#R-function for cross-validated R²

 

 

# Below is given a function that may compute cross-validated R².

# Copy the commands and paste them into the command-window of R:

 

cv.R2=function(lmobj,y=lmobj$y,x=lmobj$x)

{

   a=t(x)%*%x

   d=diag(1/eigen(a)$values)

  e=eigen(a)$vector

   inv.a=e %*% d %*% t(e)

   v=diag(x%*%inv.a%*%t(x))

   SSkryss=sum((lmobj$res/(1-v))^2)

   SStot=sum((y-mean(y))^2)

   R2kryss=1-SSkryss/SStot

   R2kryss

}

 

 

# To use the function, you first fit a linear regression model by a command of the form lmobj=lm(y~x1+x2+x3+x4+x5, x=T, y=T)

# (note that you need to include the options "x=T" and "y=T")

# Then you compute cross-validated R² by the command cv.R2(lmobj)