Adding pch on ROC curves and labelling AUC values

Question

I want to plot ROC curves for 5 estimated models using R's pROC application. Because I'm using gray scales, which may look indistinguishable visually, so I want to accentuate different line types by assigning different pch to different ROC curves, however, I found the lines() function is quite unresponsive to pch options.

Also, when I want to label their respective AUC values, using the print.auc option, beyond the default model (Model 1), the roc() and lines() functions do seem to work. Only AUC value for the first model is shown.

It will be very appreciated if someone could point me toward some feasible ways to achieve the desired result. Below is my current code for the plot and output.

### load the data ###
library(readxl)

dataURL <- "https://www.dropbox.com/s/vri9fx2xa1pfj7w/predict.xlsx?dl=1"
temp = tempfile(fileext = ".xlsx")
download.file(dataURL, destfile=temp, mode='wb')
predict <- readxl::read_excel(temp, sheet =1)

### plotting and labelling ###

library(pROC)

roc.pred2 <- roc(predict$IAC, predict$phat2, percent = TRUE, main = "Smoothing")
roc.pred3 <- roc(predict$IAC, predict$phat3, percent = TRUE, main = "Smoothing")
roc.pred4 <- roc(predict$IAC, predict$phat4, percent = TRUE, main = "Smoothing")
roc.pred5 <- roc(predict$IAC, predict$phat5, percent = TRUE, main = "Smoothing")

plot.roc(predict$IAC, predict$phat1, percent = TRUE, main = "ROC curves", add =  FALSE, asp = NA, print.auc = TRUE)

lines(roc.pred2, type = "l", lty = 2, col = "grey35")
lines(roc.pred3, type = "l", lty = 3, col = "grey48")
lines(roc.pred4, type = "l", lty = 4, col = "grey61")
lines(roc.pred5, type = "l", lty = 1, pch = 24, col = "grey76")

legend("bottomright", 
       legend = c("Model 1", "Model 2", "Model 3", "Model 4", "Model 5"), 
       col = c("black", "grey35", "grey48", "grey61", "grey76"),
       lty = c(1, 2, 3, 4, 1))

`pch` selects the *point character*. Obviously `lines` doesn’t respond to it. Use `points` instead. — Konrad Rudolph, Feb 03 '20 at 11:18
It says the first argument is a list object, so it `does not have components 'x' and 'y'` — Chris T., Feb 03 '20 at 11:23
Adding `lty = c(2, 3, 4, 1)` to the legend will fix the line types in the legend — Allan Cameron, Feb 03 '20 at 11:30
@Allan, I have updated that in my edit, thanks for reminding me that. — Chris T., Feb 03 '20 at 11:30
To do what @KonradRudolph suggests you need to do `points(roc.pred2$sensitivities, roc.pred2$specificities, pch = 2)` — Allan Cameron, Feb 03 '20 at 11:33
Though there are so many points, it doesn't look good. You might need to do something like `points(roc.pred2$specificities[seq(1, length(roc.pred2$sensitivities), 100)],roc.pred2$sensitivities[seq(1, length(roc.pred2$sensitivities), 100)] )` so that only every hundredth point is plotted. — Allan Cameron, Feb 03 '20 at 11:39
Fabulous! I have around 5800 data points, so this adjustment makes them more redable. Now they all work, thanks a lot @Konrad and Allan. What's left is how to label AUC values for each curve. — Chris T., Feb 03 '20 at 11:42

score 2 · Accepted Answer · answered Feb 04 '20 at 07:38

Apart from color, there are two things you can change on a line: type (lty) and width (lwd).

plot.roc(predict$IAC, predict$phat1, percent = TRUE, main = "ROC curves", add =  FALSE, asp = NA, print.auc = TRUE)

lines(roc.pred2, type = "l", lty = 2, col = "grey35")
lines(roc.pred3, type = "l", lty = 3, lwd = 4, col = "grey48")
lines(roc.pred4, type = "l", lty = 4, lwd = 8, col = "grey61")
lines(roc.pred5, type = "l", lty = 1, pch = 24, col = "grey76")

legend("bottomright", 
       legend = c("Model 1", "Model 2", "Model 3", "Model 4", "Model 5"), 
       col = c("black", "grey35", "grey48", "grey61", "grey76"),
       lty = c(1, 2, 3, 4, 1),
       lwd = c(1, 1, 4, 8, 1))

Besides of that, you'll reach the limits of what you can show on a black and white plot.

If you had fewer points on your ROC curve, I would suggest to use type = "b" which plots points and lines:

data(aSAH)
plot(roc(aSAH$outcome, aSAH$wfns), type = "b")

Then you can play with pch, bg and col on the points too. However as you noticed that's not going to work for you because your curve has too many points.

Adding pch on ROC curves and labelling AUC values

1 Answers1