In statistics the mean squared prediction error of a smoothing or curve fitting procedure is the expected value of the squared difference between the fitted values and the (unobservable) function g. If the smoothing procedure has operator matrix L, then