Вы находитесь на странице: 1из 22

4(a)(ii) Sampling Methods………….Contd.

• STRATIFIED RANDOM SAMPLING :

In this method the population is first divided in to mutually exclusive


groups or strata and then a simple random sample is chosen within
each strata/group.

For example human population can be divided in to strata/group based


on age, occupation, income, gender. After that samples can be drawn
from each stratum by simple random sampling (SRS).

The stratification is done on the basis of characteristic of interest. The


elements within a stratum should be homogeneous while those in
different strata should be as heterogeneous as possible, from the point
of view of characteristic of interest.

1
Reasons for Stratified Random Sampling:
1. When marketers want information about the component parts
of the universe.
Example: MDH Pickle sales in “large”, “medium” and “small” stores.
In such a case, separate sampling from within each of these strata
would be called for. The results might then be used to plan different
promotional activities for each store-size stratum.
2. Stratified random sampling is also used to increase the
precision of sampling estimates.
Example: To estimate monthly MDH Pickles sales per store in a
universe of 100,000 grocery stores. A typical frequency distribution of
store sales would be as depicted in Figure 11-1. Very few stores would
have very high sales. A larger number would have moderate sales, and
an even larger number would have small sales. Page.393 2
MDH Pickles sold

A small random sample from this universe would tend to yield


a sample mean with a large sampling error. The presence or absence
(by chance) in the sample of a few very large stores could impact
significantly on the sample mean value obtained via simple random
sampling.
3
We can subdivide this store universe into three strata, A, B, C, as indicated in
Figure 11-2. A relatively small sample taken within each stratum would provide a
good estimate of the mean of the stratum because of the similarly of the items
included in that stratum. The estimated means of these strata could then be
weighted together so as to provide an estimate of the mean of the entire universe.

MDH Pickles sold

Stratified random sampling will provide greater precision ( a smaller standard


error of the mean) than a simple random sample of the same total size.
4
2. ESTIMATION OF THE UNIVERSE MEAN: A stratified
random sample is a group of simple random samples, the sample mean of each
stratum is an unbiased estimate of the actual mean of the stratum. Therefore,
the individual stratum sample means can be combined (weighted ) into as
unbiased estimate of the overall universe mean. Thus, the estimate of the
overall universe mean is simply a weighted average of the strata sample means.
Store Size Sample Mean Number of store
% of store
Stratum Unit sales/ store n
Large 200 20,000 20%
Medium 80 30,000 30%
Small 40 50,000 50%
Total 1,00,000 100%
To estimate the universe mean of monthly sales, each stratum sample mean is
multiplied by its relative weight (i.e., percent of all stores), and the results are
added together. In this illustration, the estimated universe mean is
(200)(20%) + (80)(30%) + (40)(50%) = 84 units per store.

Page.3955
3. CONFIDENCE INTERVAL ESTIMATION WITH STRATIFIED
RANDOM SAMPLES:
An estimated standard error of the mean is first secured. Then the appropriate
multiple of this figure (e.g., Z=2 for 95.4 percent confidence) is added to and
subtracted from the estimated mean to provide the desired confidence limits.

It involves three elements:


a) The size of the sample within each stratum,
b) The relative size of each stratum compared to the total universe, and
c) The variance within each stratum.

Once obtained, the estimated standard error of the mean is used to construct a
confidence interval in the manner described in the preceding paragraph.

6
4. THREE ISSUES IN THE SELECTION OF STRATIFIED RANDOM
SAMPLES:
4.1 How should the Universe Be Stratified?
Homogenous among sampling units within strata, and Heterogeneous among strata.
As a result, a relatively small sample within each stratum will provide a precise
measurement of that stratum’s mean. The weighting together of the different stratum
sample means, will, generally, provide a better estimate of the universe mean than would be
provided by a simple random sample of the same total number of units.
4.2 How many Strata Should Be Constructed?
Higher the numbers of strata more homogenous will be stratum
However, practical considerations limit the number of strata such as Cost. . Secondly it
requires separate listing for all the strata and in many cases these lists are not available.
4.3 How many Observations Should be Taken in Each Stratum?
Proportional Allocation: The most obvious way is to use proportional allocation. Here, one
samples each stratum in proportion to its relative weight in the universe as a whole.
Disproportional Allocation: As a general principal, when the variability among observations
within a stratum is high, one samples that stratum at a higher rate than for strata with less
internal variation 7
4.4 Concluding Remarks on Stratified Random Sampling:
a) Information can be obtained about different parts of the universe,
b) It often provides universe estimates of greater precision than
simple random sampling.
The price paid for these advantages is greater complexity of both
design and analysis. A separate list of the items within each stratum
is required. To be of maximum value, strata should be constructed
such that the variable being studied varies only a little bit within
strata – but a great deal among strata.
It is most useful when sampling highly skewed universe where items
vary greatly in size.

8
5. CLUSTER SAMPLING:
In this method the population is divided into different clusters.
Usually the clusters are formed in terms of geographical area
(Area sampling), membership of a club/ organization. They are
widely used in the sampling of human populations.

Example: - House hold in a locality is a single stage cluster. If


divided further by roads or blocks then it is an example of
multistage clusters. After clusters are formed they are listed. Then
clusters randomly chosen based on a probability sampling technique.
From the chosen cluster either all the elements are included in the
sample or a sample of elements is drawn probabilistically from each
of the clusters chosen.

9
5.1 Relatively Efficiency of Cluster Sampling and Simple Random Sampling:
Cluster sampling will be more statistically efficient if each cluster represent most of
the possible observations obtained from the universe. If each cluster represents only
a few different universe observations, then cluster sampling will be less statistically
efficient than SRS.
Higher economic efficiency – that is, the relative cost per observation –cluster
sampling is the superior to SRS. Gain in economic efficiency usually offset the decline
in statistical efficiency .
5.2 Concluding Remarks on Cluster Sampling:
1. Cluster sampling may be either more or less statistically efficient than simple random
sampling. This depends on the degree of intracluster heterogeneity obtained.
2. In practice, clusters are often constructed in such a way that the observations within a
cluster are relatively homogeneous. When this is the case, cluster sampling will be less
efficient statistically than simple random sampling.
3. The lower relative cost of obtaining observations in cluster sampling often offsets the
loss in statistical efficiency. The net efficiency is often greater for cluster sampling.
10
Two types of cluster sampling – systematic sampling and area sampling – are
discussed below.

6. SYSTEMATIC SAMPLING:
This is a simpler method and is easier to practice than simple random sampling. In
this method at first all the units are listed (thus each is allocated a number). After
this we decide the sample size either by statistical formula or by judgment.

Let the sample size be 'n' and the population be 'N' . Then the fraction N/n is
calculated. Say it is 600/30 = 20. Then as a first step, a number is chosen from 0 to 9
at random (say 4). Then the units 4, 24, 44, 64, 84, 104 are. chosen as samples.
Example: Assume one wishes to study dentist’s attitudes towards dental insurance and decided
to sample 20 dentists from a list of 100 dentists. One way of doing this is as follows:
1. Draw a random number between 1 and 5. Assume the number chosen is 2.
2. Include in the sample the dentists numbers 2,7,12,17,22,--------,97. That is, starting with
number 2, take every fifth number.
The above is an illustration of systematic sampling. That this is a particular kind of
cluster sampling is readily seen if all possible samples produced by this procedure are
considered. This particular example has only five possible samples. 11
Sample Identification of Dentists in Sample

1 1,6,11,16,21…………………….96

2 2,7,12,17,22…………………….97

3 3,8,13,18,23…………………….98

4 4,9,14,19,24…………………….99

5 5,10,15,20,25…………………..100

Each of these samples amounts to a cluster of dentists, and each has a


chance of 1 out of 5 of being included in the sample. In effect, the original universe
of 100 dentists has been redefined as a universe of five clusters of dentists, and each
cluster is given an equal chance of being chosen by the sampling process.
Page. 402
12
6.2 Advantages of Systematic Sampling: The principal advantage of this
technique is its simplicity. When sampling from a list, it is easier to choose a
random start and select every Kth item thereafter than to make a simple random
selection. The technique is faster and less subject to error than simple random
selection. Hence, systematic sampling is often used in place of simple random
sampling.

6.3 Disadvantages of Systematic Sampling: If the order is considered random,


the standard error can be evaluated as the simple random sampling. Alphabetical
lists are usually assumed to be in random order. If the order of the items on the
list is not random, then estimating the standard error of the mean requires more
complex methods.

6.4 Concluding Comments on Systematic Sampling: Appealing, easy to


execute, and valid under most circumstances, used mush more often than simple
random sampling. It is often combined with stratification, being used in place of
simple random sampling to choose sample items within strata.
13
7. AREA SAMPLING:
* Area sampling is a special form of cluster sampling in which the sample items are
clustered on a geographic area basis. The practical motivation underlying area
sampling is that for many problems there is no current and accurate list of universe
elements.
* The original universe of interest- for which there is no list- is transformed into a
universe for which there is a list. Such a list consists of city blocks, or PIN code
areas, or countries, or other geographically defined area that can be identified on
maps.
* A rule of association, uniquely linking each item in the universe of interest to a
single physical area, is established. By drawing a probability sample of area and using
the rule of association, one obtains a probability sample from the universe of
interest.
7.1 Application of Area Sampling: It enjoys wide usage in situations where very
high quality data are wanted but for which no list of universe items exists. For
instances, many governmental agencies use area sampling. However, the practical
execution of a large scale area sample is highly complex. Typically, an area sampling is
conducted in multiple stages, with successively smaller area clusters being sub-
14
sampled at each stage.
7.2 Concluding Comments on Area Sampling: Execution of area sampling designs
are expensive and time consuming. They also require that researchers address a
number of important questions concerning the sample design. For example:
* How strata should be built; how area sampling units should be defined at each stage.
* How many area sampling units should be selected at each stage and what selection
process should used; and so on.
* For best results, substantial information (detailed maps, statistical data by areas, cost
data) is needed and expert statistical counsel is required.

8. NONPROBABILITY SAMPLING: Any sampling procedure that does not


specify the chance of selecting any universe element is a nonprobability sampling
method, no matter what else is included in the specifications.

15
8.1 Convenience Sampling: In this the choice of the sampling units is done
primarily by the interviewer. The interviewer chooses anyone who is conveniently
available during the time of interview. The elements might or might not be
representative of the target population.

An extreme example is soliciting opinions from people conveniently in camera range


(also SMS on TV). The major problem with this (and other nonprobability methods)
is that one is unable to draw objective inference about a rigorously defined universe.
Convenience sampling is sometimes useful in exploratory work, to help understand
the range of variability of response in a subject area, just taking to a few consumers
may help identify issues.
8.2 Judgment Sampling: Specialists in the subject matter of the survey choose what
they believe to be the best sample for that particular study. For example, a group of
sales managers might select a sample of grocery stores in a city that they regarded as
“representative”. This approach has been found empirically to produce unsatisfactory
results. And, of course, there is no objective way of evaluation the precision of
sample results. Despite these limitations, this method may be useful when the total
sample size is extremely small.
16
8.3 Quota Sampling: As in stratified random sampling, the researchers begins by constructing
strata, Bases for stratification in consumer surveys are commonly demographic, e.g., age, sex,
income, and so on. Often compound stratification is used- for example, age groups within sex.
Next, sample sizes (called quotas) are established for each stratum. As with stratified random
sampling, the sampling within strata may be proportional or disproportional. Field-workers are then
instructed to conduct interviews with the designed quotas, with the identification of individual
respondents being left to the field-workers.
*Concluding Comments of the Use of Quota Sampling: Owning to this relative economy and
speed of execution, quota sampling will continue to enjoy wide usage. Because it uses the principal
of stratification, this method is likely to be superior to ordinary convenience sampling or judgment
sampling.
However, a quota sample should not be mistaken to be stratified random sample, because the
two differ importantly. The difference lies in how the samples are selected within strata. With
stratified random sampling, the process is objective, based on random identification of respondents.
In quota sampling, the process is subjective, being done by field-workers using what amounts to
convenience sampling. As a result, quota sample respondents may differ- in ways very important to
the study’s purpose-when compared with respondents selected at random. Because respondents are
not selected objectively, confidence interval statement cannot be made in quota sampling with the
same legitimacy as when probability sampling has been used.
Page. 408
17
9. SPECIAL FORMS OF NONPROBABILITY SAMPLING:
9.1 Group Interview Sample: A “group” is usually a quota sample of 5-10
consumers possibly all products users, nonusers, and so on) assembled for a one to
two hour joint interview by a person specially trained in group dynamics. Usually
several such groups are conducted.
The number of “groups” used is often small for two reasons: they are expensive and
the information obtained from each group is highly unstructured because it is based
on a simultaneous, free-form interview with a number of people. Researchers usually
learn as much as this method will provide from just a few group interviews. The
number of respondents per group is restricted to about 10, because of the practical
difficulties experienced in maintaining control over a larger group.
9.2 Shopping Mall Intercept Samples: Respondents are recruited for individual
interviews (using assigned “quotas”) at fixed locations in shopping malls. It is
common in a given study to use several cities. Mall sample advantages are speed and
economy relative to a probability sample of personal interviews. Their limitations
derived from the expediency of selecting the sample and the atypical character of the
population sampled (i.e., mall customers).
Page. 410 18
9.3 Controlled Panel Samples: This technique, pioneered by national Family
Opinion, Inc. (NFO), of Toledo, Ohio, is an elaborate and highly controlled form
of quota sampling. NFO and similar organizations have developed huge files of
names, addresses, telephone numbers, and a wide array of demographic
characteristics for households willing to be interviewed by mail of telephone. Using
computers, they have constructed “panels” (often of 1,000 households each) that
approximately replicate the U.S. household universe in demographic characteristics,
such as age, income, and the like, known to be related to consumer attitudes and
behavior. In a given project one or more panels are interviewed via mail or
telephone to obtain the requisite information.
A major advantage of this approach is that large national samples can be provided
relatively cheaply and easily.
The disadvantages of controlled panel samples are those attendant on any quota
sample plus the obvious bias that such samples are comprised of people who are
willing to be included in panels and to participate in surveys from time to time.
Such bias may be particularly acute when the panel is a continuing one that
demands extensive cooperation over an extended interval.
19
10. CHOICE OF SAMPLE DESIGN IN PRACTICE:
10.1 Quality of Sample Design Required: The “Quality” of sample design required varies
from one problem situation to another. It is convenient to think of a continuum of “quality of
sample design required”, as schematized below:
Quality of Sample Design Required

Extremely Low Extremely High

(Convenience sampling, from an (Probability sampling, from most


accessible universe) relevant universe)

Page. 411
In some situations, a low-quality sample design, as represented by
convenience sampling from as accessible universe, may be adequate. An example
would be an exploratory study to help define issues when virtually nothing is known
of the subject.
At the other extreme, exceptionally high-quality data will sometimes be
necessary. If the analyst requires universe estimates with calculable precision from a
universe that corresponds closely to the universe of interest, then probability 20
sampling is essential.
10.2 Choosing a Sample Design: While it is clear that “quality of sample design required” is
the key consideration, there are no wholly objective methods of determining this quality.
Judgment and experience necessarily play a crucial role in assessing how much “quality” is
required and how best to proceed.
For large-scale studies, supported by major resources, some form of probability
sampling will be the tentative initial choice. From that beginning, other factors, will influence
the final decision.

11. COMPROMISE SOLUTIONS: PROBABILITY SAMPLING


ADAPTATIONS: Many researchers have used compromise solutions to the sampling
problems posed.
11.1 Area Sampling Modifications: One such compromise involves the selection of
interview areas, using probability sampling, but identifying respondents for interview in the
selected areas by nonprobability methods. For example, a several stage area probability sample
might be used to identify areas (e.g., blocks) as places of interview. Then, sample respondents
could be identified for interview using a systematic pattern (e.g., every fifth house) from an
arbitrarily designated point within each sample area (e.g., southeast corner). Methods akin to
this are widely used.

21
11.2 Approximating the Relevant Universe with a Listed One:
Probability Sampling of Telephone Homes: A very common example of the
approximating approach is to assume that telephone-accessible house-holds are
equivalent to all households for a particular study. Given that some 95 percent of
households have a phone in USA and in NCR, this assumption will be satisfactory
for many consumer research purpose. The principal advantages of this approach
(versus, for example an area sample of personal interviews) are savings in cost and
time, as well as relative simplicity. The major disadvantages are those linked to
telephone interviewing.
11.3 Probability Sampling from Other Listed Universes: In similar vein,
researchers sometimes approximate probability sampling from a universe of interest
by using approximate universe based on lists obtained from trade or professional
associations, direct mail lists, and the like.A directory of professional medical
specialists, known to contain practically all members of the desired universe, may be
a virtually perfect list for probability sampling purposes. At the other end of the
spectrum, some lists (e.g., of amateur investors) are often based on periodical
subscription rosters. Typically, the composition of such lists is unknown, and their
use is generally questionable. Page 414
(Source: Boyd, Westfall & Stasch, Ch.11) (Revised by PSN in Jan 2011)
22

Вам также может понравиться