Page 1 of 1
Relative risk ratios in R2MLwiN
Posted: Thu Aug 18, 2016 3:28 pm
by timmfrerk
Hi Everyone,
I was wondering if there is an implemented option (as in Stata) or an easy workaround to get the estimates of the fixed effects parameters in Relative risk ratios when using multilevel unordered response multinomial models?
Thank you
Timm
Re: Relative risk ratios in R2MLwiN
Posted: Thu Aug 18, 2016 5:09 pm
by ChrisCharlton
There is not a built-in display option for this, however can you perform the transformation yourself by simply using the exponential function on the fixed part estimates. An example of this for the first model in chapter 10 of the user guide (
http://www.bristol.ac.uk/cmm/media/r2ml ... rGuide10.R) would be:
Code: Select all
library(R2MLwiN)
data(bang, package = "R2MLwiN")
(mymodel1 <- (runMLwiN(logit(use4) ~ 1 + lc, D = "Unordered Multinomial", data = bang)))
exp(mymodel1@FP)
Re: Relative risk ratios in R2MLwiN
Posted: Fri Aug 19, 2016 7:03 am
by timmfrerk
Thanks for the quick reply and easy solution.
Re: Relative risk ratios in R2MLwiN
Posted: Tue Aug 23, 2016 2:21 pm
by adeldaoud
Hi
Observe, that when you take the exponential of the log-odds you will get the odds ratio not the relative risk ratio. The relative risk ratio is more complicated to obtain: see e.g. Osborne, 2006, Bringing balance and technical accuracy to reporting odds ratios and the results of logistic regression analyses.
I am myself in the midst of figuring out how to calculate relative risk ratios from a logistic regression in Mlwin. Chris do you or the CMM-team have any suggestions for this?
Best,
Re: Relative risk ratios in R2MLwiN
Posted: Wed Aug 24, 2016 2:35 pm
by GeorgeLeckie
Hi Timm,
Just to clarify. Stata refers to the exponentiated regression coefficients from a multinomial logistic regression as "relative-risk ratios". This is the what the original post requested and Chris's proposed solution gives the user these ratios.
http://www.stata.com/manuals14/rmlogit.pdf
You are right in saying that the exponentiated regression coefficients from a multinational logistic regression are also odds ratios. Note, however, that the odds involved in these ratios are "improper odds" that is strictly speaking they are conditional odds (conditional that the response is either the numerator focal category or the denominator base category). This is why Stata uses the terminology they do as they reserve the term odds ratio for ratios of "proper" odds.
Best wishes
George
Re: Relative risk ratios in R2MLwiN
Posted: Wed Aug 24, 2016 3:00 pm
by adeldaoud
Thanks for the clarification, George.
Sounds that the language that STATA is using is a bit confusing/misleading compared to what the statistics community tend to use.
Nevertheless, is it possible to obtain relative risk from an ordinary logistic regression from Mlwin?
Best,
Adel