Random Sampling
When determining cause-and-effect relationships, and in many other
questions, science requires random sampling.
"Random" here means that every individual in a population is equally likely
to be included in any sample taken, or at least that individuals with a
certain intersting property are no more likely to be included than
individuals without that property. That's all it means.
Say we're trying to assess the side effects of Upslap, a drug that is widely
used in the treatment of mopery. Some doctors have noticed that some of
their patients who take Upslap also experience chronic hiccups. Scientists
decide to find out if Upslap really causes hiccups and, if so, how strongly
it does so.
Three studies are conducted. Each one studied a small proportion of the
people who take Upslap and compared their hickuppery with the hickuppery of
the general population, and then generalized their result to all Upslap
takers.
Study one found that their sample of Upslap takers had 20% more chronic
hiccupers than the general population, and concluded that Upslap takers in
general were 20% more likely to have hiccups than non-Upslap takers.
Study two found that their sample of Upslap takers had 3% more chronic
hiccupers than the general population, and concluded that Upslap takers in
general were 3% more likely to have hiccups than non-Upslap takers.
Study three found that their sample of Upslap takers had no
chronic hiccupers, and concluded that Upslap actually prevented
hiccups.
Upon investigation, it was found that:
Study one obtained its sample by asking randomly selected consumers of
Coldkey, a popular remedy for grouchery, snappishness and hiccups if they
also took Upslap, and included them in its sample if they did.
Study two obtained its sample by asking randomly selected residents of
Bakersfield if they took Upslap, and included them in its sample if they
did.
Study three obtained its sample by asking randomly selected professional
actors if they took Upslap, and included them in its sample if they did.
So which study is random with respect to incidence of hiccups?
Study one is bad because it picks out people who are likely to have hiccups
already, so it runs the risk of including people who have hiccups from some
other cause than Upslap.
Study three is bad because people who suffer from hiccups have a very hard
time working as professional actors, and so excludes hiccupers from its
sample.
Bakersfield has no known relationship to hiccups, so it's random with
respect to hiccups.
This is what random sampling means.
Copyright © 2005 by Martin C. Young
This Site is Proudly Hosted
By:
