|
|
Experimental design: generalization answers
In practice, the results of an experiment on a sample generalize only to
the population that it represents
A sample of ten people at a checkout queue with tins of MoggyScoff in their
shopping baskets represents the population ``people with a prior
preference for MoggyScoff or with an inclination to try it at least once'', if
it represents anything. While the result may be accurate for this population,
it becomes increasingly difficult to trust as the population widens. As a cat
food buyer, I am a member of the population ``people who need to buy cat food
but with no particular preference for any given brand''. It is impossible to say to
what degree the survey results generalize to my population.
This leads to the following principle:
A good experimental group will be chosen to represent the population which
is of interest
Ideally, an experimenter will decide in advance for what population a result is
required, and then select a sample that represents that population.
|
