# Compute area under the ROC curve

# Define storage vector for ROC
roc <- rep(NA, times = length(taxa.names))

for (i in 1:length(taxa.names)) {
                 # Compute mean predicted probability of occurrence
     predout <- predict(modlist.glm[[i]], type = "response")

                  # Generate logical vector corresponding to presence/absence
     resp <- dfmerge[, taxa.names[i]] > 0

                   # Divide predicted probabilities into sites where
                   # species is present ("x") and sites where the species is
                   # absent ("y").
     x <- predout[resp]
     y <- predout[! resp]

                    # Now perform all pairwise comparisons of x vs. y
                    # and store results in a matrix
     rocmat <- matrix(NA, nrow = length(x), ncol = length(y))
     for (j in 1:length(x)) {
          rocmat[j,] <- as.numeric(x[j] > y)
     }

                    # Summarize all comparisons to compute area under ROC
     roc[i] <- sum(rocmat)/(length(x)*length(y))
}
names(roc) <- taxa.names
print(roc)