2007), ”A Note on Penalized Spline Smoothing with Correlated Errors”, mimeo

This note investigates the behavior of data driven smoothing parameters for penalized spline regression in the presence of correlated data. It has been shown for other smoothing methods before, that mean squared error minimiz-ers, such as (generalized) cross validation or Akaike criterion, are extremely sensitive to misspecifications of the correlation structure over or (under) fit-ting the data. In contrast to this, we show that a maximum likelihood based choice of the smoothing parameter is more robust and for moderately mis-specified correlation structure over or (under) fitting does not occur. This is demonstrated in simulations, data examples and supported by theoretical investigations