I have a set of data that is not normally distributed and for which I need to build a model. So, I tried the lrm function from the design-package. The first run went well, and I got the following results: Wald Statistics Response: RVCL2PROC.mott Factor Chi-Square d.f. P TTV.mott (Factor+Higher Order Factors) 69.01 4 <.0001 All Interactions 12.13 3 0.0069 BEHANDLING (Factor+Higher Order Factors) 14.94 6 0.0208 All Interactions 12.13 3 0.0069 TTV.mott * BEHANDLING (Factor+Higher Order Factors) 12.13 3 0.0069 TOTAL 69.76 7 <.0001 Now, how is it to be interpreted? Does it mean that the p-value for the interaction (TTV.mott*BEHANDLING) is 0.0069? And what does the "TOTAL" p-value signify? Then, I ran exactly the same script on another dataset and got the following error message: singular information matrix in lrm.fit (rank= 0 ). Offending variable(s): Error in est[z$pivot[nvi:(irank + 1)] - kint] : only 0's may be mixed with negative subscripts Does anyone know? I suspect that the first question is probably rather easy for you clever guys but I'm a statistics noob so... looking forward to your help. Martin Kellner
Frank E Harrell Jr
2009-Jul-29 12:18 UTC
[R] lrm-function: Interpretation and error message
Martin Kellner wrote:> I have a set of data that is not normally distributed and for which I > need to build a model. So, I tried the lrm function from the > design-package. The first run went well, and I got the following > results: > > Wald Statistics Response: RVCL2PROC.mott > > Factor Chi-Square d.f. P > TTV.mott (Factor+Higher Order Factors) 69.01 4 <.0001 > All Interactions 12.13 3 0.0069 > BEHANDLING (Factor+Higher Order Factors) 14.94 6 0.0208 > All Interactions 12.13 3 0.0069 > TTV.mott * BEHANDLING (Factor+Higher Order Factors) 12.13 3 > 0.0069 > TOTAL 69.76 7 <.0001 > > Now, how is it to be interpreted? Does it mean that the p-value for the > interaction (TTV.mott*BEHANDLING) is 0.0069? And what does the "TOTAL" > p-value signify? >To understand all that, first make sure you understand all the parameters in the model - as printed when you print the fit object. Then get print.anova.Design to annotate the output to display exactly which parameters are being tested in each line: a <- anova(fit) print(a, which='subscripts') # or 'names' or 'dots'> Then, I ran exactly the same script on another dataset and got the > following error message: > > singular information matrix in lrm.fit (rank= 0 ). Offending > variable(s): Error in est[z$pivot[nvi:(irank + 1)] - kint] : > only 0's may be mixed with negative subscripts >Please follow the posting guide and submit a small example that reproduces this problem. Frank> Does anyone know? I suspect that the first question is probably rather > easy for you clever guys but I'm a statistics noob so... looking forward > to your help. > > Martin Kellner > > ______________________________________________ > 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. > >-- Frank E Harrell Jr Professor and Chair School of Medicine Department of Biostatistics Vanderbilt University