Evaluating the Robustness of Multivariate Tax Models to Errors: A Section 162(a)(2) Illustration.
提出一种评估多元税收模型对数据输入误差稳健性的通用方法，以100个税务法院案例为例，通过蒙特卡洛模拟测试线性判别模型在随机和系统误差下的表现。
Abstract A general procedure is described for evaluating the robustness of multivariate empirical tax models to data input error. Our study focuses on 100 Tax Court cases concerning Section 162(a)(2)-location of the "tax home." The sensitivity of several linear discriminant models to random and systematic data error contamination, ranging between three and 30 percent, was tested in 600 Monte Carlo iterations. Robustness was measured with regard to three factors affected by errors: (1) the order in which variables enter the discriminant models, (2) the impact on standardized discriminant coefficients, and (3) the deterioration in model classification accuracy. Implications and extensions for future research are discussed.