Professional Documents
Culture Documents
Sampling in Marketing Research
Sampling in Marketing Research
Sampling in Marketing Research
Author(s): Thomas T. Semon, Reuben Cohen, Samuel B. Richmond and J. Stevens Stock
Source: Journal of Marketing, Vol. 23, No. 3 (Jan., 1959), pp. 263-273
Published by: American Marketing Association
Stable URL: http://www.jstor.org/stable/1248266
Accessed: 23-06-2016 10:37 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.
American Marketing Association is collaborating with JSTOR to digitize, preserve and extend access to
Journal of Marketing
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
SAMPLING IN MARKETING RESEARCH
THOMAS T. SEMON, REUBEN COHEN, SAMUEL B. RICHMOND,
AND J. STEVENS STOCK
This condensation was prepared especially for THE JOURNAL OF MARKETING from the
report of the same title published as No. 3 of the Marketing Research Techniques Se-
ries by the American Marketing Association. It is a survey of problems and practices in
sampling for marketing research.
The report is part of the program of the Marketing Research Techniques Committee
created in 1955 under Everett R. Smith, Chairman, to provide discussion and clarifica-
tion of current marketing-research methods.
FOR drawing conclusions about any THE SAMPLE AND THE "POPULATION"
group-people, plants, stores-sam-
Samples provide otherwise unknown
pling offers a practical compromise be-
information about some whole "popula-
tween certainty (actual knowledge) and
tion" such as "all residents of Nebraska,"
expediency (use of a single observation,
or a guess). or "all gasoline stations in cities of
In general, certainty is too expensive, 100,000 or larger in the U. S.," or "all
and the single observation, or guess, too users of neoprene gaskets more than .030
unreliable. thick," etc. The information provided by
Generalizations regarding the whole the sample is an estimate of what would
group can come from study of a small be found if the whole population were
part, properly selected. These generaliza- studied in the same manner.
tions are subject to certain degrees of
doubt, whose probable limits can some-
Need for Precise Terms
times be stated mathematically, some-
times not. A precise definition of the elements
Like marketing research as a whole, that make up the population to be
sampling is not a science, but rather a studied is the first step in any sampling
craft or discipline, using scientific tech- problem. Does the phrase "gasoline sta-
niques wherever applicable. tions" really describe the universe for
the particular problem? Does it include
both owned and leased stations? Are
* About the Authors. Thomas T. Semon, Group
Chairman, is a Research Consultant on the staff of marine stations included, and pumps
Stewart, Dougall & Associates, Inc., and Vice-Chair- operated by private companies for their
man of the Marketing Research Techniques Com- own vehicles? Or is the definition of the
mittee.
Reuben Cohen, Chief Statistician at Opinion Re- universe "all outlets that sell gasoline at
search Corporation, was for several years associated retail to the public"? If so, it will include
with Federal Government survey research activities.
Samuel B. Richmond, Associate Professor of many garages, automobile dealers, mo-
Economics and Statistics, Graduate School of Busi- tels, etc., not usually classified as gasoline
ness, Columbia University, is the author of Princi-
ples of Statistical Analysis (Ronald Press, 1957), station operators. Even the apparently
and is active as a consultant. clear phrase "in cities of ioo,ooo or
J. Stevens Stock collaborated in the development larger" can cause confusion, since "city"
of the first national probability sample twenty
and "market" are terms often used
years ago; he now heads his own research firm, J.
Stevens Stock Research Company. loosely and interchangeably.
263
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
264 THE JOURNAL OF MARKETING January 1959
264 TH JUNAO ARETNJnur15
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
SAMPLING
SAMPLINGIN MARKETING RESEARCH
IN MARKETING 265
RESEARCH 265
being sampled is inversely proportional lation. But the need for care in the selec-
to the number of eligible persons in the tion of sampling units is the same
household, and members of large fam- whether the sample is stratified or not.
ilies are thus seriously undersampled.
From a practical point of view, the TWO CLASSES OF DESIGN
interviewing of only one person per
household is usually sound, to avoid con- There is a bewildering array of sam-
pling methods currently used and talked
ditioning, and improve dispersion. But
about in marketing research. But all of
the adjustment for household size is es-
them fall into one of two classes: Proba-
sential, and often forgotten.
bility and Non-Probability.
STRATIFICATION According to statistical theory, a sam-
ple should be so drawn that every ele-
Most samples can be improved by ment in the whole has a known positive
splitting the population into smaller chance of being drawn. Knowledge of
populations according to factors relevant the probability of each element's selec-
to the subject being studied. These
tion permits one to determine the prob-
smaller populations ("strata") are then ability of the results being correct,
sampled separately. These important within stated limits, as an estimate for
factors are then not subject to sampling the whole from which the sample is
error in the study, and the sampling drawn.
error of the results is reduced.
Efficient use of stratification may re- Problems of Probability Sampling
quire that the different strata be sampled One might wonder: Why bother with
at different rates. Some parts of the popu-
non-probability samples whose probable
lation are then knowingly favored in statistical error cannot be determined?
the sample selection, and proper com- In marketing research, this is quickly
pensation is made statistically in the answered: We cannot spread families or
tabulation of the survey results. The plants out on a table like pennies, and
choice of disproportionate sampling rates pick them out with our fingers. We can-
depends on cost factors, and on the rela- not ask them to come to a convenient
tive variability or homogeneity of the hall so that we can line them up for the
population in the different strata with purpose; we have to get at them where
respect to the subject studied. they are. Usually there is no complete
Stratification is often confused with list available, and a geographic "frame"
has to be used.
the requirement that a sample be repre-
sentative of the whole. Stratification does To avoid selection bias (the choice of
not insure a representative sample ex- the cleaner-looking store, or the home
cept with respect to the criteria used. with no dog in the front yard), a "blind-
The confusion stems from the concept fold" is essential to true probability
that a sample represents the population sampling. Unfortunately a practical
if certain group characteristics-for ex- "blindfold" system requires expensive
preparations and field procedures.
ample, the proportions of men and
Once we have an unbiased system for
women-are the same in the sample as drawing sample units, we may still fail
in the whole.
to get an interview. If we substitute or
Stratification may guarantee this type change the selection, we deny any chance
of similarity between sample and popu- of selection to those homes where no one
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
266 THE JOURNAL OF MARKETING January 1959
2TEV6T U A OM K IJn
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
SAMPLING
SAMPLINGIN MARKETING RESEARCH
IN MARKETING 267
RESEARCH 267
stairs, tenant upstairs, or some other ter, and count the number of dwelling
arrangement), a systematic interval that units in the whole block. Then the sur-
is an even number, such as every other, vey results from that cluster can be
every fourth, etc., will yield either ten- weighted in proportion to the change in
ants only or owners only, unless proper size since the Census data were collected.
precautions are taken. This latter method requires that the in-
terviewer use the precise Census defini-
The Size-of-Block Problem tion of "dwelling unit," a difficult prob-
Blocks vary in size. Some have no lem in field operation.
homes in them, while other blocks with If probability of selection of a block
large apartment houses may have thou- was proportional to "size," a block of
sands of dwelling units. If one of these size zero had no chance of being selected.
large blocks were selected, too much of Therefore, homes newly built in formerly
the sample would be concentrated by vacant blocks must be represented by
chance into a limited, possibly homoge- use of a supplementary sample of "zero
neous, area. One solution is to stratify blocks." Omission of this step excludes
the blocks by size, selecting a higher pro- recent housing developments.
portion of these large blocks for inclu-
Cluster Size
sion in the sample, but within each of
these large blocks selecting a propor- Generally speaking, a sample is more
tionately smaller fraction of the dwelling precise the smaller the cluster. On the
units. other hand, a sample is less expensive to
Another plan is to make the chance of execute the larger the cluster, because of
a block being included in the sample less interviewer travel. This gives rise to
proportional to the number of dwelling the concept of an optimum-sized cluster.
units in that block, and let each block The fixed cost of interviewer travel is
selected yield the same number of sam- often so large per cluster that it remains
ple homes. The simplest way to do that best to adjust the size of the cluster to
is to assign a fixed number of homes by one or more whole interviewer's work-
some random or systematic method. days. This puts a premium on being able
However, unless the statistics used in the
to select clusters of about equal size, such
area selection are very recent, there is an as are selected with probability propor-
error due to obsolescence; that is, the tional to the estimated number of dwell-
population is sampled as it was distrib- ing units in the block.
uted when the statistics were compiled.
To avoid this bias, instead of assigning
TYPES OF SAMPLES
a fixed number of interviews, the sample
assignment is a fraction or segment such There is great variety in sample types,
that, according to the available statistics, and frequently the same type of sample
it would yield the desired number of can be either a probability or a non-
homes. Then, if new homes have been probability design. "Area Samples" may
built, the sample will reflect that growth or may not be probability samples, and
automatically. This is called "half-open the same goes for "List Samples." These
interval" sampling. terms are thus incomplete descriptions;
Another way of dealing with this to identify a sample merely as an "Area
problem is to have the interviewer make Sample" reveals that a frame of geo-
a fixed number of interviews in the clus- graphic units was used, but tells nothing
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
268 THE JOURNAL OF MARKETING lanuarv 1959
2J /
about the method of selection at any researchers claim that they are often bet-
stage of the sampling process. ter than probability samples, especially
in small (less than 2,500 interviews) na-
Non-Probability Samples tional samples.
In non-probability sampling, the no- High-grade area-quota samples are ex-
menclature of sample types is anything pensive compared with most other non-
but standardized. The word "quota" fell probability types, but still cheaper than
into disrepute some time ago, and many most probability samples.
euphemisms were invented, such as
"Judgment," "Purposive," or "Con- Samples Without Area Control
trolled" samples. Whatever it is called, Quota samples without area control
the quota sample is the main type of leave the interviewer far wider discre-
non-probability sample. tion in the choice of respondents, and do
An exception is the directed-route not insure dispersion of interviews, ex-
type, sometimes mislabeled as a proba- cept for the nature of the quota controls
bility sample. This sample uses geo- and possible restrictions on the maxi-
graphic units that may be selected by mum number of interviews to be ob-
probability methods, but usually are not. tained in any one block or section.
Within these area units (blocks, or road- Where the population to be studied is
way segments, or simply starting points, small and widely dispersed (for example,
with a path of canvassing prescribed) the owners of i6-mm. movie projectors, per-
interviewer has no discretion in the se- sons who have traveled to South Amer-
lection of the respondents. This resem- ica, etc.), an ordinary area-quota sample
bles a probability design, but the selec- may be drawn to find the incidence and
tion probabilities are unknown. Under distribution of such persons, but the
some conditions, though, there is little number found may be too small for anal-
practical difference between this type ysis. In that case, a list of eligible persons
and a "probability" sample without can be generated by asking respondents
call-back.
in the over-all sample for the names and
Non-probability area samples predes- addresses of persons they know who fill
ignate areas within which the interview- the specifications, and these can then be
ing must be done, and a fixed number
interviewed as a supplementary sample.
of interviews to be completed in each This device is frequently the only eco-
area. For tight control, the areas can be
nomical way of studying special groups.
very small in relation to the assigned It is sometimes called a "self-generating"
interviews per area. At the other ex- sample, and tends to be biased in favor
treme, to prevent excessive concentra- of persons well-known in their commu-
tion of interviews, the "areas" may be nity.
large sections of a city. Quotas on re- Traffic samples-persons, or cars pass-
spondent characteristics usually are su- ing a stationary interviewer, or a roving
perimposed on a city-wide basis, to avoid interviewer picking respondents in pub-
unwarranted assumptions about the lic places or streets-can be probability
block-by-block distribution of the char- samples, but these require elaborate or-
acteristics controlled. ganization, rigid supervision, and usiu-
Tightly-controlled area-quota samples, ally large crews. In all traffic samples a
when properly designed and carried out, person's opportunity of being selected is
can yield very accurate results. Some proportional to his frequency of passage
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
SAMPLING
SAMPLINGIN MARKETING RESEARCH
IN MARKETING 269
RESEARCH 269
of the sampling station during the sam- data; the definitions used may differ; the
pling period. published data may be inaccurate; the
List samples use the simplest frame questions may have been asked in a dif-
there is for probability sampling; but ferent context.
unclustered probability list samples are This discussion relates only to sam-
often uneconomical, unless the interview pling errors, and in no way concerns
can be conducted by telephone. A non- itself with errors that may be due to poor
probability design allows some leeway in questionnaire design, faulty coding or
the selection of respondents. This lee- tabulation, misunderstandings between
way can be narrow or broad; the inter- interviewer and respondent, etc.
viewer can be instructed to interview
"three out of these five," or "three out of Two Classes of Errors
these twenty." The best compromise be-
Errors in sampling can be divided into
tween economy, dispersion, and selec-
tion bias will depend on the nature of
two classes: Inherent Sampling Error,
and Errors of Procedure.
the list, the length of the interview, and
the funds available. Inherent Sampling Error cannot be
avoided, but it can be measured and
Panels cannot be probability samples
controlled at tolerable level at known
of the general population, but can repre-
risk by probability design. Thus, while a
sent only that part of the public willing
to cooperate in a panel. certain percentage of samples of a given
design will lead to wrong conclusions
about the population, the analyst must
ERRORS IN SAMPLING
rely on the low probability of occurrence
Accuracy vs. Precision of such aberrant samples and act on the
The words "accuracy" and "precision" assumption that his particular sample is
not one of these.
are often used rather loosely and inter-
changeably, but there is a very important Errors of Procedure in sampling can
difference between the two concepts. Ac- be avoided, in theory. In practice, they
curacy means freedom from error. Pre- are more serious than the Inherent Sam-
cision refers to the reproducibility of pling Error. If they are part of the de-
sample results; if many samples of the sign, they tend to occur in all samples
same size were taken from the same pop- which that design might yield. In effect,
ulation by the same methods, the con- the difference between the target popu-
sistency among the results obtained from lation and the population actually sam-
pled is an Error of Procedure.
these various samples indicates the pre-
cision of that particular result based on On the other hand, poor field work, an
samples of this type. Error of Procedure in execution rather
The statistical "sampling error" of than design, is likely to result in incon-
probability sample results is often mis- sistent biases that are hard to allow for.
takenly used as a measure of accuracy. The elaborate field-control systems used
Actually it is virtually impossible to by competent research agencies serve to
measure the accuracy of sample results, minimize errors in execution that could
whether probability samples or not. otherwise have serious effects.
Comparison with published data,
sometimes called "validation," is unsatis- Non-Response
factory in most cases. The sample results Non-response bias is the error of pro-
are more up-to-date than the published cedure that has received the most atten-
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
270 THE JOURNAL OF MARKETING270 TH ORA FMREIGJnay
January 1959
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
SAMPLING
SAMPLINGIN MARKETING RESEARCH
IN MARKETING 271
RESEARCH 271
COSTS, SAMPLE SIZE, AND EFFICIENCY such results may be taken to define a dis-
Marketing research costs have in- proportionate sampling plan.
creased markedly in the past decade, to- In probability samples, therefore, the
gether with the quality of the work. One bases for setting the precision require-
simple reason lies in the upgrading of ment are: (i) How small a difference in
the survey result would call for a differ-
interviewers' pay scales, necessitated in
part by the fact that a higher caliber of ence in action taken? (2) How sure
should we be that this difference is not
personnel is needed to carry out today's
more refined sampling and question- just a matter of sampling error?
naire designs in a reliably consistent These questions would have to be an-
manner. swered in advance if the survey were part
Today an average interview based on of an Operations-Research program. To
a probability sample including sample be sure that the survey will meet its ob-
design, processing, and tabulation, is jectives without excess expenditure, they
likely to run $1o per interview, or should be answered in any event.
How much risk is allowable in the
higher. In so-called motivation research,
the cost of "depth" interviews may be reliance on the survey findings? Tradi-
three to five times as high. Chances are tionally, a tolerance of two standard de-
that in the future these costs will rise viations has been used in marketing re-
further. search. Translated into odds, that is 19
An offsetting factor lies in the increas- to i-a pretty high degree of certainty. It
ing realization that size of sample is a is doubtful that much in the way of busi-
ness venture would be undertaken if
consideration usually secondary to the
design itself. A well-designed sample odds of 19 to 1 were demanded before
with few interviews still can provide use- proceeding. A less stringent criterion
ful approximations, while a poorly de- might be practical in many cases.
signed sample of l o,ooo will probably Any decision on sample size must be
yield misleading data. Still, in order to tempered by a practical consideration.
have a usable base for analysis, a certain Regardless of the precision formulas,
minimum number of interviews must be anyone would be reluctant to take action
made. Note that the criterion is "num- on the basis of a difference that depends
ber of interviews," not "number of on two or three interviews, so that small
punch cards" or "tabulation cases." sample sizes are not likely to be accepted
as conclusive unless the difference found
Determining Sample Size is markedly greater than that we have
accepted as a criterion; and cost will
In probability sampling, the number
probably preclude use of very large sam-
of interviews needed can be determined
ple sizes. These latter considerations also
mathematically, once the desired preci- apply to determination of size of non-
sion and estimated variability of the criti-
probability samples.
cal result are decided on. The critical
In sampling-error quotations, there is
result in any study is that important re-
often confusion between percentages
sult requiring the greatest sample size to
and percentage points. The formulas for
achieve the desired precision in view of
standard error yield answers in percent-
the estimated variability. The formulas age points. The standard error of a sta-
are found in most textbooks on sam-
tistic of 30 per cent incidence might be
pling. If the critical result pertains to +3 per cent. That means three percent-
only a section of the population, a set of age points. But the projection of the
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
272 THE JOURNAL
272 THE JOURNAL OF MARKETING lanuary
OF MARKETING January 1959
1959
30 per cent incidence to the population our object is to study the screwdriver
results in an estimate with a standard market, it is obviously the latter, and the
error of +? io per cent of itself. population includes a wide miscellany
of retail stores, from dime stores to su-
Statistical Efficiency permarkets.
Because clustering reduces dispersion, Projectability of results is essential for
it reduces sample efficiency as well as estimating share of market, or forecast-
cost. The efficiency of a sample is meas- ing industry needs. Sampling require-
ured by the answer to: "How much ments may be less stringent for studies of
larger (or smaller) would a plain random attitudes. The sample needed depends on
sample have to be for equal precision?" the reason for the study and its nature.
A random sample (unclustered, unstrati- What is an establishment? As used by
fied) has an efficiency of i.o. Clustering the Census of Manufactures, it is not the
reduces efficiency, stratification usually in- same as a "reporting unit" according to
creases it. A common efficiency factor for County Business Patterns. In turn, both
stratified probability cluster samples is of these may differ from the definition of
0.5. (That means that for a given result, a "company," "plant," "chain," or "buy-
the statistical error is i: V/o.5, or about ing office" in a private directory or list.
1.4 times as great as in a plain random In drawing samples of establishments,
sample of the same size.) it is of paramount importance to define
Because of the effect of clustering, exactly what is meant by a given type of
most research executives have strong establishment, and to make sure that this
reservations about national population definition corresponds to the "popula-
samples below a certain size. Generally tion" of interest for the problem at
speaking, bases of less than 1,5oo are hand. In the case of industrial plants, the
viewed as too small; and 2,500 is often complications multiply.
considered to be an acceptable mini-
Procedures
mum, although the size must depend
largely on what is being studied. For It is useful to distinguish between
those types of establishments that are
some inquiries, l o,ooo might be just
barely adequate. widespread and numerous (such as most
retail stores, garages, dry cleaners, and
SAMPLING OF ESTABLISHMENTS other personal services) and less numer-
ous types. The area-sampling procedures
In theory, there is no difference be- used in consumer research can often be
tween sampling of people and sampling applied where the population of estab-
of establishments such as manufacturers,
lishments is widespread, but they tend
supermarkets, hospitals, gasoline sta-
to be less practical in the smaller popu-
tions, and so on. In practice, the diffi- lations.
culties are varied and vexing.
In sampling establishments, personnel
of considerably higher caliber and train-
The Importance of Definitions
ing than the usual field interviewer is
The basic design may be "probability" often needed, and the over-all cost per
or "non-probability," but the definition interview can easily reach a level ten or
of the target population is crucial, and more times that of the average consumer
the purpose of the research must be kept interview. As a result, small samples are
in mind. Are we to study hardware the rule rather than the exception. A
stores, or outlets selling screwdrivers? If partial reason is that, for sampling pur-
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms
SAMPLING
SAMPLING IN MARKETING RESEARCH
IN MARKETiNG 273
RESEARCH 273
poses, establishments are far more com- problem is raised by studies designed to
plex units than persons or households. measure a change in time.
In studying the market for an indus- If it is possible to re-interview all the
trial product, an establishment may be same persons, no sampling error exists
a customer or potential customer; but it as a result of differences between succes-
may be difficult to determine whether an sive samples studied. However, it is often
interview is required with the purchas- felt that the respondent once inter-
ing agent, design engineer, maintenance viewed has been "conditioned," and is
engineer, vice president in charge of pro- unduly aware or critical with regard to
duction, division manager, controller, or the subject matter of the inquiry. There-
with several or all of these. It is not al- fore, a new sample is needed in most
ways possible to specify in advance the cases.
This content downloaded from 132.236.27.217 on Thu, 23 Jun 2016 10:37:00 UTC
All use subject to http://about.jstor.org/terms