7 - 9
a certain percentage, and that a certain percentage will decline to participate. All this effort, and we have
not even begun to talk about collecting data from these children.
From this example, it is clear that random sampling can require an incredible amount of financial
resources. As noted earlier in the chapter, we have two options. We can define the population more
narrowly (perhaps the 5- to 15-year-olds in a particular school district) and conduct random sampling
from this population, or we can turn to a sampling technique other than probability sampling. Before we
discuss these nonprobability sampling techniques, let’s look at one other form of probability sampling.
Stratified Random Sampling
This procedure known as stratified random sampling is also a form of probability sampling. To stratify
means to classify or to separate people into groups according to some characteristics, such as position,
rank, income, education, sex, or ethnic background. These separate groupings are referred to as subsets or
subgroups. For a stratified random sample, the population is divided into groups or strata. A random
sample is selected from each stratum based upon the percentage that each subgroup represents in the
population. Stratified random samples are generally more accurate in representing the population than are
simple random samples. They also require more effort, and there is a practical limit to the number of
strata used. Because participants are to be chosen randomly from each stratum, a complete list of the
population within each stratum must be constructed. Stratified sampling is generally used in two different
ways. In one, primary interest is in the representativeness of the sample for purposes of commenting on
the population. In the other, the focus of interest is comparison between and among the strata.
Let’s look first at an example in which the population is of primary interest. Suppose we are
interested in the attitudes and opinions of university faculty in a certain state toward faculty unionization.
Historically, this issue has been a very controversial one evoking strong emotions on both sides. Assume
that there are eight universities in the state, each with a different faculty size (faculty size = 500 + 800 +
900 + 1,000 + 1,400 + 1,600 + 1,800 + 2,000 = 10,000). We could simply take a simple random sample of
all 10,000 faculty and send those in the sample a carefully constructed attitude survey concerning
unionization. After considering this strategy, we decide against it. Our thought is that universities of
different size may have marked differences in their attitudes, and we want to be sure that each university
will be represented in the sample in proportion to its representation in the total university population. We
know that, on occasion, a simple random sample will not do this. For example, if unionization is a
particularly “hot” issue on one campus, we may obtain a disproportionate number of replies from that
faculty. Therefore, we would construct a list of the entire faculty for each university and then sample
randomly within each university in proportion to its representation in the total faculty of 10,000. For
example, the university with 500 faculty members would represent 5% of our sample; assuming a total
sample size of 1,000, we would randomly select 50 faculty from this university. The university with 2,000