Re: npindex: fitted values of the function itself?
I realized my mistake!
For those who are interested: what I predicted was in fact G*(X'b*): A single-index model assumes a linear index function for which I obtain the estimated coefficients b*. The predicted probabilities are then G*(X'b*).
Indeed, this is equivalent to the probit case, where I only need "G_hat=pnorm(x_fit) " if x_fit is the linear prediction.