摘要：In this paper, we examine the small sample properties of R 2 based on the Stein-rule estimator of coefficients (say, R 2S) when relevant regressors are omitted in a specified model. The following is shown, when the model is correctly specified, the bias of R 2S is smaller than that of R 2 based on the OLS estimator (say, R 2S), and the mean square error (MSE) of R 2S is smaller than or comparable with that of R 20. But, as the magnitude of specification error increases, both bias and MSE of R 2S become larger than those of R 20.
