Conditionally specified gaussian models for spatial statistical analysis of field trials
Spatial dependence in data from one- and two-dimensional field trials has been recognized since the mid 1930s. Although originally employed in an agricultural context, the general purpose of field trials is to compare the effects of a number of treatments applied to a collection of proximate experimental units. For example, in the manufacture of integrated-circuit chips from silicon ingots, the experimental units may be contiguous wafer slices from the same ingot or regions of the same wafer. An iterated version of Papadakis' nearest-neighbor method for estimating treatment effects is shown to yield maximum likelihood estimates when the spatial dependence is a conditionally specified Gaussian model. Models of both the mean and the covariance in two-dimensional space are featured in this article. 1996 American Statistical Association and the International Biometric Society.
Please refer to publisher version or contact your library.