Multivariate Calibration
研究了如何用一组解释变量预测多个响应变量，比较了经典抽样理论和贝叶斯方法下的点估计与置信区域，并给出了响应变量子集选择方法。
Summary A set of q responses Y = (Y 1 ,…, Yq)T are determined by a set of p explanatory variables x = (X 1,…, Xp)T. A set of l observed vectors Y are available at a single unknown x and it is desired to draw inferences about this unknown vector X. In order to do this calibrating data is available jointly on (Y, X) where two situations are distinguished (i), x is controlled, (ii) x is random. Using orthodox sampling theory on the one hand and Bayesian methods on the other hand point estimators and confidence regions are derived and contrasted. A procedure for selection of a subset of responses is given. Finally a comparison is made of the methods on data from (a) a random calibration experiment of wheat quality using an infrared spectrometer and (b) a controlled experiment of point finish.