Skip to content

Relative Risk in logistic regression

10 messages · aminreza Aamini, nalluri pratap, Kevin E. Thorpe +3 more

#
Hi all,
I am very grateful to all those who write to me
1) how i  can  obtain relative risk (risk ratio) in logistic regression in R.
2) how to obtain  the predicted risk for a certain individual using
fitted regression model in R.

Many thanks, in advance, for your help.

Amin.
#
On 01/30/2013 09:02 AM, nalluri pratap wrote:
Only if you fit using the log link.  Using the logit link, this gives 
odds ratios.

  
    
#
I am not sure why one would want a relative risk from a logistic regression. The measure of association from a logistic regression is the odds ratio, not the relative risk.
John

 
John David Sorkin M.D., Ph.D.
Chief, Biostatistics and Informatics
University of Maryland School of Medicine Division of Gerontology
Baltimore VA Medical Center
10 North Greene Street
GRECC (BT/18/GR)
Baltimore, MD 21201-1524
(Phone) 410-605-7119
(Fax) 410-605-7913 (Please call phone number above prior to faxing)>>> "Kevin E. Thorpe" <kevin.thorpe at utoronto.ca> 1/30/2013 11:14 AM >>>
On 01/30/2013 09:02 AM, nalluri pratap wrote:
Only if you fit using the log link.  Using the logit link, this gives 
odds ratios.

  
    
#
On 01/30/2013 11:17 AM, John Sorkin wrote:
Yes, the natural measure, when using the logit link, is the OR.  I 
always use that myself, but if you use the log link, you get RR.  As for 
why you would, I would suggest it's because physicians think they 
understand a RR better than and OR.

  
    
#
If you use a log link, you are not, I believe, performing a logistic regression!

 
John David Sorkin M.D., Ph.D.
Chief, Biostatistics and Informatics
University of Maryland School of Medicine Division of Gerontology
Baltimore VA Medical Center
10 North Greene Street
GRECC (BT/18/GR)
Baltimore, MD 21201-1524
(Phone) 410-605-7119
(Fax) 410-605-7913 (Please call phone number above prior to faxing)>>> "Kevin E. Thorpe" <kevin.thorpe at utoronto.ca> 1/30/2013 11:22 AM >>>
On 01/30/2013 11:17 AM, John Sorkin wrote:
Yes, the natural measure, when using the logit link, is the OR.  I 
always use that myself, but if you use the log link, you get RR.  As for 
why you would, I would suggest it's because physicians think they 
understand a RR better than and OR.

  
    
#
On 01/30/2013 11:26 AM, John Sorkin wrote:
I guess strictly speaking, that is true.  I was being a little sloppy in 
terminology.

  
    
#
On Jan 30, 2013, at 5:49 AM, aminreza Aamini wrote:

            
You obtain the predicted probabilities with something like:

predict(model, data.frame(x1="a", x2=30), type = "response")

See ?predict.glm

This would give the odds ratios (similar but larger than the risk  
ratios):

exp(coef(model))
#
I am curious why one would want risk ratios.  Unlike odds ratios, they are
not interpretable without reference to the base risk.  For example a risk
ratio of 2 cannot possibly apply to anyone with a starting risk exceeding
1/2.

I think it is most helpful to use one of the existing nomograms to show
someone how the base risk and odds ratio translate to final risk, for a
range of base risk.

Frank

David Winsemius wrote

            
-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context: http://r.789695.n4.nabble.com/Relative-Risk-in-logistic-regression-tp4657040p4657163.html
Sent from the R help mailing list archive at Nabble.com.