Abstract
In field experiments with large numbers of treatments, inference can be affected by 1) local variation, and 2) method of analysis .
The standard approach to local, or spatial, variation in the design of experiments is blocking. While the randomized complete block design is obviously unsuitable for experiments with large numbers of treatments, incomplete block designs - even apparently well-chosen ones - may be only partial solutions. Various nearest neighbor adjustment procedures are an alternative approach to spatial variation .
Treatment effects are usually estimated using standard linear model methods. That is, linear unbiased estimates are obtained using ordinary least squares or, for example when nearest neighbor adjustments are used, generalized least squares. This follows from regarding treatment as a fixed effect. However, when there are large numbers of treatments, regarding treatment as a random effect and obtaining best linear unbiased predictors (BLUP) can improve precision .
Nearest neighbor methods and BLUP have had largely parallel development. The purpose of this paper is to put them together .
Keywords
linear model, fixed effect, random effect, generalized least squares, best linear unbiased prediction, nearest neighbor, spatial correlation
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.
Recommended Citation
Stroup, Walter W.
(1990).
"NEAREST NEIGHBOR ADJUSTED BEST LINEAR UNBIASED PREDICTION IN FIELD EXPERIMENTS,"
Conference on Applied Statistics in Agriculture.
https://doi.org/10.4148/2475-7772.1442
NEAREST NEIGHBOR ADJUSTED BEST LINEAR UNBIASED PREDICTION IN FIELD EXPERIMENTS
In field experiments with large numbers of treatments, inference can be affected by 1) local variation, and 2) method of analysis .
The standard approach to local, or spatial, variation in the design of experiments is blocking. While the randomized complete block design is obviously unsuitable for experiments with large numbers of treatments, incomplete block designs - even apparently well-chosen ones - may be only partial solutions. Various nearest neighbor adjustment procedures are an alternative approach to spatial variation .
Treatment effects are usually estimated using standard linear model methods. That is, linear unbiased estimates are obtained using ordinary least squares or, for example when nearest neighbor adjustments are used, generalized least squares. This follows from regarding treatment as a fixed effect. However, when there are large numbers of treatments, regarding treatment as a random effect and obtaining best linear unbiased predictors (BLUP) can improve precision .
Nearest neighbor methods and BLUP have had largely parallel development. The purpose of this paper is to put them together .
