R help - Oct 2002 - More Logistic Regression Tools?

I've been using R to do logistic regresssion, and that's working
well, but there are two things I haven't figured out how to do.

(1) Is there some pre-existing function that will let you compute the
    odds ratios and confidence intervals for them for a specific fit. I know
    how to do this manually or even write a function that I can call with
    the coefficients and se, but extending the printing function was a
    little more work than I expected.

    I.e. I'm looking for something where you'd type
    odds.ratios(fit)

    and get:

    Coefficient		Odds ratio	95%	CI
    XXX			5.2		1.5	7.2
    
    etc.

(2) I'd like to compute goodness-of-fit statistics for my fit
    (Hosmer-Lemeshow, Pearson, etc.). I didn't see a package that
    did this. Have I missed one?

Thanks,
-Ekr

    
    
  

	


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

Eric Rescorla <ekr at rtfm.com> writes:
> I've been using R to do logistic regresssion, and that's working
> well, but there are two things I haven't figured out how to do.
> 
> (1) Is there some pre-existing function that will let you compute the
>     odds ratios and confidence intervals for them for a specific fit. I
know
>     how to do this manually or even write a function that I can call with
>     the coefficients and se, but extending the printing function was a
>     little more work than I expected.
> 
>     I.e. I'm looking for something where you'd type
>     odds.ratios(fit)
> 
>     and get:
> 
>     Coefficient		Odds ratio	95%	CI
>     XXX			5.2		1.5	7.2
My favourite version of this is

library(MASS)
example(glm)
exp(cbind(coef(glm.D93),confint(glm.D93)))

(OK, so that's a Poisson regression, not a logistic one, but you get
the picture.) Notice that rather than est.+-2SE it gives you
likelihood ratio profiling, which is hardly worth not doing these
days.
     
>     etc.
> 
> (2) I'd like to compute goodness-of-fit statistics for my fit
>     (Hosmer-Lemeshow, Pearson, etc.). I didn't see a package that
>     did this. Have I missed one?
As I understand it, some of these methods are rather unreliable. It
might be a good idea to hear if Frank Harrell has anything to say.
Last time someone requested a g-o-f statistic, he said that he
regretted ever implementing it in SAS....

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

On Mon, 21 Oct 2002 06:26:46 -0700
Eric Rescorla <ekr at rtfm.com> wrote:
> I've been using R to do logistic regresssion, and that's working
> well, but there are two things I haven't figured out how to do.
> 
> (1) Is there some pre-existing function that will let you compute the
>     odds ratios and confidence intervals for them for a specific fit. I
know
>     how to do this manually or even write a function that I can call with
>     the coefficients and se, but extending the printing function was a
>     little more work than I expected.
> 
>     I.e. I'm looking for something where you'd type
>     odds.ratios(fit)
> 
>     and get:
> 
>     Coefficient		Odds ratio	95%	CI
>     XXX			5.2		1.5	7.2
>     
>     etc.
> 
library(Design)  # see hesweb1.med.virginia.edu/biostat/rms
d <- datadist(mydataframe); options(datadist='d') # defines data
ranges etc.
f <- lrm(y ~ ....)
summary(f, age=c(21,65))
# 65:21 year OR (even if age is nonlinear)
# other continuous variables: use quartiles

See all the contrast function in Design.
> (2) I'd like to compute goodness-of-fit statistics for my fit
>     (Hosmer-Lemeshow, Pearson, etc.). I didn't see a package that
>     did this. Have I missed one?
Hosmer-Lemeshow has low power and relies on arbitrary binning of predicted
probabilities.  The Hosmer-Le Cessie omnibus test is unique and has more power
usually.  To get it:

f <- update(f, x=T, y=T)
resid(f, 'gof')  # uses residuals.lrm

Frank Harrell
> 
> Thanks,
> -Ekr
> 
>     
>     
>   
> 
> 	
> 
> 
>
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
> r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
> Send "info", "help", or "[un]subscribe"
> (in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
>
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

-- 
Frank E Harrell Jr              Prof. of Biostatistics & Statistics
Div. of Biostatistics & Epidem. Dept. of Health Evaluation Sciences
U. Virginia School of Medicine  http://hesweb1.med.virginia.edu/biostat
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._