Analysis of Spatial Cross-Classified Data

Abstract

It can often happen that we collect cross-classified data at many locations. Such spatial cross-classified data can arise in many settings in the agricultural, biological and environmental sciences. Because of possible spatial correlation in the data, the usual chi-squared tests can fail in these settings. In this talk first I review the statistical literature that addresses this problem and discuss their drawbacks. Then by examing the truncated Gaussian random field I establish the variogram model for spatial correlations. My approach based on this spatial correlation model combined with the idea of quasi-likelihood estimation in then introduced. I prove that the quasi- likelihood estimate is consistent and asymptotically normal. Some examples are used to illustrate my methods.