Paul Jones writes:
>There was some research recently linking heart attacks with
>Marijuana smoking.
>
>I'm trying to work out the correlation and, most
>importantly, its statistical significance.
With all due respect, it might be worthwhile to look at something else
first. When you are trying to interpret medical research, you ought to begin
with how the data was collected, and only after that has been done should
you worry about how it was analyzed. Most of the problems and limitations
with medical research come from data collection rather than data analysis.
First thing you need to do is to carefully define your two groups. Are you
comparing marijuana smokers to non-smokers? Are you comparing heart attack
patients to non-heart attack patients? It makes a huge difference. The first
is a cohort design and the second is a case-control design. Sometimes there
is only a single group of patients, but there is usually still some level of
comparison, because the single group is split into categories which are then
compared to one another.
When you are comparing two groups that were not created by randomization,
you need to look at the possibility of covariate imbalance. Was there an
important prognostic factor (other than your exposure variable) which was
more likely to be present in one group rather than the other?
You should then identify the outcome measure (in this case, it probably
involves a proportion of some kind). Is it a valid outcome measure? Was it
assessed in the same way in both the exposure and control group (or between
cases and controls)?
This doesn't mean that you ignore the data analysis, but the most
sophisticated statistic computed from a flawed sample is still worthless.
Furthermore, a flaw in the data analysis is easily corrected. A flaw in the
data collection is often irreparable.
A good general resource is the users guides to the medical literature. The
one most relevant to your question is available at
http://www.cche.net/principles/content_harm.asp
If you assess the questions listed in this guide, you should have an idea
about the quality of the evidence. It may make the calculation of a
statistic moot. Even if it doesn't, knowing the type of design and the type
of outcome measure will help you describe the problem well enough for us to
help.
Steve Simon, [EMAIL PROTECTED], Standard Disclaimer.
STATS: STeve's Attempt to Teach Statistics. http://www.cmh.edu/stats
Watch for a change in servers. On or around June 2001, this page will
move to http://www.childrens-mercy.org/stats
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
