Estimating effective sample size for spatially correlated data.

摘要

Correlated data has long been difficult or impossible to analyze with traditional statistical methods that typically require independent observations. One cannot simply use the sample size since the information content is clearly reduced due to the relationships between the observations. In this thesis, we propose a method to estimate the effective sample size (ESS) in common scale, in correlated data using a function called the measure of equivalent exchange (MEE), which has the eigenvalues of the correlation matrix of the data and a tuning parameter, tau, as its arguments. The efficacy of the proposed method will be demonstrated empirically via simulations of one-, two-, and three-dimensional spatially correlated fields, and by application. We conclude with possible future implementations in a more generalized covariance settings.