The Two-Stage approach to optimal \textit{non-linear} predictions via Generalized Ridge Regression is again illustrated. This time, we use a model with six $x-$predictors and more than $2,500$ observations. Unbiased estimates and predictions are then compared with their corresponding ``optimally biased'' estimates and predictions most likely to have minimal MSE risk under Normal distribution theory. Again, we find that lower residual standard errors and lower MSE risks relative to those lower errors predominate.

翻译：本文再次阐释了通过广义岭回归实现最优非线性预测的两阶段方法。此次，我们使用了一个包含六个$x$预测变量及超过$2{,}500$个观测值的模型。在正态分布理论框架下，将无偏估计与预测同其对应的“最优有偏”估计与预测（最可能具有最小均方误差风险）进行比较。再次发现，残差标准误差更低且均方误差风险低于这些更低误差的情况占据主导地位。