Posted to sci.stat.consult, sci.stat.math, sci.stat.edu  where the
same note was separately posted --

On Thu, 27 Jul 2000 21:14:31 GMT, "Terry Chan" <[EMAIL PROTECTED]>
wrote:

> I have numbers (arranged in a 2x2 contingency table - smokers/nonsmokers,
> respiratory disease/no disease) for our study population (of about 1,000).
> I also have numbers (arranged the same way) for the general U.S. population
> (from the U.S. Office of Smoking and Health).
> 
> Is there a statistical test that allows me to compare both populations (i.e.
> a test that compares the chi-square value from each population)?

No, that is certainly not what you compare.  "chi-squared" is a
function of N (and design) so it is never appropriate to compare it
unless the Ns (and design) are the same.  If there is a strong
association in both samples, the U.S.  would have a test statistics
200,000 times larger; your first conclusion would have to be, "The
U.S. sample has more than  N= 1000."


> 
> Is there a statistical test that assesses if our study population is biased
> (compared to the U.S. population) and if it is biased, associates some type
> of value to this bias?
> 
> Any help will be greatly appreciated.  Thanks.

Q1.  Does the US have more/less Smokers reported?
Q2.  Does the US have more/less respiratory disease reported?

Those are preliminary and related questions.
 - keep in mind that the reporting standards are not necessarily fixed
by someone's proclamation, and the reporting is not 100% accurate even
when the standards are the same.  Having conventional overall rates in
your sample provides some encouragement that you are measuring the
same thing.

What you are aiming for, I think, is 
Q3.  Is the Odds ratio for smoking  and resp. disease the same in your
sample?

You can set this up as a log-linear model, and get the whole set of
tests.  I don't know what point you are trying to make, but if you
claim that your ratio is DIFFERENT, the first thing you have to defend
is whether your measurements are responsible.  

I suppose you might be proposing that you have  new-and-improved
criteria for "smoking" and "disease" and that is why your OR is
higher.  Other arguments might be tougher to make.

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html


=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
                  http://jse.stat.ncsu.edu/
=================================================================

Reply via email to