[R] Results from clogit out of range?

Charles C. Berry cberry at tajo.ucsd.edu
Thu Apr 22 04:11:28 CEST 2010


On Wed, 21 Apr 2010, Frank E Harrell Jr wrote:

> Thomas Lumley wrote:
>>  On Tue, 20 Apr 2010, Noah Silverman wrote:
>> 
>> >  I just read the help page for predict.coxph.
>> > 
>> >  It indicates that the risk score is just exp(lp)
>> > 
>> >  What I'm trying to find, and have seen with some other implementations
>> >  is the "conditional probability within group".  Neither the lp or the
>> >  risk options seem to deliver this.
>>
>>  exp(lp)/(1+exp(lp))
>>
>>  or, as Martin Maechler would prefer,  qlogis(lp)
>
> Minor correction: plogis

In what context is this reasonble?

It seems to me that rather strong assumptions are needed - like that the 
strata are independent of the covariate(s).

The OP requested the "conditional probability within group", which I 
interprete to mean something like

 	exp(lp)/ave( exp(lp), stratvar, FUN=sum )

when there is just one case in each stratum.

??

Chuck


>
> Frank
>
>>
>>     -thomas
>>
>> 
>> >  What am I missing?
>> > 
>> >  -N
>> > 
>> > 
>> >  On 4/20/10 4:22 PM, Noah Silverman wrote:
>> > >  Thanks David,
>> > > 
>> > >  That explains a lot.  I appreciate it.
>> > > 
>> > >  -- 
>> > >  Noah
>> > > 
>> > > 
>> > >  On 4/20/10 3:48 PM, David Winsemius wrote:
>> > > 
>> > > >  On Apr 20, 2010, at 5:59 PM, Noah Silverman wrote:
>> > > > 
>> > > > 
>> > > > >  Hi,
>> > > > > 
>> > > > >  I'm calculating a conditional logit on some data stratified by 
>> > > > >  group.
>> > > > >  My understanding was that a conditional logit by definition 
>> > > > >  returns a
>> > > > >  value between 0 and 1 a a probability.  Can anyone suggest why I'm
>> > > > >  seeing results outside of the {0,1} range??
>> > > > > 
>> > > >  Probably because you did not read the help page for the function
>> > > >  predict.coxph. Pay special attention to the type argument.
>> > > > 
>> > > >  "type=c("lp", "risk", "expected", "terms")"
>> > > > 
>> > > > 
>> > > > >  The call in R is:
>> > > > > 
>> > > > >  m <- clogit(score ~ val_1 + val_2 + strata(group), data=data)
>> > > > > 
>> > > > >  Then
>> > > > > 
>> > > > >  prediction <- predict(m,newdata)
>> > > > > 
>> > > >  I do not see that you have defined any newdata.
>> > > > 
>> > > > 
>> > > > > 
>> > > > >  A sample of the data with resulting predictive values is:
>> > > > >      group          score   val_1               val_2 
>> > > > >  prediction
>> > > > >  1  2009-01-04_1     1 0.5913962     -1.121589  1.62455210
>> > > > >  2  2009-01-04_1     1 0.6175472     -3.249820 -0.20093346
>> > > > >  3  2009-01-04_1     1 0.5439640     -2.424501  0.46651849
>> > > > >  4  2009-01-04_1     0 0.3745209     -2.477424  0.31263855
>> > > > >  5  2009-01-04_1     0 0.6329855     -3.424174 -0.34200448
>> > > > >  6  2009-01-04_1     0 0.4571999     -2.770247  0.11190788
>> > > > >  7  2009-01-04_1     0 0.3822623     -2.259422  0.50627534
>> > > > >  8  2009-01-04_1     0 0.2605742     -4.424806 -1.44566070
>> > > > >  9  2009-01-04_1     0 0.4449604     -2.357060  0.46174993
>> > > > >  10 2009-01-04_1     0 0.6595178     -2.246518  0.69427842
>> > > > >  11 2009-01-04_1     0 0.5260032     -2.977887 -0.02393871
>> > > > > 
>> > > >  By default (which is what you have implicitly chosen) you are
>> > > >  requesting "lp" = linear predictors rather than "risk".
>
>>  Thomas Lumley            Assoc. Professor, Biostatistics
>>  tlumley at u.washington.edu    University of Washington, Seattle
>
>
> -- 
> Frank E Harrell Jr   Professor and Chairman        School of Medicine
>                      Department of Biostatistics   Vanderbilt University
>
> ______________________________________________
> R-help at r-project.org 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.
>

Charles C. Berry                            (858) 534-2098
                                             Dept of Family/Preventive Medicine
E mailto:cberry at tajo.ucsd.edu	            UC San Diego
http://famprevmed.ucsd.edu/faculty/cberry/  La Jolla, San Diego 92093-0901



More information about the R-help mailing list