Additive Models with Spatio-Temporal Data

• Activity Number: 29
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Xiangming Fang*+ and Kung-Sik Chan
• Companies: East Carolina University and The University of Iowa
• Keywords: Additive models ; Penalized likelihood ; Matérn class ; Spatio-temporal data ; ML ; REML

Abstract

The current methods for fitting additive models with correlated data are not completely satisfactory. We propose a new approach to fit additive models with spatio-temporal data via the penalized likelihood approach which estimates the smooth functions and covariance parameters by iteratively maximizing the penalized log likelihood. Both ML and REML estimation schemes are developed. The asymptotic distribution of the estimates is studied in a Bayesian framework. Conditions for asymptotic posterior normality are investigated for the case of separable spatio-temporal data with fixed spatial covariance structure and no temporal dependence. We also propose a method to check the assumption of temporal independence and a new model selection criterion for comparing models with and without spatial correlation. The proposed methods are illustrated by both simulation study and real data analysis.