[R] Odds Ratios in rms package

[Frank Harrell]

Peter- for summary.rms, high and low columns are supposed to be the
quartiles.  Perhaps Sebastian, the original poster, could run age through
the ordinary summary function or the describe function and show us the
result.
Frank


Peter Dalgaard-2 wrote
> 
> On Jun 20, 2012, at 21:05 , David Winsemius wrote:
> 
>> 
>> On Jun 20, 2012, at 12:12 PM, Sebastian Pölsterl wrote:
>> 
>>> Hi,
>>> 
>>> I'm using the rms package to do regression analysis using the lrm
>>> function. Retrieving odds ratios is possible using summary.rms. However,
>>> I could not find any information on how exactly the odds ratios for
>>> continuous variables are calculated. It doesn't appear to be the odds
>>> ratio at 1 unit increase, because the output of summary.rms did not
>>> match the coefficient's value.
>>> 
>>> E.g. print gives me:
>>> 
>>>               Coef    S.E.   Wald Z Pr(>|Z|)
>>> age              0.1166 0.0289  4.04  <0.0001
>>> 
>>> whereas summary gives me:
>>> 
>>> Factor      Low     High     Diff.   Effect S.E. Lower 0.95 Upper 0.95
>>> age         27.0000 37.00000 10.0000  0.78  0.20  0.40        1.17
>>> Odds Ratio 27.0000 37.00000 10.0000  2.19    NA  1.49        3.22
>>> 
>>> Does anybody know how these values are obtained, especially in the
>>> presence of interactions?
>> 
>> It is explained in the first paragraph of ?summary.rms, :
>> 
>> " By default, inter-quartile range effects (odds ratios, hazards ratios,
>> etc.) are printed for continuous factors,"
>> 
>> ... and the labeling makes it fairly clear (at least it was for me)  that
>> it is an odds ratio for a change in predictor value from the 25th to the
>> 75th percentile (which are the values in the Low and High columns)
> 
> Begparding? 
> 
> Seems to me that would require (37-27)*0.1166 == 0.78, which doesn't quite
> hold in the arithmetic that I was taught...
> 
> Rather the Effect/Coef ratio is about 6.69 on both the effect and its SE,
> so I think it is a fair guess (being too lazy to check the actual code)
> that Low and High columns define the full range of ages, not the
> quartiles. 
> 
> -pd
> 
>> In the presence of interactions you should not be looking at the
>> coefficients, but rather at the predictions.
>> 
>> ?Predict
>> 
>> -- 
>> David.
>> 
>> 
>>> 
>>> Best regards,
>>> Sebastian
>>> 
>>> ______________________________________________
>>> R-help@ mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>> PLEASE do read the posting guide
>>> http://www.R-project.org/posting-guide.html
>>> and provide commented, minimal, self-contained, reproducible code.
>> 
>> David Winsemius, MD
>> West Hartford, CT
>> 
>> ______________________________________________
>> R-help@ mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-help
>> PLEASE do read the posting guide
>> http://www.R-project.org/posting-guide.html
>> and provide commented, minimal, self-contained, reproducible code.
> 
> -- 
> Peter Dalgaard, Professor
> Center for Statistics, Copenhagen Business School
> Solbjerg Plads 3, 2000 Frederiksberg, Denmark
> Phone: (+45)38153501
> Email: pd.mes@  Priv: PDalgd@
> 
> ______________________________________________
> R-help@ mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 


-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context: http://r.789695.n4.nabble.com/Odds-Ratios-in-rms-package-tp4634004p4634100.html
Sent from the R help mailing list archive at Nabble.com.