Nearly Orthogonal Randomized Designs
提出一种新的实验随机化方法，适用于加性一般线性模型，通过模拟退火算法生成设计，并引入诊断指标平衡有效性与效率，对田野实验中的连续分段线性模型有潜在应用。
SUMMARY A new method of experimental randomization is presented that is suitable under additive general linear models for treatment and unit effects. The method is designed for a standard normal theory analysis of the data. The method is motivated by thinking of the randomization strategy, which may be represented as a distribution on the group of n × n permutation matrices, as a discrete approximation to a fictitious, ideal randomization strategy corresponding to a continuous distribution on the group of n × n orthogonal matrices. Designs are generated by using a simulated annealing algorithm. The method requires the selection of a tuning constant that determines a trade-off between validity and efficiency. The tuning constant is chosen with the aid of a new diagnostic measure of validity. The ideas are illustrated by using continuous, piecewise linear models of potential use in field experiments.