Convenience Sample: You grab individuals or a group that is handy, and measure all of them. No attempt at a probability sample.
Sampling frame: A list of individuals that is not the whole population, but that you treat as the whole population. You draw a random sample from the sampling frame. (Using the Campus Directory to sample students, or a Medical Society membership list to sample doctors, would be using a sampling frame.) The sampling frame idea and the convenience sample overlap a little, since the sampling frame is probably some convenient list. But you do try to get a probability sample from your sampling frame, even if the frame is not perfectly representative of the population.
Voluntary response sample: You make a wide appeal for answers and let whoever wants to answer it (no attempt to get a probability sample). Especially prone to Nonresponse bias. Send-in, phone-in, magazine surveys are the paradigm. Mail surveys to specific lists can slip into this category if there is no attempt to encourage response or to follow up on the non-responders. (Is this undercoverage or nonresponse bias? Fits both ideas.)
---------------
No matter how careful you are of your sampling method, you can still
incur Nonresponse bias if the people in your sample don't
respond.
Nonresponse bias: Your sampled individuals don't all give you
answers, and the individuals who don't answer are different from those
who do (that's what makes for bias.) Of course it's very hard to
know
if the individuals who don't answer are different, because they don't
answer.
Nonresponse bias can happen even if you are careful about choosing the
sample you ask.
(Response bias: People answer but their answers are "off," conscious or unconscious lying. Sometimes from form of question.)
- - - - - - - - - - - - - - - - - -
Mail surveys that attempt to cope with the nonresponse problem use
several strategies:
Paying you to respond--a dollar bill in the envelope
creates a sense of obligation; a gift certificate if you answer; entry
into a sweepstakes if you answer; a promise of an analysis of the
results
of the survey if you answer.
Reinforcement--phoning you to urge you to send it
in.
Coding the surveys so that they can track who
responds,
and sending second or even third and fourth attempts to the
nonresponders.
If they do this, they can compare the people who responded first to the
people who responded on later appeals, and see if they are
different.
They still don't know for sure about those who never respond,
of
course.
| Sievers home | Math151-Fall04/Samplekinds.htm | 10/12/04 |