|
Activity Number:
|
173
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 1, 2016 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics and the Environment
|
|
Abstract #321269
|
View Presentation
|
|
Title:
|
Approximate Optimal Weighted Composite Likelihood Approach for Large Spatial Data
|
|
Author(s):
|
Furong Li* and Huiyan Sang
|
|
Companies:
|
Texas A&M University and Texas A&M University
|
|
Keywords:
|
Composite likelihood ;
Pairwise likelihood ;
Estimation equation ;
Tapering
|
|
Abstract:
|
Maximum likelihood (ML) method is attractive for the estimation of covariance parameters in spatial Gaussian process. However,the complexity of inverting a matrix to evaluate likelihood makes it computationally intractable. Composite likelihood (CL) method becomes popular as it avoids matrix inversion and model misspecification. But it cannot avoid the loss of statistical efficiency with respect to ML method. We proposed an approximate optimal weighted composite likelihood method for estimation. This method combined block-diagonal approximation and tapering strategies to obtain optimal weight for each component in CL. Gain of statistical and computational efficiency was illustrated in the simulations. We applied this method to the yearly total precipitation anomalies data. The BT weighted method exhibited estimation closed to MLE but much faster computation speed.
|
Authors who are presenting talks have a * after their name.
