UNIT 12 SIMPLE RANDOM SAMPLING AND

SYSTEMATIC SAMPLING

Structure Page No.

12.1 Introduction 5
Objectives
12.2 Sampling -What and Why? 6
12.3 Preliminaries 7
12.4 Simple Random Sampling 9
12.5 Estimation of Population Parameters 13
12.6 Systematic Sampling 18
Linear Systematic Sampling
Circular Systematic Sampling
Advantages and Limitations of Systematic Sampling.
12.7 Summary 21
12.8 Solutions/Answers 21
Appendix-A

12.1 INTRODUCTION

In our day to day routine we often have to make certain judgements about a large bulk
or population after studying a small portion, or a sample of it. For example, a house
wife tastes a spoonful of soup to see whether a little more salt is required before it is
served to guests; a quality control inspector inspects a sample from a tin of oil before
passing the whole tin as acceptable quality; or a doctor takes a few drops of blood
from a patient to decide if the patient has malarial infection. These are typical
examples where it is not practical to examine the entire lot or population and decision
making is done on the basis of sample information. Essentially this is sampling. It
saves time and if you have based your judgement on judicious observations, it serves
your purpose. Think of situations, when decisions are to be taken at macro levels, say,
national level. Sampling has become an effective tool for generating information for
policy formulation at different administrative levels. The above examples have the
special phenomenon that a spoon of soup, tinful of oil and the blood in the patient are
known to be pefiectly homogeneous material so that every part of the material
represents the material exactly. Often, however, we are not in this simple situation.
For example, suppose we are interested in knowing the average weight of an adult
Indian male. Obviously, it will not be satisfactory to measure just a few adult males,
as not all adult males are of the same weight. They show considerable
hetereogeneity. So how do we take a part of a large heterogeneous mass to draw
valid conclusions from it? However, careful considerations are needed for selection of
samples and making valid inferences from these samples. In the subject of 'sampling',
these considerations and criteria are developed on a scientific basis.

In this unit, we shall start by introducing the basic concepts of sampling. We shall
then discuss the simplest procedure for sample collection i.e., simple random sampling
(SRS). In the process you will get introduced to the concept of random numbers,
simple random sampling as a method of selection, estimation as well as considerations
for determination of sample size.

Further, the concept of systematic sampling, which is also a method of selection based
on random sampling procedure, will be introduced alongwith the estimation approach.
Objectives
After reading this unit, you should be able to
select samples using simple random sampling and systematic sampling
procedures;
estimate population mean, populatioTkvariance, proportion, alongwith their
efficiency in SRS;
select samples using systematic sampling by linear and circular systematic
procedures.

12.2 SAMPLING - WHAT AND WHY?

You have already learnt in Block 1, the concept of a statistical population. Often we
are interested in studying a specified characteristics of the individuals in a finite
population. For example, we may be interested in studying the annual income and
size of the households in Delhi for the year 2000-2001. In this case our population is
the collection of all households in Delhi during the year 2000-01 and the individuals
are the households. It may be enough for our purpose to find out the average size and
annual income of the house-holds in Delhi. How do we go about obtaining this
information? One way is to visit each household, note down the number of members
and find out from the members how much was their total income during the year
2000-01 and calculate a simple average of all the figures obtained. You can very well
imagine the difficulties and expenses involved in undertaking this huge task. The task
of enumerating all the households in Delhi, visiting them and noting their sizes and
annual income will be termed a census of the population under study. As you may all
be aware, such a census is held once in ten years and information is obtained on a
large number of characteristics including the ones mentioned above. Given this
complexity and the huge expenses involved, this method cannot be adopted every time
we need information on a population. Here are a few more instances where we may
need to collect information on a characteristic from all the individuals of a population:

i) A telephone company may be interested in figuring out the average

number of calls and average duration of a call made by the households in
a locality or a city
ii) A large fruit store may be interested in the quality of the truck load of
peaches packed in crates received at the stores from a farm
iii) A company producing electric bulbs may be interested in the average life
of bulbs produced by them during a shift.
-

El) In the cases mentioned above identify the population and the individuals.

You may notice that observing or inspecting all the individuals in a population can be
very expensive in terms of time and money. In certain instances, like observing the
life of bulbs, inspection may be destructive, so that observing all the bulbs of a
population for calculating their average life is a meaningless exercise. so an
alternative approach is called for. Can you think of an alternative? One alternative is
to observe or measure only some individuals of a population, but estimate the average
forthe whole population based on the few measurements made. The idea is roughly
as follows: Suppose you want to find the total number of mangoes on a tree. First, it is
easy to count the nuniber of mangoes on a "typical" branch. Multiply the number
obtained by the number of branches and you get an estimate. Do you think this
method will yield a good enough estimate? The method adopted in this approach is
called the metbod of sampling. You may realise that in this methodology your final
estimate of the total may depend on the number of fruits on the selected branch. Of
course, if all the branches happen to bear the same number of fruits, it does not matter Simple Random Sampling
which branch you happen to select. You may therefore want to know if all the
branches bear nearly the same number of fruits or, if they differ then by how many
they may differ. In order to do this, again using the principle of sampling, you may
select another branch to get an idea of the difference. If you select the branches with a
priori known probabilities, (for example, by ensuring that any set of two branches
have the same probability of being selected in the sample as any other set), you may
be able to use the theory of probability and statistical inference you have learnt in the
earlier blocks to calculate the likely error in your estimate. The theory and method of
sampling deal with the issue of how to select the sample, how to estimate the
population total or average and how to estimate the error in observing only a sample
instead of the entire population. The error that results from estimation based only on a
sample of observations is called sampling error. In contrast, there may also be
nonsampling errors while observing or measuring an individual. A census approach,
while free from the sampling error may suffer from nonsampling errors. In fact, in
measuring a large number of individuals, due to 'inspection fatigue' the nonsampling
error may be large. The theory of sampling may allow one to estimate the sampling
error but will not be able to assess the extent of nonsampling errors.

In order to have a better grasp of what we have discussed above, you may try this
exercise.

E2) List the advantages and disadvantages of using a sampling approach instead of
a census approach for studying a characteristic.

You may also recall at this stage that in Block-2, you learnt about sampling
distributions or the derived distribution of functions of a sample of observations. The
issues studied in that block are to be distinguished from the issues we will be
considering in this block. In Block-2, you had a conceptual infinite population, such
as the drying time required for a formulation of paint, and you had ten realizations of
the random variable 'drying time' on ten pieces of wood on which this formulation of
paint was applied. These ten observations were assumed to be ten independent
realizations of the same random variable having a theoretical distribution and you
derived the distribution of their average and other functions. The issues to be studied
in this block relate to sampling from a finite population and no theoretical distribution
of the characteristic is assumed.

Before proceeding further to discuss the methods of simple random sampling, we

shall introduce some concepts and definitions, which we shall be using frequently in
our discussicn.

12.3 PRELIMINARIES

Let us assume that we wish to find out the proportion of votes a particular party A is
expected to get in an election in a particular constituency.

An element is a unit for which information is sought. In this example the element is
a registered voter in the constituency. The study variable will be measured as one if
the voter prefers to vote for the party A, otherwise, the measurement will be taken as
zero.

The population as you already know, is an aggregate of elements about which the
inference is to be made; the collection of all the registered voters of the constituency
in this case constitutes the population. A population is finite if it consists of a finite
number of elements. In this unit we shall consider the case of finite population only.

For studying a population we select some of the elements or a collection of elements

(i.e. a sample) of the population on which observations are made. These elements are
called the sampling units. In our example, if households, which has got a number of
elements i.e., individual voters, are to be selected then households are sampling units.
Sampling units are non-overlapping collections of elements of the population.
For selection purposes, identity of sampling units is necessary. Usually, a list of
sampling units of the population provides such an identity. A complete list of
sampling units which represents the population to be covered is called a sampling
frame. The number of units in the sample is the sample size.
In the entire theory of sampling, the approach for selecting a representative sample
and making a good estimate from the sample is addressed. In the example considered
here, preference for the party is asked only from the registered voters selected in the
sample. This information is then used to determine the proportion of all votes that
party A is expected to get in the election. It is therefore necessary to exercise a great
deal of care in selecting sampling units for a sample survey. Other examples of results
based on sample surveys are quite common in practice. Whenever you read about
important figures like production of important crops, or average income of people in
rural or urban areas, it must be realised that such figures are invariably based on
results of well planned sample surveys.

If the units in the sample are selected using some random mechanism then such a
procedure is called random sampling or probability sampling. In this method
samples are selected according to certain laws of probqbility in which each unit of the
population has some definite probability of being selected in the sample. All other
sampling procedures, which are not based on random procedures but are based on
subjective judgement or convenience of the sampler, are known as non-random
sampling or non-probability sampling. They are also termed as purposive sampling
or judgement sampling. Clearly, inferences drawn on the basis of a purposive sample
can often be subjective and biased.
Random sampling is preferred over non-random sampling for a variety of reasons.
Besides eliminating the subjectivity in selection, it provides a measure of reliability
associated with the estimates developed from the samples Thus, one can make
inferences from the sample with a known level of confidence. As stated earlier, in
random sampling procedure, every unit in the population is assigned definite
probability of selection. The randomness associated with the sampling procedure is the
key to make valid inferences from the sample.

Samples are often selected by adoptinghe procedure of one after the other draw
procedure or unit by unit selection. If the units'selected at one draw are replaced in
the population before the next draw then the procedure is called with replacement
(WR) procedure. If the units are not replaced in the population and the selection is
made from the remaining units then the selection is called without replacement
(WOR).If a population consists of N units and a sample of size n is to be selected,
number of possible samples for with replacement procedure is Nn.In case of without

replacement sample, if the order of sample units is ignored then there are

possible samples.
(3
You may now try to solve the following exercises to see whether you have grasped the
basic concepts of sampling discussed above.
 Simple Random Sampling
83) Define population, sampling unit and sampling frame for conducting surveys
on each of the following subjects.
a) Measurement of the volume of timber available in a forest.
b) Annual yield of apple fruit in a hilly district.
c) Study of nutrient contents of food consumed by the residents in a city
E4) Consider a population consisting of 5 villages, the areas (in hectares) of
which are given below
Village A B C D E
Area 760 343 657 550 480
Enumerate all possible WOR samples of size 3. Also write the values of the study
variable (area) for the sampled units.

List all the WR samples of size 3 along with their area values.

Now, we consider the simplest of the random sampling procedures i.e., simple random
sampling.

12.4 SIMPLE RANDOM SAMPLING

If each sample among the all possible samples has the same chance of being selected,
then the associated method is called simple random sampling.

In simple random sampling procedure with unit by unit selection, every unit has got
equal chance (probability) of selection at every draw. However, the converse is not
true i.e. there are sampling schemes in which every unit gets the same chance of
selection but they are not simple random sampling methods e.g. the systematic
sampling. You shall learn about such sampling methods later in this unit. We now try
to answer the question which may be occurring in your mind.
How to select a simple random sample (SRS)?
We consider the selection of a simple random sample through unit by unit selection
method. At every draw equal probabilities fro be assigned to the available sampling
units of the population. Thus, a pre-requisite for the selection is a random device by
which selections are to be made. The most commonly used procedures for selecting a
SRS are (1) lottery method, (2) through the use of random number tables. Let us
discuss these methods one by one..
Lottery tnethod
As the name suggests, units from the given frame (of size N) are selected using any
procedure of generating a number randomly through lottery procedures. The simplest
method may be writing down N numbers on identical slips of papers and drawing one
of the slips after thoroughly mixing the slips. The number on the selected slip
indicates the unit selected. For instance, suppose we have a population of 400
individuals and we wish to draw a random sample of 40 individuals. We can number
the individuals of the population serially from 1 to 400. We can than take 400
identical slips of paper, write numbers 1 to 400 on them, put them in a box, mix them
thoroughly and pick out 40 slips, one by one without looking. This gives us a random
sample of 40 individuals. In with replacement procedure, the slip is replaced before
the next draw while in case of without replacement it is not replaced. The sampling is
continued till desired number of units are selected.
Any other randomization device such as pack of cards or random disc etc, may be
used. However, the procedure becomes cumbersome if large number of selections is to
be made as numbering of the slips become inconvenient and one has to be careful to
see that the slips are thoroughly mixed after each draw. This method is not so common
 in random selections. We now discuss another method which makes use of random
number tables.
Through random number tables
Before discussing the method based on the use of random number tables you may like
to know what random numbers are? Random numbers are numbers generated by a
random procedure involving repeated independent trials. Such numbers are generated
with the help of random digits 0 through 9. When we say random digits 0 through 9 it
is assumed that a trial of the procedure yields each of the ten digits with probability
0.1. One simple way of generating these random digits is to take ten cards of the same
size and write the digits 0 through 9 on the cards so that each card has a different digit.
Then take a large hat, say, toss in the cards and mix them well. Now choose a card at
random from the hat. Write down on a piece of paper the digit appearing on the card
you have chosen. Put the card back into the hat and mix the cards again. Repeat the
procedure by choosing a card at random, writing down the digit appearing on the card,
replacing, mixing, choosing again, and so on. The string of digits we write down
constitutes a string of random digits because it has been produced by a random device
supposed to yield each digit with probability 0.1 in independent trials. Random digits
can also be produced by using a modified roulette wheel in which the wheel is divided
into ten equal parts, each one corresponding to one of the ten digits.

Given random digits, we can get more complicated random numbers. Suppose we
have ge%erated the sequence 3217900597 of ten digits. Then each digit is random,
and also the two digit numbers 32, 17,90,05,97 obtained by taking the numbers two
at a time are random numbers because they have been produced by a random
procedure so that each of one hundred two-digit numbers 00 through 99 has the
probability 0.01 of appearing, and moreover selection of these two digit numbers are
independent. Taking the original ten digit sequence and choosing the two digits at a
time going backward to get 79,50,09,71,23 also gives random two-digit numbers. In
the same way you might think of some other ways to get two-digit numbers using the
generated string, as long as the method does not use the same selection more than
once.

E5) Give two more sequences of 5 two digit random numbers obtained by using
the string 3217900597 of ten digits.

In a similar manner, random numbers with three, four or even more digits can be
obtained by using the given string of random digits. There are several standard
random number tables available which give the arrangement of these numbers in a
rectangular manner. Some of these which are commonly used are prepared by
Tippett (1927), Fisher and Yates (1938), Kendall and Smith (1939), Rand Corporation
(1955), and Rao et al. (1974). One such random number tables is reproduced in
Appendix A. You may note that this random number table can be used as single
digited numbers or two digited numbers or, three or four digited numbers depending
on the size of the population you are sampling from. We shall now illustrate the use
. of random number tables for selecting samples.
We shall discuss here three commonly used methods of using random number tables
for selection of simple random samples.

Direct Approach. The first step in the method is to assign serial numb~rs1 to N to the
N population units. If the population size N is made up of K digit., then consider K
digit random numbers, either row wise or column wise, in the random number table.
The sample of required size is then selected by drawing, one by one, random numbers
from 1 to N, and including the units bearing these serial numbers in the sample.
 Simple Random Sampling
Problem 1: Consider a population of 56 households. Select a simple random sample
of 10 households by with replacement as well as without replacement methods.

Solution: Here the sampling unit is a household. The first step is to serially arrange
the households if they are not already arranged so. Since, the population size is a
number consisting of two digits we have to use two digited random number table.
Alternatively, in the table sf random numbers (Appendix A) the first two digits of any
column can be used randomly. While using these tables, it is advisable to take a blind
start on the table by placing your finger on the table with closed eyes - let it be
column 7, row 6 of the first page of Appendix A. Then the first number is 20 and
going down the page subsequent random numbers between 1 to 56 are 12,03 etc.

By selecting first 10 random numbers from 1 to 56, without discarding repetitions for
with replacement procedure (WR), we obtain the serial numbers of the households
in the sample. These are given below:

For without replacement procedure (WOR), repetitions have to be avoided and the
number 15, when appears again at the tenth draw, should be dropped. The next two
digits are then chosen. Thus, a WOR sample will consist of:
--
20 12 03 16 30 15 24 37 01 07

This procedure may involve number of rejections of random numbers, since zero and
all the numbers greater than 56 appearing in the table are not considered for selection.
The use of random numbers has, therefore, to be modified. We now discuss two of
the commonly used modified procedures,

Remainder Approach. In this method if the population size N is a K digit number,

then first we have to determine the highest K digit multiple of N. Let it be N'. Then a
random number r is selected, such that 1 S r IN'. This number r is then divided by N
and let the remainder be R. The unit bearing the serial number equal to the remainder
R, is then considered as selected. If remainder is zero, the last unit is selected. As an
illustration, let N=24. Here N is a two digit number. The highest two digit multiple of
24 is 96. Let us now choose a number between 1 and 96 say, 83. On dividing 83 by
24, we get the remainder as 11. Therefore, the unit bearing serial number 11 is
selected in the sample. Then another number between 1 and 96 is selected and the
process is repeated till the sample of required size is selected. As before, the repeated
selections of population units in the sample are permitted for W R sample, whereas
they are rejected and only distinct units selected for a WOR sample. With a little
variation in this method we consider another method.

Quotient Approach. As before, let N be a K digit number and N' be the highest K
digit multiple of N, such that N' = Nm for some integer m. Select a random number r
from 0 to N'-1. Then the unit having serial number (Q+l) is included in the sample,
where Q is the quotient when r is divided by m.For instance, if N=24 then N'=96 and
m=4. Let a random number 1=49 be chosen from 0 to 95. Then dividing 49 by 4 one
gets the quotient Q=12. The unit bearing serial number (Q+1)=13 is then serected in
the sample. The process is repeated by selecting each time a new number r from 0 to
95 till a sample of required size is obtained.

As we have mentioned earlier while using the random number tables, any starting
point can be used, and one can move in any predetermined direction along the rows or
columns. However, normally as a convention, column-wise selection is followed. If
Sampling
more than one sample is to be selected in any problem, each should have an
independent starting point.

Besides the methods discussed above, some more methods for sample selection are
available in the literature. However, being operationally inconvenient, these are
usually not employed in practice.

In order to get conversant with the methods discussed above for selecting a simple
random sample, you may try the following exercise. While doing this exercise you
will also get convinced that the number of rejections in quotient and remainder
approach are much less as compared to the direct approach.

E6) Select a simple random sample of 10 households from the same population of 56
households by WR and WOR methods, using remainder approach and quotient
approach.

The whole purpose of sampling is to collect information about the population from
which the sample is drawn. It is used to study the unknown characteristics in the
population called parameters. However, a sample cannot tell us about the population
parameters exactly, it can only estimate parameters of the population. In the next
section we shall see how these estimations are done.

12.5 ESTIMATION OF POPULATION PARAMETERS '

In order to infer about population parameters, we compute various quantities from the
sample. These computed quantities from the sample are called statistics. In general,
we can say that a statistic is any quantity computed from a sample. Values such as
mean, variance and standard deviations derived from samples are sample statistics
which are then used to estimate population parameters and hence are called
estimators.

Some of the important population parameters required to be estimated are population

mean, variance and proportion. When the population is of size N, comprising of units
with variate values Yl,YZ,....,YN,then population mean and variance are

2
- 1 N N
Y = - yi = N - ~ Y ,where Y = Total population and o2 = (yi - Y ) (1)
N i=1 i=l ,

respectively. The corresponding formulas for sample mean and variance are given
by

respectively for a sample of size n with variate values yl,yz, ...y,. . You may also
note here that a paramter is a fmed unknown quantity. For example, the average
height of the population of adult Indian males at a given time, has a single fixed value.

A statistic on the other hand is a variable quantity. The value of a statistic (or, an
estimator) computed from different samples would differ from sample to sample. We
know that the heights of adult Indian males vary. If two samples of the same size are
drawn from this population then it may happen that one sample has a few more taller
people than the other. Hence, the average height computed from one sample is likely
to be different from that computed from the other. Thus, there is a need to study this
variability in the statistic if it is to be used as an estimator. In other words, we can say Simple Random Sampling,
that there is a need to study the distribution known as sampling distribution of an
estimator.

The sampling distribution of an estimator helps in defining certain desirable criteria

for goodness of an estimator. One of the most important criteria is unbiasedness. The
estimator is said to be unbiased for the parameter t, if E(t) = t. where E(.) stands for
expectation. This expectation is computed by averaging the value o f t over all
possible samples. The criterion of unbiasedness ensures that on an average the
estimator will take value equal to the unknown population parameter t. We now
illustrate the concept of a sampling distribution through an example.

Problem 2 :Consider a simple random sample (WOR) of two households from a

population of five households having monthly income (in rupees) as follows :

Household 1 2 3 4 5

Income (rupees) 1560 1490 1660 ' 1640 1550

Enumerate all possible samples (WOR) of size 2 and show that the sample mean
gives an unbiased estimate of population mean.

Solution: Population mean Y = 1560 + 1490+ 1660

5
+ 1640 + 1550 ---7900 = 1580
5

All possible samples and their corresponding sample means in this case are presented
in Table-2.

Table 2 :All samples and their corresponding sample means in SRSWOR

Average 1 1580
Sampling It may be seen that the average of sample means (1580) is equal to the population
mean. This shows the unbiased nature of the sample mean as an estimator of
population mean.

In the same way unbiasedness can be shown in the case when all possible samples of
size 2 are drawn with replacement. We are leaving this for you to do it ~ourself.

E7) In Problem 2 above, enumerate all possible samples (WR) of size 2 and show
that the sample mean is an unbiased estimator of the population mean.

The sampling variance is the variance of thesampling distribution of the estimator. It

A

measures the divergence of the estimator from its expected value. If 8 is an estimator
of 0 then,

The positive square roof of sampling variance is termed standard error (SE).

Thus, standard error is the standard deviation of the sampling distribution. It measures
the precision of the estimator particularly in view of the fluctuations due to specific
sampling design.

In case of simple random sampling with replacement (SRSWR). Variance of ;

can be written as

and estimated variance is

V(Y)=;
sL
where, s2 = --x
1
n-li=l
(Yi --y)Z

For simple random sampling without replacement (SRSWOR)

-
variance of y is

and estimated variance is

v (Y)=(+ -N) 'where,x 's2 = n-1 .

n
(yi -7y
1=1

To have a better understanding of the above formulas, you may try the following
exercises.
 -,
Simple Random Sampling
E8) sippose we have a population of 5 students enrolled for statistics course and a
counsellor wants to find the average amount of time spent by each student in
preparing for classes each week. The amount of time (in hours) each student
spends per week is given by 7,3,6,10 and 4. If the counsellor takes a sample
of three students WOR, obtain the sampling distribution of the sample mean.
Compute the population mean and the mean and standard error of the
sampling distribution.

E 9) In the data of Example 2, show that the sample mean square (s2)is an
unbiased estimator of the population mean square (s2).

Estimation Of Proportion
Sometimes interest lies in estimating the population proportion. Examples, such as,
proportion of persons below poverty line or proportion of female members in a
particular group or proportion of persons getting degrees through distance education
etc. are very common. In all these examples the population is considered as divided in
two parts on the basis of an attribute.

For instance, a crop field may be irrigated or not irrigated. If it is irrigated, we say
that it possesses the characteristic 'irrigation'. If it is not irrigated, we say that it does
not posses the particular characteristic of irrigation. If we are interested in estimating
the proportion of irrigated fields, the population of N fields can be defined with variate
yi as

yi = 1, if the field is irrigated

= 0,otherwise

If the totql number of irrigated fields be N1 out of N then

N 1 N
Thus, Y = - Yi =I= P = proportion of irrigated fields.
Ni=l N

Thus, the problem of estimating a population proportion becomes that of estimating a

population mean by defining the variate as above. Now if a simple random sample of
size n is taken from the population and if nl units out of n possess that characteristic,
then sample proportion is given by

- 1 n "1
Thus, Y =; x ~ i,=P-
=
i=l

It follows that, in SRSWR as well as SRSWOR, p is an unbiased estimator of P. The

variances and estimator of variances in case of WR and WOR procedures are given as
follows.

For SRSWR

Variance of p is
Sampling

and estimated variance is

For SRSWOR

Variance of p is

and estimated variance is

We now take up an example to illustrate what we have discussed aboye.

Problem 3: Obtain the sampling dismbution of the sample proportion and the
standard error of the proportion of households having monthly income more than
Rs.1550 in the population given in Problem 2 by considering simple random sample
(WOR) of two households.

Solution: In the population of 5 households the income of household 1,3 and 4 is

more than Rs.1550. Population proportion of interest to us is P = 315.

Let us now work out the sampling distribution of the sample proportion by getting the
sample proportion of households with income exceeding Rs.1550 fiom each of 10
possible samples listed in Table 2. In each sample we score a household as 0 if its
income is Rs.1550 or less and as 1 if it exceeds Rs.1550. Then the mean score in each
sample shown in Table3 below gives the sample proportions.

Table-3: Computation of the sample proportions

 Simple Random Sampling
Standard error SE(p) of the proportion given by Formula (9) is

>(
Why don't you try this exercise now
E10) IT facility committee of IGNOU has a total of eight members whose ages in
years are 27,32,33,26,43,52,28 and 25. The committee has a rule which
requires a minimum age of 33 for a member to be the chairperson. Assume
that a simple random sample of size 4 is selected to provide an estimate of the
population proportion eligible to be chairperson. Find the mean and standard
deviation of the sampling distribution.

Sample size determination

While planning a sample survey, a decision has to be made regarding the sample size
in the initial stages. Having decided about the method of selection, one has to
determine about:the sample size in view of the resources available as well as the
desired level of precision of the estimators. Larger sample sizes will involve more cost
in data collection as well as data analysis while smaller sample sizes will reduce the
precision of estimate. Therefore, a balance has to be struck between cost and
precision, while deciding about the sample sizes. We consider a simple example to
explain the idea behind principles involved in determination of sample size.

An anthropologist wants to study the inhabitants of some island. He wishes to estimate

the percentage of inhabitants belonging to blood group 0.A simple random sample is
to be selected. How large should the sample be?

Some related questions are pertinent here. How accurately does the anthropologist
wish to know the percentage of people with blood group O? Suppose, he answers that
he will be contented if the percentage is correct within a tolerance limit d of _+ 5%. It is
also to be understood that even with this specification of tolerable limit of error it is
not possible to ensure that the estimates are obtained in this margin in 100% of cases.
A level of confidence has therefore to be attached with the estimates. Let confidence
level 1 - a be 95% associated with the estimates.

Let us assume that p is normally distributed about P. It will then lie in the range
(P & 20,) where op is the standard deviation of p, apart from a one in twenty chance
(i.e. apart from a probability of a ) .

In case of SRSWR, ap F.
- - Hence we can put

At this stage some idea about P is needed in order to determine the sample size n.
Fortunately, we do not need very accurate estimate of P for this purpose. In fact, if P
 lies between 0.3 to 0.5 then, the determined sample size lies between 336 to 400. To
be on the safe side, 400 may be taken as the initial estimate of n.
Note that the maximum value of PQ with 0 5 P I 1, Q = 1-P is attained at P = -5 and
its value is 0.25

In case of SRSWOR, for estimation of P, the formula for sample size is given as

where d is the margin of tolerable error and t is the abscissa of the normal curve that
cuts of an area of a at the tails, (1-a) being the confidence level of the estimate. If N is
large, a first approximation to n is no = t2p %.
In the example considered above tolerable error is within 5% so d = 0.05. We want
this at the level of confidence of 95%, or in other words a-0.05. Assume that P =
0.5. From the standard normal distribution, the value of the variate corresponding to
the two-sides tail of 5% is 1.96 or approximately 2 and hence

t=2 ~ h u sn, o d x Oe5 x".5 = 4 x 100 =4oo.

(95)

In simple random sampling, units are selected rando'mly at each draw. Now we shall
discuss a sampling technique which has a nice, feature of selecting the whole sample
, with just one random start.

12.6 SYSTEMATIC SAMPLING

In simple random sampling units are drawn randomly at every draw. In many
situations, it may be desirable to select a sample in a systematic way. For example, if
we want to have an even spread in terms of spatial distribution, a systematic selection
may ensure that units maintain a uniform distance between selected units. In
systematic sampling, one unit is selected randomly and subsequent units are selected
according to a pre-determined system. Invariably uniform distance is adopted for pre-
assigned system. Systematic samples actually provide an improvement over simple
random samples as the samples are spread more evenly over the entire populations.
We now discuss sample selection procedures for systematic sampling.

12.6.1 Linear Systematic Sampling

The most commonly adopted procedure of systematic sampling is linear systematic

sampling. We shall explain the method through an example.
Problem 4 :Consider a population of 12 households from which a sample of 3
households is to be selected.
-
Solution: Let the households be arranged serially from 1 to 12. These households are
now rearranged in 3 rows of 4 columns as follows:
 Simple Random Sampling
Then, for selecting a systematic sample of size 3, we select a random number r (say)
between 1 to 4. Starting with r, every 4~ unit is selected. Thus, if r=3, then the units
selected are 3,7(=3+4), and 11(=7+4). Thus, if r is selected the entire colurnn,of
units consisting of r is selected.

In general, this method is applicable if the population size N is a multiple of the

sample size n i.e. N = nk where k is an integer. The random number r is selected
between 1 to k. Here, r is called a random start and k is called sampling interval.
The sample then comprises of the units r, r+k, r+2k + ..... + r + (n-1)k. The
technique will generate k systematic samples with equal probability. The method is
known as linear systematic sampling as N units are assumed to be arranged
sequentially on a line.

The method is specially suitable in forestry where for estimating the volume of timber
this method is used for selection of area units. Some other applications are in
industries where items for sample checks are selected systematically in a production
process. The concept of systematic sampling is not only confined to spatial
distributions. It can also be done over time. In fact in one of the applications in
estimation of fish catch from marine resources, sampling of boats on landing centres is
carried out systematically over time. Boats arriving every two hourly on selected
landing centres are observed.

In the method described above, if N is not a multiple of n, then it may not be possible
to get samples of equal size. For example, if N = 14 and n = 3 then the method
described above would lead to following arrangement of units in a n x k table as
follows:

In this case if randomly selected number r between 1 :o 4 is 1 then the sample is

1,5,9,13 while if r=3, then the sample is 3,7,11. Thus, samples are not of the same
size. Sample size is either 4 or 3 depending on the value of r. As an jmprovement to
this method we shall discuss circular systematic sampling

12.6.2 Circular Systematic Sampling

To overcome the difficult of varying sample size in a situation when N f 11k the
procedure is modified slightly by which a sample of constant size is always obtained.
This procedure is known as circular systematic sampling

In this method, the N units may be regarded as arranged round a circle. A random
start is taken between 1 to N and thereafter every !k unit, k being an integer nearest
N
to -,in a circular manner is selected until a sample of n units is chosen. Suppose
n
that a unit with random number i is selected. The sample will then consists of the
units corresponding to the serial numbers.
Sampling

This method is applicable, even if N # nk To illustrate the method, we consider the

following example:

Problem 5: Consider a population of 14 households from which a sample of size 5 is

to be selected .

Solution: Here n=5, k=3. Consider Fig. I:

4

Let the random start be 7. Then the selected sample is 7,10,13,2,5. If we start from 9
then the selected sample is 9, 12, 1,4,7. Like this we can have 12 more samples as
the total number of possible samples in this case is N = 14.

The above method has got the advantage of providing samples of the given size
irrespective of the random start. In case of linear systematic sampling the number of
possible samples is k while in case of circular systematic sampling it is N. When N is
a multiple of n then linear systematic sampling is normally preferred although one
could also go for circular systematic sampling. However, when N is not a multiple of
n then one should necessarily go for circular systematic sampling.

12.6.3 Advantages and Limitations of Systematic Sampling

The systematic sampling has the nice feature of operational convenience because the
selection of the first unit determines the whole sample. This operation is easier to
understand and can be speedily executed in relation to simple random sampling.
Secondly, systematic samples are.wel1 spread over the population and there is no risk
that any large part of the population will be left u~epresented.For populations with
linear trend, systematic sampling is more efficient in comparison to simple random
sampling.

Systematic sampling should, however, be cautiously used in case the population

exhibits a periodic trend. For periodic populations, the efficiency of the systematic
sampling depends upon the value of the sampling interval. If the sampling interval
coincides with the period, the sample will contain identical units and consequently the
systematic sampling performance becomes very poor. If, however, the sampling
interval is an odd multiple of half the period, the systematic sampling becomes most
effective.
 Simple Random Sampling
A serious limitation of this scheme lies in its use with populations having unforeseen
periodicity, which may substantially contribute to the bias in the estimate of
meanttotal. Another serious limitation of the sampling scheme, as mentioned earlier, is
that the variance of the estimator cannot be estimated unbiasedly.

You may now try this exercise.

-
E l 1) A sample of size 4 is to be selected from a population of 11 households. List
all the possible sample by (i) linear systematic sampling ii) circular systematic
sampling

We now end this unit by giving a summary of what we have covered in it.

12.7 SUNIMARY

In this unit, we have learnt

1) The preliminary concepts and definitions for simple random sampling
2) Method of selecting a simple random sample
3) How to estimate the population meanttotal
4) The method of estimating population proportions
5) How to determine sample size in case of simple random sampling
6) The basic concept of systematic sampling
7) How to select a linear systematic sample
8) The method of selecting a circular systematic sample
9) The advantages and limitations of systematic sampling

12.8 SOLUTIONSIANSWERS

El). i) collection of all the households in a localitylcity is the population and

the individuals are the households
ii) Total number of crates loaded in a truck is the population and the
individuals are the crates.
iii) All the bulbs produced by the company during a shift is the population
and the individuals are the bulbs.
E2) Advantages: less expensive in terms of time, money and energy, less
cumbersome, free of nonsampling errors. Also in case of destructive testing,
sampling is the only method.
Disadvantages: may suffer from sampling errors. The estimate is only an
approximation to the true value.
E3) a) The collection of all trees in the forest is the population, a tree is the
individual sampling unit and a list of all trees is the sampling frame.
b) Total number of apple trees in a district is the population, an apple tree
is the sampling unit and list of trees is the sampling frame.
 c) Collection of all the household in a city consuming the food is the
population, an individual household is the sampling unit and a list of
all households selected for a sample is the sampling frame.
E4) a) Table 3: All samples and their corresponding sample means in
SRSWOR (N=5, n=3) (order of sample units is ignored)

) Average 1 558
b) Similarly make a table for all samples of size 3 with replacement.
There will be s3samples in all.
E5) One sequence could be 21,79,00,59,73. Similarly give another.
E6) Remainder approach: For N=56, the highest two digit multiple of N is N .
itself. Using Appendix A select a two digit random number r, s.t. 1 < r 5 56.
-
I = = is one possibility. Also r 1 N =
44
56
gives quotient as 0 and remainder is
44. Thus select the unit with serial number 44. Likewise you can select other
9 units also. One such simple random sample (WOR) of 10 households
selected could be a sample of households with serial numbers

and a sample of 10 households (WR) could be

Quotient approach: In this case m=l. So if a random number r is selected s.t

1 Ir 5 56, say I==,then dividing r by m we get Q = 44 and selected unit is
the unit with serial number 44+1 = 45. Likewise you can select a SRS of 10
households (WR) and (WOR) from the population of 56 households.
E7) Average of sample means (1580) is equal to the population mean and thus the
sample mean is an unbiased estimator of the population mean. Calculations
are shown in Table 4 in the next page.
 Simple Random Sampling
Table 4: All samples and their corresponding sample means in SRSWR (N=5, n=2)
(order of sample units ignored)

Sample No. Units in Probability Sample observations Sample mean

Sample
Y1 Y2

1 1.2 1/10 1560 1490 1525

2 1.3 1/10 1560 1660 1610
3 1.4 1/10 1560 1640 1600
4 1.5 1/10 1560 1550 1555
5 2,3 1/10 1490 1660 1575
6 2,4 1/10 1490 1640 1565
7 2.5 1/10 1490 1550 1520
8 3-4 1/10 1660 1640 1650
9 3,5 1/10 1660 1550 1605
10 4,5 1/10 1640 1550 1595
11 1.1 1/10 1560 1560 1560
12 2,2 1/10 1490 1490 1490
13 3,3 1/10 1660 1660 1660
14 4,4 1/10 1640 1640 1640
15 5,5 1/10 1550 1550 1550

Average 1580

E8) of Sample mean in sample size 3

_~istribution

I Total 10 1
Sampling
7 + 3 + 6 + 1 0 + 4 30
Population mean = p = -
-- = 6.
5 5

The mean of the above distribution

S.E = -
J;; 5-1
/--
2.45 5-3
= 1.00 (approximately)

S.E = b
J;T
\l
N-1
c
(Usin g formula (5))

where o is the population standard deviation

average of 10 sample mean squares gives ~(s').

Here ~ ( s ' )= 48500/10 = 4850
Also, population mean square =

1
= - x 19400 = 4850
4
Thus from (i) and (ii)
~ ( s ' )= S'
i.e. sample mean square provides an unbiased estimator of the population
mean square.
E10) The mean of the Sampling distribution is the population proportion P,

S.E for the sampling distribution

I2
 Simple Random Sampling
11
Ell) i) N=l 1, n=4, k =- = 3 (approx). Arranging the units in 4 rows of 3
4
columns each (except for the last row)' we get table as follows:

selecting a number r between 1 and 3, possible samples are 1,4,7,10;

2,5,8,11; 3,6,9 of size 4 or 3.
ii) Here N= 11, n=4, k=3. Consider Fig.2
1. .

Let the random start be 2. Then sample selected is 2,5,7,10. If

random start be 5 then sample selected is 5,8,11,3. Likewise you can
write the remaining 9 samples as the total number of possible samples
is N=ll.
Sampling
Appendix A :Random Numbers
 Simple Random Sampling

Source: Rao, CR,Mitra S.K., Maitthai, A. and Ramamurthy, K.G. (1974). Formulae and Tables for Statistical
Work. Statistical Publishing Society. Indian Statistical Institute Calcutta.