R help - Jun 2010 - "Unable to fit" error message from the lrm function in the rms library

Hi all,

I have another question about the lrm function (from the rms library) that I
cannot find the answer to.

I get an error message when I try to fit a model, and I don't know what to
make of it. Please forgive me for not having a toy example, but it appears the
size and complexity of my data is somehow causing the error. The best I can do
is show you what I type and what errors I get.

Here's what I'm doing that causes the error. I fit this model:
> model.lrm24 <- lrm(x.all[,156] ~ pol(PC1, 3) + pol(PC2, 3) + pol(PC3, 3)
+ PCNM1 + PCNM2 + PCNM3 +PCNM4 + PCNM5 + PCNM6 + PCNM7 + PCNM8 + PCNM9 + PCNM10
+ PCNM11 + PCNM12 + PCNM13 + PCNM14 + PCNM15 + PCNM16 + PCNM17 + PCNM18 + PCNM19
+ PCNM20 + PCNM21 + PCNM22 + PCNM23 + PCNM24 + PCNM25 + PCNM26 + PCNM27 + PCNM28
+ PCNM29 + PCNM30 + PCNM31 + PCNM32 + PCNM33 + PCNM34 + PCNM35 + PCNM36 + PCNM37
+ PCNM38 + PCNM39, data=x.all)
...and I get this error:

Error in lrm(x.all[, 156] ~ pol(PC1, 3) + pol(PC2, 3) + pol(PC3, 3) +  : 
  Unable to fit model using “lrm.fit”

The weird thing is that the model still seems to have been "fit"
anyway. For instance:
> model.lrm24$stats
         Obs    Max Deriv   Model L.R.         d.f.            P 
3.610000e+02 5.352815e-02 1.251547e+02 4.800000e+01 8.340322e-09 
           C          Dxy        Gamma        Tau-a           R2 
8.795642e-01 7.591284e-01 7.631040e-01 2.578793e-01 4.522090e-01 
       Brier            g           gr           gp 
1.113480e-01 3.489163e+00 3.275851e+01 2.494908e-01 

So it still seems it was able to run, at least to some extent, despite it's
protests.

Any ideas what is going on? Why does the model run if it is "unable"
to be fit? That seems contradictory on its face. What does that even mean?

Here's a hint to the source of the problem. If I fit a simplified model, the
error message goes away:
> model.lrm24 <- lrm(x.all[,156] ~ pol(PC1, 3) + pol(PC2, 3) + pol(PC3,
3), data=x.all)
> 
No problem at all -- no error.

Here's what the same summary output looks like, from the simplified model:
> model.lrm24$stats
         Obs    Max Deriv   Model L.R.         d.f.            P 
2.030000e+02 2.861290e-01 7.180027e+01 9.000000e+00 6.747713e-12 
           C          Dxy        Gamma        Tau-a           R2 
8.195579e-01 6.391159e-01 6.424134e-01 2.975662e-01 4.077021e-01 
       Brier            g           gr           gp 
1.583826e-01 2.032737e+00 7.634955e+00 2.869544e-01 

So what's going on here? Any ideas? And why does the model still seem to be
"fit" despite the claim it was "unable" to be fit? Thanks
very much in advance -- you guys are great.

-----------------------------------
Josh Banta, Ph.D
Center for Genomics and Systems Biology
New York University
100 Washington Square East
New York, NY 10003
Tel: (212) 998-8465
http://plantevolutionaryecology.org



      
	[[alternative HTML version deleted]]

Hi,

The error message you are getting (probably) means that the algorithm did not
converge.  Did you check for convergence? (Look at the "fail" element
of the returned lrm object)

Christos


 		 	   		  
_________________________________________________________________
Hotmail: Powerful Free email with security by Microsoft.

	[[alternative HTML version deleted]]