Tukukkk

I am attempting to plot a ROC curve using the plotROC package for ggplot2, but I am not sure how to transform the data I have into the M1 marker format. The documentation provided the following example:

# plotROC documentation example

library(plotROC)

library(ggplot2)

D.ex <- rbinom(200, size = 1, prob = .5)

M1 <- rnorm(200, mean = D.ex, sd = .65)

M2 <- rnorm(200, mean = D.ex, sd = 1.5)



D         M1         M2

1  1.4995932  0.5508204

1  0.4181619  1.6339181

0 -0.3620614 -1.0428972

1  0.7991132 -1.6396751

0  0.9574047  2.1159753

1  1.3440595  1.3026485



test <- data.frame(D = D.ex, D.str = c("Healthy", "Ill")[D.ex + 1], 

               M1 = M1, M2 = M2, stringsAsFactors = FALSE)



ggplot(test, aes(d = D, m = M1)) + 

   geom_roc()

Sample ROC plot output by plotROC

My data is logistic regression scores on a test subset:

# Example starting point

test <- rbinom(200, size = 1, prob = 0.2)

scores.prob <- runif(200, min = 0 , max = 1)

scores.class <- ifelse(scores.prob > 0.5, 1, 0)



# Example generated data

test scores.prob scores.class

 0   0.7323306            1

 0   0.7860687            1

 0   0.9535123            1

 1   0.3082551            0

 0   0.5762784            1

 1   0.4613730            0

I would like to know what M1 is and how to transform my data to get that field.

library(plotROC) 

library(ggplot2) 

test <- rbinom(200, size = 1, prob = 0.2)

scores.prob <- runif(200, min = 0 , max = 1) 

test <- data.frame(D = test,

           M1 = scores.prob, stringsAsFactors = FALSE)

ggplot(test, aes(d = D, m = M1)) + 

geom_roc()

Your marker/predictor is the fitted values of your glm model. The ROC will give you the idea of how your model work (by means of AUC) and the best probability threshold (the ROC cutoff) for assigning persons to classes.
It is a usefull approach if you want to visualize the added value of fdifferent multivariate/univariate approaches.
Here a full example with the mtcars dataset. Hope it helps.

# Loading data

data(mtcars)

# Manual transmission (am = 1) depends on 1/4 mile time (qsec) and miles/(US) gallon (mpg)

glmfit <- glm(am ~ qsec + mpg, data = mtcars, binomial)

mtcars$fitted_am <- glmfit$fitted.values

# Loading packages

library(plotROC)

library(ggplot2)

library(pROC)

# Calculating ROC curve, AUC and threshold according to Youden index

rocfit <- roc(mtcars$am, mtcars$fitted_am)

auc(rocfit)

coords(rocfit, x = "b")

basicplot <- ggplot(mtcars, aes(d = am, m = fitted_am))

basicplot + 

  geom_roc() + 

  style_roc(theme = theme_grey) +

  theme(axis.text = element_text(colour = "blue")) +

  ggtitle("Automatic transmission prediction") +

  scale_x_continuous("1 - Specificity", breaks = seq(0, 1, by = .1))

plot(rocfit)

prop.table(table(mtcars$am))