Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Question regarding metaregression

From   Austin Nichols <>
Subject   Re: st: Question regarding metaregression
Date   Thu, 10 Feb 2011 11:39:52 -0500

Chan Yoon <>:
I assume you are referring to a relative risk (ratio) point estimate y
and a confidence interval with lower bound x and upper bound z which
are not equidistant from y, i.e. they were computed as the antilog of
the endpoints of a confidence interval around ln(y).

The log of the relative risk ln(y) has a sampling distribution that is
approximately normal, and the resulting confidence interval (lnx,lnz)
is symmetric around ln(y), i.e.,
ln(x) = ln(y) - SE*1.96
ln(z) = ln(y) + SE*1.96
approximately, where 1.96 should be replaced with the appropriate
critical value. So your SE is
SE = [ln(y)-ln(x)]/1.96 = [ln(z)-ln(y)]/1.96
and if you have x, y, and z measured with finite precision, maybe you want
SE = [ [ln(y)-ln(x)]/1.96 + [ln(z)-ln(y)]/1.96 ]/2

On Thu, Feb 10, 2011 at 5:06 AM, Chan Yoon <> wrote:
> I have a problem using metareg commend.
> To use commend. wsse (selogrr) I have to have selogrr value.
> But I only have rr with 95% confidence interval( 95% ci)
> Is there a way I can use metareg with rr and upper limit and lower limit of 95% ci? Or is there a way to obtain selogrr from rr and ci?
> I would very much appreciate your help. Thank you.
> Chan.
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index