REPARAMETRIZATION OF AUTO REGRESSIVE MODELS WITH COEFFICIENTS DEPENDING ON COVARIABLES - APPLICATION TO EEG SPECTRUM MATURATION

To describe the spectral characteristics of the EEG development throug h autoregressive (AR) time series models it is necessary to perform re gression analysis of the AR parameters with regards to the age of the subject. A major difficulty in this approach is the very complex natur e of the admissible region of the AR coefficients, which impedes the s traight use of regression techniques. The present paper overcomes this difficulty by first applying the Barndorff-Nielsen and Schou reparame trization of AR models, followed by Fisher's transformation, and then carrying out age regression analysis of the transformed parameters. We apply this approach to real EEG data obtained from a normative sample of subjects in the age range from 5 to 95 years. (C) 1997 by John Wil ey & Sons, Ltd.