sampling error
(b) Non-sampling error
(c) Both(a)&(b)
(d) None
20. If the sample sizes are large from the population, then which error will contribute
less errors
(a) Sampling error
(b) Non-sampling error
(c) Both (a)&(b)
(d) None
21. Simple random sample can be drawn with the help of (a)random number
tables(b)Chit method (c) roulette wheel (d) all the above
22. If each and every unit of a population has an equal chance of being included in
the sample, it is called(a) restricted sampling (b) unrestricted sampling (c)
purposive sampling (d) subjective sampling
23. If the observations recorded on five sampled items are 3,4,5,6,7,then sample
variance is equal to (a)0 (b) 1(c)2 (d) 2.5
24. If all the observations in a set of observations are the same, then variance of set
of values is (a) 0 (b) 2 (c) infinite (d) none
25. If the sample values are 1,3,5,7,9 then S.E. of sample mean is (a) 2 (b) √2 (c) 3
(d) √3 ℎ ( )
26. As a normal practice sampling fraction is considered to be negligible if it is (a)
more than 5%(b) less than≤ 5% (c) More than 10% (d) none
27. Systematic sampling is used when (a) when data are on cards (b) when the
items are in row (c) when the items are given in a sequential order (d) all the
above
28. If the population units N is multiple of n and k, then we use (a) linear
systematic sampling (b) circular systematic sampling (c) random systematic
sampling (d) all the above
29. Circular systematic sampling is used when (a) N is a whole number (b) N is not
divisible by n (c) N is a multiple of n (d) All the above.
30. Problem of non-response has (a) no solution (b) can be solved (c) no meaning
(d) none
31. If sample sizes increase, then sampling error will (a) increase (b) decrease (c)
both (a) &(b) (d) none
32. If sample sizes increase, then non-sampling error will (a) increase (b) decrease
(c) both (a) &(b) (d) none
33. A population is divided into clusters and it has been found that all the units within
a cluster are same. In this situation which sampling will be adopted (a)
SRSWOR (b) Stratified random sampling (c) Cluster sampling (d) Systematic
sampling
34. A population N is divided into k strata. A sample of size n is to be chosen and Ni
is the size of the ith stratum. Then sample size n as per proportional allocation is
given by (a) ni=nN (b) ni/Ni=n/N (c) niNi=nN (d) none
35. In case of iniverse sampling, the proportion p of m units of interest contained in a
sample of n units is equal to (a) m/n (b)(m-1)/n (c) (m-1)/(n-1) (d) (m-1)/(n+1)
36. If the respondents do not provide the required information to the researcher,
then it is known as (a) non-sampling error(b) the problem of non-response (c)
both (a) &(b) (d) none
37. The errors falling under faulty planning of survey, it is called (a) non-sampling
errors (b) non-response errors (c) Sampling errors (d) Absolute error
38. If there is a certain number of very high values in the sample, it is preferable to
compute (a) Standard error (b) Standard deviation (c) variance (d) all the
above.
39. For estimating the population mean T,let T1 be the sample mean under
SRSWOR and T2 sample mean under SRSWR, then which relationship is true
(a) Var(T1)< Var(T2) (b) Var(T1)> Var(T2) (c) Var(T1)≤ Var(T2) (d) none
40. The magnitude of the standard error of an estimate is an index of its (a) accuracy
(b) precision (c) efficiency (d) none
41. Which of the following statement is true (a) population mean increases with
increase in sample size (b) population mean decreases with increase in sample
size(c) population mean decreases with decrease in sample size(d) population
mean is a constant value.
42. A sample of 25 units from a population with standard deviation10 results into a
total score of 450. Then the mean of sampling distribution is equal to (a) 45(b) 18
(c) 50 (d) none
43. A population is perfectly homogeneous with respect to a characteristic, what size
of sample would you need (a) no sample (b) a large sample (c) a small sample
(d) a single sample.

SOME MORE QUESTIONS

44. A sample constant representing a population parameter is called estimate.

45. An estimator Tn which is most concentrated to the parameter θ is called best
estimator
46. If we have purposive sample, estimation is not possible.
47. Estimation is possible only in case of random sample.
48. An estimator itself a random variable.
49. A value of an estimator is called an estimate.
50. If the value of an estimator Tn is equal to the parameter θ, then Tn is called
unbiased estimator of the parameter θ.
51. Sampling is useful in many situations.
52. In a hypothetical population , all sampling units are imagined.
53. If the number of units in a population is limited, it is called finite population.
54. If the number of units in a population is unlimited, it is called infinite population.
55. If all the units of a population are surveyed, it is called complete enumeration.
56. The errors other than sampling errors is called non sampling errors.
57. The discrepancy between parameter and its estimate due to sampling process is
known as sampling errors.
58. Any population constants is known as parameter.
59. A function based on sample values for estimating a parameter is called an
estimator.
60. The specific value of an estimator is called an estimate.
61. Standard deviation of all possible estimates from sample of fixed size is called
standard error.
62. Standard error of sample mean based on n sizes with standard deviation s is
known as s/√ .
63. The list of all items of a population is known as sampling frame.
64. Another name of population is universe.
65. Number of possible samples of size n out of N using SRSWOR is N c n
66. Probability of selection of any one sample out of N c n is equal to 1/ N c n
67. Number of possible samples of size n out of N using SRSWR is Nn.

68. Under SRSWOR, the same item can occur more than once.
69. The sampling procedure in which the population is divided into homogeneous
groups and sample drawn from each group is called stratified random
sampling.
70. Stratified random sampling is useful when population is heterogeneous.
71. Stratification is done with respect to certain characteristics.
72. Deciding the sample size for each stratum is known as allocation problem.
73. If the sample size of each stratum is in proportion to stratum size, it is called
proportional allocation.
74. Stratified random sampling falls under the category of restricted sampling.
75. More heterogeneous is the population, larger sample sizes are required.
76. In usual notation (N-n)/N is known as finite population correction.
77. In usual notation n/N is called sampling fraction.
78. For a high precision of estimates, larger samples are required.
79. Estimators and estimates are different.
80. Determination of sample sizes for each stratum subject to the cost constrained is
called optimum allocation.
81. Optimum allocation is also known as Neyman allocation.
82. Double sampling is termed as two phase sampling.
83. Cluster sampling ordinarily leads to the loss of precision.
84. Cluster sampling helps to reduce cost of survey.
 85. Larger the cluster size, less efficient it is relative to the element as sampling
unit.
86. Two stage sampling is less efficient rather than that of single stage sampling.
87. A sampling procedure in which units are selected with chance of selection in
proportion to some measure of their size is known as PPS sampling.

SECTION B AND SECTION C

Q1.Comparison of stratified random sampling with simple random sampling without

replacement

We know that
 1 1
V ( yn )  (  ) S 2
n N
 1 1 k
and V ( yst ) p  (  ) pi Si 2
n N i 1

Now, to express S2 in terms of Si2

We know that

1 K Ni 
S2   i N )2
N  1 i 1 j 1
( y  Y

K Ni   
(N-1)S2=  ( yi  Y Ni  YNi  YN ) 2 after addition and subtraction of Y¯Ni
i 1 j 1

K Ni  K Ni  
=  (y  Y
i 1 j 1
i Ni )   ( YNi  YN ) 2
2

i 1 j 1
+ Third term

The third term is zero due to the algebraic property of arithmetic mean.
K K  
( N  1) S 2   (Ni  1) Si 2   Ni ( YNi  YN ) 2 ……………(1)
i 1 i 1

We assume that Ni’s are large such that

Ni  1 N  1
 1
Ni N
Dividing equation (1) by N, we have

N  1 2 K ( Ni  1) 2 K Ni  
( )S   Si   ( y Ni  Y N ) 2
N i 1 N i 1 N

K
( Ni  1) 2 K Ni  
S2   Si   ( y Ni  Y N ) 2
i 1 N i 1 N

1 1
Multiplying both sides by (  ) we have
n N

1 1 1 1 K 1 1 K  
(  ) S 2  (  ) pi Si 2  (  ) pi ( y Ni  YNi ) 2
n N n N i 1 n N i 1

It implies that V random =V prop + positive term

Certainly, there is some gain in precision due to stratification. A gain can be increased
by making the stratum means difference among themelves.

If we see the difference between V prop - V Ney= a positive term

Therefore, Neyman allocation is certainly more efficient than that of proportional

allocation.

If we combine these two results, then we have V random - V Ney= a positive term

Ultimately, we can say that

V Ney ≤V prop ≤ V random

Hence proved.

Q2. Define ratio estimator for a population mean of character Y giving its conditions.
Write the expression for its bias and variance. Find the condition under which ratio
estimator is more efficient than usual simple mean estimator.

we have considered only estimates based on simple arithmetic means of the observed
values in the sample. We shall consider other methods of estimation which make use
of the auxiliary /ancillary information and which, under certain conditions, give more
reliable estimates of the population values than those based on the simple averages.
One of these methods is of particular importance. It is called the ratio method of
estimation.
    
Ratio estimator is defined as y R  R X N where R is ratio of sample mean of y and

 
yn
sample mean of x i.e. R  
and X N is the population mean of auxiliary characteristics
xn
X.

Conditions under Ratio Method of Estimation:

(i) The regression line of Y on X should be linear.

(ii) The regression lines passes through the point (0,0) i.e. origin.

Expression of bias in ratio estimator:

   
Bias of ratio estimator= E ( y R )  YN which may be simplified as Cov( R, xn )

N n
= [C X 2  2  C X CY ] where Cx and Cy are the coefficient of variation of auxiliary
Nn
variable and study variable respectively.

Expression of variance in ratio estimator:

N n  N n 2
= [CY 2  C X 2  2  C X CY ] = V ( y R )  [S y   2 S x 2  2 S x S y ]
Nn Nn

Comparison of Ratio estimator with simple mean estimator:

Here, we will compare ratio estimator with simple mean estimator. For this we require
the variance of ratio estimator and variance of sample mean estimator in SRSWOR. We
know that variance of ratio estimator is

N n 2

V ( yR ) [ S y   2 S x 2  2  S x S y ] and variance of sample mean in the case of
Nn
SRSWOR is
 N n 2
V ( yn )  Sy .
Nn

The efficiency of ratio estimator to sample mean estimator is equal to

1 1

/ 
which is equal to 1+ ρ2[Sx/Sy]2-2RρSx/Sy. It implies that the ratio
V ( y R ) V ( yn )
estimator is more efficient in comparison to sample mean estimator if ρ>1/2 [Cx/Cy]
where Cx and Cy are the coefficient of variation of auxiliary variable and study variable
respectively.

Q3.Derive the condition when regression estimator and ratio estimator are equally
efficient.

Ans,

The regression estimator will be more efficient than the ratio estimator provided
 
MSE ( ylr )  MSE ( yR )

i.e. S y 2 (1   2 )  S y 2  R 2 S x 2  2  RS x S y

Sy
or (  S y  RS x ) 2  0 which is always true. R= (  S y  RS x ) 2  0  R  in which both
Sx
regression and ratio estimator are equally efficient.

Q4. In SRSWOR, the probability of drawing any specified unit at rth draw is equal to the
probability of drawing it at the first draw.

Proof: In order to prove this property two probability statements will be multiplied
together since they are mutually exclusive.

(i) The unit is selected at the previous (r-1)th draw, and

(ii) The unit is selected at the rth draw.

Let us suppose that there are N units in the population. The probability of selecting a
unit at the first draw is 1/N and the probability of its not selection at this draw is 1-
(1/N)= (N-1)/N. Similarly, the probability of not selecting the unit at the second draw
is (N-2)/(N-1)

In the same way, the probability of not selecting the unit at the third draw is (N-3)/(N-
2).

In general, above statements say that the probability of not selecting the unit at the
(r-1)th draw is (N-r+1)/(N-r+2).

The second statement was that the unit should be selected at the rth draw.

If we proceed in the same way, we can deduce this probability. The probability of
selecting a unit at the first draw is 1/N. The probability of selecting a unit at the
second draw is 1/(N-1). Similarly, the probability of selecting a unit at the third draw
is 1/(N-2).
 Then, the probability of selecting a unit at the rth draw is 1/(N-r+1).

Multiplying the above two statements we have,

( N  1) ( N  2) ( N  3) ( N  r  1) 1
......... which is equal to 1/N
N ( N  1) ( N  2) ( N  r  2) ( N  r  1)

It shows that in SRSWOR, the probability of drawing any specified unit at rth draw is
equal to the probability of drawing it at the first draw.

Hence proved.

Q5.(i) Estimator and estimate;

Estimator is a function of sample values which can be used for estimating the
population parameter. It is also called statistic. For example sample mean
and sample mean square are always based on sample values and they are
able to estimate population mean(µ) and population mean square(S 2).
Estimate is a specific value of the estimator which is always fixed for an
estimator.
(i) Lahiri’s method in PPS sampling:

(a) Let M=max Xi i.e. maximum of the sizes of N units in the population or some
convenient number greater than M. We can write the following steps in a nutshell
in order to select our desire samples:
(b) Select a pair of random number (i, j) such that 1≤i≤N and 1≤j≤M
(c) If j ≤ Xi then ith unit is selected otherwise rejected and another pair of random
number is chosen.
(d)To get a sample of size n, this procedure is repeated till n units are selected.
(iii) In inverse sampling (sometimes called standard inverse sampling), you continue to
choose items until an event has occurred a specified number of times. It is often
used when you don’t know the exact size of the sample you want to take. For example,
let’s say you were conducting a wildlife management survey and wanted to capture
20 banded birds. Inverse sampling is often performed when a certain characteristic
is rare. For example, it is a good method for detecting differences between two different
treatments for a rare disease.

(iv) Double sampling:

Double sampling is a two-phase method of sampling for an experiment,
research project, or inspection. An initial sampling run is followed by
 preliminary analysis, after which another sample is taken and more analysis is
run.
It is used in three main ways: acceptance/rejection double sampling, ratio
double sampling, and stratification double sampling.
Used primarily in industrial sampling inspection, this is the most simple type of
two phase sampling.

For example, if you were studying how many books the average city dweller read
in a week, you might stratify the population by college graduates and less educated
people. If an initial quick survey told you that, in 100 residents, 60 were college
graduates and 40 were not, you could decide to sample college graduates and non-
graduates at a 6/4 ratio during phase two of your study.

(v) Optimum allocation:

It is an allocation based on cost function involved in conducting a survey for various

items of the data. It is related to the stratified random sampling which is one of the more
important allocation of choosing samples from heterogeneous population. If the cost of
surveying is equal for all the items in the survey, it is called as Neyman allocation.
Optimum allocation is more efficient rather than that of equal and proportional
allocation.

Q6. Define simple random sampling. Differentiate between SRSWOR and SRSWR.

Ans. The simplest of the methods of probability sampling which is usually called the
method of random sampling. In this method, an equal probability of selection is
assigned to each available units of the population at the first and each subsequent
draw. Thus, if the number of units in the population is N, then the probability of selection
of any unit at first draw is 1/N and at the second draw is 1/N-1 etc, which are ultimately
equal to 1/N. The sample obtained using the above method is called “ Simple Random
Sampling”. Since this result is independent of the specified unit it follows that every one
of the units in the population has the same chance of being included in the sample
under the procedure of simple random sampling.

Difference Between SRSWOR and SRSWR:

(i) If the selected units are not being replaced back in the population before the
second draw, it is called SRSWOR and if the selected units are being
replaced back in the population before the second draw, it is called SRSWR
(ii) In SRSWOR, at each draw ,new information on the units will be generated
while it may be possible to have the same kind of information on the units in
SRSWR.
 (iii) SRSWOR method will cover the whole population units while it is not true in
the case of SRSWR.
(iv) The variance of sample mean in SRSWOR is found to be smaller rather than
that of SRSWR providing more efficiency in SRSWOR.

Q7. Define systematic sampling with four examples.

we have considered methods of sampling in which the successive units

(whether elements or clusters) were selected with the help of random
numbers. We shall now consider a method of sampling in which only the first
unit is selected with the help of random numbers, the rest being selected
automatically according to a predetermined pattern. The method is known as
systematic sampling. The pattern usually followed in selecting a systematic
sample is a simple pattern involving regular spacing of units. Thus, suppose a
population consists of N units, serially numbered from 1 to N. Suppose further
that N is expressible as a product of two integers k and II, so that N = kn.
Draw a random number less than k, say i, and select the unit with the
corresponding serial number and every k-th unit in the population thereafter.
Clearly, the sample will contain the n units i, i + k, i + 2k, ... ,( i + n -1)k, and is
known as a systematic sample. The selection of every k-th strip in forest
sampling for the estimation of timber, the selection of a corn field, every k-th
mile apart, for observation on incidence of borers, the selection of every k-th
time-interval for observing the number of fishing craft landing on the coast,
the selection of every k-th punched card for advance tabulation ,after the first
unit is chosen with the help of random numbers less than k, are all examples
of systematic sampling.

Q8. Sampling and Non –Sampling errors:

Sampling error:

If complete accuracy can be ensured in the procedures such as

determination, identification and observation of sample units and the
tabulation of collected data, then the total error would consist only of the error
due to sampling, termed as sampling error. Measure of sampling error is
mean squared error (MSE). The MSE is the difference between the estimator
and the true value and has two components: - square of sampling bias. -
sampling variance.
 Non Sampling Errors

It is a general assumption in the sampling theory that the true value of each
unit in the population can be obtained and tabulated without any errors. In
practice, this assumption may be violated due to several reasons and
practical constraints. This results in errors in the observations as well as in
the tabulation. Such errors which are due to the factors other than sampling
are called non-sampling errors.

Sources of non-sampling errors:

Non sampling errors can occur at every stage of planning and execution of
survey or census. It occurs at planning stage, field work stage as well as at
tabulation and computation stage. The main sources of the non-sampling
errors are
(i) lack of proper specification of the domain of study and scope of
investigation,
(ii) incomplete coverage of the population or sample,
(iii) faulty definition,
(iv) defective methods of data collection and
(v) tabulation errors.

More specifically, one or more of the following reasons may give rise to nonsampling
errors or indicate its presence

(i) The data specification may be inadequate and inconsistent with the objectives
of the survey or census
(ii) Due to imprecise definition of the boundaries of area units, incomplete or
wrong identification of units, faulty methods of enumeration etc, the data may
be duplicated or may be omitted.
(iii) The methods of interview and observation collection may be inaccurate or
inappropriate.
(iv) The questionnaire, definitions and instructions may be ambiguous.
(v) The investigators may be inexperienced or not trained properly.
(vi) The recall errors may pose difficulty in reporting the true data.
(vii) The scrutiny of data is not adequate.
(viii) The coding, tabulation etc. of the data may be erroneous.
(ix) There can be errors in presenting and printing the tabulated results, graphs
etc.
(x) In a sample survey, the non-sampling errors arise due to defective frames
and faulty selection of sampling units.
Q9. Explain the situations when cluster sampling is used.

A sampling procedure, pre-supposes the division of the population into a finite

number of distinct and identifiable units called the sampling units. Thus a
population of fields under wheat in a given region might be regarded as
composed of fields or groups of fields on the same holdings, villages, or other
suitable segments. A human population might similarly be regarded as
composed of individual persons, families, or groups of persons residing in
houses and villages. The smallest units into which the population can be
divided are called the elements of the population, and groups of elements the
clusters. When the sampling unit is a cluster, the procedure of sampling is
called cluster sampling. When the entire area containing the population under
study is subdivided into smaller areas and each element in the population is
associated with one and only one such small area, the procedure is
alternatively called area sampling. For many types of population a list of
elements is not available and the use of an element as the sampling unit is
therefore not feasible. The method of cluster or area sampling is available in
such cases. Thus, in a city a list of all the houses is readily available, but that
of persons is rarely so. Again, lists of fields are not available, but those of
villages are. Cluster sampling is, therefore, widely practiced in sample
surveys. The size of the cluster to be employed in sample surveys therefore
requires consideration. In general, the smaller the cluster, the more accurate
will usually be the estimate of the population character for a given number of
elements in the sample. Cluster sampling is used for conducting a larger
survey when the sampling frame (List of the units in the population) is not
available.

Q10. Define PPS sampling. Explain about cumulative total method.

In simple random sampling without replacement, we assume that samples are selected
with equal probability for all the units in the population. If the units vary considerably in
size, SRSWOR may not be appropriate since it does not take in to account the possible
importance of the varying sizes of the units in the population. In fact, a larger unit for
variable Y may contribute more to the population total rather than the smaller units. For
example, villages having larger geographical areas are likely to have larger population
and larger areas under food crops. It is therefore natural to expect that a scheme of
selection which provide the probability of selection in the sample to larger units also
than to smaller units giving more efficient estimators in comparison to equal probability.
Such type of sampling which vary from probability to probability according to the size of
units. It is called “Probability Proportional to Size” (PPS ) sampling.
In the cumulative total method , the following steps are considered:

(i) A random number M is selected from” Random Number Table” from 1 to T N

where TN is the total of all units in the population.
(ii) If this random number M falls between T i-1 <M ≤ Ti, then ith unit is selected.
(iii) This process is repeated for n times for getting n samples from the lot.
(iv) It is to be noted that here if a random number gets repeated in the procedure,
the corresponding same unit is taken in the sample, then the selection of the
units is called sampling with replacement.

Q11. Explain the important points for planning and organization of a sample survey.

Write three –four lines in each of the following points:

(i) Objectives
(ii) Data to be gathered
(iii) Population under investigation
(iv) Sampling frame
(v) Methods of collecting data
(vi) Organization and supervision of field work
(vii) Tabulation of data
(viii) Analysis of data
(ix) Precision
(x) Writing reports and conclusions

Q12. Derive a relationship between mean square error, sampling variance and bias.

Let us suppose that  be the estimator of the parameter θ. Then mean square error is

defined as MSE= E (   ) 2


subtracting & adding E ( ) in the right hand side, we have

  
MSE= E[  E ( )  E ( )   ]2

Squaring R.H.S. taking two terms together , we have

     
MSE= E[  E ( )]2  E[( )   ]2  2 E[  E ( )(   ]

The third term in R.H.S. vanishes equal to zero,then

  
MSE= E[  E ( )]2  E[( )   ]2 , which is equal to
   
MSE= S.V. + Bias2 ,Since S.V.= E[  E ( )]2 and Bias= E[( )   ]2

Q13. . Define stratified random sampling and write the advantages of stratification. What
are the choice of sample sizes in different allocations?.Derive the expression of
variance of sample mean in proportional and Neyman allocations.

STRATIFIED SAMPLING (definition)

We know that the precision of a sample estimate of the population mean depends
upon two factors: (1) the size of the sample, and (2) the variability or heterogeneity of
the population. Apart from the size of the sample, therefore, the only way of increasing
the precision of an estimate is to devise sampling procedures which will effectively
reduce the heterogeneity. One such procedure is known as the procedure of stratified
random sampling. It consists in dividing the population of N units into K groups of sub
population of N1,N2,…Nk units respectively. These sub population are non-overlapping
and together they can comprise the whole of the population, so that N 1+N2+…Nk =N.
These sub-population are called strata and the single group is called stratum. The
sample size within the strata are denoted by n1,n2,…nk respectively such that
n1+n2+…nk=n. K is called the number of strata or groups.

If a simple random sample is taken in each stratum using SRSWOR, the whole
procedure is described as” Stratified Random Sampling”

Advantages of stratification:

(i) If the admissible error is given, a small sample should be taken so that our
expenditure may be reduced.
(ii) There is a reduction of error due to stratification, if the cost of survey is fixed.
(iii) Stratification provides the individual means of stratum and then for the whole
population.
(iv) Stratification may provide the administrative convenience.

Choice of Sample Sizes in Different Strata

There are four types of choice of sample sizes in different strata.

(i) Equal allocation: ni=n/k where n is the total sample size and k is the
number of strata.
(ii) Proportional allocation: ni=npi where pi= Ni/N stratum weight.
(iii) Optimum allocation: It involves cost per unit in the stratum
(iv) Neyman allocation: A particular case of optimum allocation when ci=c ,a
constant cost for all the units ni= npisi/∑ .
Derivation of the variance of sample mean in proportional and Neyman allocation:

 1 1 k 2 2
We know that V ( yst )  (  ) pi Si
ni N i i 1

Where pi=Ni/N stratum weight and Si2 is the population mean square.

Substituting ni=npi in the above expression, we have the variance of sample mean in
proportional allocation.

 1 1 k p i
2
Si 2
1 k
V ( yst ) p  (  ) pi 2 Si 2  ( i 1  p i
2
Si 2 ) which is written in simplified form
npi N i i 1 n N i 1

 1 1 k
as V ( yst ) p  (  ) pi Si 2 .
n N i 1

Similarly, substituting ni= npisi/∑ . in the above expression, we have the variance
of sample mean in Neyman allocation .

( pisi ) ( pisi )
2 2
 1 1 k 1 k 1 k
V ( yst ) Ney  (  ) pi 2 Si 2  [   pi 2 Si 2 ]  [  pS i i
2
]
ni Ni i 1 n Ni i 1 n N i 1

Q14. Procedure of Selecting a Random Sample

A practical procedure of selecting a random sample is by using a table of random

numbers, such as those published by Tippett (1927), a page from which is reproduced
in the Appendix to this chapter. The procedure takes the form of (a) identifying N units in
the population with the numbers 1 to N, or what is the same thing, preparing a list of
random numbers, and (c) taking for the sample the n units whose numbers correspond
to those drawn from the table of random numbers. The following examples will illustrate
the procedure: units in the population and serially numbering them, (b) selecting
different numbers from the table

Example:

Select a sample of 34 villages from a list of 338 villages.

Using the three-figure numbers given in columns 1 to 3, 4 to 6, etc., of the table given
in the Appendix and rejecting numbers greater than 338 (and also the number 000), we
have for the sample:
 125, 326, 12, 237, 35, 251, 165, 131, 198, 33, 161, 209, 51, 52, 331, 218, 337, 263,
223, 241, 277, 42, 14, 303, 40, 99, 102, 173, 137, 321, 335, 155, 163, 81.

The procedure involves the rejection of a large number of random numbers, nearly two-
thirds. A device commonly employed to avoid the rejection of such large numbers is to
divide a random number by 338 and take the remainder as equivalent to the
corresponding serial number between 1 to 337, the remainder zero corresponding to
338. It is, however, necessary to reject random numbers 677 to 999 and also 000 in
adopting this procedure as otherwise villages with serial numbers 1 to 323 will get a
larger chance of selection equal to 3/999 while those with serial numbers 324 to 338 will
get a chance equal to 2/999. If we use this procedure and also the same three-figure
random numbers as given in columns 1 to 3, 4 to 6, etc., we will obtain the sample of
villages with serial numbers given below:

125, 206, 326, 193, 12, 237, 35, 251, 325, 338, 114, 231, 78, 112, 126, 330, 312, 165,
131, 198, 33, 161, 209, 51, 52, 331, 218, 337, 238, 323, 263, 90,

Q15. BASIC IDEAS IN SAMPLING

A sampling method is a method of selecting a fraction of the population in a way that

the selected sample represents the population. Everyone of us has had occasion to use
it. It is almost instinctive for a person to examine a few articles, preferably from different
parts of a lot, before he or she decides to buy it. No particular attention is, however, paid
to the method of choosing articles for examination. A wholesale buyer, on the other
hand, has to be careful in selecting articles for examination as it is important for him to
ensure that the sample of articles selected for examination is typical of the
manufactured product lest he should incur in the long run a heavy loss through wrong
decision. Similarly, in obtaining information on the average yield of a crop by sampling,
it is not sufficient to ensure that the fields to be included in the sample come from
different parts of the country, for, the sample may well contain a very much larger (or
smaller) proportion of fields of a particular category like irrigated, manured or growing
improved variety, than is present in the population. If any category is consistently
favoured at the expense of the other, the sample will cease to represent the whole.
Even if the sample is selected in such a way that the proportions in the sample under
different categories agree with those in the population, the sample may not still
represent the population. A sampling method, if it is to provide a sample representative
of the population, must be such that all characteristics of the population, including that
of variability among units of the population, are reflected in the sample as closely as the
size of the sample will permit, so that reliable estimates of the population characters can
be formed from the sample.
Q16.Advantages of sampling.

(i) There is a reduction of cost either in terms of money or in terms of man hours.
Although the cost per individual may be larger in a sample survey but the total
cost that is expected to be smaller.
(ii) There is a greater scope in a sample survey rather than that of complete
enumeration. We should employ highly trained field staff for collection of data.
(iii) There is a chance of getting better quality data rather than that of complete
enumeration because in sample survey we gather information after probing.
(iv) Accuracy and efficiency is increased along with greater speed and more
importantly the quantification of uncertainty i.e.errors.

Q17. STRATIFIED SAMPLING

We have seen that the precision of a sample estimate of the population mean depends
upon two factors: (1) the size of the sample, and (2) the variability or heterogeneity of
the population. Apart from the size of the sample, therefore, the only way of increasing
the precision of an estimate is to devise sampling procedures which will effectively
reduce the heterogeneity. One such procedure is known as the procedure of stratified
sampling. It consists in dividing the population into k classes and drawing random
samples of known sizes, one each from the different classes. The classes into which the
population is divided are called the strata and the process is termed the procedure of
stratified sampling as distinct from the procedure considered in the previous chapters,
called unrestricted or un-stratified sampling. An example of stratified sampling is
furnished by the survey for estimating the average yield of a crop per acre in which
administrative areas are taken as the strata and random samples of predetermined
numbers of fields are selected from each of the several strata. The geographical
proximity of fields within a stratum makes it more homogeneous than the entire
population and thus helps to increase the precision of the estimate. In this chapter we
shall consider the theory applicable to the procedure of stratified sampling. Stratified
sampling is a common procedure in sample surveys. The procedure ensures any
desired representation in the sample of all the strata in the population. In un-stratified
sampling, on the other hand, adequate representation of all the strata cannot always be
ensured and indeed a sample may be so distributed among the different strata that
certain strata may be over-represented and others under-represented. The procedure of
stratified sampling is thus intended to give a better cross-section of the population than
that of un-stratified sampling. It follows that one would expect the precision of the
estimated character to be higher in stratified than in un-stratified sampling. Stratified
sampling also serves other purposes. The selection of sampling units, the location and
enumeration of the selected units and the distribution and supervision of field work are
all simplified in stratified sampling. Of course, stratified sampling presupposes the
knowledge of the strata sizes, i.e., the total number of sampling units in each stratum
and the availability of the frame for the selection of the sample from each stratum. It is
not necessary that the strata be formed of geographically continuous administrative
areas. Thus, in yield surveys, the fields may be stratified according as they are irrigated
or un-irrigated and separate samples selected from each. In a survey for estimating the
acreage under crops, strata may be formed by classifying the villages according to their
geographical area instead of on the basis of geographical proximity. The principles to be
followed in stratifying a population will become clear in the subsequent sections.

Choice of Sample Sizes in Different Strata

There are four types of choice of sample sizes in different strata.

(i) Equal allocation: ni=n/k where nis the total sample size and k is the number
of strata.
(ii) Proportional allocation: ni=npi where pi= Ni/N stratum weight.
(iii) Optimum allocation: It involves cost per unit in the stratum
(iv) Neyman allocation: A particular case of optimum allocation when ci=c ,a
constant cost for all the units ni= npisi/∑ .
(v) From above allocations, it can be concluded that (i) the larger the size of the
stratum, the larger should be the size of the sample to be selected there from;
(ii) the larger the variability within a stratum, the larger should be the size of
the sample from that stratum; and
(vi) (iii) the cheaper the labour in a stratum, the larger the sample from that
stratum.

Q18. Selection of the Samples with Probability Proportional to Size

In simple random sampling the chance of each selecting units are equal. But often
units may vary in sizes. If simple random sampling is being used in this case the
expected result or desired information may not get. Under this circumstance, such
subordinate information can be utilized in selecting the sample so as to more precise
estimators of the population parameters. The probabilities of selecting samples to
different units depend on their sizes. The probability of selection may be assigned to
the proportional of the sizes are known to us probability proportion to sampling size
(PPS).
Let consider y is the variable under study. We are considering Y is the shopping malls
in a town and X is an auxiliary variables or number of workers who works in these
factories. The most commonly used varying probability scheme. The shopping malls
are selected with proportional to their number of workers. This term is known to us
Probability proportion to sampling (PPS).
 PPS Sampling Procedure with Replacement
We are discussing here the two methods to draw sample PPS with replacement.

The Cumulative Total Method

Let the size of the unit be Xi (i=1, 2, 3 …N). The total being X= . We associate the
numbers 1 to X1 with the first unit. Then we associate the numbers (X1+1) to
(X1+X2) with second unit, the numbers (X1+X2+1) to (X1+X2+X3) with third unit and
so on. A number K is being chosen at random from 1 to X. Now we observe the unit
with which this random number is concomitant is selected. Clearly the ith units are
being selected with probability proportional to Xi. It is clearly can be stated that if a
sample of size n is required, the procedure is repeated n times with replacement of
the units selected.

Cumulative Total Method on the Basis of Practical Example

Table 1. The data set of the workers in the shopping mall and their output.
Shopping mall no Number of workers Sell commodities Cumulative totals
sizes
1 2 30 C1=2
2 5 60 C2=2+5=7
3 11 12 C3=7+11=18
4 13 6 C4=18+13=31
5 7 8 C5=31+7=38
6 3 13 C6=38+3=41
7 9 4 C7=41+9=50
8 16 17 C8=50+16=66
9 6 13 C9=66+6=72
10 4 8 C10=72+4=76

We will describe the cumulative total method on the basis of example. We consider a
district Kushtia that contains 10 shopping malls and where following number of
workers work 2, 5, 11, 13,7,3,9,16,6 and 4. We select a sample of four shopping
malls with replacement method for knowing the life pattern of workers.
The first step of selecting factories is to form cumulative totals. So we will determine
the cumulative totals in order to compare with selected random numbers.

Selection of Sample Using Cumulative Total Method

It is necessary to bear in mind that that To select 4 shopping malls, we have to
choose such random numbers from 1 to 76 that don’t exceed 76.
 First draw:
Suppose we select a random number k=23. It can be seen from the cumulative total
that this selected random number associated with C4. So unit Y is selected and
Y4=6.
Second Draw:
Now further we select a random number k=38. It can be easily seen from above table
that it associated with C5. So clearly the shopping mall Y5=8 is being selected
corresponding to the random number 38.
Third and fourth Draw:
Similarly, we select two more random numbers from random table. Suppose these
numbers are 15 and 45. Then the selected shopping malls are c3=12 and c6=13
corresponding to its random numbers. Now we could make analysis or our desire
work on these selected 4 shopping malls.
So at length we may deduce from here that the probability of selecting unit in this
procedure is proportional to its size.
Drawback:
The major draw backs of cumulative total is finding successive cumulative total must
be found. This is time consuming process and wearisome.
This problem could be solved by the help of Lahiri’s method.

Lahiri’s Method
Let M=max Xi i.e. maximum of the sizes of N units in the population or some
convenient number greater than M. We can write the following steps in a nutshell in
order to select our desire samples:
Select a pair of random number (i, j) such that 1≤i≤N and 1≤j≤M
If j ≤ Xi then ith unit is selected otherwise rejected and another pair of random
number is chosen.
To get a sample of size n, this procedure is repeated till n units are selected.

Lahiri’s Method on the Basis of Practical Example

We’ll describe the Lahiri’s method on the basis of practical instance. This is an
alternative procedure in which cumulation is avoided wholly. We consider the
previous example of shopping malls of Kushtia district. We select a sample of four
shopping malls by Lahiri’s method of PPS.
Table 2. The data set of the workers in the shopping mall and their output without
cumulative total.
Shopping mall no Number of workers Sell commodities Cumulative total are avoided.
1 2 30
2 5 60
3 11 12
4 13 6
5 7 8
6 3 13
7 9 4
8 16 17
9 6 13
10 4 8

Selection of Sample Using Lahiri’s Method

The given number of shopping malls N=10, First of all we select the random number
between 1-N. It means we have to choose such random numbers those are less than
10. Suppose 3 is selected. It is noting down that unit with corresponding serial
number provisionally selected. We select another random number between 1 to M
where M= Max Xi=16. Suppose our second random number 7 is selected. Now, if the
second random number that we have selected smaller than the size of unit
provisionally selected. Then the unit is selected into the sample. If not then entire
procedure will be repeated until is finally selected. We are considering these selected
random numbers into table.

The table of selected and rejected shopping mall.

1st random 2nd random
Observation Selection Shopping malls
numbers 1≤i≤10 numbers 1≤j≤16
3rd number shopping malls
3 7 j=7
selected
6 13 J=13>X6=3 rejected
4rth number shopping mall
4 5 j=5
selected.
2 3 J=3 2nd number selected
9 7 J=7>X9=6 Rejected
7 8 J=8 7th number shopping mall selected

The pairs (3,7),(4,5),(2,3),(7,8) are selected. Hence the samples will consist of the
shopping malls with serial number 3, 4, 2 and 7
 The sum and substance of this method is that we will repeat the procedure until our
desire samples are selected.
The basic difference between simple random sampling and varying probability
scheme:
In simple random sampling the probability of selecting unit at any drawn is the same.
But in varying probability scheme the probability of selecting any unit diff er from unit
to unit. It appears in PPS sampling that such procedure would give biased estimators
as the larger units are over-represented and the smaller units are under-represented
in the sample. This will happen in case of sample mean as an estimator of population
mean.

The Cumulative Total Method for without Replacement

For selecting a sample of size n without replacement, the first unit is selected by the
above cumulative total method and then it is deleted from the population and for the
reminder population new cumulative totals are calculated and again the same
procedure is used to select a second unit. The procedure is continued until a sample
of n units is obtained.
The procedure is illustrated in table. We consider the first example for explaining the
cumulative total method for without replacement. We will describe the cumulative total
method on the basis of example. We consider a district Kushtia that contains 10
shopping malls and where following number of workers work 2, 5, 11, 13,7,3,9,16,6
and 4. We select a sample of 2 factories without replacement method for knowing
their life pattern.
The first step of selecting factories is to form cumulative totals. So we will determine
the cumulative totals in order to compare with selected random numbers.
The data set of 10 numbers of workers in the shopping mall and their output.
Shopping mall no Number of workers sizes Sell commodities Cumulative totals
1 2 30 C1=2
2 5 60 C2=2+5=7
3 11 12 C3=7+11=18
4 13 6 C4=18+13=31
5 7 8 C5=31+7=38
6 3 13 C6=38+3=41
7 9 4 C7=41+9=50
8 16 17 C8=50+16=66
9 6 13 C9=66+6=72
10 4 8 C10=72+4=76
 Suppose we wish to draw a pps sample of 3 factories without replacement for the
selection of the first unit. We choose such random numbers that don’t exceed 76. We
select a random number k=37. We could see from the table it lies in 6 th no unit. So it
will be selected. Now we remove this unit and rearrange the shopping mall and
calculate the cumulative total.

The rearrangement of shopping mall and their output.

Shopping mall no Number of workers Sell commodities Cumulative totals
sizes
1 2 30 2
2 5 60 7
3 11 12 18
4 13 6 31
5 7 8 38
7 9 4 47
8 16 17 62
9 6 13 68
10 4 8 72
Now for the second draw we select another unit that doesn’t exceed 72. So we select
a random number, suppose it is 65 and lies in 9 th unit. So 9th unit will be selected then
again we remove this selected unit and rearrange the unit and calculate the
cumulative total.
The rearrangement of shopping mall and their output for the new unit.
Shopping mall no Number of workers Sell commodities Cumulative totals
sizes
1 2 30 2
2 5 60 7
3 11 12 18
4 13 6 31
5 7 8 38
7 9 4 47
8 16 17 62
10 4 8 66
Thus the PPS sample of 2 shopping malls without replacement consists of shopping
mall with serial numbers 6 and 9.
The drawback:
The main disadvantage of cumulative total method is that it involves writing down of
the cumulative total Ti which is tedious and time consuming.
The procedure of selection of a pps sample by using Lahiri’s method:
 For selection of PPS of n units without replacement the units selected in the earlier
draws are removed from the population. From the remaining units another PPS
sample of size one is taken as before and the selected unit removed from the
population. This procedure is repeated until n selections are made. In this case
Lahiris’s method is being used.

The Practical Example of Choosing Samples by Using Lahiri’s Method

We consider previous example of shopping mall. Suppose we wish to select 2
shopping malls with probability proportional to the number of workers in the shopping
malls without replacement.
We will select a pair of random numbers (i, j), (i≤10, j≤16). Where i are the number of
units (shopping malls) and j are the sizes(number of works in different shopping
malls). Now using the random number table we get the pair (3, 10). Since the number
of workers Xi for shopping mall 3 is greater than the second number 10 of the
selected random pair. So the 3rd shopping mall is selected in the sample. For
selecting the second unit by probability proportional to the number of workers in the
shopping malls, we prepare the following arrangement once again after deleting the
3rd shopping mall (unit).
As in pair (3, 10), a pair of random numbers (i, j), (i≤9, j≤16) has to be selected, using
the random number table , referring to the table of random numbers, the pair selected
is (4,6). As the size of the 4 unit in the above arrangement is greater than the second
number of pair random number, so the shopping mall 4 will be selected into the
sample.
The rearrangement of shopping mall and their output for Lahiri’s method.
Shopping mall no Number of workers Sell commodities Cumulative total are avoided.
1 2 30
2 5 60
3 13 6
4 7 8
5 3 13
6 9 4
7 16 17
8 6 13
9 4 8

Conclusion
It is known to us that Hansen and Hurwitz first introduced the use of probability
proportional to size (PPS) sampling; it goes without saying that a number of
procedures for selecting samples without replacement have been developed by the
help of statisticians. Survey statisticians have found probability proportional to size
(PPS) sampling scheme more useful for selecting units from the population as well as
estimating parameters of interest especially when it is clear that the survey is large in
size and involves multiple characteristics. So finally we could say that the selection of
samples are being done on the basis of its unit sizes.

Q19. Objective: Showing the unbiased estimator for population mean and biased
estimator for population mean square in SRSWR with the help of an example

Kinds of data: Consider a finite population of size N=4 including the values of sampling
units as ( 1,2,3,4). Enumerate all possible samples of size n=2 using SRSWR.

(i) Show that sample mean provides an unbiased estimator of population mean
(ii) Show that sample mean square does not provide an unbiased estimator of
population mean square.
 N 1 2
(iii) Show that V ( yn )  S is correct .
Nn
(iv) Compute sampling errors and show that their sum is equal to zero.

S.No. Possible Sample mean Sample mean Sampling error

samples yn square s2
1. 1,2 1.5 0.50 -1.0
2. 1,3 2.0 2.00 -0.5
3. 1,4 2.5 4.50 0.0
4. 2,3 2.5 0.50 0.0
5. 2,4 3.0 2.00 0.5
6. 3,4 3.5 0.50 1.0
7. 2,1 1.5 0.50 -1.0
8. 3,1 2.0 2.00 -0.5
9. 4,1 2.5 4.50 0.0
10, 3,2 2.5 0.50 0.0
11. 4,2 3.0 2.00 0.5
12. 4,3 3.5 0.50 1.0
13. 1,1 1.0 0.00 1.5
14. 2,2 2.0 0.00 -0.5
15. 3,3 3.0 0.00 0.5
16. 4,4 4.0 0.00 1.5
Total 40.0 20.00 0.0

E[yn] =40.0/16=2.5

Population mean= YN =(1+2+3+4)/4 = 2.5

 (i) It indicates that sample mean in SRSWR provides an unbiased estimator of
population mean i.e. sample mean is so strong and true that we can have true
information regarding population mean. Similarly

E[ s2] =20/16=1.25

1
And population mean square S2= (Yi  YN ) 2 =5/3=1.67
N 1

(ii) It shows that in SRSWR, sample mean square is not an unbiased estimator of
population mean square, because 1.25 is not equal to 1.67.
 N 1 2
(iii) V ( yn )  S , we will compute L.H.S. and R.H.S.
Nn

1 n  
L.H.S.= V ( yn ) =  i n
n i 1
( y  y ) 2
where y 
n is based on n =16 observations.

 10
Then, V ( yn ) = =5/8
16

4 1 5 5
R.H.S= x 
4.2 3 8
 N 1 2
It shows that V ( yn )  S is correct.
Nn

(iv) In the last column of the respective table, sampling errors are computed and
shown their sum is equal to zero.

Q20.

Objective: Showing the unbiased estimator for population mean and population mean
square in SRSWOR with the help of same example given above

Kinds of data: Consider a finite population of size N=4 including the values of sampling
units as ( 1,2,3,4). Enumerate all possible samples of size n=2 using SRSWOR

(i) Show that sample mean provides an unbiased estimator of population mean
(ii) Show that sample mean square provides an unbiased estimator of population
mean square.
 N n 2
(iii) Show that V ( yn )  S is correct
Nn
(iv) Compute sampling errors and show that their sum is equal to zero.
S.No. Possible Sample mean Sample mean Sampling error
samples yn square s2
1. 1,2 1.5 0.50 -1.0
2. 1,3 2.0 2.00 -0.5
3. 1,4 2.5 4.50 0.0
4. 2,3 2.5 0.50 0.0
5. 2,4 3.0 2.00 0.5
6. 3,4 3.5 0.50 1.0
Total 15.0 10.0 0.0

E[yn] =15.0/ 6=2.5

Population mean= YN =(1+2+3+4)/4 = 2.5

(i) It indicates that sample mean in SRSWOR provides an unbiased estimator of

population mean i.e. sample mean is so strong and true that we can have true
information regarding population mean. Similarly

E[ s2] =10/6=1.67

1
And population mean square S2= (Yi  YN ) 2 =5/3=1.67
N 1

(ii) It shows that in SRSWOR, sample mean square is an unbiased estimator of

population mean square, because 1.67 is equal to 1.67.
 N n 2
(iii) Show that V ( yn )  S is correct
Nn
we will compute L.H.S. and R.H.S.

1 n  
L.H.S.= V ( yn ) = 
n i 1
( yi  yn ) 2 where yn is based on n = 6 observations.


Then, V ( yn ) = 2.50/6=5/12

4  2 5 10 5
R.H.S= x  
4.2 3 24 12
 N n 2
It shows that V ( yn )  S is correct.
Nn
 (i) In the last column of the respective table, sampling errors are computed and
shown their sum is equal to zero.

Q21.Bias of the Ratio and regression estimator:

We define ratio estimator as

  
yR  R X N

         
Bias of ratio estimator = E ( y R )  Y N = E ( R X N )  Y N = E ( R) E ( x n )  E ( yn ) = Cov( R, xn )

 
Similarly, bias of regression estimator = E ( y l )  YN =

           
E ( y n )  E (  ( X N  x n ) YN  E ( y n )  E (  ) E ( x n )  E (  x n )  E ( y n )  Cov(  x n )

Q22. Sample size determination:

We use the following formula to determine sample size from a population:

z 2 p(1  P)
n where z is the value of Z-score at 95% or 99% confidence levels.
e2

p is proportion of events which are to appeared in the population in which we are

interested and (1-p) is the proportion of events not to be occurred in the population,

e is the margin of errors. That is 5% or 1%.

For example:

Suppose in a population the event in which we are interested consists of proportion as

0.5 and we choose 95% confidence levels, then sample size required for this population
is given below:

(1.96) 2 0.5(1  0.5)

n  384.16  384 to be needed from this population.
(0.05) 2

Procedure of selecting a random sample:

A practical procedure of selecting a random sample is done by using the random
numbers which are given in any Statistics Book.

The procedure takes the following form

 (i) Identifying N units in the population with the numbers 1,2,…N. or preparing a
list of the units in the population and serially numbered them.
(ii) Selecting different numbers from the table of random numbers
(iii) Taking for the sample the n units whose numbers correspond to those drawn
from the table of random numbers.

Example: Select a sample of 34 villages from a list of 338 villages.

Using the three digits number taking either from a row or a column, we have the 34
selected samples

125,326,12,237,35,251,165,131,198,33,161,209,51,52,331,218,337,283,223,241,27
7,42,14,303,40,99,102,173,137,321,335,155,163,81

The procedure involves the rejection of large number of random numbers, nearly two
thirds. A device is commonly used to avoid the rejection of such large
numbers(known as Remainder Method of Selection) is to divide a random
number by 338 and take the remainder as equivalent to the corresponding serial
number between 1 to 377, the remainder zero corresponding to 338.It is, however,
essential to reject random numbers 677 to 999 and also 000 in adopting this
procedure as otherwise villages with serial numbers 1 to 323 will get a larger chance
of selection.

Using the random number tables, the following samples have been made.

125,206,326,193,12,237,35,251,325,338,114,231,78,112,126,330,312,165,131,198,
33,161,209,51,52,331,218,337,238,323,263,90,11,223

Q23. Sampling for proportion and percentage:

We estimate the population mean or population total through considering samples of

appropriate size from a population. Sometimes, we are usually interested to estimate
the proportion or the number of units in the population possessing a particular attribute.
For example in a household survey to assess the extent of damage caused by a
hailstorm, earthquake, hurricane etc., it would be of more interest to know the number of
households that have damaged extensively on the proportion of households that have
spent more money on repairs. In others, one may estimate the number of persons
engaged in different occupations.

We know that sample proportion provides an unbiased estimator of population

proportion.

i.e. E(pi)= Pi
It has been shown with an example given below:

Let us consider a hypothetical population N=(1,2,3,4,5). Draw a sample proportion of

size n=( 3,5) using SRSWOR.

S.No. No.of possible samples Sample proportion pi

1 1,2 0
2 1,3 1/2
3 1,4 0
4 1,5 1/2
5 2,3 1/2
6 2,4 0
7 2,5 1/2
8 3,4 1/2
9 3,5 1
10 4,5 1/2
Total 4.0

Then,E(pi)=4.0/10=0.4

Population proportion Pi=[0+0+1+0+1]/5=2/5=0.4

It shows that sample proportion provides an unbiased estimator of population

proportion.

If the sampling is done using SRSWR, then, E(pi)=10.0/25=0.4 and Pi=0.4.Here also
sample proportion provides an unbiased estimator of population proportion.

Q24. Simple and Stratified Random Sampling ( For practical and Theory)

Simple Random Sampling(SRS): It is the process of selecting a sample from given population
according to some law of chance in which each unit of population has an equal and independent
chance of being included in the sample.
SRSWR(With Replacement): A selection process in which the unit selected at any draw is
replaced to the population before the next subsequent draw is known as Simple random sampling
with replacement. In this case the number of possible samples of size n selected from the
population of size N is . The samples selected through this method are not distinct.
SRSWOR(Without Replacement): A selection process in which the unit selected at any draw
is not replaced to the population before the next subsequent draw and the next sample is
selected from the remaining population is known as Simple random sampling without
replacement. In this case the number of possible samples of size n selected from the population
of size N is . The samples selected through this method are distinct.
Note: Sample mean is an unbiased estimate of population mean in SRSWR and SRSWOR,
whereas sample variance is an unbiased estimate of population variance in case of SRSWOR
only.
SRSWOR is more efficient than SRSWR because V( ) < V( ) .

Stratified Random Sampling: When the population is heterogeneous and we wish that every
section of population is represented in the sample. We divide the whole population into different
number of strata so that the one stratum is much different from one another whereas the samples
within each stratum are more homogeneous. This technique of selecting a representative sample
of whole population is known as stratified random sampling.
In stratified random sampling allocation of sample size to different strata is based on the
staratum sizes (Ni), the variability within the stratum Si2 and the cost of surveying per sampling
unit in the stratum.
Methods for allocation of sample size to different strata are
Equal Allocation : ni =

Proportional Allocation: ni =

Neyman Allocation: ni = ∗∑

Optimum Allocation (based on cost) : ni = ∗∑

Objective: In simple random sampling, show the sample mean and sample mean square is an
unbiased estimate of population mean and population mean square with the help of an
hypothetical population in SRSWOR and to determine its variances and S.E.
Kinds of data: The data relate to the hypothetical population whose units are 1, 2, 3, 4 and 5.
Draw a sample of size n=3 using SRSWOR.
Solution: Number of all possible samples of size n=3 under SRSWOR is given by =
5 =10.
∑
Compute the mean of each sample = and sample mean square = ∑( − ) .
 ∑
Similarly the mean of population = = =3 and population mean square =
∑( − )
S2 = [(1-3)2 + (2-3)2 +(3-3)2 + (4-3)2 + (5-3)2]= =2.5
The 10 possible samples are given below in the table.
S.No. Possible Sample mean Sample mean Sampling error
samples square (s2) ( − )

1. 1,2,3 2.0 1.0 -1.0

2. 2,3,4 3.0 1.0 0.0
3. 3,4,5 4.0 1.0 1.0
4. 4,5,1 3.33 4.33 0.33
5. 5,1,2 2.67 4.33 -0.33
6. 1,3,4 2.67 2.33 -0.33
7. 2,4,5 3.67 2.33 0.67
8. 3,5,1 3.0 4.00 0.0
9. 4,1,2 2.33 2.33 -0.67
10. 5,2,3 3.33 2.33 0.33
Total 30.0 24.98=25 0.00

Now we have to check whether E ( )= and E (s2) = S2 ,

∑ ∑
E ( )= = =3 = and E (s2)= = =2.5=S2,
then we can say that sample mean and sample variance s2 are an unbiased estimator of
population mean and population variance S2 respectively.
In order to find out the variance of sample mean in SRSWOR, we know that
V( )SRSWOR= S2 = ∗
*2.5 = 0.33

We can verify that this variance is correct.

∑( ( )) (∑ )
V( )= = [∑ – ] = [93.33-90]=0.33
This shows that V( ) is correct.
Standard Error of ( ) = V( ) = √0.33 =0.57
We can also compare the two variances, one in SRSWOR and the other in SRSWR.
V( )SRSWR= S2 = ∗ *2.5 = 0.66
Since V( ) < V( )
Hence we can say that SRSWOR is more efficient than SRSWR.
Objective: Showing the unbiased estimator for population mean and biased estimator for
population mean square in simple random sampling with replacement (SRSWR) with the help of
an hypothetical example and determination of its variance and standard error (S.E.)
Kind of data: Consider a finite population of size N=5 including the values of sampling units
as (1,2,3,4,5). Enumerate all possible samples of size n=2 using SRSWR. find the estimate of
V( ) in 9th sample.
Solution: Number of all possible samples of size n=3 under SRSWOR is given by = 5 =25.
∑
Compute the mean of each sample = and sample mean square = ∑( − ) .
∑
Similarly the mean of population = = =3 and population mean square =
∑( − )
S2 = [(1-3)2 + (2-3)2 +(3-3)2 + (4-3)2 + (5-3)2]= =2.5

S.No. Possible Sample Sample Sampling S.No. Possible Sample Sample Sampling
samples mean mean error samples mean mean error
square ( − ) square ( − )
(s2) (s2)
1 1,2 1.5 0.50 -1.5 13 4,1 2.5 4.50 -0.5
2 1,3 2.0 2.00 -1.0 14 5,1 3.0 8.00 0.0
3 1,4 2.5 4.50 -0.5 15 3,2 2.5 0.50 -0.5
4 1,5 3.0 8.00 0.0 16 4,2 3.0 2.00 0.0
5 2,3 2.5 0.50 -0.5 17 5,2 3.5 4.50 0.5
6 2,4 3.0 2.00 0.0 18 4,3 3.5 0.50 0.5
7 2,5 3.5 4.50 0.5 19 5,3 4.0 2.00 1.0
8 3,4 3.5 0.50 0.5 20 5,4 4.5 0.50 1.5
9 3,5 4.0 2.00 1.0 21 1,1 1.0 0.00 -2.0
10 4,5 4.5 0.50 1.5 22 2,2 2,0 0.00 - 1.0
11 2,1 1.5 0.50 -1.5 23 3,3 3.0 0.00 0.0
12 3,1 2.0 2.00 -1.0 24 4,4 4.0 0.00 1.0
25 5,5 5.0 0.00 2.0
Total 75.0 50.00

Now we have to check whether E ( )= and E (s2) = S2 ,

∑ ∑
E ( )= = =3 = and E (s2)= = =2 ≠S2,

then we can say that sample mean is an unbiased estimate of population mean whereas and
sample variance s2 is not an unbiased estimate of population variance S2 in case of SRSWR.
In order to find out the variance of sample mean in SRSWR, we known that

V( )= = S2 = ∗
*2.5 = 1.0

Standard Error of ( )= V( ) = √1 =1
In order to find the estimate of V( ) based on 9th sample, we have

V( )= = ∗
*2.0 = 0.8
Standard Error of ( )= V( ) = √0.80 =0.894
Objective : Drawing of samples in stratified random sampling under different allocation along
with determination of their variances and standard errors.
Kinds of data: A hypothetical population of N= 3000 is divided into four strata, their sizes of
population and standard deviations are given as follows :
Strata I II III IV
Size Ni 400 600 900 1100
SD Si 4 6 9 12
A stratified random sample of size 800 is to be selected from the population
Soultion : In case of
(i) Equal allocation the sizes of sample allocated to different strata will be the same. Hence the
different sample sizes will be ni = = = =200 samples from each
allocation.
(ii) In case of proportional allocation ni (i=1,2,3,4) is given by ni = npi where pi =Ni/N
ni =
∗
Hence n1 = =106.67≈107 samples from stratum I
∗
n2 = =160 samples from stratum II
∗
n3 = =240 samples from stratum III
∗
n4 = =293 samples from stratum IV
Thus, n1 + n2 + n3 + n4 = 800 constitute the samples required from all the strata.

(iii) The sample size in Neyman allocation is given by ni = ∗∑ = ∗∑

Here ∑ =400*4+600*6+900*9+1100*12= 26500

∗ ∗
Hence, n1 = 800 ∗ =48, n2== 800 ∗ =109,

∗ ∗
n3 = 800 ∗ =245, n4== 800 ∗ =398,

In Neyman allocation, the sample sizes from four strata are 48, 109, 245 and 398 which
constitute the required sample size.
∑ ∑
Variance of in equal allocation V( ) = − ,
from above data ∑ p S = 8.83, ∑pisi2= 86.43 and ∑pi2 si2 = 28.37
∗ . .
V( ) = − , =.141-.028= 0.1130
Standard Error of ( )= V( ) = √0.1130 =0.336
 Variance of in proportional allocation V( ) =( − ) ∑ =( -
)*86.43 =0.0792
Standard Error of ( )prop = V( ) = √0.0792 =0.2815
(∑ ) ∑ . .
Variance of in Neyman allocation V( ) = − = − =.068
Standard Error of ( )ney = V( ) = √. 068 = 0.262

6. Ratio and Regression Estimator

Ratio Estimator: Ratio method of estimation is based on the information available for
auxiliary variable. When the correlation coefficient between the study variable and the
auxiliary variable is positive and high, the ratio method of estimation can be used to study the
population parameters of study variable Y.
The equation of ratio estimator is given by = , where and are sample means of

y and x respectively and is population mean.

In case of ratio estimator sample mean is not an unbiased estimate of population mean. The
bias will be zero only when there is a perfect positive correlation between y and x.
The bias of ratio estimator to the first order of approximation is given by
( )
( )= ( − ) , where = and =
( )
The variance of ratio estimator is given by V ( ) = ( + − 2 ) where

R=

Regression Estimator: Ratio estimator is used if y and x are linearly related and the line of
regression between y and x are passes through origin. But when this is not the case and the
variate y is approximately a constant multiple of an auxiliary variate x, the regression
estimator is used.
The regression estimator can be defined as = + ( − )
Regression estimator is also a biased estimate of population mean.
( )
The variance of regression estimator is given by V( )= (1- ), here rxy =

(∑ ) ∑ ∑
and = where = [∑ - ] and = [∑ - ]
  Regression estimator is more efficient than Ratio Estimator V( ) < V( )
 If correlation coefficient is equal to zero , we should not apply regression estimator.

Objective : Estimation of the average number of bullocks per acre using ratio estimator and
show that it is a biased estimator of population mean. Compute bias and variance along with its
standard error.
Kinds of data : A bivariate population of size N=6 is given below :
No. of bullocks(Y) 3 4 8 9 6 9
Farm Size (acre)(X) 15 20 40 45 25 42

Enumerate all possible samples of size n=2 using SRSWOR.

Solution :Here it is given that N=6 and n=2.
The total number of possible samples of size n=2 is =6 =15
S.No. Possible Possible Sample Sample =
Samples Samples ( ) mean mean =
( )
1. 3,4 15,20 3.5 17.5 6.233 0.20
2. 3,8 15,40 5.5 27.5 6.233 0.20
3. 3,9 15,45 6 30 6.233 0.20
4. 3,6 15,25 4.5 20 7.013 0.225
5. 3,9 15,42 6 28.5 6.561 0.211
6. 4,8 20,40 6 30 6.233 0.20
7. 4,9 20,45 6.5 32.5 6.233 0.20
8. 4,6 20,25 5 22.5 6.930 0.222
9. 4,9 20,42 6.5 31 6.535 0.210
10. 8,9 40,45 8.5 42.5 6.233 0.20
11. 8,6 40,25 7 32.5 6.713 0.215
12. 8,9 40,42 8.5 41 6.461 0.207
13. 9,6 45,25 7.5 35 6.679 0.214
14. 9,9 45,42 9 43.5 6.448 0.207
15. 6,9 25,42 7.5 33.5 6.978 0.224
Total 467.5 97.716 3.135

∑ ∑
= = =31.17, = = =6.50
∑ .
E( )= = = 6.514,
Since E( ) ≠ , ℎ ratio estimator is not an unbiased estimator of population mean .
The bias of ratio estimator to the first order of approximation is given by
( )
( )= ( − ) , where = and =

Now, SX = ∗ (6639 − ) = 12.73 and SY =2.588,

 Cx= 0.408, Cy =0.397
To find out the value of correlation coefficient between X and Y, we have to make the
following values :
∑y=39, ∑x=187, ∑xy=1378, ∑x2=6639, ∑y2=287
∗
∗
= = 0.9859
( ) ∗ ( )
( )
Hence ( ) = ∗ * 6.50 * (0.408 − 0.9859 ∗ 0.408 ∗ 0.397)=0.014
.
The variance of ratio estimator is given by
( )
V( )= ( + − 2 ) where R = =0.208
( )
= ∗ (2.58 + 0.208 ∗ 12.732 -2*0.208*0.9859*12.73*2.58) =0.065 =0.0625.
The above formula of variance in terms of coefficient of variation can be written as :
( )
V( )= ( + − 2 )
= ( ∗ ) * 6.50 * (0.397 + 0.408 − 2 ∗ 0.9859 ∗ 0.408 ∗ 0.397 = .0660
Both values of variances of ratio estimator are approximately equal.
Standard Error of Ratio Estimator ( ) = V( ) = √0.0660 =0.256

Objective: Determination of the regression estimator, comparison with the ratio estimator, and
its sampling variance and standard errors.
Kinds of data: A bi-variate population of size N=85 with population mean = 6.55 and =
8.55, a random sample of size n=10 was drawn using SRSWOR scheme and was recorded as
Y 11 8 7 6 4 5 3 2 9 10
X 10 7 6 5 3 4 2 1 8 9

Solution: First we will calculate the , and

= = 6.5, = = 5.5, and = 8.55
The equation of regression estimator = + ( − ),
Where
(∑ ) ∑ ∑
= where = [∑ - ] and = [∑ - ]
Total
Y 11 8 7 6 4 5 3 2 9 10 65
X 10 7 6 5 3 4 2 1 8 9 55
YX 110 56 42 30 12 20 6 2 72 90 440
Y2 121 64 49 36 16 25 9 4 81 100 505
X2 100 49 36 25 9 16 4 1 64 81 385
.
By putting the values we get = (385- )= = 9.16, = 9.16 and =9.16
.
So the value of = = .
=1
 Now the equation of regression estimator = + ( − ) = 6.5 + 1*(6.55-5.5)r=6.55
Estimate of the sampling variance of the estimator
( ) .
V( )= (1- ), here rxy = = .
=1
( )
By putting the values we get V( )= ∗
* 9.16*(1-12) =0
Hence we can say that in case of perfect positive correlation the variance of linear regression
estimator for population mean will always be equal to zero.

The variance of ratio estimator is given by

( ) .
V( )= ( + − 2 ) where R = = .
=1.18
( )
V( ) = ∗
(9.16+1.18 *9.16-2*1.18*1*9.16)=.0262
The estimated sampling variance of the sample mean is given by

( ) ( )
V( )SRSWOR = = ∗
*9.16 = 0.808

Here V( )< V( ) < V( )SRSWOR

This shows that sample mean is less efficient than the ratio and regression estimator.

Q25. Stratified Random Sampling: Definition

Stratified random sampling is a type of probability sampling where the entire population is
divided into non-overlapping, homogeneous groups (strata) and randomly choose samples
from the various strata for research which reduces cost and improves efficiency. Members
in each of these groups should be distinct so that every member of all groups get equal
opportunity to be selected using SRSWOR. This sampling method is also called “restricted
sampling”.

Age, socioeconomic divisions, nationality, religion, educational achievements and other

such classifications fall under stratified random sampling.

Let’s consider a situation where a research team is seeking opinions about religion amongst
various age groups. Instead of collecting feedback from 326,044,985 U.S citizens, random
samples of around 10000 can be selected for research. These 10000 citizens can be
divided into strata according to age,i.e, groups of 18-29, 30-39, 40-49, 50-59, and 60 and
above. Each stratum will have distinct members.

Steps to select a stratified random sample:

1. Define the target population.
2. Recognize the stratification variable or variables and figure out the number of strata to be
used. These stratification variables should be in line with the objective of the research.
 Every additional information decides the stratification variables. For instance, if the
objective of research to understand all the subgroups, the variables will be related to the
subgroups and all the information regarding these subgroups will impact the
variables. Ideally, no more than 4-6 stratification variables and no more than 6 strata
should be used in a sample because an increase in stratification variables will increase
the chances of some variables canceling out the impact of other variables.
3. Use an already existent sampling frame or create a frame that’s inclusive of all the
information of the stratification variable for all the elements in the target audience.
4. Make changes after evaluating the sampling frame on the basis of lack of coverage,
over-coverage, or grouping.
5. Considering the entire population, each stratum should be unique and should cover
each and every member of the population. Within the stratum, the differences should
be minimum whereas each stratum should be extremely different from one another.
Each element of the population should belong to just one stratum.
6. Assign a random, unique number to each element.
7. The researcher can then select random elements from each stratum to form the
sample. Minimum one element must be chosen from each stratum so that there’s
representation from every stratum but if two elements from each stratum are
selected, to easily calculate the error margins of the calculation of collected data.

Types of Choice of Sample Sizes in Stratified Random Sampling:

 Equal allocation: ni=n/k

 Proportional allocation: ni = npi=( Ni / N ) * n

ni= Sample size for ith stratum

Ni= Population size for ith stratum

N = Size of entire population

n = Size of entire sample

Example: A hypothetical population of N=2000 is divided into four strata. Their sizes of
population and standard deviations are as under:
Strata I II III IV
Size(Ni) 300 400 600 700
S.D.(Si) 6 10 12 15
Sample 5 10 15 20
means(yni)
A stratified random sample of size 400 is to be chosen from the population using
SRSWOR.
(i) Equal allocation; ni=n/k, means 400/4=100 samples will be tken from
each stratum.
 (ii)Proportional allocation: ni=npi
n1=400x300/2000=60 samples from I stratum
n2=400x400/2000=80 samples from II stratum
n3=400x600/2000=120 samples from III stratum
n4=400x700/2000=140 samples from IV stratum
 Neyman allocation:ni=npisi/Σpisi

n1=400x(300/2000)x6/(241/20)=40 samples from I stratum
n2=400x(400/2000)x10/(241/20)=66 samples from II stratum
n3=400x(600/2000)x12/(241/20)=120 samples from III stratum
n4=400x(700/2000)x15/(241/20)=174 samples from IV stratum
In this way ,we have 400 sample from the entire population using three allocations.

Advantages of Stratified Random Sampling:

 Better accuracy in results in comparison to other probability sampling methods such
as cluster sampling, simple random sampling, and systematic sampling .
 Convenient to train a team to stratify a sample due to the exactness of the nature of this
sampling technique.
 Due to statistical accuracy of this method, smaller sample sizes can also retrieve highly
useful results for a researcher.
 This sampling technique covers maximum population as the researchers have complete
charge over the strata division.

When to use Stratified Random Sampling?

 Stratified random sampling is an extremely productive method of sampling in situations

where the researcher intends to focus only on specific strata from the
available population data.
 Researchers rely on this sampling method in cases where they intend to establish a
relationship between two or more different strata. If this comparison is conducted
using SRSWOR there is a higher likelihood of the target groups .
 Samples with a population which are difficult to access or contact, can be easily be
involved in the research process using the stratified random sampling technique.
 The accuracy of statistical results is higher than simple random sampling since the
elements of the sample and chosen from relevant strata. The diversification within the
strata will be much lesser than the diversification which exists in the target population.
Due to the accuracy involved, it is highly probable that the required sample size will be
much lesser and that will help researchers in saving time and efforts.

Q26. Show that the sample mean in stratified random sampling provides an unbiased
estimator of population mean.

We know that the sample mean in stratified random sampling

  K 
yst   pi yni
i 1

Taking expectation of both sides of above equation, we have

 K 
E ( yst )   pi E ( yni )
i 1

Since samples have been taken using SRSWOR from each stratum from stratified
 
random sampling, therefore E ( yni ) provides an unbiased estimator of Y Ni from ith
stratum.
 K  
Hence, E ( yni )   piYNi which is equal to YNi
i 1

It shows that sample mean in stratified random sampling provides an unbiased

estimator of population mean.

Hence proved.

Q27. Important Theorems in SRSWOR:

Hence proved. Theorem1: In SRSWOR, the probability of drawing any specified unit at
rth draw is equal to the probability of drawing it at the first draw.

Proof: In order to prove this property two probability statements will be multiplied
together since they are mutually exclusive.

(i) The unit is selected at the previous (r-1)th draw, and

(ii) The unit is selected at the rth draw.

Let us suppose that there are N units in the population. The probability of selecting a
unit at the first draw is 1/N and the probability of its not selection at this draw is 1-
(1/N)= (N-1)/N. Similarly, the probability of not selecting the unit at the second draw
is (N-2)/(N-1)

In the same way, the probability of not selecting the unit at the third draw is (N-3)/(N-
2).

In general, above statements say that the probability of not selecting the unit at the
(r-1)th draw is (N-r+1)/(N-r+2).

The second statement was that the unit should be selected at the rth draw.
 If we proceed in the same way, we can deduce this probability. The probability of
selecting a unit at the first draw is 1/N. The probability of selecting a unit at the
second draw is 1/(N-1). Similarly, the probability of selecting a unit at the third draw
is 1/(N-2).

Then, the probability of selecting a unit at the rth draw is 1/(N-r+1).

Multiplying the above two statements we have,

( N  1) ( N  2) ( N  3) ( N  r  1) 1
......... which is equal to 1/N
N ( N  1) ( N  2) ( N  r  2) ( N  r  1)

It shows that in SRSWOR, the probability of drawing any specified unit at rth draw is
equal to the probability of drawing it at the first draw.

Theorem 2: In SRSWOR , the probability of inclusion any specified unit in the sample is
equal to n/N.

Proof: In order to prove this property,we have to sum all the probability of selecting the
units. The probability of selecting the unit at the first draw ,then for the second draw ,
third draw and so on and so forth

1 1 1 1
Thus,    ……n samples which is equal to n/N.
N N N N

Hence proved.