LOCATION:  1252 Howe Hall
                  

SPEAKER: Goeran Kauermann, University of Bielefeld, Germany

TITLE: Penalized Spline Smoothing Mixed Models and Correlated Errors

ABSTRACT:

The presentation illuminates the relation between Penalized spline smoothing (commonly known as P-spline smoothing) and Linear Mixed Models. In fact, P-spline estimates are equivalent to posterior Bayes estimates in a Linear Mixed Model where the spline basis coefficients are treated as normally distributed. This connection can be exploited for bandwidth selection, since (Restricted) Maximum Likelihood (REML) based estimates of the variance of the random effects are conveniently used as smoothing parameter estimates. The talk illuminates this choice in more depth and compares it to standard routines like an Akaike criterion. In a next step REML smoothing parameter selection is investigated in the case of correlated errors. In this case REML shows a more robust behavior than an Akaike criterion. COFFEE: 3:45 p.m., 104 Snedecor Hall