Regression Strategies When Multicollinearity Is a Problem: A Methodological Note
指出会计研究中应对多重共线性的两种常见策略（正交变换和两阶段法）并未真正解决问题，因为它们得到的系数估计要么与普通最小二乘法相同，要么有偏。
Researchers in accounting often find that the high degree of multicollinearity in financial data prevents them from obtaining precise estimates of coefficients of specific variables in regression applications. In response to this problem, several investigators have adopted regression strategies which presumably circumvent the problem of multicollinearity. One approach, employed by Brown and Ball [1967] and Beaver, Griffin, and Landsman [1982], among others, transforms the correlated explanatory variables into orthogonal variables.1 A second strategy, used recently by Wilson and Howard [1984], involves a two-stage procedure. Unfortunately, these strategies do not solve the multicollinearity problem since they lead to estimated coefficients which either are identical to those from ordinary least squares, or they are biased. Assume that the model we wish to estimate is: