Data Collection and

Measurement
Activity 1
Define population and sampling unit in each of the following problems.
i) Popularity of family planning among families having more than two children.
………………………………………………………………………………….
ii) Election for a political office with adult franchise.
……………………………………………………………………………….......
iii) Measurement of volume of timber available in forest.
…………………………………………………………………………………...
iv) Annual fruit in a hilly district.
…………………………………………………………………………………...
v) Study of birth rate in a district.
…………………………………………………………………………………...
5.2 ADVANTAGE OF SAMPLING OVER CENSUS
The census or complete enumeration consists in collecting data from each and
every unit from the population. The sampling only chooses apart of the units from
the population for the same study. The sampling has a number of advantages as
compared to complete enumeration due to a variety of reasons.
Less Expensive
The first obvious advantage of sampling is that it is less expensive. If we want to
study the consumer reaction before launching a new product it will be much less
expensive to carry out a consumer survey based on a sample rather than studying the
entire population which is the potential group of customers. Although in decennial
census every individual is enumerated, certain aspects of the population are studied
on a sample basis with a view to reduce cost.
Less Time Consuming
The smaller size of the sample enables us to collect the data more quickly than to
survey all the units of the population even if we are willing to spend money. This is
particularly the case if the decision is time bound. An accountant may be interested to
know the total inventory value quickly to prepare a periodical report like a monthly
balance sheet and a profit and loss account. A detailed study on the inventory is
likely to take too long to enable him to prepare the report in time. If we want to
measure the consumer price index in a particular month we cannot collect data of all
the consumer prices even if the expenditure is not a hindrance. The collection of data
on all the consumer items and their processing in all probability are going to take
more than a month. Thus when ready, the price index will not serve any .meaningful
purpose.
Greater Accuracy
It is possible to achieve greater accuracy by using appropriate sampling techniques
than by a complete enumeration of all the units of the population. Complete
enumeration may result in accuracies of the data. Consider an inspector who is
visually inspecting the quality of finishing of a certain machinery. After observing a
large number of such items he cannot just distinguish items with defective finish
from good one's. Once such inspection fatigue develops the accuracy of examining
the population completely is considerably decreased. On the other hand, if a small
number of items is observed the basic data will be much more accurate. It is of course
true that the conclusion about a population characteristic such as the proportion of
defective items from a sample will also introduce error in the system. However, such
errors, known as sampling errors, can be studied, controlled and probability
statements can be made about their magnitude. The accuracy which results due to
fatigue of the inspector is known as non sampling error. It is difficult to recognise
the pattern of the non sampling error and it is not possible to make any comment
about its magnitude even probabilistically.
20
Destructive Enumeration Sampling is
 indispensable if the enumeration is destructive. If you are interested in computing the Sampling and Sampling
average life of fluorescent temps supplied in a batch the life of the entire batch cannot Designs
be examined to compute the average since this means that the entire supply will be
wasted. Thus, in this case there is no other alternative than to examine the life of a
sample of lamps-and draw an inference about the entire batch.
Activity 2
Name a few research studies carried out in your organisation or the organisation you
know of where sample surveys were conducted.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
Activity 3
List out the advantages of sampling over complete enumeration (relative it to the
studies in Activity 2).
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………

5.3 SIMPLE RANDOM SAMPLING

The representative character of a sample is ensured by allocating some probability
to each unit pf the population for being included in the sample. The simple random
sample assigns equal probability to each unit of the population. The simple random
sample can be chosen both with and without replacement.
Simple Random Sampling with Replacement
Suppose the population consists of N units and we want to select a sample of size n.
In simple random sampling with replacement we choose an observation from the
population in such a manner that every unit of the population has an equal chance of
1/N to be included in the sample. After the first unit is selected its value is recorded
and it is again placed back in the population. The second unit is drawn exactly in the
swipe manner as the first unit. This procedure is continued until nth unit of the sample
is selected. Clearly, in this case each unit of the population has an equal chance of
1/N to be included in each of the n units of the sample.
Simple Random Sampling without Replacement
In this case when the first unit is chosen every unit of the population has a chance of
1/N to be included in the sample. After the first unit is chosen it is no longer replaced
in the population. The second unit is selected from the remaining N-1 members of the
population so that each unit has a chance of 1/N-1 to be included in the sample. The
1
procedure is continued till nth unit of the sample is chosen with probability
N-n+1
We now consider the operational aspect of selecting a simple random sample.
Random Sampling Numbers
The random sampling numbers are collection of digits generated through a
probabilistic mechanism. The numbers have the following properties:
i) The probability that each digit 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, will appear at any
particular place is the same, namely 1/10.
ii) The occurrence of any two digits in any two places is independent of each other.
A Table of Random Sampling Numbers are given below:
21
Data Collection and
Measurement
 An Example
Suppose a class has 115 students and you have been asked to selected a sample of
size 15 with replacement.
The students are numbered in some order from 1 to 115 so that each student is
identified uniquely by a serial number. Since the total number of students 115 is a
three digit number we take numbers in groups of three. The student bearing serial
number 1 is identified with the number 001, the student bearing the serial number 2 is
identified with the number 002 and so on. Proceeding in this manner the digit 115
will correspond to the student bearing the serial number 115. The number 116 will
again correspond to the student bearing the serial number 1, the number 117 again to
the student bearing the serial 2. This process will continue upto 920, the highest
multiple of 115 less than 1000. The digits 921-999 and 000 will be rejected.
It is possible to select three digit numbers from the random number table anywhere.
However, once a starting position has been selected you should continue to choose
numbers according to the given sequence. For the sake of simplicity we start at the
beginning row wise. The Table 2 below shows the selection of a sample of size 15
with replacement.

22

Note that the student bearing the serial number 49 has been selected twice. This is
permissible since the sampling is with replacement.
Activity 4 Sampling and Sampling
 Designs
i) In the example considered select a simple random sample of size fifteen without
replacement starting at the beginning of the random sample numbers.
……………………………………………………………………………………
……………………………………………………………………………………
……………………………………………………………………………………
ii) In the example considered what is the probability that each of the 115 students be
included in the sample ?
……………………………………………………………………………………..
iii) Why is it necessary to reject numbers from 921 ?
…………………………………………………………………………………….
iv) The data below indicate the number of workers in the factory for twelve
factories.

Select a simple random sample without replacement of size four. Start at the
beginning of the third row. Compute the average number of workers per factory
based on the sample. Compare this number with the average number of workers per
factory in the population.
…………………………………………………………………………………………
…………………………………………………………………………………………

5.4 SAMPLING FRAME

A Sampling frame is a list of all the units of the population. The preparation of a
sampling frame is sometimes a major practical problem. The frame should always be
made upto date and be free from errors of omission and duplication of sampling
units.
A perfect frame identifies each element once and only once. Perfect frames are
seldom available in real life. A frame is subject to several types of defect which may
be broadly classified on the following lines.
Incomplete Frame
When some legitimate sampling units of the population are omitted the frame is said
to be incomplete. If we want to collect information regarding the political opinion of
a group of voters on a sample basis, a complete list of voters is necessary to select a
sample. But instances are known when such a list is incomplete. 23
Data Collection and
Inaccurate Frame
Measurement
When some of the sampling units of the population are listed inaccurately or some
 units which do not actually exist are included, the frame is said to be inaccurate. If
you use the list of ration cards as a frame to select persons obviously such a frame
will be inaccurate as the details about the persons such as age are never updated.
Inadequate Frame
A frame which does not include all units of the population by its structure is an
inadequate frame. If you use the list of names included in the telephone directory of
a city as the frame for selecting a sample to collect information about a consumer
product, obviously it will be an inadequate frame. It will include the names of only
those persons who have a telephone omitting the majority of the residents of the city.
Out of Date Frame
A frame is out of date when it has not been updated although it was accurate,
complete and adequate at the time of preparation. The use of census blocks as a
frame to select a sample of households is a fairly accurate frame immediately after
the decennial census. But thereafter, its reliability as a frame deteriorates in a rapidly
growing or rapidly declining area.
Activity 5
Consider the sampling problems indicated in Activity 1. Suggest a suitable frame in
each case.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
5.5 PROBABILISTIC ASPECTS OF SAMPLING
Simple random sampling is based on the concept of probability. The use of
probability in sampling theory makes it a reliable tool to draw inference or
conclusion about the population. Although the types of conclusion or inference can
be quite diverse, two particular types of decision making are quite prevalent in
problems of business and government.
On various occasions, the management would like to know the percentage or
proportion of units in the population with a certain characteristic. An organisation
selling consumer product may like to know the proportion of potential consumers
using a certain type of cosmetic. The government may like to know the percent of
small farmers owning some cultivable land in a rural region. A manufacturer
planning to export some product may be interested to ascertain the proportion of
defect free units his system is capable of manufacturing.
The performance level of many products in a mass production shows statistical
variation. This aspect of quantitative analysis has been discussed in MS-8 Block 3.
Thus, we would only indicate the results without details. A manufacturer of
fluorescent lamps would like to know the average life of these lamps keeping in mind
that there will be variation- in life between individual lamps. An investor could be
interested to know the prices of a number of stocks. But as the prices vary over the
year, he may look for the average price of these stocks over the year.
In the first case, it is assumed that the proportion of units in the population with the
specific characteristic is P (unknown) (0 < P < 1) which is also referred to as
population proportion. You select a simple random sample (with or without
replacement) of sized and find out the number of units in the sample possessing the
particular characteristic under consideration. Suppose p is the proportion of units in
the sample possessing the characteristic. Then the numerical value of p gives us an
indication of the unknown population proportion P. Likewise, the mean characteristic
of all the units of the population is unknown. Suppose you select a simple random
sample (with or without replacement) of size n and denote by y1, y2,….,yn the
characteristic of these selected units of the sample (such as life of this lamp or the
price of the stock on that particular day). The sample mean y is
1
y= (y1 +y2 +.. .+yn )
24 n
The numerical value of y provides an indication of the population mean m. a statistic. Likewise y
The unknown quantities P or m are commonly referred to as parameters of the is also a statistic.
population. The corresponding sample counterparts p or y are known as estimators. It must be recognised
The quantity p is based on sample observations only. This is why it is also known as
 that an estimator is also a statistical variable with a sampling distribution. The Sampling and Sampling
variance of this sampling distribution is an indicator of the extent of error committed Designs
by using an estimator in place of a parameter. In particular, the
variance of p, V(p) and the variance of y , V ( y ) are known to be

The quantity  2 is the variance of all the units of the population. This quantity is
unknown but its estimator on the basis of a simple random sample of size n with
observed units (y1, y2,.....yn ) is
1
S2  [(y  y)2 ............... (y  y)2 ]
1 n
1 n
PQ
The estimate of
n is pq/n-l.

It may be observed from the expressions of variances of V(p) and V( y ) that they are
virtually identical for sampling with and wi thout replacement when the population
size is very large compared to the sample size.
The square root of the variance of an estimator is usually known as its standard
error. The standard error is generally taken as a measure of precision to be
achieved by sampling and will be used for developing suitable sampling design in a
subsequent section.
Activity 6
A list of 3000 voters of a ward in a city was examined for measuring the accuracy of
age of individuals. A random sample of 300 names was taken, which revealed that 51
citizens were shown with wrong ages. Estimate the proportion of voters having a
wrong description of age. Estimate the standard error of the estimate if the sampling
is done (a) with replacement (b) without replacement.

5.6 STRATIFIED RANDOM SAMPLING

The simple random sampling may not always provide a representative miniature of
the population. Certain segments of a population can easily be under represented
when an unrestricted random sample is chosen. Hence, when considerable
heterogeneity is present in the population with regard to subject matter under study,
it is often a good idea to divide the population into segments or strata and select a
certain number of sampling units from each stratum thus ensuring representation
from all relevant segments. Thus for designing a suitable marketing strategy for a
consumers durable, the population of consumers may be divided into strata by
income level and a certain number of consumers can be selected randomly from each
strata. 25
Data Collection and
Measurement
Speaking formally, the population of N units is subdivided into k sub-populations
or strata the ith sub-population having N i unit (i = 1, 2, ………., k). These sub-
populations are non overlapping so that they comprise the whole population such that
N1 + N2 + ………. + Nk = N
A simple random sample (with or without replacement) is selected independently
in each stratum, the sample size in the ith stratum being
ni (i = 1, 2, in such a manner that the strata are homogeneous within themselves with respect to
………., the characteristic under study. On the other hand, strata should be heterogeneous
k).Thus the total between themselves. Sometimes administrative convenience is taken into
sample size is n consideration to stratify the population. For instance, in order to study the problems
= n1 = n2 =.......................................................................................................
of railway commuters each railway division may be considered to be a different
stratum. In rural areas, the region covering adjacent districts are likely to be
The
homogenous with respect to socio-economic and cultural pattern. Hence they could
strati
be included in a common strata. Distribution of consumer products may face different
ficati
types of problems in rural, urban or hilly areas. These may be considered as separate
on
stratum from the point of view of management.
shoul
d be Allocation of' Sample Size in Different Strata
perfor
In stratified sampling, the sample to different strata is allocated on the basis of three
med
considerations.
i) The total number of units in the stratum i.e. stratum size
ii) The variability within the stratum
iii) The cost of taking observation per sampling unit each stratum.
From the point of view of management the most effective utilisation of the resources
will be achieved if the variance of the estimator is minimised for a fixed budget or
the cost of sampling is minimised for a fixed variance of the estimator. In practice,
however, two different types of allocation are mostly prevalent.
Proportional Allocation
This procedure of allocation is very common in practice because of its simplicity.
The number of sampling units ni allocated to the ith stratum is proportional to the
number of units in the population. Symbolically
Ni
n n ; (i  1, 2,........k )
i
N
Optimum Allocation
Let Ci be the cost of sampling one unit from stratum i and σ 2 be the variance of the

units belonging to stratum i. We assume that knowledge about σ 2 is available from

i
a previous consumer survey. Then, symbolically

According to this principle of ascertaining the sample size, the sample size in a
particular stratum is larger if (i) the stratum size is larger (ii) the stratum has larger
variability and (iii) the sampling cost in the stratum is lower.
An Example
The Table 3 below shows the allocation of a sample of size 400 in seven strata. The
26 strata have been formed according to monthly income of the consumers.
 Sampling and Sampling
Designs

As a market heterogeneity is present among the income strata the optimum allocation
differs sharply from the proportional. allocation, placing more emphasis on
representation from the higher income strata. This is because the higher income
group has a large variability.
Activity 7
A bank wants to select a sample of 500 borrowers who have borrowed money during
the last one year. The borrowers who have taken loan can be divided into four strata
according to occupation with stratum sizes N1 = 10000, N2 = 5000, N3 = 3000, N4 =
2000. What should be the sample sizes in the four stratum with (i) Proportional
allocation (ii) Optimum allocation ? In the second case the cost of sampling each unit
is Rs 20 irrespective of the occupational classes. But the standard deviations of the
characteristic under study are Rs 420, 262.5, 603 and 355 respectively for the four
strata.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
………………………………………………………………………………………….

5.7 OTHER METHODS OF SAMPLING

Cluster Sampling : In this method of sampling a collection or a Cluster of sampling
units are selected in a random manner. Then each unit of the cluster is included in the
sample.
In order to motivate the use of a Cluster, we consider a survey where the sampling
units are households in a rural area. If simple random sampling is used to select
households they will be located over several villages. On the other hand, a village can
be regarded as a Cluster of households. We select a few villages randomly and
include every household in the selected villages in our sample. Such a sampling
procedure will be an illustration of Cluster Sampling. It has a number of advantages
over simple random sampling.
i) If the households are chosen using simple random sampling, they are likely to be
distributed over several villages. Hence from administrative point of view such a
selection will involve more cost, more field supervision and more travelling. On
the other hand, if a selected village is completely enumerated, the cost involved
will be lower and the supervision exercised will be better.
ii) If the households in the sample are distributed over several villages then a frame
containing the list of households of each of these villages is necessary for proper
identification and selection of the household in the sample. On the other hand, if
every 27
Data Collection and
Measurement
 h village is included in the sample, no sampling frame listing the households in a
o village is necessary.
u iii) If the type of question is of intimate nature an isolated household selected in a village
s is unlikely to cooperate with the investigators. On the other hand, if every household
e in a village is visited, a particular household after observing that his neighbours are
h also being interviewed are likely to offer greater cooperation and as such the quality
o of the basic data will be more reliable.
l
d In the example presented in this section clusters have been formed based on geographic
subdivisions. A village is a natural geographic Cluster of households. A Cluster sample with
i clusters based on geographic subdivisions is known as area Sample and the procedure is
n known as area Sampling.
A Cluster sample need not be always formed on geographic consideration. Suppose we want
a to estimate the proportion of defective machine parts in the inventory. If there are 20,000
items stored in 400 cases of 50 each, then those 400 cases can be considered as clusters. A
s Cluster sample of ten cases may be randomly selected resulting in a sample of 500 items.
e For a given sample size, the variance of an estimator increases or equivalently the estimator
l becomes less precise with a larger cluster size. On the other hand, the cost decreases with a
e larger cluster size. We may balance the two conflicting aspects of the problems by finding out
c the Optimum Cluster size which minimises the variance of the estimator for a fixed cost or
t minimise the cost for a given level of precision.
e
d Systematic Sampling
The mechanics of taking a systematic sample are very simple. Suppose the population
consists of ordered N units (numbered from 1 to N) and a sample of size n is selected from
N
the population in such a way that =k (rounded to the nearest integer). Here k is called
n
sampling interval. In systematic sampling from the first k of the units a unit is selected
randomly. Then every kth unit is selected in the sample. As for illustration, suppose k = 20.
Then among the first 20 units coming out through the production line one unit is selected
randomly. Suppose this is unit number 9. Then we select unit number, 29, 49, …..in our
sample till the requisite sample size has been reached.
The main advantage of systematic sampling is its simplicity of selection, Operational
Convenience and even spread of the sample over the population. The investigators using this
sampling method need not be familiar with the use of random sampling numbers. It has been
found very useful in forest surveys for estimating the volume of timber or in fisheries for
estimating the total catch of fish. The method can be used even when no formal list of the
units of the population is available. As for illustration, a five per cent sample of the household
in a locality can be selected by selecting every twentieth household after all the households
are ordered in a systematic manner.
Although a systematic sample is not a random sample in the strict sense of the tern, its
performance is quite comparable to a random sample. However, if there is a hidden
periodicity in the population, systematic sampling will prove to be an inefficient method of
sampling. As for illustration, if every twentififth unit of a production process is defective a
four per cent systematic sample will select all defective or all non defective items depending
upon the starting point. Such an orientation is, however, very rare.
Multistage Sampling : The multistage sampling procedure is used for large scale enquiry
covering large geographical area such as a state. As for illustration, a bank may like to gather
information regarding the quality of customer service it is offering in a state. A random
sample of districts is selected from the list of districts. From each of the selected districts a
number of branches are randomly selected. From each of the selected branches a number of
depositors which is the ultimate sample sampling unit is selected randomly for collecting
information. The districts are called first stage units, the branches are known as the second
stage units and the depositors are regarded as the third stage units. This is an illustration of
three stage sampling, the third stage units being the ultimate sampling units.
The multistage sampling procedure has the advantage that the frame of second stage units is
necessary only for the selected first stage units. Likewise, the frame of the third stage units is
necessary only for the selected second stage units. The procedure is quite flexible and it
28 permits the use of different selection procedures in different stages. It may also be mentioned
that multistage sampling is the only sampling procedure available in a number of practical situations,
 such as collecting information about consumers in a large geographical region. This Sampling and Sampling
is because no satisfactory sampling frame is available for the ultimate stage units, Designs
namely, consumers and the cost of preparing such a frame is prohibitive.
5.8 SAMPLING DESIGN
A Sampling design is a definite plan for obtaining a sample from a given population.
It refers to the technique or the procedure the business researches would adopt to
select units for the sample. It will also indicate the number of units to be included in
the sample also known as Sample size. Sampling design is determined before data
are collected. While developing a sampling design, the management must pay
attention to the following points.
Type of Population
The first step in developing any sampling design is to clearly define the aggregate of
sampling units, namely, the population. Sometimes the structure of the population is
quite unambiguous such as the population of a city or the number of workers in a
factory. However, there may be occasions where the composition of the population is
not so simple. For example, in the context of advertising a company may like to
collect information about the number of television viewers watching the
advertisement.
Sampling Unit
The sampling unit must be identified before selection of a sample. A sampling . unit
may be a natural geographical unit such as a state, a district, a village or a
constructed unit such as a L use or a flat. It may be a social entity such as a family
or a school. It may also be an individual. The management should be able to select
appropriate unit so that its objective is achieved.
Type of Sample
You have observed in the previous subsections that starting from simple random
sampling, the business researcher has a wide choice of sampling procedures. Each
procedure has its suitability for the relevant occasion. The type of frame necessary
also depends upon the sampling procedure being used. As for illustration, the use of
simple random sampling will require a comprehensive list of the sampling units.
But a two stage sampling will require the list of all first stage units and the list of all
second stage units in the selected first stage units only.
Size of the Sample
There are two basic requirements for the sampling procedure to fulfill. A sample
must be representative and it must be adequate. When it is representative, a sample
will be relatively small piece of the population that mirrors the various patterns and
subclasses of the population. A sample is adequate if it provides an estimator with
sufficiently high precision. It should be remembered in this context that the higher is
the precision, the larger is the sample size and more is the cost.
An Example
A market research unit wants to conduct a survey in order to estimate the proportion
of smokers smoking a particular brand of cigarette. If it is required to estimate with
90% probability within .02 of the true proportion, how many observations should be
taken ?
Suppose, n is the sample size to be determined and p is the observed proportion of
smokers. Then if the population size is sufficiently large, from section 5.5
irrespective of sampling with or without replacement
V(p) p(1-p)
 n
The notation P is the unknown population proportion of smokers. Further, as
indicated in MS-8, section 14.6 for large n, p is normally distributed with expectation
P (unknown population proportion of smokers) and variance V(p). Since the
estimator p should be within .02 of the true proportion P with 90% probability
p{ p-p  .02} = .90

29
Data Collection and
Measurement
 where z is normally distributed with expectation zero and variance one. From the
Table of Normal Distribution (Mustafi, 1981) i

i.e. n = 6800 P (1-P)

1
Since the maximum value of P (1-P) = , the sample size necessary is 1700.
4

Activity 8
What should be the size of the sample ( assumed to be large) if in a large
consignment of items the population proportion defective is to be estimated within
.02 of true value with probability .95?
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
5.9 NON PROBABILITY SAMPLING METHODS
Non probability sampling is the sampling procedure which does not provide any
basis for estimating the probability that each item in the population possesses to be

included in the sample. In such a case, the sampling error is not measurable and the
error in the estimator tends to increase sharply because the representativeness of the
sample members is questionable. Nevertheless, non probability samples are useful in
certain situations. This is the case when the representativeness is not particularly the
primary issue. In pretesting a questionnaire, we are interested to ascertain its
suitability for the problem under consideration. Non probability sampling methods
can be used to contact respondents in this case as the primary aim is to probe for the
possible range of answers.
In general, there are three types of non probability sampling that may, under
appropriate conditions, be useful in business and government. These are
Convenience, Judgement and Quota sampling.
1) Convenience Sampling : Under convenience sampling, as the name implies, the
samples are selected at the convenience of the researcher or investigator. Here, we
have no way of determining the representativeness of the sample. This results into
biased estimates. Therefore, it is not possible to make an estimate of sampling error
as the difference between sample estimate and population parameter is unknown both
in terms of magnitude and direction. It. is therefore suggested that convenience
sampling should not be used in both descriptive and causal studies as it is not
possible to make any definitive statements about the results from such a sample.
This method may be quite useful in exploratory designs as a basis for generating
hypotheses. The method is also useful in testing of questionnaire etc. at the pretest
phase of the study. Convenience sampling is extensively used in marketing studies
and otherwise. This would be clear from the following examples.
i) Suppose a marketing research study aims at estimating the proportion of Pan
(Beettle leave) shops in Delhi which store a particular drink say Maaza It is
decided to take a sample of size 100. What investigator does is to visit 100 Pan
shops near his place of residence as it is very convenient to him and observe
30 whether a Pan shop stores Maaza
iii) or not. This is definitely not a
representative sample as
most Pan shops in Delhi had
ii) no chance of being selected.
It is only those Pan shops
which were near the
iv) residence of the investigator
had a chance of being
selected. c make ups are approximately the same as national average. While conducting Sampling and Sampling
The other marketing tests for new products, the researcher may take samples of consumers from Designs
example such cities and obtain consumer evaluations about these products as these are
where supposed to represent "national" tastes.
convenience A ball pen manufacturing company is interested in knowing the opinions about the
sampling is ball pen (like smooth flow of ink, resistance to breakage of the cover etc.) it is
often used presently manufacturing with a view to modify it to suit customers need. The job is
is in test given to a marketing researcher who visits a college near his place of residence and
marketing. asks a few students (a convenient sample) their opinion about the ball pen in
There might question.
be some As another example a researcher might visit a few shops to observe what brand of
cities whose vegetable oil people are buying so as to make inference about the share of a particular
demographi brand he is interested in.
2) Judgement Sampling : Judgement sampling is also called purposive sampling. Under
this sampling procedure, a researcher deliberately or purposively draws a sample from
the population which he thinks is a representative of the population. Needless to mention,
all members of the population are not given chance to be selected in the sample. The
personal bias of the investigator has a great chance of entering the sample and if the
investigator chooses a sample to give results which favours his view point, the entire
study may be vitiated.
However, if personal biases are avoided, then the relevant experience and the
acquaintance of the investigator with the population may help to choose a relatively
representative sample from the population. It is not possible to make an estimate of
sampling error as we cannot determine how precise our sample estimates are.
Judgement sampling is used in a number of cases, some of which are mentioned below.
i) Suppose we have a panel of experts to decide about the launching of a new product in
the next year. If for some reason or the other, a member drops out from the panel, the
chairman of the panel, may suggest the name of another person whom he thinks has
the same expertise and experience to be a member of the said panel. This new
member was chosen deliberately - a case of Judgement sampling.
ii) The method could be used in a study involving the performance of salesmen. The
salesmen could be grouped into top-grade and low-grade performer according to
certain specified qualities. Having done so, the sales manager may indicate who in
his opinion would fall into which category. Needless to mention this is a biased
method. However in the absence of any objective data, one might have to resort to
this type of sampling.
3) Quota Sampling : This is a very commonly used sampling method in marketing
research studies. Here the sample is selected on the basis of certain basic parameters such
as age, sex, income and occupation that describe the nature of a population so as to make
it representative of the population. The investigators or field workers are instructed to
choose a sample that conforms to these parameters. The field workers are assigned quotas
of the numbers of units satisfying the required characteristics on which data should be
collected. However, before collecting data on these units the investigators are supposed to
verify that the units qualify these characteristics.
Suppose we are conducting a survey to study the buying behaviour of a product and it is
believed that the buying behaviour is greatly influenced by the income level of the
consumers. We assume that it is possible to divide our population into three income strata
such as high income group, middle income group and low income group. Further it is
known that 20% of the population is in high income group, 35% in the middle income
group and 45% in the low income group. Suppose it is decided to select a sample of size
200 from the population. Therefore, samples of size 40, 70 and 90 should come from high
income, middle income and low income groups respectively. Now the various field
workers are assigned quotas to select the sample from each group in such a way that a
total sample of 200 is selected in the same proportion as mentioned above. For example,
the first field 31
Data Collection and
Measurement
worker may be assigned a quota of 10 consumer from the high income group, 25 from
the middle income group and 40 from the low income group. Similarly the 2nd field
worker may be given a different quota and so on such that a total sample of 200 is
obtained in the same proportion as discussed earlier.
The above example was a very simple one, suppose we are told further that the buying
behavio
ur is not
only
influenc
ed by
his
income
but also T
by his he above table indicates that in this universe there are 12% of people in the high
age income group and fall in the age group of 45 and above, there are 25% people in
(categor middle income group and below the age of 45 and so on. Suppose, it is decided to take
ised as a sample of size 200. Therefore the distribution of the sample conforming to these two
45 and parameters (in the same proportion as population) would be as follows :
above
or
below
45).
With
this
addition
al
characte
r,
suppose T
the he above table shows that a sample of 30 should be taken from the population with low
distribut income and below the age of 45. Similarly a sample of 20 should be taken from the
ion of population with middle income and having age of 45 and above, and so on.
populati Now, having decided the size of sample falling under each of the six cells ["high
on income and below 45" "middle income with 45 and above" and so on], we fix the
(univers quotas for each of the field worker to collect data conforming to the above norms so as
e) is as to obtain a total sample of size 200.
follows At the outset, the Quota sampling procedure might look similar to stratified sampling.
: However, there is a difference between the two. Under stratified sampling, the field
worker selects a random sample from each cell of the population, whereas under Quota
sampling the selection of sample is not random. It is left to the judgement of the field
worker.
The Quota sampling method has some weaknesses. These are listed below :
i) It is usually difficult to obtain an accurate and up to date proportion of respondent
assigned to each cell.
ii) As the number of parameters (control characteristics) associated with the objectives
of the study become large, the total number of cells increase. This makes the task
of field staff difficult as it may not be easy to get a desired respondent.
iii) It is very important that all of the proper parameters (control characteristics) related
to the study in question must be incorporated while taking a sample. The results of
the study could be misleading if any relevant parameter is omitted for one reason or
the other.
iv) The field workers might like to visit those areas where the chances of the
32 availability of a respondent satisfying certain desired parameters is very high.
Further, the field workers
might avoid certain respondents who look unfriendly and live in houses which ………………………
may not be of good appearance. These factors are likely to make the findings of ………………………
the study less reliable. ………………………
The Quota sampling method has some advantages too. The method has a lower cost …………………
and field workers have a free hand to select respondents for each cell to fill their ………………………
quota. The samples, if selected with care would result into more definitive findings. ………………………
………………………
Activity 9 …………………
Make a list of some research studies where some of the non probability methods
could be used. Also justify the choice of a particular sampling method you have
5.10 SUMMAR
selected for a study. Y
………………………………………………………………………………………… A sample is a part of a
 group or aggregate selected with a view to obtaining information about the whole Sampling and Sampling
group also known as population. The population is composed of a number of units. Designs
The total number of units in the population and in the sample are known as the
population size and the sample size respectively. The technique of sampling has been
used in traditional problems such as estimating the yield of crop or examining the
quality of an outgoing product. In addition to these traditional spheres, sampling has
been applied successfully in management problems such as estimation of inventory
by sampling or in socio-economic problems such as ascertaining the trend of a
political process through opinion polls. As compared to census or complete
enumeration sampling is less expensive, less time consuming and more accurate.
Further sampling is the only method for collecting information when the
measurement of the sampling unit is destructive in nature.
A probabilistic sampling attaches some probability to each unit of the population to
be included in the sample and in this sense it is a representative sample of the
population. A simple random sample attaches equal probability to each unit of the
population to be selected in the sample. Operationally, the selection of a random
sample is based on a sampling frame containing a list of all the units of population
and a table of random sampling numbers.
When the population is heterogeneous we divide the units into several groups each
known as a stratum. The strata are so selected that each stratum is as homogeneous as
possible while the compositions of two different strata are as heterogeneous as
possible. The sampling units are selected from each stratum using simple random
sampling. This procedure known as stratified sampling which improves the precision
of the estimator.
Another sampling procedure known as cluster sampling is frequently used in
consumer survey in rural areas where a number of sampling units are grouped
together as clusters. A number of clusters is selected randomly and the units of each
cluster are completely enumerated. A cluster sample incurs less expenditure and
provides more reliable data. Other sampling methods used in business research are
systematic sampling and multistage Sampling.
A sampling design is a suitable scheme for obtaining a sample from a given
population. It also indicates the size of the sample to be used keeping the cost and
precision in view.
Non probabilistic sampling such as Convenience Sampling, Judgement Sampling and
Quota Sampling are sometimes used although representativeness of such a sample
cannot be ensured.

5.11 KEY WORDS

Sampling is the selection of a part of a group with a view to obtaining information
about the whole.
Population is the aggregate from which a sample is taken.

33
Data Collection and Measurement
 Sampling Units are units or members which constitute the population.
Population size is the total number of units present in the population.
Sample size is the total number of units in the sample.
Simple Random Sample is a sample where each unit of the population has an equal
chance of being included in the sample.
Random Sampling Numbers are a collection of digits generated through a
probabilistic mechanism.
A Sampling frame is a list of all the units of the population.
Stratified Random Sample is a sample collected by dividing the entire population
into a number of subgroups and selecting a simple random sample from each
subgroup.
Cluster Sample is a sample where a collection or cluster of sampling units is
selected randomly.
Systematic Sample is a sample where the units are selected at equal interval.
Multistage Sample is a sample where the sampling units are selected through several
stages.
A Sampling Design is a definite plan for obtaining a sample from a given
population.
Convenience Sampling is a process of selecting a sample based on convenience.
Judgement Sampling is a procedure where the sample is selected through judgement
or expertise.
Quota Sampling is a procedure where the sample is selected according to a number
of key characteristics.

5.12 SELF-ASSESSMENT EXERCISES

1) Define Population and Sampling unit for selecting a simple random sample in
each of the following cases.
i) Fifty voters of a constituency.
ii) Twenty-five stocks from the Bombay Stock Exchange.
iii) Twenty students enrolled in University `1'.
iv) One hundred depositors of a branch of a bank.
2) Discuss the advantages of sampling method over census method of data
collections.
3) "Sampling is a necessity under certain conditions." Explain with illustrative
examples.
4) What is meant by "representativeness" in a Sample? Explain in what sense a
simple random sample is representative of the population ?
5) Show that in simple random sampling without replacement each sample of size

n from a population of size N has a probability of being selected is 1  N 

 .
n
6) What are random numbers ? Select with the help of the table of random numbers
a simple random sample of size ten without replacement from a population of
121 households.
7) Discuss the importance of sampling frame. What are the possible errors resulting
from a faulty sampling frame ?
8) What do you mean by sampling design ? What points should be taken into
consideration while developing the sampling design for studying in the problems
34 of domestic airline passengers ?
9) State the formulas for the standard errors of sample mean and sample proportion. Explain how
 standard error of sample proportion can be used to determine the sample size. Sampling and Sampling
Designs

10) What should be sample size so that the proportion of TV owners during the last
one year opting for hire purchase scheme is estimated within an error of .03 with
probability 0.98 ?
11) Under what circumstances stratified random sampling procedure is considered
appropriate ? How would you select such a sample ? Explain by means of an
example.
12) A certain population is divided into five strata so that N1 = 2000, N2 = 2000, N3 =
1800, N4 = 1700, N5 = 2500. The respective standard deviations are : σ1 = 1.6,
σ2 = 2.0, σ3 = 4.4, σ4 = 4.8 and σ5 = 6.0. Further, the sampling cost in the first
two strata is Rs. 4 per interview and in the remaining three strata the sampling
cost is Rs. 6 per interview. How a sample of size 226 should be allocated to five
strata with (i) proportional allotment (ii) optimal allotment.
13) Under what circumstances would you recommend (i) cluster sampling (ii)
systematic sampling (iii) multistage sampling (iv) non probability sampling. Give
an example in each case.
14) Answer true or false
i) In order to gather opinion about a television show I asked the views of my
friends, the friends asked about the show to their friends and in this way a
sample is selected. The sampling procedure is judgement sampling.
…………………………………………………………………………………
ii) A cluster sample provides more reliable data because of complete
enumeration within a cluster.
…………………………………………………………………………………
iii) A systematic sample can be used even if a formal list of all the units of the
population is not available.
…………………………………………………………………………………
iv) The main difficulty with multistage sampling is that we have to prepare
along frame.
…………………………………………………………………………………
v) If the population size and the sample size increase in such a manner that the
sample size is always one tenth of the population size, simple random
sampling with or without replacement is equivalent.
…………………………………………………………………………………
vi) In stratified sampling with optimum allocation, if the cost of sampling is
equal in each stratum then the strata with larger size will always have a large
allocation of sample size.
…………………………………………………………………………………

5.13 ANSWERS
Activity 1
i)
Population : All couples living together and whose wife is in the child bearing
age group (commonly known as Eligible Couple) with two children.
Sampling Unit : Eligible couple with two children.
ii)
Population : All eligible voters for the political office i.e. all adults.
Sampling Unit : Eligible voter.
35
Data Collection and
Measurement
iii) iv)
v) Pop the forest.
ulati Sampling Unit : Tree
on Population All the Apple trees in the district.
All
the Sampling Unit : Apple tree.
trees Population All eligible couples.
of Sampling Unit : Wife of the eligible couple.

iii) If the numbers from 921 ... 999, 000 are not rejected, the students bearing serial
9
numbers 1 to 80 have a chance of of being selected. The remaining students
1000
8
have a chance of of being selected. So each unit of the population does not
1000
have equal probability of being included in the sample.

Average No. of workers per factory in the sample 5908.25

36
Average No. of workers per factory in the population 3118.08
Activity 5 Sampling and Sampling
Designs
 i) The frame of all eligible couples with more than two children should be obtained
by census method.
ii) The frame is the voters list which is the list of all adults.
iii) Each tree in the forest should be serially marked. The marked trees form the
sampling frame.
iv) Each apple tree in the district is serially marked. The marked apple trees form the
sampling frame.
v) The frame of all eligible couples should be obtained by census method.
Activity 6
51
Estimate =
300
Estimate of the standard error
a) With replacement = .0217
b) Without replacement = .0206
Activity 7
i) n1 = 250, n2 = 125, n3 = 75, n4 = 50.
ii) n1=261, n2 = 82, n3 =113, n4 = 44.
Activity 8
n = 2401
Self-assessment Exercises
1) i) Population : All the voters.
Sampling unit : A voter.
ii) Population : All the stocks listed in Bombay stock exchange.
Sampling unit : A stock.
iii) Population : All the students enrolled in the University ‘I’.
Sampling unit : A student.
iv) Population : All the depositors of the branch.
Sampling unit : A depositor.
5) Suppose p1, p2,......pn be any n units of population. The probability that the first
unit selected is p1, …., nth unit selected is pn when the sampling is without
replacement is given by
1 1 1 1
  .......
N N-1 N-2 N-n+1
(N-n)!
 N!
But any permutation of p1, p2,... pn will result in an identical sample. Hence, the
probability of selecting the sample whose composition is (p1, p2,….pn) is
(N-n)! n!
1
 N
N!  
n
10) n = 1508
12) Proportional Allocation
n1 = 45, n2 = 45, n3 = 41, n4 = 38, n5 = 57 37
Data Collection and
Measurement
Optimal nl = 22, n2 = 28, n2 = 45, n4 = 46, n5 = 85
Allocation
14) i) False, it is Convenience Sampling.

ii) True.

iii) True.

iv) False, the sampling frame of the second stage units are required only for the
selected first stage units.

v) False, n/N = 1/10.

vi) False, the sample size will depend upon both the stratum size and the stratum
standard deviation.

5.14 FURTHER READINGS

Ferber, R. and P.J. Verdoom, 1962. Research Methods in Economics and Basiness,
Macmillan: USA.

Haber, A. and R. P. Runyon, 1972. General Statistics, Addison-Wesley: USA.

Kothari, C.R. 1985. Research Methodology Methods and Techniques, Wiley Eastern :
New Delhi.

Mustafi, C.K. 1981. Statistical Methods in Managerial Decisions, Macmillan : New

Delhi.

Singh, D. and F.S. Chaudhary, 1986. Theory and Analysis of Sample Survey Designs,
Wiley Eastern : New Delhi.

Tull, DS and G.S. Albaum, 1973. Survey Research a Decisional Approach, Intext
Educational Publishers : USA.

38
 Attitude Measurement and
UNIT 6 ATTITUDE MEASUREMENT Scales

AND SCALES
Objectives
After reading this unit, you should be able to :

 describe the type of managerial research problems which utilise the tools of
attitude measurement;
 discuss the type of issues which come up when one attempts the measurement of
attitudes;
 explain the different attitude measurement scales, alongwith their strengths and
limitations;
 decide for which type of research problems one should go in for specific scale or
think of using multidimensional scaling.
Structure
6.1 Introduction
6.2 Attitudes, Attributes and Beliefs
6.3 Issues in Attitude Measurement
6.4 Scaling of Attitudes
6.5 Deterministic Attitude Measurement Models : The Guttman Scale
6.6 Thurstone's Equal-Appearing Interval Scale
6.7 The Semantic Differential Scale
6.8 Summative Models : The Likert Scale
6.9 The Q-Sort Technique
6.10 Multidimensional Scaling
6.11 Selection of an Appropriate Attitude Measurement Scale
6.12 Limitations of Attitude Measurement Scales
6.13 Summary
6.14 Key Words
6.15 Self-assessment Exercises
6.16 Further Readings
6.1 INTRODUCTION
There are a number of management decisions which are taken in an organisation,
from time to time. The decisions may relate to the acquisition or disposal of
materials/machines, manufacturing or marketing of products, hiring or firing of
employees, opening or closedown of plants, promotion or reversion of personnel, and
so on. Some of these decisions may rely on data for which the units of measurement
are capable of statistical manipulation. Such data largely refer to quantifiable
parameters or numerical properties of a given population. However, there are
illustrations of other decisions which may rely primarily on behavioural data or data
which is statistically not manipulatable, in the strict sense of the word. The units of
measurement of such data are not interchangeable and are not susceptible to rigorous
statistical analysis. The major area of utilisation of such data lies in the discipline of
marketing where the-manager is interested in knowing the attitudes of the current and
potential users of his/her product or service towards his/her product or service
concept or idea. This knowledge of attitudes could result in decisions which would be
sensible and effective. Some illustrations of managerial decisions which rely on
attitude measurement are product positioning and market segmentation, advertising
message decisions etc.
6.2 ATTITUDES, ATTRIBUTES AND BELIEFS
Before one plunges into the topic of attitude measurement, it will be worthwhile to
understand the key terms which figure repeatedly in this topic. 39
Data Collection and
Measurement
 Each object/product/service is believed to be composed of certain characteristics
which fulfil certain needs of its user. These needs may be of psychological, physical
or social nature. The characteristics of the object under consideration are called its
attributes. The term belief refers to judgements made by a user regarding the object
possessing certain attributes or not. Finally, the term attitude refers to the
predisposition/mental state of individuals/users towards a product/idea/attributes of
an object. It also implies the mental readiness to act in a particular manner and
influences the individuals's behaviour towards the object/group/organization/person
under consideration. The salient factors that go into the building of the overall
attitude of the individual towards the object are a) his/her beliefs about the attributes
possessed by the object, b) his/her preference or otherwise for those attributes, and c)
the relative importance of each attribute to the individual's decision making process.
6.3 ISSUES IN ATTITUDE MEASUREMENT
Measurement implies the process of obtaining information which can be subject to
analysis. Attitude measurement relates to the process of measuring an individual's
attitude towards an object. When we go for measurement of attitudes or any other
parameter, one has to clearly sort out the following :
 "what" has to be measured ?
 "who" is to be measured ?
 the accuracy desired in the measurement
 the costs permissible
 the choices available in the measurement/data collection techniques.
In attitude measurement, the researcher is primarily interested in measuring the "state
of mind" of the respondent (s). It may include factors such as awareness, attitudes
and decision processes. An interesting characteristics of these measures is that their
verification is rather difficult. There is no way to determine whether the answer given
by a respondent to the level of liking for a new product, such as ice-cream mix,
represents the "truth" or not. The researcher, unless he is a "telepathist", cannot
actually observe the states of mind like preference, likes and dislikes, etc. Such things
can only be inferred.
It has been stated in the previous section, that attitudes are affected by attributes and
beliefs. So, the first step, before embarking on an attitude-measurement exercise, is
selecting the relevant attributes of the object under investigation. For instance, the
salient attributes of a product like "Shrikhand" may be price, shelf life, flavour, and
pack size. For a public distribution system they may be quality of grains, prices,
outlet working timings, and assurance of availability. It is clearly impossible to
measure every attribute of the process/object under consideration. The researcher
should settle for the relevant ones only. It is advisable to measure only those
attributes which can be related to actions by the respondents. Exploratory research
can be helpful in identifying attributes. The methods used could include nondisguised
ones like depth interviews and disguised ones like projective techniques. The depth
interviews are the most commonly used technique. They use no structured framework
for gathering information. The respondents are. encouraged to talk about the object
under investigation and the investigator tries to uncover its salient attributes in this
process. This procedure requires skilled investigators. It is also considered costly and
the results are prone to bias errors. The projective techniques attempt to uncover the
information from the respondent in an indirect manner. The subject is requested to
respond to incomplete stimuli here. In doing so, he/she is believed to reveal elements
of attitude towards the object that will not be revealed in response to direct queries.
The projective techniques used may include a cartoon test, word association test,
sentence completion test, etc. Though these techniques also have some
disadvantages, they are used more than the nondisguised methods.
The next important issue in attitude measurement is that "who" is to be measured. It
involves people. The question to be posed now is of what kind ? Their education,
age, sex, occupation, religion etc. may have a bearing on the choice of the
40 measurement method. The measurement procedure must be designed with the
characteristics of the respondents under consideration. For instance, using a mail
questionnaire for disinterested or hostile respondents would hardly be the right choice
as a research instrument.
The third major issue in attitude measurement is the choices in data collection and The approaches for
measurement techniques. The data collection techniques can be categorised into (a) measuring attitudes are
Questionnaire methods, and (b) Observational methods. Usually questionnaires are as follows :
used for measuring the attitudes. 1) Self-report
 inventories Attitude Measurement and
2) Using psychological measures like galvanic skin response or pupillary response Scales
3) Projective techniques like thematic a perception test.
The self-report inventories, also known as attitude scales, involve presenting the
subjects with a list containing favourable and unfavourable statements about the
research topic and ask whether they agree or disagree with each of them.
Most attitude measurement methods use the self-report technique. However they
differ in terms of the way the scales are constructed and used.
The weaknesses of the self-report measures are that :
1) The results are limited to what the individuals know about their attitudes and are
willing to relate,
2) The validity of the verbalised attitudes is questionable.
We will be discussing some of these scales, alongwith their characteristics, in the
subsequent sections of this unit.
Finally, the last major issue for a managerial researcher here relates to the costs and
accuracy desired in the measurement. As has been stated earlier, these type of
measurements are never entirely free of inaccuracy. Moreover, cost and accuracy are
generally reciprocal properties in measurement. The intimate knowledge of a
research instrument can go a long way in the correct interpretation of the results.
Activity 1
List out the salient attributes of the following products.

41
Data Collection and
Measurement
 Activity 2
You may conduct depth interviews to identify attributes of the products mentioned in
Activity 1. Compare it with the one you have already listed out in Activity 1.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
Activity 3
List out the three important issues in attitude measurement.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
………………………………………………………………………………………….

6.4 SCALING OF ATTITUDES

Researchers in management have dipped into the bag of tricks of sociologists and
psychologists to develop the techniques for measurement of attitudes. Basically, what
is done here is to confront the respondent with a number of favourable and
unfavourable statements about the subject and find out the extent of his/her
agreement or disagreement with them. The object here is, to measure people 's attitude
42 to ideas, products or service. The scales are usually highly structured. The underlying
assumption in attitude scaling is that attitudes determine opinions and decisions about
the product/service/idea. To understand the human behaviour, and to influence it in a
given direction, it is necessary to understand the opinions and the basic attitudes
underlying them.
There are many types of scales and scaling techniques. We have to choose the most
appropriate technique to suit the research being done. The statements contained
therein have to be prepared in such a way that the answers received can easily be
converted into numerical values. The three most commonly used scales are the
Nominal scale, the Ordinal scale and the Interval scale. You will note later on that the
appropriateness of a statistical method depends upon the type of scale used in data
collection.
l) Nominal Scale : The nominal scale simply allows the categorization of
responses into a number of mutually exclusive categories. There are no relationships
between the categories, implying that there is no ranking or ordering. The typical
applications of the nominal scale is in classification of responses by social class,
"like" or "dislike", "yes" or ''no”, sex, and so on. The statistical operation possible for
nominally scaled data is counting only.
2) Ordinal Scale : The ordinal scale, allows the respondents to rank some found but. It is only
alternatives by some common variable. An illustration of this would be the ranking of possible to compute
three brands of pasteurised milk by a group of consumers on the basis of the positional statistical
perceived quality. Here it is feasible for a user of the product to rank the brands from measures like median
the best to the worst. However the amount of difference between the ranks cannot be and. mode for such
 data. Attitude Measurement and
Scales

3) Interval Scale : The deficiencies of the nominal and the ordinal scales are taken
care of in the interval scale. The scale has an arbitrary zero point with numbers
placed at equally appearing intervals. A number of statistical operations including
addition, subtraction, and computation of the mean can be done on intervally scaled
data.
Mostly the nominal and the ordinal type of scales are used in attitude measurement.
However, there are attempts to treat them or convert them into interval scales so as to
make the data more amenable to statistical operations. But the reasonability of this
assumption should be always tested before making inferences.
Most of the attitude measurement scales used are ordinal in nature, though there is
attempt to treat the resulting data as intervally scaled. The simplest possible type of
such scale has the respondent classifying the object/issue/product/himself into one
among two dichotomous categories. Further refinements may include the provision of
additional alternatives denoting the degrees of liking or disliking. These are listed in
sequence so that the alternatives form a type of scale. These scales are basically self-
report inventories, with a list of favourable and unfavourable statements towards the
subject.
The attitude measurement scales can be categorised into those which are
unidimensional in nature and those which are multidimensional. The different type of
single dimensional attitude measurement scales which are available are graphical and
numerical scales, summated scales, paired comparisons, equal-appearing intervals,
etc. Some scales of the above-mentioned type will be discussed in the subsequent
sections of this unit.
Activity 4
Identify the scale to which the following statements/responses belong.

6.5 DETERMINISTIC ATTITUDE MEASUREMENT

MODELS: THE GUTTMAN SCALE
In the deterministic attitude measurement techniques the underlying assumption is
that each statement has a perfect relationship, of one type or another, with the
particular dimension of the attitude being investigated. For instance, let us consider a
research study where one is interested in investigating a community's attitude to
family planning. The items in the questionnaire which relate to this could consist of
the following :
Yes No
1) Family planning is the best hope for our country …………………
2) Family planning would lead to healthier children …………………
3) We should all participate in the family planning programme. ………………… 43
Data Collection and
Measurement
 Usually a person who answers YES to 1, would have a high probability of answering
as YES to the subsequent statements. Any person who does not answer as YES to 1
but does answer as YES to 2 would have a high probability of answering YES to the
later items. Any set of items that produces a pattern of responses as we have
described here is called a Guttman Scale.
Guttman scale analysis is usually applied to dichotomous data, i.e., data with only
two values, YES or NO, 0 or 1, agree or disagree, etc. However, a number of reasons
have made the Guttman scale and impractical tool for the measurement of attitudes.
First, the construction of the scale requires a lot of time and effort. Secondly, there
may be very few items existing that may fit the model. Since such scales seldom have
more than eight items, they can make only rather gross distinctions among
respondents.
Activity 5
Define Dichotomous data.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
Activity 6
Mention two reasons which make Guttman scale an impractical tool for the
measurements of attitudes.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………

6.6 THURSTONE'S EQUAL-APPEARING INTERVAL

SCALE
In this scale we are interested in scaling respondents and not statements. The first
step in the scale construction is to scale the attitude statements along the attitude
continuum. This is done by asking some "judges" to evaluate the items along some
continuum.
The statements are printed on some cards and the judges are asked to sort the
statements into 11 groups. The extreme piles represent the most favourable and the
most unfavourable statements. The judges are expected to make the intervals between
the groups equal. The mean rating by judges is taken as the scale point for each item.
Items which are found to be ambiguous or irrelevant are dropped. The items selected
44 for the final scale are such that (a) each item has a small standard deviation of ratings
over judges and (b) the mean ratings spread evenly from one end of the rating
continuum to the other. The selected items are listed in a random order to form the
final scale.
The administration of the scale for the measurement of the attitude of a respondent
needs the latter to mark only the items with which he/she agrees. The score for the
respondent then is taken as the scale value of the median item endorsed or the
average scale value of the items endorsed. For instance, suppose a respondent agrees
with items which have scale values as 9, 10, and 11. This would imply that he has a
favourable attitude to the object (assuming that score of 11 implies most positive
attitude !).
The Thurstone scales are prepared with an odd number of positions, the usual number
being 11. The scale has some drawbacks such as the time requirement being fairly
high, the influencing of scale positions by the attitudes of the judges, and no
information on the degree or intensity of agreement with the different items.
Activity 7 its service, location
and timings, etc.
Construct a Thurstone’s equal-appearing interval scale for a bank interested in
knowing about its image in public by developing a number of statements relating to ………………………
 ………………………………………………………………… Attitude Measurement and
………………………………………………………………………………………… Scales
…………………………………………………………………………………………
…………………………………………………………………………………………

6.7 THE SEMANTIC DIFFERENTIAL SCALE

The term Semantic differential scale refers to any collection of rating scales anchored
by bipolar adjectives. It is a very flexible approach to obtaining measures of attitudes.
The, object that is rated is called the "concept" and almost anything can be rated
including family planning, cosmetics, Shrikhand, political parties, etc.
107nally, a semantic differential scale is based on a seven-point rating scale for each
of a number of attributes relating to the research topic. The extreme point represent
the bipolar adjectives with the central category representing neutral. In the semantic
differential scale only the extremes have names. The in-between categories have
either blank spaces or sometimes a number. Some examples of the scale are as
follows
Good..............................................................................................Bad
Honest............................................................................................Dishonest
Progressive.....................................................................................Behind the times
The preparation of a semantic differential scale for a study requires expressing the
things that could he used to describe the object, and thus serve as a bets for attitude
formation, in terms of positive and negative statements. The negative phrase is
sometimes put on the left side of the scale and sometimes on the right. This prevents
a respondent with a positive attitude from simply checking either the left or right
hand sides without reading the describing words.
The scale can be used for a variety of purposes. 11 can be used to check whether a
respondent has a favourable attitude towards the object, which out of three
neighbourhood banks has the most appealing profile for housewives, etc.
It is possible to assign points to individual cells in the scale. Then one could arrive at
the scores for comparisons of different objects. The Figure 1 gives an example based
on image study of three neighbourhood banks among a sample of 100 housewives.

45
Data Collection and
Measurement
The study shown in Fig.1 indicates to a bank manager those aspects in the bank's
image which are relatively weak or strong in the eyes of the customers, vis-à-vis what
he/she was planning to achieve and how this bank compared the other two
neigh ndents provides an index of the overall image rating. The maximum score possible
bourh for each bank is +21 and minimum possible score is -21. The bank A scored +10,
ood bank B -8, and bank C - 4. These scores are based on each attribute having the same
banks weighting factor of 1. However, in case the researcher wants to weigh the attributes
. The in a different way, it is feasible. This may lead to different image score for the rival
image banks. For the manager of bank A, we can conclude that the attributes which need
profil immediate attention are service reliability, followed by modernisation and staff
es are attitudes towards the customers.
based Activity 8
on the Use semantic differential technique to develop the profile of three talcum powder
scores producing companies in India.
of
…………………………………………………………………………………………
each
…………………………………………………………………………………………
respo
…………………………………………………………………………………………
ndent
…………………………………………………………………………………………
on
………………………………………………………………………………………….
each
dimen 6.8 SUMMATIVE MODELS: THE LIKERT SCALE
sion The summative models assume that the individual items in the scale are
and monotonically related to the underlying attributes and a summation of the item scores
the is related linearly to the attitude. In a summative model, one obtains the total score by
avera adding scores on individual items. For the statements that imply negative attitudes,
ge the scoring is reversed. The scales allow an expression of the intensity of feeling.
total These scales are also called Likert scales. Here, instead of having just "agree" and
scores "disagree" in the scale, we can have intensities varying from "strongly agree" to
for all "strongly disagree".
respo
The scale construction consists of the following steps:
1) Write a large number of statements that-concern the particular attitudinal object
being investigated. For instance one may be looking at the role of voluntary
agencies in providing health services . in rural areas. Most of these statements
should either be moderately positive or moderately negative. Neutral items are
generally avoided in these scales. The items should be evenly divided between
positive and negative statements.
2) Administer the pool of statements on a group of respondents who are similar to
the population on whom the scale will be used. For example, if we want to study
the attitude of housewives the pool should be administered on a group of
housewives with similar background to our final population.
3) Assign scale values to the degrees of agreement or disagreement with each item.
The particular values may differ from one researcher to another. Sometimes one
may adopt the values 1, 2, 3, 4, 5 and sometimes +2, +1, 0, -1, -2. For negative
items the directions should be reversed.
4) Calculate a total attitude score for each respondent using the same scaling
procedure. The distribution of total , scores is then used to refine the list of items.
This step is called item analysis.
5) Item analysis : Analyse the responses and select for the scale those items which
most clearly differentiate between the highest and lowest scores. This can be
done by dividing the respondents into the high and the low scoring categories.
The high scorers can be assumed to be with favourable attitudes and the low
scorers can be taken as having the least favourable attitudes. If the statement is a
good one, then it is safe to expect that the mean score for the favourable group
would be greater than the mean score for the unfavourable group. If the mean
scores across the two groups, for an item, are found nearly equal or equal, then
that statement can be dropped from the scale. One can take the high group as the
46 top twenty-rave per cent of all total scores and the low group as
the lowest twenty-five per cent. Alternatively we can divide the respondents into sum of his scores
Attitude Measurement and quartiles and compute the median score for each item on each statement.
for the highest twenty-five per cent States and the lowest twenty-five per cent of
scale scores. The summated scales
have certain
6) The statements remaining in the pruned fast are randomly_ ordered on the scale advantages. They are
form. The positive and negative ones are mixed. easy to construct, are
highly reliable, and can
7) The scale is now administered on the respondents who are asked to indicate their be adapted to the
degree of agreement with the items. A respondent's total score is generated as the
 measurement of many different kinds of attitudes. Attitude Measurement and
Scales
Activity 9
How would you use a Likert scale to ascertain the image of Colgate toothpaste
among some consumers ?
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
………………………………………………………………………………………….

6.9 THE Q-SORT TECHNIQUE

The Q-sort technique grew out of a more general methodology for the study of
verbalized attitudes, preferences etc. The peculiar characteristic of this methodology
is that here it is more important to make comparisons among different responses
within respondents than between respondents. Thus it is a comparative rather than an
absolute rating method.
In the area of management, the application of Q-sort has come up in marketing
research. Here the respondents may be requested to enunciate their images of their
ideal brands, specified brands and their current brand. The respondent may be given a
large number of statements (50-100) describing the characteristics of a product. For
instance for a cosmetic product like shampoo he may be asked to compare
alternatives with adjectives like "easy_ to use", "economical", and "safe for children",
with instructions to array them along the continuum "least preferred - the ideal
shampoo". Essentially what the respondents have done here is to array the attributes
along the scale.
The Q-sort technique is faster and less tedious for subjects than paired comparison
measures. It also forces the subject to conform to quotas at each point of the scale so
as to yield a normal or quasi-normal distribution.
The utility of Q-sort in marketing research is to derive clusters of individuals who
display similar preferences, thus representing unique market segments. The objective
of Q-sort, thus, is intensive study of individuals.

6.10 MULTIDIMENSIONAL SCALING

In the attitude measurement scales which have been discussed till now the object is
measured against each characteristic, one at a time. The measurement process tells
little about the relative importance of different characteristics or how the
characteristics relate to each other. When these aspects become important, one takes
recourse to multidimensional scaling. It is a term used to describe a group of
analytical techniques used to study attitudes, specially those relating to perceptions
and preferences. These techniques attempt to identify the object attributes that are
important to the respondents and to measure their relative importance. The major
application of multidimensional scaling in managerial research comes in marketing
research. Some of the uses which have already been illustrated in the literature are as
follows :

47
Data Collection and
Measurement
 Advertising : It has been applied here to answer questions such as: Which media
should be used for getting the desired reach ? If written media is selected then which
magazines newspapers to advertise in ?
Market Segmentation Vendor Evaluations
A detailed discussion of Multidimensional Scaling is outside the purview of this unit.
Generally the use of this method requires the use of a computer and a number of
computer programmes, both on PCs and the larger systems are available. For a
detailed discussion on Multidimensional Scaling, please refer Handbook of
Marketing Research, Robert Ferber (ed.), pp. 3-44 to 3-61.

6.11 SELECTION OF AN APPROPRIATE ATTITUDE

MEASUREMENT SCALE
We have examined in the earlier sections a number of different techniques which are
available for the measurement of attitudes. Each has some strengths and some
weaknesses. Almost every technique can be used for the measurement of any
component of attitudes. But at the same time all techniques are not suitable for all
purposes. Then selection of the scale depends on the stage and the size of the
research project. The costs of developing and implementing the instrument, reliability
and validity of the instrument and the statistical analysis necessary.
Generally, Thurstone's scale, Q-sort and the Semantic differential scale are preferred
for preliminary investigation. The Likert scale is used for item analysis. For specific
attributes the semantic differential scale is very appropriate.
Overall the semantic differential is simple in concept and results obtained are
comparable with more complex, one dimensional methods. Hence, it is widely used.

6.12 LIMITATIONS OF ATTITUDE MEASUREMENT

SCALES
The main limitation of these tools is the emphasis on describing attitudes rather than
predicting behaviour. This is primarily because of a lack of models that describe the
role of attitudes in behaviour.

6.13 SUMMARY
We have briefly examined the role of attitude measurement and scales in managerial
research. We started by looking at the type of managerial decisions which need
quantification of attitudinal data. Subsequently, we defined some key terms such as
attributes, beliefs and attitudes. This was followed by a discussion on the critical
issues in attitude measurement. Then the different types of scales, viz., nominal,
ordinal, and interval, which feature in this field, were described. This was followed
by five important tools/scales of attitude measurement, viz., Guttman, Thurstone's
equal-appearing interval, Semantic Differential, Likert's and the Q-sort technique.
Simultaneously, the steps in scale construction and their use were also highlighted. A
brief discussion of multidimensional scaling followed. Finally, the issues of the
selection of an appropriate attitude measurement scale and the limitations of these
research tools were discussed.

6.14 KEY WORDS

Attributes : Characteristics of the object, tinder investigation.
48
Beliefs : Judgements made by a user regarding the object possessing certain
attributes or not.
Attitude : Predisposition of individuals) towards the attributes of an object.
Measurement: The process of obtaining information, which can be subjected to
analysis.
Scale : A collection of statements which can be used for measuring attitudes. Bipolar adjectives: A
pair of words which
Respondent/Subject : Person on whom an attitude measurement scale is have opposite
administered. meanings.
 Attitude Measurement and
6.15 SELF-ASSESSMENT EXERCISES Scales

1) What do you understand by the terms attitude and attitude measurement ?

Explain.

2) Which type of managerial research and decisions utilise attitude measurement ?

Explain with examples.

3) Review briefly the different types of issues in attitude measurement.

4) Compare and contrast the various attitude measurement techniques. When will
you use each of them ? Discuss briefly.

5) In which type of study will you use multidimensional scaling ? Discuss.

6.16 FURTHER READINGS

Bailey, Kenneth D., 1978, Methods of Social Research, The Free Press, New York.

Churchill, Gilbert A., 1983, Marketing Research: Methodological Foundations, The

Dryden Press, New York.

Nunally, Jum C., 1978, Psychometric Theory, Tata McGraw-Hill, New Delhi. Feber,
Robert 1974, Handbook of Marketing Research, McGraw-Hill, New York.

Meister, David, 1985, Behavioural Analysis and Measurement Methods, John Wiley,
New York.

Rodger, Lesile W., 1984, Statistics for Marketing, McGraw-Hill (UK), London.

Boyd, H.W., Westfall, Ralph, and S.F. Statch, 1986, Marketing Research: Text and
Cases, Richard D. Irwin, Illinois.

Aaker, David A. and George S. Day, 1983, Marketing Research, John Wiley; New
Yak. Luck, D.J., et al., 1978, Marketing Research, Prentice Hall (India), New
Delhi.

Lundstrom, William J.;et. al. November 1976. "The development of a scale to

measure consumer discontent", Journal of Marketing Research, Vol. 13, pp. 373-
381.

Balasubramanian, Siva K. and Wagner A. Kamakura, August, 1989. "Measuring

Consumer attitudes towards the market place with tailored interview", Journal of
Marketing Research, Vol. 26, pp. 311-326.

49
 Data Processing
UNIT 7 DATA PROCESSING
Objectives

After studying this unit, you should be able to:

 explain the significance of data presentation

 describe various types of data classification
 develop appropriate data tabulation and presentation devices.
Structure
7.1 Introduction
7.2 Editing of Data
7.3 Coding of Data
7.4 Classification of Data
7.5 Statistical Series
7.6 Tables as Data Presentation Devices
7.7 Graphical Presentation of Data
7.8 Summary
7.9 Self-assessment Exercises
7.10 Further Readings
7.1 INTRODUCTION
The data collected from the field has to be processed and analysed as laid down in the
research plan. The processing of data primarily mean editing, coding, classification
and the tabulation of the data collected so that they are amenable to analysis. In this
unit, therefore, we. shall concentrate on the various aspects of processing of data.
Further, the presentation of data can be either in tabulation forms or through charts.
In Unit 6, Block 2 (Data Collection and Analysis) of Quantitative Analysis for
Managerial Applications (MS-8), you have been introduced to the basic concepts of
presentation of data. You are, therefore, advised to refer to it for better understanding
of this unit.
7.2 EDITING OF DATA
The editing of data is a process of examining the raw data to detect errors and
omissions and to correct them, if possible, so as to ensure completeness, consistency,
accuracy and homogeneity. This will facilitate coding and tabulation of data. In fact,
the editing involves a careful scrutiny of the completed questionnaires.
The editing can be done at two stages: field editing and central editing.
Field editing : The field editing consists of review of the reporting forms by the
investigator for completing or translating what the latter has written in abbreviated
form at the time of interviewing the respondent. This form of editing is necessary in
view of the writing of individuals, which vary from individual to individual and
sometimes difficult for the tabulator to understand. This sort of editing should be
done as soon as possible after the interview, as it may be necessary sometimes to
recall the memory. While doing so; care should be taken so that the investigator does
not correct the errors of omission by simply guesing what the respondent would have
answered if the question was put to him.
Central Editing : Central editing should be carried out when all the forms of
schedules have been completed and returned to the headquarters. This type of editing
requires that all the forms are thoroughly edited by a single person (editor) in a small
field study or a small group of persons in case of a large field study. The editor may
correct the obvious errors, such as an entry in a wrong place, entry recorded in daily
terms whereas it should have been recorded in weeks/months, etc, Sometimes, 5
Data Presentation and Analysis inappropriate or missing replies can also be recorded by the editor by reviewing the
other information recorded in the schedule. If necessary, the respondent may be
contacted for clarification. All the incorrect replies, which are quite obvious, must be
deleted from the schedules.

The editor should be familiar with the instructions and the codes given to the
interviewers while editing. The new (corrected) entry made by the editor should be in
some distinctive form and they be initialed by the editor. The date of editing may also
be recorded on the schedule for any future references.

Activity 1

Define the following

a) Field editing

……………………………………………………………………………………
……………………………………………………………………………………
……………………………………………………………………………………

b) Central editing

……………………………………………………………………………………
……………………………………………………………………………………
……………………………………………………………………………………

Activity 2

Your organisation is conducting a survey to determine the consumption pattern of

food items by households in Delhi. You are the head of computer division
responsible for editing the raw data from the questionnaires and analysing the same.
A filled up set of questionnaires have been sent to you. List out the points on which
you would like to concentrate while editing the raw data.

…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………

7.3 CODING OF DATA

Coding is the process of assigning some symbols (either) alphabetical or numerals or
(both)to the answers so that the responses can be recorded into a limited number of
classes or categories. The classes should be appropriate to the research problem being
studied. They must be exhaustive and must be mutually exclusive so that the answer
can be placed in one and only one cell in a given category. Further, every class must
be defined in terms of only one concept.

The coding is necessary for the efficient analysis of data. The coding decisions
should usually be taken at the designing stage of the questionnaire itself so that the
likely responses to questions are pre-coded. This simplifies computer tabulation of
the data for further analysis. It may be noted that any errors in coding should be
eliminated altogether or at least be reduced to the minimum possible level.

7.4 CLASSIFICATION OF DATA

In most research studies, voluminous raw data collected through a survey need to be
reduced into homogeneous groups for any meaningful analysis. This necessitates
classification of data, which in simple terms is the process of arranging data in groups
or classes on, the basis of some characteristics. Classification helps in making
comparisons and drawing meaningful conclusions. Classification can either be
6 according to attributes or according to numerical characteristics.
In case of classification according to attributes, the data is classified by descriptive qualitative
i.e. sex, caste, education, land holding, etc. The descriptive characteristics refer to phenomenon
 which cannot be measured quantitatively, only their presence or absence in an Data Processing
individual item can be observed. This classification can either be simple or manifold.
In case of simple classification, also referred to as classification according to
dichotomy, each class is divided into two sub-classes and only one attribute is
studied, such as employed or unemployed, resident or non- resident, married or
unmarried, etc. In case of manifold classification, more than one attributes are
considered, and therefore, the classification leads to the formation of a number of
classes and sub-classes. For example, industries may be classed as belonging to
public sector or private sector. Among each of these broad classes, there can be a
second level classification in terms of large and small, which can further be classified
(third level) as profit making and loss making, and so on. This process of
subclassification is carried keeping in view our basic purpose for which the
classification is required. The objectives have to be clearly defined before the
classification can be carried out. Further, the attributes should be defined in such a
way that there is least possibility of ambiguity concerning the said attributes.

When individual observations possess numerical characteristics, such as height,

weight, marks, income, etc, they are classified on the basis of class intervals . For
example, persons whose monthly income is between Rs. 1001 and Rs. 1500 may
form one group, those whose income is within Rs. 150.1 and Rs. 2000 may form
another group, and so on. In this manner, the entire data may be divided into a
number of groups or classes, which are usually called class-intervals. The number of
items in each class is called the frequency of the class. Every class has two limits: an
upper limit and a lower limit, which are known as class limits. The difference
between these two limits is called the magnitude of the class or the class interval.
Several aspects of classification of data has been detailed in Unit 6 (Presentation of
Data), Block 2 (Data Collection and Analysis) of Quantitative Analysis for
Managerial Applications (MS-8). You are advised to refer to that study material.

Activity 3
Following data refer to monthly salary of 40 employees of an organisation. Tabulate
the data using the inclusive and exclusive methods:

…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………

7.5 STATISTICAL SERIES

A series is defined as a logical or systematic arrangement of observations or items.
When the attributes or things are counted, measured or weighed and arranged in an
orderly manner, say either discending or ascending order, they constitute a series.
When the statistical data pertains to time, the series is said to be historical or time
series. The important factor in such series is the chronology. When the data pertains
to space, the series is referred to as spacial, and is also known as geographical series.
When the data refers to physical conditions such as height. weight, age, etc., the
series is referred to as condition series. The following series, for example, refers to
special series of regionwise sales of a firm during 1989-90.
7
Data Presentation and Analysis
 The series can also be classified as individual observations, discrete series and
continuous series. In case of series of individual observations, the items are listed
singly as distinguished from listing them in groups. In case of discrete series, items
are arranged in groups (frequency distribution) showing definite breaks from one
point to another and are exactly measurable. The marks obtained by 30 students in a
subject grouped together in a series in the following way.

In case of continuous series, the items are arranged in class and they can be arranged
either in ascending order or distending order of magnitude and their continuity is not
broken. At the point at which a class ends, the next begins and thus the continuity is
maintained. Monthly consumption of edible oil by 40 families is given below in the
form of a continuous series.

Activity 4
Develop a sample continuous series for data collected from your own work situation.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
7.6 TABLES AS DATA PRESENTATION DEVICES
Statistical data can be presented in the form of tables and graphs. In the tabular form,
the classification of data is made with reference to time or some other variables. The
graphs are used as a visual form of presentation of data.
The tabulation is used for summarization and condensation of data. It aids in analysis
of relationships, trends and other summarization of the given data. The tabulation
may be simple or complex. Simple tabulation results in one-way tables, which can be
used to answer questions related to one characteristic of the data. The complex
8 tabulation usually results in two way tables, which give information about two
interrelated characteristics of the date; three way tables which give information about
three interrelated characteristics of data, and still higher order tables, which supply
information about several interrelated characteristics of data.
Following are the important characteristics of a table:
i) Every table should have a clear and concise title to make it understandable
without reference to the text. This title should always be just above the body of
the table.
ii) Every table should be given a distinctnumber to facilitate easy reference. Data Processing
 iii) Every table should have captions (column headings) and stubs (row headings)
and they should be clear and brief.
iv) The units of measurements used must always be indicated.
v) Source or sources from where the data in the table have been obtained must be
indicated at the bottom of the table.
vi) Explanatory footnotes, if any, concerning the table should be given beneath the
table along with reference symbol.
vii) The columns in the tables may be numbered to facilitate reference.
viii) Abbreviations should be used to the minimum possible extent.
ix) The tables should be logical, clear, accurate and as simple as possible.
x) The arrangement of the data categories in a table may be a chronological,
geographical, alphabetical or according to magnitude to facilitate comparison.
xi) Finally, the table must suit the needs and requirements of the research study.
Activity 5
Using your own data source, present the data on any agricultural commodity in a
tabular form.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
7.7 GRAPIHICAL PRESENTATION OF DATA
Several types of graphs or charts are used to present statistical data. Of them, the
following are commonly used: bar chart, two dimensional diagrams, pictograms,
piecharts and arithmetic chart or line chart. Several of these have been discussed in
detail in Unit 6, (Presentation of data). Block 2 (Data Collection and Analysis) of
Quantitative Analysis for Managerial Applications (MS-8). You may therefore refer
to the said study material. Therefore, charts and graphs not covered are discussed
below.
Two dimensional diagrams: The commonly used two dimensional diagrams are
rectangular diagrams and squares. In rectangular diagrams the rectangles are used to
present the data in the graphic form. These diagrams are used for comparing two sets
of data. The height of the rectangle is proportional to the ratio of the data which bear
to each other in a given series and the width of the rectangle varies in proportion to
the aggregate. Before constructing the rectangular diagram, the data is converted into
percentages. A rectangular diagram to the data on cost of production and profits, in
two firms A and B is given below.

When the difference between two quantities is very large, one bar would become too
big and the other too small in a rectangular diagram. To overcome this difficulty,
squares are used to present the data. The size of the square is the square roots of the
given data.
9
Data Presentation and Analysis
 Pictograms : In this form of presentation, data are represented by a picture. For
example, population figures are presented by the picture of a human being,
production figures of, say motorbikes, is presented by the picture of a motorbike,
cattle population is presented by a picture of a cattle and so on. The following figure
presents a pictograph showing troops strengths of two countries A and B.

Pie Chart : In a pie chart, different segments of a circle represent percentage

contribution of various components to the total. It brings out the relative importance
of various components of data. For drawing a pie chart, we construct a circle of any
diameter and then the circle is broken into a desired number of segments, angle 360 0
represents 100 per cent. The cost of production in an industry is presented below in
the form of a pie chart.

10
Line or Arithmetic Chart : The line or arithmetic chart is used to identify the
changes or the trend that exist in a series of data. The data on export sales of a
company between 1978 and 1989 is presented below in the form of a line chart.
Although we can see changes in the data, the presentation of the same on a line chart
gives a better
 Data Processing

picture of the information. The other forms of presentation of the data have already
been discussed in Unit 6, Block 2 of Quantitative Analysis for Managerial
Applications (MS-8).
Activity 6
With the help of your own data, draw an appropriate chart to present the same.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
7.8 SUMMARY
We have in this unit discussed the various aspects of data processing. The importance
of editing and coding of data before any processing has been emphasised here. In any
research studies, the voluminous data can be handled only after classifying the same.
The various aspects of classification, summarisation and presentation of data have
been discussed.
7.9 SELF-ASSESSMENT EXERCISES
1) Describe, in brief, the importance of editing, coding, classification, tabulation
and presentation of data in the context of research study.
2) Discuss the different aspects of classification of data. What are the likely
problems encountered in the classification and how they can be handled?
3) Why tabulation is considered essential in a research study? Give the
characteristics of a .good table.
4) Write briefly about the different forms of data presentation devices.
5) Monthly salary of 32 employees of a firm are given below. Tabulate the data
after selecting a suitable class interval.

11
Data Presentation and Analysis
 7.10 FURTHER READINGS

Enns, P.G., 1985. Business Statistics, Richard D. Irwin Inc.: Homewood.

Kothari, Cr., 1986. Research Methodology-Methods and Techniques, Wiley Eastern;

New Delhi.

Robert Sandy, 1990. Statistics for Business and Economics, Me Graw-Hill

International: Singapore.

Shenoy, G.V., Srivastava, U.K. and Sharma, S.C. 1988.Business Statistics, Wiley
Eastern: New Delhi.

12
 Statistical Analysis and
UNIT 8 STATISTICAL ANALYSIS AND Interpretation of Data :
Nonparametric Tests
INTERPRETATION OF DATA:
NONPARAMETRIC TESTS
Objectives

On successful completion of this unit, you should be able to

 explain the differences between parametric and nonparametric tests

 describe the relevance of nonparametric tests in data analysis
 discuss the steps involved in the nonparametric test
 design and conduct one sample, two samples and more than two samples
nonparametric tests.
Structure
8.1 Introduction
8.2 One Sample Tests
8.3 Two Sample Tests
8.4 K Sample Tests
8.5 Summary
8.6 Self-assessment Exercises
8.7 Further Readings
8.1 INTRODUCTION
In Unit 15, Block 4 (Sampling and Sampling Distributions) of Quantitative Analysis
for Managerial Decisions (MS-8), you have been introduced to the general class of
testing of hypotheses. These tests, popularly known as parametric tests, assume that
parameters such as mean, standard deviation, etc. exist and are used to develop a test.
For example, a t-test is based upon the comparison of means of two samples. The
parametric tests are developed with an assumption that the form of the population
distribution was known and that a test concerning a parameter of the distribution was
to be made. Tests of hypotheses which deal with population parameters are called
parametric tests.
There are situations, particularly in psychological or in market research studies,
where in the basic assumptions underlying the parametric tests are not valid or one
does not have the knowledge of the distribution of the population parameter being
tested. The tests which handle problems of these types are known as nonparametric
tests or distribution free tests.
In the recent past, the nonparametric tests have gained importance basically for three
reasons :
(i) These tests require no or less restricting assumptions than the corresponding
parametric tests,
(ii) these tests are more suitable for analysing ranked, scaled or rated data, and
(iii) these tests involve very few arithmetic computations.
However, it must be understood that when basic assumptions about the parametric
tests are valid, the nonparametric tests are less powerful than the parametric tests.
Thus, there is a greater risk of accepting a false hypothesis and thus committing a
type II error. Secondly, in nonparametric methods, the null hypothesis is somewhat
loosely defined thereby, when the null hypothesis is rejected, the nonparametric tests
yield less precise conclusions compared to the parametric tests. For example, in
parametric tests we have a test for equality of two population means compared to the
two population distributions are identical in a nonparametric test. In this case, a
rejection of null hypothesis in a parametric test would mean that the two population
means are not equal while the rejection of the null hypothesis in the nonparametric
test leads to the conclusion that the two population distributions are different, the 13
specific form of the difference between the two distributions is not clearly specified.
Data Presentation and Analysis
The following ale some i) In a consumer behaviour survey for new package design, the responses are not
of the typical situations likely to be normally distributed but clustering around two extreme positions,
for using nonparametric with a very few respondents giving a neutral response to the package design.
tests :
ii) Sometimes, the responses to a question are given in terms of names (nominal
data), which cannot be treated as numbers. For example, if we ask young
graduates "in which part of the country would you like to take up a job and
live", the replies could be north, north-west, west or south, etc. Nominal data
can be analysed only by nonparametric methods.
iii) In mailed questionnaire method of survey, more often partially filled
questionnaires are received. Nonparametric tests are designed to take the
missions data and make necessary adjustments to extract maximum
information from the available data.
iv) Nonparametric tests can be used to provide reasonably good results even for
very small samples.
8.2 ONE SAMPLE TESTS
The one sample tests are used to answer the questions such as:
i) Is there a significant difference between the observed and expected
frequencies?
ii) Is it reasonable to believe that the sample has been drawn from a specified
population?
iii) Is it reasonable to accept that the sample is a random sample from some known
population?
The tests which are used to obtain answers to the above questions are classified as
tests for goodness of fit.
Kolmogorov-Smirnov One Sample Test
This test is used for comparing the distribution on an ordinal scale. The test is
concerned with the degree ,of agreement between the distribution of observed values
and some specified theoretical distribution. It determines whether the scores in a
sample can reasonably be thought to have come from a population having the
theoretical distribution.
Example : A leading paint manufacturer is interested in developing a new colour
shade. He is interested in testing four different shades: dark, bright, light and very
light. A sample of 160 customers were shown these four shades and following results
were obtained: 30 liked the dark shade, 45 liked the bright shade, 60 liked the light
shade and the remaining liked the very light shade. The manufacturer wants to know
whether the sample results indicate any preference towards the shade?
The test procedure to test the above hypothesis involves specifying the cumulative
frequency distribution which would occur given the theoretical distribution under H.
and comparing that with the observed (from the sample) cumulative frequency
distribution. The point at which these two distributions show the greatest absolute
difference is determined. Reference to the sampling distribution would indicate
whether such a difference is likely to be on the basis of chance. The Kolmogorov-
Smirnov test assumes that the distribution of the variable being tested in continuous
as specified by the cumulative frequency distribution. The test statistic is given by
D = maximum { F0 (x)  Sn (x) }
Where Fo (x) is the theoretical cumulative frequency distribution (under Ho) and Sn
(x) is the observed cumulative frequency distribution, with a sample of size n.
Statistical Table 1 at the end of this unit gives the critical value for D. If the
calculated value of D exceeds this value, the null hypothesis is rejected, i.e. the
observed difference is significant.
In our above example, the calculated value of D is obtained as follows :

14
 Statistical Analysis and
Interpretation of Data :
Nonparametric Tests

For large n and at 5 per cent significance level, the critical value of D is given by
1.36
This value is equal to 0.1075. As the calculated value of D is less than the
n
critical value, the null hypothesis is not rejected, i.e., any difference in shade
preference is only due to chance.
Runs Test for Randomness
In order to draw conclusions about the population on the basis of the sample
information, it is necessary that the sample drawn must be random or unbiased. The
runs test is used to test the sample for randomness. The test is based on the order or
sequence in which the individual observations originally were obtained. A run is
defined as a sequence of identical symbols or elements which are followed and
proceeded by different types of symbols or elements or by no symbols on either side.
For example, in studying the arrival pattern of customers in a large departmental
store, we might observe the following sequence of male (M) and female (F) arrivals
MMFFFMMFFFMMMMFMMFFM
For this sample of 20 customers, we can observe that there are in all 9 runs. The total
number of runs in a sample is an indication of whether or not the sample is random.
Too few runs or too many runs indicates lack of independence or nonrandomness. In
order to use the runs test, we observe the sequence of occurrence of the n 1 + n2 = n
events (observations) and determine the number of runs, r. If both n 1 and n2 are equal
to or less than 20, then we use the statistical Table 2 at the end of this unit, to find out
the critical value for r under H0 with 0 = 0.05. If the observed value of r falls within
the critical value, we accept H 0. Otherwise, the null hypothesis (H0) is rejected. If
either n1 or n2 is larger than 20, we use normal distribution as an approximation to the
sampling distribution of r, with

The normal probability table is used to obtain the critical value of Z and the
hypothesis H0 is rejected when the calculated value of Z is greater than the critical
value.
In order to study the arrival pattern of customers at a Supermarket, the manager noted
down the arrival sequence of customers, sex-wise (M and W denote man and woman
arrival). The sequence is:
MM WWW M WW MM WWWW MMM WW MM W MMM WWW MM WW
MM WW M WW MM WW M WW MM WW
In order to conclude the randomness of the arrival pattern of the customers, we can
use the runs test.
15
The hypothesis to be tested here is:
Data Presentation and Analysis
Ho : The arrival pattern, sex-wise, of the customers at the supermarket is random.
H1 : The arrival pattern, sex-wise, of the customers at the sup market is not random.
Here
n1 = 23, nz = 27, r = 24

As α = 0.05, the critical value of Z is 1.96. As the calculated value of Z is less than
the critical value, the null hypothesis is accepted. Thus, the arrival pattern of the
customers (sex-wise) is random.
One-Sample Sign Test
In Block 4 of MS-8, all tests concerning means that you have studied are based on the
assumption that the samples are taken from a population having roughly the shape of
a normal distribution. When this assumption is untenable, the standard tests can be
replaced by nonparametric tests. Here we will discuss one such nonparametric tests,
namely, sign test.
The one-sample sign test is applicable, when sample is taken from a continuous
symmetrical population. In this case, the probability that the sample value is less than
1
mean and the probability a sample value is greater than mean are both . Suppose
2
we are testing our null hypothesis H0:   against a suitable alternative
0
hypothesis by using a sample of size n. The procedure adopted in the sign test is very
simple. Each sample value greater than µ o is replaced by a plus sign and each sample
value less-than is replaced by a negative sign. We then test the null hypothesis that
these plus and minus signs are the values of random variable having a binomial
1
distribution with p = . In case any sample value exactly equals µo, we simply omit
2
it.
For a small sample, the test is performed by computing the binomial probabilities (or
you may refer to binomial probabilities table). For a large sample, the normal
distribution is used as an approximation to the binomial distribution.
Example : It is desired to test the hypothesis that the median value µ of a continuous
distribution is 15 against the alternative hypothesis µ>15. Twenty observations were
taken and the following results were obtained :
17, 18, 16, 16, 17, 19, 14, 13, 19, 21, 22, 11, 9, 12, 14, 17, 23, 18, 17, 16
You may use  = 0.05 level of significance.
Solution : Replacing each value greater than 15 with a plus sign and each value less
than 15 with a minus sign we get:
++++++--+++---- ++++++
Now the question is whether 14 plus signs observed in 20 trials supports the null
1 1
hypothesis p = or the alternative hypothesis p > .
2 2
16
1 Statistical Analysis and
We can find that for n = 20 and p =
2 the probability of 14 or more successes is 0.05 Interpretation of Data :
Nonparametric Tests
and since this is less than or equal to  = 0.05, we reject the null hypothesis
Therefore, the median value of the distribution is greater than 15.
You may please note that in the above example both np and n (1-p) are greater than
 5. Therefore, we could have as well used normal approximation to binomial
distribution. We will illustrate this with a different example.
Example : The data given below is on a large industrial plant's daily emission of
sulphur oxides (in tons)

Use the one-sample sign test to test the null hypothesis that the plant's true average
daily emission of sulphur oxide is µ = 23.5 tons against the alternative hypothesis µ
< 23.5 tons at the 0.05 level of significance.
Solution : There are 11 plus signs and 29 minus signs.
Ho : p = 1/2
H1 : p < 1/2
1
X = 11, n = 40, P =
2

Therefore, we conclude that plant's true average daily emission of sulfur oxides is
less than 23.5 tons.
CHI-Square Test
In certain market research field studies, we have responses which can be classified
into two mutually exclusive classes, such as like-dislike, favour-not favour, etc. The
Chi-square (X2) test of goodness of fit is used here tests whether a significant
difference exists between the observed number of responses and an expected number
based on the null hypothesis in each category or class. The test statistic used here is

Where Oi and Ei are the observed and expected frequencies in the ith class, there
bring k classes. For a given level of significance (  ), if the calculated value of X 2 is
greater than the critical value, the null hypothesis is rejected.
As an illustration, let us consider a textile manufacturer who is interested in studying
the customer preference for designs before commencing a commercial production.
He conducts a survey of 100 customers all over the country and classifies their
responses in the following way:
Prefer: design I 20, design II 30, design III 18 and design IV 32.
X2 test can be used to test the hypothesis that the customers have no preference for 17
any particular design.
Data Presentation and Analysis
 Under the null hypothesis of no preference for any design, the expected (E i) and the
observed (0i) frequencies can be given as follows:

The critical value of X2 at 5 per cent level of significance with 3 (= k - 1) degrees of

freedom is 7.81473 (see table of X 2 distribution). As the calculated value of X 2 is less
than the critical value, the null hypothesis is rejected. Thus, the survey of customers
indicates no particular preference for any design.
Activity 1
The following arrangement indicates whether 25 consecutive persons interviewed by
a social scientist are for (F) or against (A) an increase in the number of crimes in a
certain locality :
F, F, F, F, F, F, A, F, F, F, F, F, A, F, F, F, F, A, A, F, F, F, F, F, F,.
Test whether this arrangement of A's and F's may be regarded as random at 5% level
of significance.
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
Activity 2
An unbiased coin was tossed 30 times and the following sequence of heads (H) and
tails (T) was observed;
H H H TT T T H H T T T H H H T T H H H T T T T H H T H H T
Find out the number of runs in the sample.
…………………………………………………………………………………………
18 …………………………………………………………………………………………
…………………………………………………………………………………………
………………………………………………………………………………………….
Activity 3
In a given year, the average number of days that a (small) sample of 15 wholesalers
of drugs and drug sundries required to convert receivables into cash were 33.9, 35.4,
37.3, 40.9, 27.8, 35.5, 34.6, 41.1, 30.0, 43.2, 33.9, 41.3, 32.0, 37.7 and 35.2 days. Use
the one-sample sign test to test the null hypothesis µ = 32.0 days against the
alternative hypothesis µ > 32 days at the 0.01 level of significance.

…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………

8.3 TWO SAMPLE TESTS

The two sample tests are used to evaluate the effectiveness (or response) of two
treatments, two methods, say, teaching or two promotional strategies in marketing. In
each of these cases, the sample which has been administered with the treatment is
compared with the one which has been administered with another treatment. For social status, etc. In
example, while comparing the two advertising strategies on two groups of problems of this
consumers, one group might react differently compared to the other not because of type, there can be
the advertisement alone, but also because of other factors such as their background, two independent
 samples, or the same sample studied again, i.e., before and after the advertising Statistical Analysis and
campaign. Interpretation of Data :
Nonparametric Tests
The Sign Test
This test is based on the sign of a pair of observations. The test is of particular use
when the measurements of observations are on a qualitative basis. Suppose, for
example, in a market research study, the researcher selects a panel of 35 people and
obtains their attitudes towards the company's product. He then shows an
advertisement and gets their response towards the product. For every person whose
rating of the product increased, we have a plus sign and for those whose ratings have
declined, we have a negative sign and for the one which remains unchanged, the
corresponding sample observation is dropped from the analysis.
The null hypothesis to be tested by the sign test is that
P (XA > XB) P (XA < XB) = 1/2
where XA is the rating before the advertisement (treatment) and X B is the rating after
the advertisement.
In case of small samples, the probability associated with the occurrence of a
particular number of +s and -s is determined with the help of the binomial
distribution with p = q = 1/ 2, n is the sample size and x is the number of fewer signs.
We will illustrate the test with the help of an example.
Example : To determine the effectiveness of a new traffic control system, the
number of accidents that occurred at a random sample of eight dangerous
intersections during the four weeks before and the four weeks following the
installation of the new system
e observed with the following results:
9 and 5 7 and 3 3 and 4 16 and 11
12 and 7 12 and 5 5 and 5 6 and 1
Use the sign test at the level of significance α = 0.10 to test the null hypothesis that
the new traffic control system is as effective as the old system against the alternative
hypothesis that the new system is more effective.
Solution : There are 6 plus signs, 1 minus and we discard one observation. We have
1
to find the probability of 6 or more success in 7 trials with p = It can be found out
2
using
binomial probability distribution that this probability is 0.063 which is less than α =
0.10. Therefore, the null hypothesis must be rejected.
Therefore, the new traffic control is more effective.
Please note that if our sample is large, we may use normal approximation to binomial
distribution and carry out Z test as explained earlier.
The Median Test
The sign test discussed above can be used only when the observations are paired, i.e.,
the elements are studied before and after the treatment. This may not always be
possible, as we may have samples of different sizes from two populations. In such a
situation, median test is appropriate. The median test is used to give information
whether the two independent samples belong to the population with the same median.
The procedure involved in the median test is to pool both the samples and then
determine the combined median. Then, the two sample observations are tabulated in 19
the form of a 2 X 2 table with respect to the median in the following way:
Data Presentation and Analysis
 If both the samples are from the population with the same median, 50 per cent of
sample's elements will be above the median and the other 50 per cent below the
median. If the number of elements (n1 + n2) in the, two samples is small, we find out

If this probability is greater than a the level of significance, the null hypothesis is
accepted, i.e., the two samples have been drawn from the same population.
In order to demonstrate the use of median test, let us consider the following example:
Example : Two different fertilizers were used to a sample of eight plots (of same
size) each. The farm yield from these plots are given below:

The researcher would like to test the hypothesis that the two fertilizers yield the same
median output. Let the level of significance be 5 per cent.
Solution : The hypothesis to be tested here is that yield by the two fertilizers have the
same median.
The median of the combined sample is 44.5 (i.e. the average of 44 and 45) Grouping
the sample elements into elements above and below the median, we get

Since the number of dements (i.e. n1 + n2) is small, we find

20

As this probability is greater than 0.05, level of significance, the null hypothesis is
accepted. Thus the yield by the two fertilizers have the same median.
When nl + n2 is large, we use X2 (Chi-square) test for accepting or rejecting the null
hypothesis. The test incorporating the correction for continuity is givers below:

Under null hypothesis, the above statistics follows a Chi-square with 1 degree of
freedom. Large values of X 2 are significant for the null hypothesis. The tabulated Example : The
values of X2 are obtained from Table 3 given at the end of this unit. Let us illustrate table below gives
the test with the help of an example. the arithmetic
addition scores for
 27 individuals. belonging to two groups. We want to test the hypothesis of no Statistical Analysis and
difference between these two sets of score. We may choose a5 per cent level of Interpretation of Data :
significance. Nonparametric Tests

Table : Scores on an addition Test.

X Y
12 7
16 12
18 14
7 18
6 5
4 16
11 9
12 10
8 14
20 3
18 18
16 9
10 7
4
Solution : Group X contains 13 scores and group Y, 14. For the set of 27 scores, we
compute the median which is found to be 11.

In case both of these distributions come from the same population, half of the X
values and half of the Y values would lie above median and half of the X values and
half of the Y values would lie below it. A contingency table is set up as follow:

Table value of X2 with 1 degree of freedom = 6.63. Therefore, X 2 value is not

significant. Therefore both these distributions come from same population.

21
Data Presentation and Analysis
 The Mann-Whitney U Test
This test is used. to test whether the two samples have been drawn from the same
population. This is a most powerful nonparametric test as it could be used for both
qualitative and quantitative data. It is a very useful alternative to the t test when you
may wish to avoid the assumption of t test.
We will discuss the following three cases for this test.
Case 1 : Very small samples (when neither of the two samples is greater than 8)
Let n1 and n2 (n1 < n2) be the number of elements in the two samples. These samples
are pooled together and the elements are arranged in the ascending order of
magnitude, the smallest element first and the largest element the last in the ordering.
The value of the test statistic U is given as below: Focusing on the sample with lesser
in size (i.e., n1 in number), U is given by the number of times that a score in the
group with n2 elements precedes a score in the group with n 1 elements, in the ranking.
For example, let the elements of the two samples be S 1 : 11, 14, 16 and S2 : 8, 10, 12
and 15, the sample sizes being 3 and 4. These elements are pooled together and are
arranged: 8(S2), 10(S2), l1(S1), 12(S2), 14(S1), 15(S2), 16(S1). To obtain the value of
U, we consider the elements of the first sample (smaller in size). In the pooled
arrangement, no element of the first sample precede elements 8 or 10 of the second
sample while one element (11) of S 1 precede 12 of S2 two elements (11 and 14) of S 1
precede 15 of S2 thus, U = 0 + 0 + 1 + 2 = 3.
To test the null hypothesis that the samples are obtained from the same population
against a suitable alternative hypothesis, you may refer to Tables 4.1 to 4.6 at the end
of this unit. These six tables are one for each value of 112, from n 2 = 3 to n2 = 8.
These tables are used when neither n1 or n2 is larger than 8.
To determine the probability under Ho, you need to know only n1 (the size of the
small group), n2 and U. In the present example, n1 = 3, n2 = 4, and U = 3. For n2 = 4
in Table 4.2 it is given that U  3 has a probability of occurrence under Ho of p =
0.200. Since this value is greater than α = 0.05, the level of significance, we have
not got enough evidence to reject Ho. You may please note that the probabilities
given to Table 4.1 to 4.6 are one-sided. For a two-tailed test, the value of p given in
the table should be doubled.
Case 2 : When n2 is between 9 and 20.
When n2, the size of the larger sample is between 9 and 20, the significance tests are
not made using Tables 4.1 to 4.6. Instead, we make use of Tables 5.1 to 5.4. The test
procedure will be explained with the help of an example given below:
Example : Suppose we want to compare the mean lifetimes of two kinds of 9 - volt
batteries on the basis of the following lifetime (in hours)
Brand A : 6.9, 11.2, 14.0, 13.2, 9.1, 13.9, 16.1, 9.3, 2.4, 6.4, 18.0, 11.5
Brand B : 15.5, 11.1, 16.0, 15.8, 18.2, 13.7, 18.3, 9.0, 17.2, 17.8, 13.0, 15.1
Using the Mann-Whitney U test, test the hypothesis that there is no difference in the
mean lifetime of the two kinds of batteries. You may choose a 5 per cent level of
significance.
Solution : We arrange the data jointly, as if they comprise one sample, in an
increasing order of magnitude. For the data given in the above example, we get the
following

22

For each value, we have indicated whether it belongs to brand A or brand B. The
rank of 1 is given to the lowest value which is 2.4 in our case. The value 6.4 gets a
rank of 2 and so on. You will find that the lifetimes of the brand A batteries occupy
ranks 1,2,3,5,6,8,9,11,13,14,19 and 22 while those of brand B occupy ranks samples.
4,7,10,12,15,16,17,18,20,21,23 and 24. However, if there
You may note that there are no ties here between values belonging to different were ties, each of
 the tied observations would be assigned the mean of the ranks which they Statistical Analysis and
jointly occupy. Suppose, if the fifth and sixth values were same, we would Interpretation of Data :
Nonparametric Tests
56
assign each the rank  5.5 , and if
2
the seventh, eight, and ninth values were the same, we would assign each the
789
rank 8
3
The null hypothesis to be tested in this case is :
Ho : Two samples come from identical populations.
The alternative hypothesis is :
H1 : Two samples come from different populations.
We obtain the value of U as follows:

Where R1 is the sum of the ranks of the values of the first sample and R 2 is the sum of
the ranks of the values of the second sample.
For our example, we may compute the values of U1 and U2 as

U1 12(13)
 12.12  113  109
and 2

U2 12(13)
 12.12  187  35
2
Please note that R1 = 1 + 2 + 3 + 5 + 6 + 8 + 9 + 11 + 13 + 14 + 19 + 22 = 113
(sum of the ranks of the values of the first sample)
and
R2 = 4 + 7 + 10 + 12 +15 + 16 + 17 + 18 + 20 + 21 + 23 + 24 = 187
Now the smaller of the two values U1 and U2 is our U which equals 35. Please note
that U1 + U2 = 109 + 35 = 144, which equals n1 n2 == 12.12 == 144.
Finally, since U = 35 is less than 37, the value of U 0.05 for two sided test given n 1 =
12 and n2 = 12, (Table 5.3), we find that the null hypothesis must be rejected.
Therefore, there is difference in the lifetime of two kinds of batteries.
Case 3 : Large sample (n2 larger than 20)
For large sample, under the null hypothesis, U has a normal distribution with

This value of Z is compared with the critical value (from the normal probability 23
table) and a de vision to accept or reject the null hypothesis is taken.
Data Presentation and Analysis
Example : Following are the scores obtained by two groups of students in a
competitive examination. Test whether the two groups belong to the same population.
Solution :
Since these
two sample
observations
are
independent, population.
we use the
Wilcoxon Matched-Pairs Signed Rank Test
Mann-
Whitney U test You would recall that in the sign test the value of the difference of scores was used
to test the only to determine its sign but its magnitude was not taken into account. The
hypothesis Wilcoxon Matched-Pairs signed rank test incorporates this additional information and
(Ho) that the attaches a greater weight to matched pair which shows a larger difference. This is
two samples therefore, a more powerful test than the sign test.
belong to the
same It is very frequent to come across two-related samples. The common a examples are-
population. We a study where wife and husband are matched or when some subjects are studied
pool both the before and after experiment or we are trying to compare the output of two machines.
samples and The null hypothesis to be tested in this case is that there are no differences in the two
rank them. groups with respect to the characteristics under study. The test procedure is as
This is done as follows:
follows 1) Let di represents the difference score for any matched pair. We rank all the di's
We shall consider the
sum of the ranks of
elements of group A.
Thus, we have R1=
148.5. Therefore,

At 5 per cent
significance,
the critical
value of Z is
1.96. As the
calculated
value of Z is
greater than the
critical value,
the null
hypothesis is
rejected. Thus,
the two groups
of students do
not belong to
without regard to their sign, i.e. rank all di
the same
's.

2)
To each rank, prefix the sign of the difference.
24 3)
If any di = 0, delete it and reduce the sample size accordingly.