Dear R-help and Prof. Harrell: My question concerns the baseline state for continuous variable in lrm() within the RMS package. I have a model which can be reduced to: lrm(FT ~ rcs(V1, c(0, 1,5)) The model makes perfect sense if the baseline state is where V1>=5 but the model makes no sense if the baseline category is 0 (which I had expected). Can someone point me to a reference, or suggest a command/test, to get the model to fess up to the baseline state. The variable V1 and V1' are shown in the model output, but there is no clear association between these pseudo-variables and the ranges specified in the cut. Thus, it is seems possible to misidentify the baseline value for the term. Many thanks, Rob
I should rephrase my question. When using the notation rcs(V1, N) within the lrm function in the rms package, the associated model reports out N-1 values for the associated variable. Fine. The rcs-constructed temporary variables are denoted V1, V1', V1'' when one looks at the model output, but the model output is not explicit about which values of V1 are associated with variables V1, V1', V1'' . Yes, specs() tells you what the splits are for rcs, but is there a way to explicitly generate the associations between the model outcome variables, and the associated ranges, Here's the idea: e.g: Variable Low High V 0 2 V' 3 6 V' 7 22 ... Is there existing functionality to do this? If not, it suggest it would be a helpful addition. Rob
On Jan 18, 2011, at 2:51 PM, Rob James wrote:> Dear R-help and Prof. Harrell: > > My question concerns the baseline state for continuous variable in > lrm() within the RMS package. > > I have a model which can be reduced to: > > lrm(FT ~ rcs(V1, c(0, 1,5)) > > The model makes perfect sense if the baseline state is where V1>=5 > but the model makes no sense if the baseline category is 0 (which I > had expected).You are using terminology that suggests you ar working with a categorical variable but asking a question about a function that only makes sense when applied to contiguous variables. I think you need to provide results from str and describe() on V1.> Can someone point me to a reference, or suggest a command/test, to > get the model to fess up to the baseline state. The variable V1 and > V1' are shown in the model output, but there is no clear association > between these pseudo-variables and the ranges specified in the cut.Further suggestive evidence that you are working with categorical data if there was construction of the variable with cut() function.> Thus, it is seems possible to misidentify the baseline value for the > term. > > Many thanks,David Winsemius, MD West Hartford, CT