Re: [R] Nagelkerkes R2N
I am interested Andrea is whether you ever established why your R2 was 1. I have had a similar situation previously. My main issue though, which I'd be v grateful for advice on, is why I am obtaining such negative values -0.3 for Somers Dxy using validate.cph from the Design package given my value of Nagelkerke R2 is not so low 13.2%. I have this output when fitting 6 variables all with p-values0.01 I am wondering what the interpretation should be. I know my Nagelkerke R2 isn't very good but I compare my results with the example from ?validate.cph and although I have a better R2 (13% v 9%) the Somers dxy from the example data set is much better, 38%, so certainly not negative ! So my main question is : Why such a difference between explained variation, R2, and predictive ability: somers dxy ?? Obs Events Model L.R. d.f. P ScoreScore P R2 471228 66.36 6 0 73.41 0 0.132 validate(f, B=150,dxy=T) # normally B=150 index.orig training test optimism index.corrected n Dxy -0.3022537331 -0.3135032097 -0.292492573 -0.021010636 -0.2812430968 150 R2 0.1319445685 0.1431179294 0.122599605 0.0205183240.1114262446 150 Slope 1.00 1.00 0.923340558 0.0766594420.9233405576 150 D 0.0250864459 0.0276820092 0.023163167 0.0045188420.0205676038 150 U -0.0007676033 -0.0007725071 0.000610456 -0.0013829630.0006153598 150 Q 0.0258540493 0.0284545164 0.022552711 0.0059018050.0199522440 150 I also calculated the Schemper and Henderson V measure and obtained v=10.5% I was using the surev package of Lusa Lara; Miceli Rosalba; Mariani LuigiEstimation of predictive accuracy in survival analysis using R and S-PLUS.http://www.biomedexperts.com/Abstract.bme/17601627/Estimation_of_predictive_accuracy_in_survival_analysis_using_R_and_S-PLUS Computer methods and programs in biomedicine 2007;87(2):132-7. And my code was library(surev) pred.accuracy-f.surev(f) pred.accuracy sorry if my question isn't clear - should I have included my sessionInfo for a methodological question ? (I'm a newbie) many thanks for any advice [[alternative HTML version deleted]] __ R-help@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.
[R] Nagelkerkes R2N
Hello All, as I´m new to R and survival analysis, I´ve got a question about the Design::validate function: My Code: cox - cph(Surv(t,status) ~ var1 + var2 + var3, data=data, x=TRUE, y=TRUE, surv=TRUE) cox.val - validate(cox, B=10, dxy=TRUE, pr=TRUE); My output (cox.val): index.orig training test Dxy -0.3639222921368090891 -0.3591157308750822175 -0.3634294047761231106 R2 1.000 1.000 1.000 Slope 1.000 1.000 1.0055508323397084336 D 0.0232804472888947744 0.0226998668193014774 0.0232190381679612834 U -0.607553318187988 -0.610134584621832 0.254159617147094 Q 0.0233412026207135703 0.0227608802777636665 0.0231936222062465713 optimism index.corrected n Dxy0.0043136739010409269 -0.36823596603785002657 10 R2 0.000 1. 10 Slope -0.0055508323397084336 1.00555083233970843359 10 D -0.0005191713486598047 0.02379961863755457596 10 U -0.864294201768926 0.2567408835809379 10 Q -0.0004327419284829055 0.02377394454919647515 10 And my question ist about the R2: Why ist the value always 1.0. That doesn´t seem to me like a realistic value. And so I tried to calculate R2 with my own formula: LR - -2*cox$loglik[2] L0 - -2*cox$loglik[1] n - length(data[,ID]) R2N - (1-exp(-LR/n)) / (1-exp(L0/n)) R2N calculated that way is -0.00132314024559236. Can anybody help me to understand the formula to R2 and why the validate-function results in 1.0? Thanks, Andrea. __ R-help@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.
Re: [R] Nagelkerkes R2N
A new version of Design will be posted to CRAN in the next 2 days. After than, update your system, including an update to the survival package. Then re-try. Your formula is wrong as it can't be negative. LR should be the likelihood ratio chi-square stat : -2 times the difference in the two loglik values. Frank Andrea Weidacher wrote: Hello All, as I´m new to R and survival analysis, I´ve got a question about the Design::validate function: My Code: cox - cph(Surv(t,status) ~ var1 + var2 + var3, data=data, x=TRUE, y=TRUE, surv=TRUE) cox.val - validate(cox, B=10, dxy=TRUE, pr=TRUE); My output (cox.val): index.orig training test Dxy -0.3639222921368090891 -0.3591157308750822175 -0.3634294047761231106 R2 1.000 1.000 1.000 Slope 1.000 1.000 1.0055508323397084336 D 0.0232804472888947744 0.0226998668193014774 0.0232190381679612834 U -0.607553318187988 -0.610134584621832 0.254159617147094 Q 0.0233412026207135703 0.0227608802777636665 0.0231936222062465713 optimism index.corrected n Dxy0.0043136739010409269 -0.36823596603785002657 10 R2 0.000 1. 10 Slope -0.0055508323397084336 1.00555083233970843359 10 D -0.0005191713486598047 0.02379961863755457596 10 U -0.864294201768926 0.2567408835809379 10 Q -0.0004327419284829055 0.02377394454919647515 10 And my question ist about the R2: Why ist the value always 1.0. That doesn´t seem to me like a realistic value. And so I tried to calculate R2 with my own formula: LR - -2*cox$loglik[2] L0 - -2*cox$loglik[1] n - length(data[,ID]) R2N - (1-exp(-LR/n)) / (1-exp(L0/n)) R2N calculated that way is -0.00132314024559236. Can anybody help me to understand the formula to R2 and why the validate-function results in 1.0? Thanks, Andrea. __ R-help@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 __ R-help@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.