中文摘要：

偏最小二乘法（PLS）模型对数据没有做任何的分布假设，因此基于分布假设的统计检验方法无法实现对PLS模型预测效度的度量。采用非参数检验方法——Stone-Geisser检验，通过盲目（Blindfold）过程进行预测效度的度量。使用MATLAB软件实现了两种Stone-Geisser检验——公因子方差检验与冗余检验。在地税部门顾客满意度应用中，Stone-Geisser检验结果与PLS模型其他结果保持内在一致，从而表明检验方法是有效的。因为Stone-Geisser检验基于交叉验证技术，因此非常适合作为PLS模型的基准检验方法。

英文摘要：

Partial least square （ PLS ） model has no distribution hypothesis for the data, therefore the statistical test method based on distribution hypothesis cannot measure the predictive relevance of PLS model.The paper presents a method which uses the Stone-Geisser test to measure the predictive relevance through Blindfold procedure.Using MATLAB software,two kinds of Stone-Geisser tests, communality test and redundancy test are completed. In the application to the customer satisfaction mcasurement of a local taxation bureau, internal consistence is found existing between the results of the Stone-Geisser test and other results of the PLS model, which proves the validity of the Stone-Geisser test.So Stone-Geisser test can be used to verify the PLS model as a basic method because it is based on cross-validation technology.