Statistic Research Group

TITLE:Applications of the Generalized Cross Validation Statistic
.

The Generalized Cross Validation (GCV) statistic provides a way of choosing the penalty parameter in penalized regression settings. The properties of the GCV statistic are examined which enable the Newton-Raphson algorithm to be implemented as opposed to the slower and less accurate Grid Search algorithm. A simulation study is conducted to identify typical shapes for the GCV as a function of the penalty parameter and to compare the two algorithms. Also, real flow data from the Nile River is analyzed to illustrate the use of the P-Spline method with the penalty chosen based on the GCV using the Grid Search and Newton-Raphson algorithms.