[R-sig-ME] geometric mean regression

Ahmad ahmadr215 at tpg.com.au
Mon Apr 2 13:20:46 CEST 2018


I forgot to say that my data are on natural log scale- and I agree with x100 if we want see the difference in %.
This is a work for a Pharma company, they are interested in geometric means rather than arithmetic means (because data is not normally distributed).  


-----Original Message-----
From: Cole, Tim <tim.cole at ucl.ac.uk> 
Sent: Monday, 2 April 2018 8:45 PM
To: Ahmad <ahmadr215 at tpg.com.au>; Kevin E. Thorpe <kevin.thorpe at utoronto.ca>
Cc: r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] geometric mean regression

Hi Kevin and Ahmad,

Back transformation is not tricky on the natural log scale. Just multiply the coefficients by 100 and view them as differences in percentage units – see https://doi.org/10.1136/bmj.j3683s .

Best wishes,
 mailto:tim.cole at ucl.ac.uk Phone 020 7905 2666 Population Policy and Practice Programme UCL Great Ormond Street Institute of Child Health,
30 Guilford Street, London WC1N 1EH, UK

Date: Sun, 1 Apr 2018 13:14:16 -0400
From: "Kevin E. Thorpe" <mailto:kevin.thorpe at utoronto.ca>
To: Ahmad <mailto:ahmadr215 at tpg.com.au>
Cc: <mailto:r-sig-mixed-models at r-project.org>
Subject: Re: [R-sig-ME] geometric mean regression
Message-ID: <mailto:bd0017d4-17b5-be6e-f7d0-0a32979a9fa1 at utoronto.ca>
Content-Type: text/plain; charset="utf-8"; Format="flowed"

Back transformation can be tricky. You should also look at smearing estimators. The package Hmisc has a function called smearingEst() that you might like to check.


On 03/31/2018 06:37 PM, Ahmad wrote:
Hi Kevin
Thanks for your email,
Yes, I almost figured out how get this done. I needed to get the exp() of intercept for the reference group and exp() of coefficient*exp(intercept) for the other group.
When I was trying this for geometric of 95%CI, the results don't seem quite right. I found an article that if I get the exp() of lsmeans (emmeans) these will produce the correct geometric outputs. Not sure why when I do these manually using the exp() of intercept and coefficient of lm- the outputs are not identical, but close enough.
-----Original Message-----
From: Kevin E. Thorpe <mailto:kevin.thorpe at utoronto.ca>
Sent: Sunday, 1 April 2018 12:22 AM
To: Ahmad <mailto:ahmadr215 at tpg.com.au>
Subject: Re: [R-sig-ME] geometric mean regression Maybe I'm missing something, but doesn't linear regression on log(y) accomplish this?
On 03/29/2018 08:25 AM, Ahmad wrote:
Hi All

I have a dataset and I have been asked to generate geometric means from the linear regression for different groups (2 groups).

In fact my data is repeated measures, and I intend to use a mixed-effects regression model with repeated measures. But I thought I can learn how to do this for a simple geometric mean regression, I should be able to translate this into a mixed model.

Any help would be greatly appreciated!



More information about the R-sig-mixed-models mailing list