|
|
Experimental design: generalization and bias
Even if a large sample is taken, it will not represent the intended
population if it is biased
In the cat food example, if the experimenters make a sample of people with
MoggyScoff in their shopping baskets, then the sample wil not represent any
useful population, however large it is. The sample is biased,
because it favours a particular outcome. A biased sample generalizes to a
biased population, but on the whole the population is not expected to be
biased, so the sample is unpresentative.
We could avoid the bias in the cat food experiment by picking cat owners at
random from a group of people we think have no reason to be biased. This is
much more difficult than it sounds. It is very difficult to determine,
let alone eliminate, all sources of bias.
|
