RE: st: metareg of failure rates or proportions

["Khairallah, Carole [khaicar]" <Carole.Khairallah@liverpool.ac.uk>]

Thank you very much for your reply, very clear. 
And now I am confident in not log-transforming my outcome.

Carole Khairallah


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of David Hoaglin
Sent: 21 June 2013 15:37
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: metareg of failure rates or proportions

Carole,

As I understand it, -metareg- expects effect estimates whose behavior can reasonably be approximated by a normal distribution.  That's the main reason for using log of odds ratio and log of risk ratio.  It might be all right to use a proportion or a rate without transformation.  Proportions or rates based on rare events may need a transformation or a different analysis.

In a meta-analysis of proportions the use of inverse-variance weights is a potential source of bias and should be avoided.

I distinguish rates from proportions.  A proportion has a counted numerator and a counted denominator and may often be modeled by a binomial distribution.  Most rates have a different type of denominator.  Their estimated variances may not be so troublesome, depending on how they are calculated.

You mentioned "betas from a previous regression model."  If those betas come from studies that have not used the same set of predictor variables in their regression models, they are generally not comparable.  For example, they may be the logs of adjusted odds ratios with adjustments for different sets of covariates.

David Hoaglin

On Fri, Jun 21, 2013 at 7:11 AM, Khairallah, Carole [khaicar] <Carole.Khairallah@liverpool.ac.uk> wrote:
> Hi all,
>
> I would like to run a meta-regression on proportions using the metareg 
> command in Stata v12.1
>
> I've seen that odds ratio, hazard ratio, etc. have to be log-transformed (log e or natural log) - which leads to use the betas from a previous regression model as effect sizes in metareg.
>
> But if the outcome of interest is a proportion, particularly failure rates computed from -strate ? A log-transformation (should write ln-transformation) doesn't make sense there?
>
> Thanks for your help
> Carole
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/