On Tue, 23 Mar 2004 13:42:43 -0600, frank wrote:

Technically I agree with you but that goes for any survey since I
don't know one that surveys the entire population including the United
States census bureau.

No survey needs to contact the entire universe to provide a valid sample.
In fact the numbers required to get a result with a 5% confidence level
are very small.

What is important is that the sample be representative of the makeup of
the universe being sampled. If someone declines to participate, they
would be replaced by a member of the subset that the original person
represented.

Basing a survey on only CR subscribers is enough to skew the sample. The
fact that it is a voluntary survey that depends on the subscriber taking
the time to fill out the lengthy questionaire and then return it further
exacerbates the problem.

Also since someone with a bad experience is more motivated than someone
with no complaints one has to wonder about the experiences being reported
through the voluntary process. If in doubt, look at this group. The
times someone posts a postive experience are very, very rare.

In summary, the CR reliability numbers are interesting but of limited
value.