Monday, 2 December 2013

Types of Sampling

Major types of probability sampling are: simple random
sampling, stratified random sampling and cluster

Simple Random Sampling

A process that gives each element in the population
an equal chance of being included in the sample
is termed as simple random sampling. The elements
are selected, using a list of random numbers appended
with most textbooks of research and statistics. Before
using the table of random numbers, it is first necessary
to number all the elements in the population to be
studied. Then the table is marked at some point
and the cases whose numbers come up as one from
this point down the column of numbers are taken
into the sample until the desired number of elements
is obtained. The selection of any given element places
no limits on other element being selected, thus
making equally possible the selection of any one
of the many possible combinations of elements.
Proportionate Stratified Random Sampling
In stratified random sampling, the population is first
divided into strata. The strata may be based on a
single criterion or on a combination of two or more
criteria. After, stratification a simple random sample
is taken from each stratum, and the sub samples
are then joined to form the total sample.
In case the researcher is interested in the study
of some characteristics of a phenomenon he uses
a proportionate stratified random sampling plan. Of
course, this sampling design presupposes that the
investigator has some knowledge concerning the
population characteristics such as age, sex, marital
status etc.
In the sampling plan the sample will have specified
characteristics in exact proportion to those same
characteristics which are distributed in the population.
To understand this sampling plan we will consider
the following example.

Let us consider the students of a College of Social
Work. The researcher wishes to have proportionate
stratified random sample of them taking year of
study in the college as basis of stratification. Let
us suppose that the students at this college are
distributed as is shown in Table below:
Table : Distribution of Students According to
Year in College

Year                     Population     Proportion of each class
BSW I                      50                        .25
BSW II                    40                         .20
BSW III                   30                         .15
MSW I                    40                         .20
MSW II                   40                         .20
Total                       200                        1.00

Further, we suppose that the researcher decides
to have a sample of 60 students. First, he determines
the proportion of students in each class (as shown
in the second column). Then he calculates the
composition of the sample taking each proportion
of the stratifying characteristics in the population and multiplying it by the desired size of the sample.
Thus, he multiplies 60, the desired sample size by
.25, the proportion of BSW first year students in
the population or
(60) (.25) = 15
As such, he has to include 15 students from the
BSW first year in his sample. This precedence is
repeated for each year as described below:

(60) (.25) = 15
(60) (.20) = 12
(60) (.15) = 9
(60) (.20) = 12
(60) (.20) =  12
Sample Size (N) = 60
Table : Distribution of Students by Proportion
Year          Sample Break-up         Proportion
BSW I       15                                .25
BSW II     12                                 .20
BSW III      9                                 .15
MSW I       12                                .20
MSW II     12                                .20

Total Sample (n) 60                        1.00

After having determined the sample size from each
subcategory, the researcher uses simple random
sampling for drawing the desired number of elements
from each category.

Disproportionate Stratified Random Sampling

This sampling plan is almost similar to proportionate
stratified random sampling except that the sub samples
are not necessarily distributed according to their
proportionate weight in the population from which
they were drawn. It is possible that some sub samples
are over represented while other sub groups are
under represented.
Let us suppose that the researcher stratifies the
population into two sub strata using sex as the criteria.
He would get the following break-up of the population:

Table : Distribution of Students by Sex

Sex                   No. of Students     Percentage
Male                160                          80
Female             40                            20
Total               200                          100

If the researcher wants to draw a disproportionate
stratified random sample of 60 from this population,
stratified by sex, then he has to draw 30 from each
substrata, this means male students (30) will be
under represented and female students (30) will be
over represented in the sample. In other words
disproportionate sampling gives equal weights to each
There is a clear improvement over simple random
sampling when the sampling is based on a stratification
of population by sex. With this kind of stratification
we get a marked increase in the size of samples
that yields statistics very close to the population
parameters. On the contrary, a reduction in the
size of sample may yield statistics that might deviate
widely from the population parameters.Cluster Sampling
In case the area of study is wide spread, a large
expenses are involved if simple and stratified random
sampling are used. For example, in the preparation
of sampling frame from the population and in covering
the widespread areas by interviewers, a large amount
of expenditure is required. The more widely spread
the area of study, the greater are the travel expenses,
the greater is the time spent in travelling, and
hence expensive — and the tasks of administering,
monitoring and supervision of the research project
and in particular supervising the field staff become
more complicated. For the reasons mentioned above
and few other reasons, large-scale research studies
make use of the methods of cluster sampling.
In cluster sampling, first the whole research area
is divided into sub area, more commonly known as
“clusters”. The simple random or stratified method
is used to select clusters. Finally, researcher arrives
at the ultimate sample size to be studied by selecting
sample from within the clusters, which is carried
out on a simple or stratified random sampling basis.
Let us suppose, for example, that we want to do
a survey of beggars in urban areas of a state. We
may proceed as follows: prepare a list of districts
and group them into clusters, and select a simple
or stratified random sample from each clusters. For
each of the districts included in the sample, list
the cities/towns and take a simple or stratified
random sample of them. If some or all of the towns/
cities thus selected for the sample have more numbers
of beggars that can be studied, we may take a sample
of these towns/cities in each district. The beggars
in these towns/cities will be the sample of the beggars.
Characteristically, the procedure moves through a
series of stages—hence the common term, “multistage”
sampling—from more inclusive to less inclusive sampling
 units until we finally arrive at the population
elements that constitute the desired sample.
The four important types of non-probability sampling
are accidental sampling, quota sampling, snowball
sampling and purposive sampling.

Accidental Sampling

Accidental sampling refers to a method of selecting
respondents who happen to meet the researcher
and are willing to be interviewed. Thus, a researcher
may take the first hundred people he meets who
are willing to be interviewed.
For example, let us consider the situations where
a programme director, wishes to make some
generalisation about the programme in progress, selects
beneficiaries who have come to the agency for a
service or a community organiser, trying to know
how “the people” feel about health status in that
community, interviews available community dwellers
like shop-keepers, daily wage earners, barbers and
others who are presumed to reflect public opinion.
In both the situations those who are available for
study are included in the samples. This is exactly
what we call accidental sampling. It is very obvious
that the sample so collected are biased and there
is no known way (other than by doing a parallel
study with a probability sample) of evaluating the
biases introduced in such samples. However, in the
situation illustrated above, most probably, accidental
sampling is the only way out because of the reason
that the population parameters of the beneficiaries
or the community people are not available with the

Quota Sampling

Quota sampling insures inclusion of diverse elements
of the population in the sample and make sure that these
diverse elements take account of the proportions
in which they occur in the population. For example,
we take a sample from a population with equal number
of boys and girls, and that there is a difference
between the two groups in the characteristic we
wish to study and we fail to interview any girl, the
results of the study would almost certainly be
extremely misleading generalisations about the
population. In practice, elements in small numbers
are frequently under represented in accidental
samples. In anticipation of such possible exclusion
of small groups, quota sampling ensures inclusion
of enough cases from each stratum in the sample.
It should be noted here that the major goal of quota
sampling is the selection of a sample that is a replica
of the population to which one wants to generalise.
Hence it should be clear that the critical requirement
in quota sampling is not that the various population
strata be sampled in their correct proportions, but
rather than there be enough cases from each stratum
to make possible an estimate of the population stratum
value (Kidder, 1981, p. 426). Quota-sampling, however,
is more or less similar to the earlier described
accidental sampling procedure except that it insures
the inclusion of diverse elements of the population.

Purposive Sampling

Purposive sampling is based on the presumption that
with good judgment one can select the sample units
that are satisfactory in relation to one’s requirements.
A common strategy of this sampling technique is
to select cases that are judged to be typical of the
population in which one is interested, assuming
that errors of judgment in the selection will tend
to counterbalance each other. For example, if a
researcher is conducting a study of patients who
are not regular in attending out patient department
it might be desirable to choose patients for the sample
 from among those who are frequently irregular.
The causes of irregularity can be described by irregular
patients only. If he selects a random sample he
would have got patients who are regular and that
might influence the findings of the study. It is also
possible that in a truly random sample, the regular
patients would nullify the effects of irregular patients.

Snowball Sampling

Snowball sampling is externally helpful in studying
some special sampling situation like getting a sample
of drug abusers, or alcoholics or pickpockets. In
snowball sampling we start with a few respondents
of the type we wish to include in our study and
who in turn are expected to guide us to get more
respondents and so on. Like the rotating snowball,
sample increases in its size as we continue to get
more units of study. The technique is especially
useful in the investigation of sensitive topics mentioned
above because this sampling technique depends on
sampled cases having knowledge of other similar
cases. Another argument in favour of using this
sampling technique is that, the victims might be
hesitant to identify themselves if approached by a
stranger but might be friendly to someone who they
know and share their experiences or deviant status
(Gelles, 1978).