Page 96 - Design for Six Sigma for Service (Six SIGMA Operational Methods)

Customer Survey Design, Administration, and Analysis 77

on the list and a sample of 1500 people is required, a random sampling
approach might be unrealistic, simply because numbering 3,000,000 people
is already a big task. If the original list of these 3,000,000 is randomly dis-
tributed, we can select sample units by selecting them from the list at fixed
intervals (every nth entry, for example, every 20th car on the highway, every
50th customer in a store). In this example, if we want to select 1500 people,
because 3,000,000/1500 = 2000, we can select 1 out of every 2000 people
in the sampling frame. In this case, if we start with a random starting point
and then select a person after we count every 2000th sampling unit, this
procedure will create a random sample of 1500 people.
Stratified Random Sampling

Stratified sampling assumes that the sampling frame consists of several
mutually exclusive groups, called strata. In stratified random sampling, the
total number of samples is divided among strata by a predetermined pro-
portion. Then, random samples are taken from each stratum. For
example, in a community, assume that 60 percent of the population is white,
15 percent is black, 15 percent is Hispanic, and 10 percent is Asian. If a
sample of 1000 people is needed, the stratified sampling method will divide
these 1000 people into four ethnic groups based on the proportion in the
population. So 600 samples will be allocated to whites, 150 samples to
blacks, 150 samples to Hispanics, and 100 samples to Asians. Then these
600 people in the “white” strata will be randomly selected from the white
sampling frame, 150 samples of blacks will be randomly selected from the
black sampling frame, and so on.

Cluster Sampling
Cluster sampling deals with the situation in which there is a hierarchy of
sampling units. The primary sampling unit is a group (or cluster), such as
counties, cities, schools, or subdivisions. The secondary sampling units are
the individual elements within these clusters from which the information is
to be collected. For example, if we want to study the needs of first and
second graders, it is difficult to directly locate the sampling frame from a
raw population list, such as a telephone directory. It is easy to identify the
clusters, such as public and private schools, in which there are first grade
and second grade classrooms. After we select a subset of classrooms, we
can randomly select sample units from these classrooms.

Nonprobability Sampling
In nonprobability sampling, the probability that a particular unit will be
selected is unknown. In this case, we cannot generalize the finding within the
sample to the population because we cannot assume any valid statistical