Weighted Composite Likelihood Approach to the Analysis of Spatial Data

• Activity Number: 210
• Type: Invited
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Tatiyana Apanasovich*+ and Yongtao Guan
• Companies: Thomas Jefferson University and University of Miami School of Business
• Keywords: composite likelihood ; efficiency ; geostatistics

Abstract

Maximum likelihood estimators are often of limited practical use in geostatistics due to intensive computations they require. An alternative is a composite likelihood (Lindsay, 1988) consisting of a combination of valid likelihood objects, related to smaller subsets of data, which leads to reduced computational complexity at the expense of efficiency. The goal of this study is to modify a particular form of composite likelihood introduced by Curriero and Lele (1999) providing a good balance between feasibility and optimality. To achieve superior statistical efficiency we rely on the concepts of optimal and adaptive estimating equations (Qu and Lindsay (2003)). To improve the computational tractability, we adopt two methodologies. First, we utilize conjugate gradient methods for computing a sparse incomplete factorization of the inverse of a symmetric positive definite matrix. Second, to reduce the multiplication complexity we exploit the fast multipole method for matrix-vector multiplications. The practical value of the methods is illustrated through simulation studies and analysis of meteorological data.