[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
RE: st:Confidence interval of difference between two proportions and -csi-
I don't think there is any question of revising
the formula. Rather, I suspect that there is an
excellent case for adding options to -cs- to give more
flexibility to the user.
I see the case as very similar to that of
-ci, binomial-. There's been a long history
of different methods here and even an increasingly
widespread realisation that the so-called exact method
(i.e. the Clopper-Pearson method) is often not
as satisfactory as other methods. (In one case, the
Wilson score method, that predates Clopper-Pearson.) On
15 July 2003 -ci- and -cii- were extended to include
new options -wilson-, -agresti-, and -jeffreys- for computing
different types of binomial confidence intervals.
I suggest that -cs- needs a similar work-over.
> Thank you Roger for your suggestion of using -exactcci-
> However, this does not calculate a risk difference and the 95% CI.
> I suppose I am suggesting to the folks at Stata that a
> revised formula be
> used in -csi- to calculate the 95% CI of the risk difference
> when there is
> a small number of observations, and one or both have a 100%
> risk. Currently
> it is possible for the upper bound to go beyond the
> theoretical maximum of
> 100% difference.
> I would appreciate opinions on this by others on the list.
> Newcombe RG (1998) Interval estimation for the difference between
> independent proportions: comparison of eleven methods. Statistics in
> Medicine 17: 873-890
* For searches and help try: