Chapter 1 "What Is Statistics?" in Wild & Seber said:
A survey is sometimes used to determine the views of a well-defined group of
people. If Shere Hite’s 4500women consisted of most or all “women chief executives
of major corporations,” we would be interested in the views of the group for their own
sake. How did Hite get her women? She sent out 100,000 questionnaires to a variety
of women’s groups ranging from feminist organizations to church groups and garden
clubs.
Basically, the 4500 were just those who replied and thus form a group too illdefined
to be of any interest in itself. Were the 4500 representative of American
women in general, as Hite appeared to believe? First she sent out her questionnaires
to women’s groups. As Time stated, this strategy “means she was getting mostly one
kind of person—‘joiners’.” People who are unhappy or unsatisfied may be more likely
to be “joiners,” perhaps because they need more companionship outside of the marriage.
We call this type of problem, in which the population being sampled is an unrepresentative
subgroup of the population of interest, selection bias. An even more
important problem with the Hite study, however, is the fact that only 4.5% of those
surveyed responded. Time quoted Regina Herzog, of the University ofMichigan’s Institute
for Social Research, as saying, “Five percent could be any oddballs. We get pretty
nervous if respondents in our own surveys go under 70%.” Respondents to surveys
differ from nonrespondents in one important way: they go to the trouble of filling
out what in this case was a very long, complicated, and personal questionnaire. They
may well differ with respect to the issues under study as well. This type of problem
is called nonresponse bias. Time quoted pollster Hal Quinley as saying, “If sex was
not very important then the woman wouldn’t answer. If it was a burning issue, she
would.”
The sampling design of the Hite study was more complicated than portrayed in
our description of it to date. After the first 1500 responses, Hite made a comparison
between her respondents and the general U.S. female population and then tried to
fill the gaps to ensure a sample that was fairly representative by age, geographic location,
education levels, religion, and economic status. For example, the proportions of
Roman Catholics and Protestants in the sample were roughly the same as the proportions
in the general population. Does this make us feel happier about the results? Not
really.What matters is that we want a sample whose members are representative with
respect to their level of satisfaction with their love relationships. Every subgroup of
the population has dissatisfied members. If we are still tending to get disproportionate
numbers of dissatisfied people from each subgroup, then obtaining a sample that
is representative with respect to these other demographic variables or characteristics
has not helped us.
By drawing on other polls, the critics argued that Shere Hite’s women were not
representative. Some of the polls were conducted by reputable polling agencies while
others were conducted by magazines. For example, a Woman’s Day survey of 60,000
women and a New Woman survey of 34,000 women were quoted by Time. We have
to be careful here. Magazine and newspaper polls are sometimes readership polls in
which the survey questions are printed in the publication and interested readers send
in the completed questionnaires (self-selection). Such polls are plagued by precisely
the same problems as those of the Hite study—a population sampled that is not the
population of interest, low response rates, and atypical respondents. Time quoted
Hite as admitting she didn’t conduct a truly scientific study. “It’s 4,500 people. That’s
enough for me.” Having a large number of respondents is of no help. If the sample isn’t
representative, we are obviously no better off with a sample of 60,000 or 100,000, or
even 1 million respondents, than we are with 100.
The biases mentioned above would ensure that a certain portion of the female
population ruled themselves out of being surveyed by the method used. This group
of women could not be represented in the final sample. Clearly any sensible sampling
method must, at the very least, allow any woman from this group to have the same
chance of being selected as any woman not in the group. Applying this to every possible
group, what is needed is a random sample, namely one in which every woman
in the population has the same chance of being selected. If the randomly selected
sample was too small, then certain groups of women may not be represented simply
by chance. However, if the survey was large enough, the sample would be representative
in the sense that the makeup of the sample would reflect the makeup of the
population.