R help - Aug 2010 - Specifying interactions in rms package... error

I am encountering an error I do not know how to debug. The error arises 
when I try to add an  interaction term involving two continuous 
variables (defined using rcs functions) to an existing (and working) model.

The new model reads:

  model5 <- lrm( B_fainting ~   gender+  rcs(exactage, 7) +  
rcs(DW_nadler_bv, 7) +           rcs(drawtimefrom8am, 7)+  DW_firsttime+ 
   DW_race_eth +
  rcs(DW_BPS, 5)+  rcs(DW_BPD, 5)+  rcs(pulse3, 5)+  locat2+
(rcs(exactage, 3) * rcs(drawtimefrom8am, 3))
   )

This gives rise to the following error:

----
Error in dimnames(X)[[2]] <- atr$colnames :
   length of 'dimnames' [2] not equal to array extent
In addition: Warning message:
In names(nmiss)[jz] <- fname[asm != 9] :
   number of items to replace is not a multiple of replacement length
----

The sessionInfo() output is below, but I think relatively unremarkable.  
The dataset is very large (about 550K observations, although the outcome 
(B_fainting)  is relatively sparse with only 1500 events in this 
sample.   I do not get this error when the interaction term is replaced 
by two categorical variables (data not shown).  I can replicate this 
error when there is just one continuous variable and one categorical 
variable.


I'd welcome suggestions. Thank you in advance.





sessionInfo()
R version 2.11.0 (2010-04-22)
x86_64-pc-linux-gnu

locale:
  [1] LC_CTYPE=en_CA.UTF-8       LC_NUMERIC=C               
LC_TIME=en_CA.UTF-8
  [4] LC_COLLATE=en_CA.UTF-8     LC_MONETARY=C              
LC_MESSAGES=en_CA.UTF-8
  [7] LC_PAPER=en_CA.UTF-8       LC_NAME=C                  LC_ADDRESS=C
[10] LC_TELEPHONE=C             LC_MEASUREMENT=en_CA.UTF-8 
LC_IDENTIFICATION=C

attached base packages:
[1] splines   stats     graphics  grDevices utils     datasets  
methods   base

other attached packages:
[1] gmodels_2.15.0  rms_3.0-0       Hmisc_3.8-2     survival_2.35-8 
odfWeave_0.7.14 XML_3.1-0
[7] lattice_0.18-8

loaded via a namespace (and not attached):
[1] cluster_1.13.1 gdata_2.8.0    grid_2.11.0    gtools_2.6.2   MASS_7.3-7

On Aug 2, 2010, at 9:06 PM, Rob James wrote:
> model5 <- lrm( B_fainting ~   gender+  rcs(exactage, 7) +   
> rcs(DW_nadler_bv, 7) +           rcs(drawtimefrom8am, 7)+   
> DW_firsttime+   DW_race_eth +
> rcs(DW_BPS, 5)+  rcs(DW_BPD, 5)+  rcs(pulse3, 5)+  locat2+
> (rcs(exactage, 3) * rcs(drawtimefrom8am, 3))
>  )
The parentheses around the last two crossed rcs() terms are  
superfluous. I don't have experience that demonstrates that is the  
cause of the problem, but you might consider removing them anyway.

-- 
David Winsemius, MD
West Hartford, CT

Hi Rob,

rms wants symmetry in the sense that the interactions need to use the 
same number and location of spline knots as the main effects.   So if 
using the * notation omit the main effects (which are generated 
automatically) and live with the equal knots.  Or use the restricted 
interaction operator %ia% to not use doubly nonlinear terms in forming 
the interactions, which is a sensible restriction.
Frank

Frank E Harrell Jr   Professor and Chairman        School of Medicine
                      Department of Biostatistics   Vanderbilt University

On Mon, 2 Aug 2010, Rob James wrote:
> I am encountering an error I do not know how to debug. The error arises
> when I try to add an  interaction term involving two continuous
> variables (defined using rcs functions) to an existing (and working) model.
>
> The new model reads:
>
>  model5 <- lrm( B_fainting ~   gender+  rcs(exactage, 7) +
> rcs(DW_nadler_bv, 7) +           rcs(drawtimefrom8am, 7)+  DW_firsttime+
>   DW_race_eth +
>  rcs(DW_BPS, 5)+  rcs(DW_BPD, 5)+  rcs(pulse3, 5)+  locat2+
> (rcs(exactage, 3) * rcs(drawtimefrom8am, 3))
>   )
>
> This gives rise to the following error:
>
> ----
> Error in dimnames(X)[[2]] <- atr$colnames :
>   length of 'dimnames' [2] not equal to array extent
> In addition: Warning message:
> In names(nmiss)[jz] <- fname[asm != 9] :
>   number of items to replace is not a multiple of replacement length
> ----
>
> The sessionInfo() output is below, but I think relatively unremarkable.
> The dataset is very large (about 550K observations, although the outcome
> (B_fainting)  is relatively sparse with only 1500 events in this
> sample.   I do not get this error when the interaction term is replaced
> by two categorical variables (data not shown).  I can replicate this
> error when there is just one continuous variable and one categorical
> variable.
>
>
> I'd welcome suggestions. Thank you in advance.
>
>
>
>
>
> sessionInfo()
> R version 2.11.0 (2010-04-22)
> x86_64-pc-linux-gnu
>
> locale:
>  [1] LC_CTYPE=en_CA.UTF-8       LC_NUMERIC=C
> LC_TIME=en_CA.UTF-8
>  [4] LC_COLLATE=en_CA.UTF-8     LC_MONETARY=C
> LC_MESSAGES=en_CA.UTF-8
>  [7] LC_PAPER=en_CA.UTF-8       LC_NAME=C                  LC_ADDRESS=C
> [10] LC_TELEPHONE=C             LC_MEASUREMENT=en_CA.UTF-8
> LC_IDENTIFICATION=C
>
> attached base packages:
> [1] splines   stats     graphics  grDevices utils     datasets
> methods   base
>
> other attached packages:
> [1] gmodels_2.15.0  rms_3.0-0       Hmisc_3.8-2     survival_2.35-8
> odfWeave_0.7.14 XML_3.1-0
> [7] lattice_0.18-8
>
> loaded via a namespace (and not attached):
> [1] cluster_1.13.1 gdata_2.8.0    grid_2.11.0    gtools_2.6.2   MASS_7.3-7
>
> ______________________________________________
> 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.
>