R/eblup.auc.R

In blebedenko/ME-AUC: Estimate AUC in ME Models - our functions

eblup.auc <- function(Y1,Y2){
  
  Y1.df <- as.data.frame(Y1) # a data frame for lmer
  Y2.df <- as.data.frame(Y2) # a data frame for lmer
  
  ############## EBLUPS #################
  
  # factor to index random subject effects
  Y1.df$subject <- factor(1:nrow(Y1))
  Y2.df$subject <- factor(1:nrow(Y2))
  
  # Reshape the data matrix for lmer eblup 
  Y1.long <- reshape2::melt(Y1.df,id.vars="subject")
  Y2.long <- reshape2::melt(Y2.df,id.vars="subject")
  
  # fit a linear mixed model
  fit1 <- lmer(value~1|subject,data = Y1.long,REML = TRUE)
  fit2 <- lmer(value~1|subject,data = Y2.long,REML = TRUE)
  
  # extract parameter estimates
  s1 <- summary(fit1)
  mu1.hat <- s1$coefficients[1]
  eblups1 <- ((ranef(fit1)$subject[1][,1]))
  
  s2 <- summary(fit2)
  mu2.hat <- s2$coefficients[1]
  eblups2 <- ((ranef(fit2)$subject[1][,1]))
  
  # this computes the ECDF of required quantiles
  estimates1 <- eblups1+mu1.hat
  estimates2 <- eblups2+mu2.hat
  
  # compute Mann-Whitney statistic for estimates2,1
  
  A.hat <- mean(outer(estimates1,estimates2,"<"))
  
  return(A.hat)
  
}