Specifying a Prior Distribution in Structured Regression Problems
研究了正态多元线性回归模型中回归系数向量的先验均值和协方差矩阵的设定问题，提出利用设计矩阵形式的预期来设定先验，并以含二阶项的模型为例说明。
Abstract The problem of assessing the prior mean and covariance matrix of the vector of regression coefficients in the normal multiple linear regression model is considered. Arguments are advanced for using expectations regarding the form of the design matrix in specifying the prior for problems with a structure that can aid in specifying these expectations. This is illustrated with models having second-order terms. An example is given. Key Words: Bayes estimationSecond-order modelsStein estimatesRidge regression