190 The Marketing Book
Probability sampling techniques
The units which constitute a probability sample
are selected randomly, with each unit having a
known chance of selection. Thus, before a
probability sample can be drawn, the project
will need to define a sampling ‘frame’ for the
population. Such a frame will need to ensure
that each unit is included only once, and that no
unit is excluded; thus, all units have an equal
chance of selection. The frame should cover the
entire population and be convenient to use.
A probability sample should attempt to be
representative of the entire population, but it
can never be an exact replica. However, by
applying the rule of probability, generalizations
concerning the population may be made and
calculations made about the degree of con-
fidence with which the results can be viewed.
Sample error, for probability samples,
stems from the variability of the sample and/or
the size of that sample.
Simple random sampling
Units are chosen at random from the popula-
tion. Individual units are assigned a number, a
sample of these numbers is then selected either
by using a ‘lottery’ system, or by the use of
random number tables.
The method is simple to use and it obeys
the laws of probability; however, it may pro-
duce samples which are not representative of
the population.
Stratified random sampling
This method accepts the variability of the
population and, by stratifying it before the
sample is taken, tries to reduce its potential
unrepresentativeness. Stratifiers, which may be
geographical, demographic etc., are imposed
on the population like a grid, dividing it into
groups whose members, inside each ‘cell’, are
as alike as possible with respect to the stratifier.
Stratified random sampling adopts the position
that each group/stratum is a population in its
own right and then extracts a sample, by simple
random means, from each of them.
Inproportionate stratified sampling, the size
of each sub-sample taken from a particular
stratum is proportionate to the size of that
stratum in the population. Thus, if 25 per cent
of the population is aged between 35 and 45,
then 25 per cent of the sample should be
composed of people in that age group. In
disproportionate stratified sampling, the propor-
tion of a characteristic, as possessed by the
population, is not reflected to the exact extent
in the size of the sub-sample. Such a deliberate
‘distortion’ of the size of the sub-sample may
improve the quality of the data if certain strata
have an unusually large influence in the situa-
tion under investigation and need to be given a
more significant role. Here, not every unit has
an equal chance of selection, but the chance of
selection is still known, thus the laws of
probability still rule and appropriate weight-
ing(s) can be used when calculating the
results.
The method’s major drawback is in finding
stratifiers relevant to the research situation.Cluster sampling
This is similar to stratified sampling in that the
total population is divided into strata, but it
differs in that instead of sampling from each
subgroup, a sampleof the strata is taken, with
simple random sampling then taking place
inside each of the selected groups. Thus, while
in stratified sampling each stratum represents a
particular subset of the population, in cluster
sampling each stratum should be a miniature
representation of the full population. It is a
method particularly useful in cases where the
population is dispersed over wide geographical
areas.
A particular form of cluster sampling is
called multistage samplingand involves more
than the single stage of the cluster sampling
system. If, after dividing a country into various
areas (counties, regions etc.), they are found to
have greatly varying sizes of populations, then