摘要
对我国六大银行688条顾客数据,分别运用两种方法(分层回归模型和经典回归模型)在两种软件(MLwiN 2.10 Beta和SPSS 15.0),进行了数学建模.结果显示,经典回归模型进行参数估计的结果不会产生严重偏差;没有足够的证据证明经典回归模型会因为低估标准误从而使得不显著的变量变得显著.结论表明收集数据时,无论采用分层抽样还是随机抽样,建模者都可以先从建立简单模型着手,获得对数据的初步认知.
Using 688 customer data from banking industry, this paper applies a Hierarchical Regression Model using WLwiN 2.10 Beta and a Classical Regression Model via SPSS 15.0. The results show that the Classical Regression Model parameter estimates are not seriously biased, and that there is no adequate evidence to support that the Classical Regression Model contains too many effects to be significant because it underestimates standard errors for the parameter estimates. This indicates that no matter we use random sampling or cluster sampling to collect data, model builders can always apply more basic methods to get an idea about the data.
出处
《数学的实践与认识》
CSCD
北大核心
2010年第8期15-24,共10页
Mathematics in Practice and Theory
基金
国家自然科学基金(70872103)
关键词
分层回归模型
经典回归模型
忠诚意向
顾客资产驱动因素
hierarchical regression model
classical regression model
customer equity drivers loyalty intentions