在Stata中搜索ridge regression，有几个ado文件可供参考。

[此贴子已经被作者于2005-10-27 4:44:26编辑过]

help for rxrmaxl STB-28: sg45 ------------------------------------------------------------------------------

Generalized ridge regression: shrinkage extent ----------------------------------------------

rxrmaxl depvar varlist [if exp] [in range]

[, msteps(#) omdmin(#) qshape(#) rescale(#) tol(#) ]

To reset problem-size limits, see help matsize.

Description -----------

rxrmaxl calculates three normal-theory, maximum-likelihood criteria (classical, empirical Bayes, and random coefficients) for a choice of shrinkage extent in the ridge regression of depvar on varlist.

Options -------

msteps(#) specifies the number of steps per unit increase in MCAL, the multi- collinearity allowance parameter; the default value is 4. The total numbe > r of steps along the generalized shrinkage path from the least squares solu- tion (MCAL=0) to all zero coefficients (MCAL=p) will thus be 1+(msteps*p).

omdmin(#) is the strictly positive minimum value to be used for calculation of (1-delta) shrinkage factors. The default is 10e-13.

qshape(#) controls the shape (or curvature) of the generalized shrinkage path through likelihood space; the default value is 0, which yields Hoerl- Kennard "ordinary" ridge regression. qshape=1 yields uniform shrinkage, and all qshape values between -5 and 5 are allowed.

rescale(#) controls the scaling of the response variable and all p predictor variables in the varlist. The default value is rescale=1 to scale all centered variables to have sample variance 1. rescale=0 should be specified to retain the original scalings.

tol(#) specifies search convergence criterion and defaults to 0.01.

Example -------

. rxrmaxl mpg cylnds cubins hpower weight, q(-1)

Author ------

Robert L. Obenchain Eli Lilly and Company

Also see --------

STB: STB-28 sg45 On-line: help for rxridge and rxrcrlq

谢谢minixi，这几个语法我也琢磨了好几天了，但还是没看懂。好象和别的软件的ridge regression计算不一样。我的问题的缘由是这样的：投了一篇外刊，现修回重改。原论文设计的回归模型的几个变量存在多重共线性，原来的做法是将膨胀因子大的变量剔除，但这样一来许多重要的变量都删除了，无法表达原有的意思，现在想通过岭回归的方法优化原模型，减小多重共线性，然后再做其它必要的检验。

现在的问题是：一是不知道如何用STATA中的岭回归语法，二是不知道如何提取出优化后的数据做各种必要的计量经济检验