When you analyze an unbalanced dichotomy, the statistical
power is determined almost entirely by the smaller N; and your
"5" is a small N, so don't expect detect anything except *strong*
effects. Does the Event show up for one gender *only*? You can
look at the means for the Cases-only and see what leaps out as extreme.
In order to have a strong enough risk factor to generate a "significant"
statistical test, for (even) one variable, your non-100% risk factor must
be almost as skewed as the outcomes. If there are such risk factors
recorded in the data, presumably you should know about them from
your review of the relevant literature.
Five cases also gives you those cases to measure against other
suggestions in the literature. I would want to assume that that the
1500 "surveyed" represents the world of other, similar surveys,
so it is important to note whether or how that full sample of 1500
is (or is not) representative, before trusting contrasts between the
5 and 1495.
--
Rich Ulrich
> Date: Wed, 25 Sep 2013 18:02:05 -0700
> From:
[hidden email]> Subject: Rare event analysis
> To:
[hidden email]>
> Hi,
>
> I have a sample of 5 cases out of 1500 people surveyed. This is the first
> time I'm handling rare events analysis. I am interested in determining the
> risk factors. Is there a univariable or multivariable analysis that can be
> done for this sample?
>
>
>
Locked
