pROC 1.3 released
I just released the version 1.3 of pROC.
This function introduces several bug fixes and speed improvements, along with some new features. You can detect if two ROC curves are paired:
library(pROC) data(aSAH) rocs100 <- roc(aSAH$outcome, aSAH$s100) rocwfns <- roc(aSAH$outcome, aSAH$wfns) are.paired(rocs100, rocwfns)
A new method for the
roc.test is also introduced: the Venkatraman's test1. Unlike bootstrap and DeLong2 methods, it doesn't assess the equality of the AUC. Instead, it looks at the shape of ROC curves. Two ROC curves with the same AUC but a different shape will be significantly different
set.seed(1) response <- round(runif(1000, 0, 1)) predictor1 <- response + rnorm(1000) predictor2 <- response + rlnorm(1000) roc1 <- roc(response, predictor1, plot=TRUE) roc2 <- roc(response, predictor2, plot=TRUE, add=TRUE) roc.test(roc1, roc2) roc.test(roc1, roc2, method="venkatraman", boot.n=200)
I generate two artificial ROC curves. Their AUC is not different, and the DeLong test gives a non significant result. However their shapes are highly dissimilar, and the Venkatraman's test finds a significant difference.
Here is the full change log:
- CI is not re-computed by default in
smooth.roc. You can still turn it on with
- New function
- Local maximas could be incorrectly detected in
predictors containing more than 2 levels.
- New method
- MASS and tcltk packages are now only suggested instead of required (R only).
...not passed correctly in
type="bars"resulting in an error (R only).
- 1. E. S. Venkatraman and Colin B. Begg (1996) “A distribution-free procedure for comparing receiver operating characteristic curves from a paired experiment”. Biometrika 83, 835–848. DOI: 10.1093/biomet/83.4.835.
- 2. Elisabeth R. DeLong, David M. DeLong and Daniel L. Clarke-Pearson (1988) “Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach”. Biometrics 44, 837–845.