Marketing Research

2
Sampling

Module 005 Data: Population and Sampling

Course Learning Outcomes:

1. Understand the process of selecting sample
2. Explain the sampling procedures
3. Understand the different types of sampling and sampling strategy

Introduction
A population is the group of units about which we want to make
judgments. These units can be groups of individuals, customers, companies,
products, or just about any subject in which you are interested. Populations
can be defined very broadly, such as the people living in Canada, or very
narrowly, such as the directors of large hospitals in Belgium. What defines a
population depends on the research conducted and the goal of the research.
Sampling is the process through which we select cases from a
population. The most important aspect of sampling is that the sample
selected is representative of the population. With representative we mean
that the characteristics of the sample closely match those of the population.
Market researchers consider it important that their sample is
representative of the population. How can we see if this is so?
– The best way to test whether the sample relates to the population is
to use a dataset with information on the population. For example, the
Amadeus and Orbis databases provide information at the population level.
We can (statistically) compare the information from these databases to the
sample selected. The Amadeus database is available at
http://www.bvdinfo.com.
– You can use (industry) experts to judge the quality of your sample.
They may look at issues such as the type and proportion of organizations in
your sample and population.
– To check whether the responses of people included in your research
do not differ significantly from non-respondents (which would lead to your
sample nor being representative), you can use the Armstrong and Overton
procedure. This procedure calls for comparing the first 50% of respondents
to the last 50% with regard to key demographic variables. The idea behind
this procedure is that later respondents more closely match the
characteristics of non-respondents. If these differences are not significant
(e.g., through hypothesis tests, discussed in Chap. 6), we find some support
that there is little, or no, response bias (see Armstrong and Overton 1977).

Course Module
This procedure is sometimes implemented by comparing the last wave of
respondents in a survey design against earlier waves. There is some evidence
this procedure is better than the original procedure of Armstrong and
Overton (Lindner et al. 2001).
– Using follow-up procedures, a small sample of randomly chosen
nonrespondents can be contacted again to ask for cooperation. This small
sample can be compared against the responses that were obtained earlier to
test for any differences

When we develop a sampling strategy, we have three key choices:

– Census,
– Probability sampling, and
– Non-probability sampling.

If we get lucky and somehow manage to include every unit of the

population in our study, we have conducted a census study (so, strictly
speaking, this is not sampling). Census studies are rare because they are very
costly and because missing just a small part of the population can have
dramatic consequences. For example, if we were to conduct a census study
among directors of banks in Luxemburg, we may miss out on a few because
they were too busy to participate.
If we select part of the population, we can distinguish two types of
approaches: probability sampling and non-probability sampling. Figure 3.2
provides an overview of the different sampling procedures, which we will
discuss in the following sections.

Figure 5.1 Sampling Procedures

Probability Sampling
Probability sampling approaches provide every individual in the
population a chance (not equal to zero) of being included in the sample. This
is often achieved by using an accurate sampling frame. A sampling frame is
a list of individuals in the population. There are various sampling frames,
such as Dun & Bradstreet’s Selectory database (includes executives and
companies), the Mint databases (includes companies in North and South
 Marketing Research
2
Sampling

Americas, Italy, Korea, the Netherlands, and the UK), or telephone

directories. These sampling frames rarely completely cover the population of
interest and often include some outdated information, but due to their ease
of use and availability they are frequently used. If the sampling frame and
population are very similar, we have little sampling frame error, which is
the degree to which sample frames represent the population. Starting from a
good-quality sampling frame, we can use several methods to select units
from the sampling frame.
The easiest way is to use simple random sampling, which is
achieved by randomly selecting the number of cases required. This can be
achieved by using specialized software, or using Microsoft Excel or SPSS.3
Specifically, Microsoft Excel can create random numbers between 0 and 1
(using the RAND() function). Next you choose those individuals from the
sampling frame where the random value falls in a certain range. The range
depends on the percentage of respondents needed. For example, if you wish
to approach 5% of the sampling frame, you could set the range from 0.00 to
0.05.
Systematic sampling uses a different procedure. We first randomize
the order of all observations, number them and, finally, select every nth
observation. For example, if our sampling frame consists of 1,000 firms and
we wish to select just 100 firms, we could select the 1st observation, the
11th, the 21st, etc. until we reach the end of the sampling frame and have our
100 observations.
Stratified sampling and cluster sampling are more elaborate
techniques of probability sampling, which require dividing the sampling
frame into different groups. When we use stratified sampling, we divide the
population into several different homogenous groups called strata. These
strata are based on key sample characteristics, such as different departments
in organizations or the area in which consumers live. Subsequently we draw
a random number of observations from each strata. While stratified sampling
is more complex and requires accurate knowledge of the sampling frame and
population, it also helps to assure that the sampling frame’s characteristics
are similar to those of the sample.
Cluster sampling requires dividing the population into different
heterogeneous groups with each group’s characteristics similar to those of
the population. For example, we can divide the consumers of one particular
country into different provinces, counties, or councils. Several of these
groups can perhaps be created on the basis of key characteristics (e.g.,
income, political preference, household composition) that are very similar
(representative) to those of the population. We can select one or more of
these representative groups and use random sampling to select our
observations from this group. This technique requires knowledge of the
sampling frame and population, but is convenient because gathering data
from one group is cheaper and less time consuming.

Course Module
 Generally, all probability sampling methods allow for drawing
representative samples from the target population. However, simple random
sampling and, in particular, stratified sampling are considered superior in
terms of drawing representative samples.

Non-probability Sampling
Non-probability sampling procedures do not give every individual
in the population an equal chance of being included in the sample. This is a
drawback, because the resulting sample is most certainly not representative
of the population, which may bias results of subsequent analyses.
Nevertheless, non-probability sampling procedures are frequently used as
they are easily executed, and are typically less costly than probability
sampling methods.
Judgmental sampling is based on researchers taking an informed
guess regarding which individuals should be included. For example, research
companies often have panels of respondents who are continuously used in
research. Asking these people to participate in a new study may provide
useful information if we know, from experience, that the panel has little
sampling frame error.
Snowball sampling is predominantly used if access to individuals is
difficult. People such as directors, doctors, or high-level managers often have
little time and are, consequently, difficult to involve. If we can ask just a few
of these people to provide names and details of others in a similar position,
we can expand our sample quickly and access them. Similarly, if you post a
link to an online questionnaire on your Facebook page (or send out a link via
email) and ask your friends to share it with others, this is snowball sampling
through referrals to people who would be difficult to access otherwise.
In quota sampling, we select observations according to some fixed
quota. That is, observations are selected into the sample on the basis of pre-
specified characteristics so that the total sample has the same distribution of
characteristics assumed to exist in the population being studied. In other
words, the researcher aims to represent the major characteristics of the
population by sampling a proportional amount of each (which makes the
approach similar to stratified sampling). Let’s say, for example, that you want
to obtain a quota sample of 100 people based on gender. First you would
need to find out the proportion of the population that is men and the
proportion that is women. If you found out the larger population is 40%
women and 60% men, you would need a sample of 40 women and 60 men
for a total of 100 respondents. You would start sampling and continue until
you got those proportions and then you would stop. So, if you’ve already got
40 women for the sample, but not 60 men, you would continue to sample
men and discard any female respondents that came along.
What makes quota sampling a non-probability technique is that the
selection of the observations does not occur randomly. That is, once the
quota has been fulfilled for a certain characteristic (e.g., females), you do not
allow any more observations with this specific characteristic in the sample.
This systematic component of the sampling approach can introduce a
sampling error. Nevertheless, quota sampling is very effective for little cost,
 Marketing Research
2
Sampling

making it the most prominent sampling procedure in practitioner market

research.
Finally, convenience sampling is a catch-all term for methods
(including the three non-probability sampling techniques just described) in
which the researcher makes a subjective judgment. For example, we can use
mall intercepts to ask people in a shopping mall if they want to fill out a
survey. The researcher’s control over who ends up in the sample is limited
and influenced by situational factors.

Sample Sizes
After determining the sampling procedure, we have to determine the
sample size. Larger sample sizes increase the precision of the research, but
are also much more expensive to collect. The gains in precision decrease as
the sample size increases. It may seem surprising that relatively small sample
sizes are precise, but the strength of samples comes from accurately selecting
samples, rather than through sample size. Furthermore, the required sample
size has very little relation to the population size. That is, a sample of 100
employees from a company with 100,000 employees can be nearly as
accurate as selecting 100 employees from a company with 1,000 employees.
There are some problems in selecting sample sizes. The first is that
market research companies often push their clients towards accepting large
sample sizes. Since the fee for market research services is often directly
dependent on the sample size, increasing the sample size increases the
market research company’s profit. Second, if we want to compare different
groups, we need to multiply the required sample by the number of groups
included. That is, if 150 observations are sufficient to measure how much
people spend on organic food, 2 times 150 observations are necessary to
compare singles and couples’ expenditure on organic food.
The figures mentioned above are net sample sizes; that is, these are
the actual (usable) number of observations we should have. Owing to non-
response, a multiple of the initial sample size is normally necessary to obtain
the desired sample size. Before collecting data, we should have an idea of the
percentage of respondents we are likely to reach (often fairly high), a
percentage estimate of the respondents willing to help (often low), as well as
a percentage estimate of the respondents likely to fill out the survey correctly
(often high). For example, if we expect to reach 80% of the identifiable
respondents, and if 25% are likely to help, and 75% of those who help are
likely to fully fill out the questionnaire, only 15% (0.800.250.75) of
identifiable respondents are in this case likely to provide a usable response.

Course Module
 Thus, if we wish to obtain a net sample size of 100, we need to send out

(desired sample size)

(likely usable responses ) =100=0:15 = 667 surveys.
We will discuss how we can increase response rates (the percentage of
people willing to help).

References and Supplementary Materials

Books and Journals
Armstrong, J. S., & Overton, T. S. (1977). Estimating nonresponse bias in mail surveys.
Journal of Marketing Research, 14(3), 396–403.
Bearden, W. O., Netemeyer, R. G., & Haws, K. L. (2011). Handbook of marketing scales.
Multi-item measures for marketing and consumer behavior research (3rd ed.). Thousand
Oaks, CA: Sage.
DeVellis, R. F. (2011). Scale development: Theory and applications (3rd ed.). Thousand
Oaks, CA: Sage.
Diamantopoulos, A., Riefler, P., & Roth, K. P. (2008). Advancing formative measurement
models. Journal of Business Research, 61(12), 1203–1218.
Diamantopoulos, A., Sarstedt, M., Fuchs, C., Wilczynski, P., & Kaiser, S. (2012).
Guidelines for choosing between multi-item and single-item scales for construct
measurement: A predictive validity perspective. Journal of the Academy of Marketing
Science, 40(3), 434–449.
Diamantopoulos, A., & Winklhofer, H. M. (2001). Index construction with formative
indicators: An alternative to scale development. Journal of Marketing Research, 38(2), 269–
277.
Erdem, T., & Swait, J. (2004). Brand credibility, brand consideration, and choice. Journal
of Consumer Research, 31(1), 191–198.
Geyskens, I., Steenkamp, J.-B. E. M., & Kumar, N. (1998). Generalizations about trust in
marketing channel relationships using meta-analysis. International Journal of Research in
Marketing, 15(3), 223–248.
Kuppelwieser, V., & Sarstedt, M. (2014). Confusion about the dimensionality and
measurement specification of the future time perspective scale. International Journal of
Advertising, 33(1), 113–136.
Lindner, J. R., Murphy, T. H., & Briers, G. E. (2001). Handling nonresponse in social
science research. Journal of Agricultural Education, 42(4), 43–53.
Netemeyer, R. G., Bearden, W. O., & Sharma, S. (2003). Scaling procedures: Issues and
applications. Thousand Oaks, CA: Sage.
Sarstedt, M., & Schloderer, M. P. (2010). Developing a measurement approach for
reputation of nonprofit organizations. International Journal of Nonprofit and Voluntary
Sector Marketing, 15(3), 276–299.