R help - Jun 2003 - Estimates for proportions

Hello,

I need to get a point estimate and SD for a proportion, but the subjects'
data
are not binary---they are proportions (of doses received). That is, I have a 
proportion for each subject. In the past I have analysed these data as a 
continuous (normal) variable, but I really don't want CIs over 100%. This 
seems like basic stuff, but I don't remember learning it and it's
proving
difficult to find (in medical statistics texts, anyway). Any pointers would be 
greatly appreciated!

Thanks!

Tanya Murphy
Dept. Epidemiology
McGill University

1.  If none of the numbers are 0 or 1, I might try a logit 
transformation log(p/(1-p)).  Then I'd make a normal probability plot of 
the transformed variables to check the transformation.  If that seemed 
OK., then I'd do the computations on logit space and back transform the 
result.

	  2.  If some of the numbers are 0 or 1, I'd shrink everything from 0 
and 1 using p0 = (c0+(1-2*c0)*p), then log(p0/(1-p0)).

	  Have you considered this?

hth.  spencer graves

Tanya Murphy wrote:
> Hello,
> 
> I need to get a point estimate and SD for a proportion, but the
subjects' data
> are not binary---they are proportions (of doses received). That is, I have
a
> proportion for each subject. In the past I have analysed these data as a 
> continuous (normal) variable, but I really don't want CIs over 100%.
This
> seems like basic stuff, but I don't remember learning it and it's
proving
> difficult to find (in medical statistics texts, anyway). Any pointers would
be
> greatly appreciated!
> 
> Thanks!
> 
> Tanya Murphy
> Dept. Epidemiology
> McGill University
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help

Thanks Spencer!

That's what I was looking for. My data are not at all normal and neither are
the transformed values because The majority are at an upper limit that is not 
1 (these are grouped data abstracted from a paper). There's nothing that can
be done about that, but it was good to compare the CIs for the original and 
back-transformed values. The SD for the transformed disitribution was huge 
compared to the one from the original values, though. Is this normal?

Tanya