Biometrical Letters vol. 43(1), 2006, pp. 19-37
|
THE INFLUENCE OF SPATIAL COVARIANCES ON THE TYPE I ERROR AND THE POWER FOR DIFFERENT EVALUATION MODELS Xiyuan Hu1, Joachim Spilke2, Christel Richter3 1College of Agriculture, Northwest-Sci-Tech University of Agriculture and Forestry Yangling, China 2Biometrics and Informatics in Agriculture Group, Martin-Luther-University Halle-Wittenberg, Germany 3 Institute of Crop Sciences, Humboldt-Universität zu Berlin, Germany |
This paper deals with the consequences of spatial covariances for the analysis of field experiments. Based on a Monte Carlo simulation, we investigated the Type I error and the power of three analysis models (block design, exponential and spherical model) in the context of two true spatial models (exponential and spherical). The block design leads to considerable bias of the nominal type I error and to a loss of power. In contrast, the application of the spatial models for analyses guarantees the maintenance of the nominal statistical type I error and a significantly higher power. Our results confirm the necessity of model selection in the analysis of field experiments.
spatial covariances, comparison of analysis models, type I error, power
