On 09-Aug-00 George Truman wrote:
> I possibly should know the answer to this myself but...
> If you wanted to do a prevalence study, and you expect the prevalence
> to be say, 15%, you can then calculate the expected CIs etc with
> different numbers of subjects in the study, but is there a rule of
> thumb (or firmer rule) which tells you of how tight you should try to
> get the CIs?
The only valid rule I know, which is as firm as you like, is: You choose.
That being said, there are several facets to it.
You may absolutely need at least a given precision, in which case you
go all out to get it, whatever the cost.
You may have an absolute limit on your budget/effort, in which case you
get it as precise as that allows; if it's not good enough, too bad.
Between the two, you will have to decide on the minimum acceptable
precision (assuming you can afford it), and then strike a compromise
between better precision and higher cost (in which case, Beware The
Management, who often care less about accuracy than about cost).
Further aspects: In something like a prevalence study, the prevalence
you are trying to estimate may depend on how extensive a population
you go fishing in; or (the other side of the same coin) you may want
to give differences between "strata" a chance to become visible.
On that front, (a) there is no point in chasing precision which is
finer than the change in true prevalence that might be associated with
arbitrarily changing the population under study; (b) there is no point
in creating strata which may differ by less than you are likely to
be able to detect. And so on.
It boils down to: make explicit (a) what you need to know, (b) what
you would like to know, (c) what you don't need to know, (d) what you
can get, (e) what it would cost to get it; and try to make explicit
(f) what are the tradeoffs between all these. At this point, you are
in a position to choose (see line 1).
I hope this helps,
Ted.
--------------------------------------------------------------------
E-Mail: (Ted Harding) <[log in to unmask]>
Fax-to-email: +44 (0)870 284 7749
Date: 09-Aug-00 Time: 10:41:04
------------------------------ XFMail ------------------------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|