Crawford J R, Howell D C
Department of Psychology, King's College, University of Aberdeen, UK.
J Clin Exp Neuropsychol. 1998 Oct;20(5):755-62. doi: 10.1076/jcen.20.5.755.1132.
Regression equations are widely used in clinical neuropsychology, particularly as an alternative to conventional normative data. In neuropsychological applications the most common method of making inferences concerning the difference between an individual's test score and the score predicted by a regression equation is to multiply the standard error of estimate by an appropriate value of z to form confidence limits around the predicted score. The technically correct method is to calculate the standard error of a new individual Y and multiply it by the value of t corresponding to the desired limits (e.g., 90% or 95%). These two methods are compared in data sets generated to be broadly representative of data sets used in clinical neuropsychology. The former method produces confidence limits which are narrower than the true confidence limits and fail to reflect the fact that limits become wider as scores on the predictor deviate from the mean. However, for many of the example data sets studied, the differences between the two methods were trivial, thereby providing reassurance for those who use the former (technically incorrect) method. Despite this, it would be preferable to use the correct method particularly with equations derived from samples with modest Ns, and for individuals with extreme scores on the predictor variable(s). To facilitate use of the correct method a computer program is made available for clinical practice.
回归方程在临床神经心理学中被广泛应用，尤其是作为传统常模数据的一种替代方法。在神经心理学应用中，推断个体测试分数与回归方程预测分数之间差异的最常见方法是将估计标准误差乘以适当的z值，以围绕预测分数形成置信区间。技术上正确的方法是计算新个体Y的标准误差，并将其乘以对应于所需置信区间（如90%或95%）的t值。在生成的广泛代表临床神经心理学中使用的数据集的数据集中，对这两种方法进行了比较。前一种方法产生的置信区间比真实置信区间窄，并且没有反映出随着预测变量得分偏离均值，置信区间会变宽这一事实。然而，对于所研究的许多示例数据集，两种方法之间的差异微不足道，从而为使用前一种（技术上不正确）方法的人提供了安慰。尽管如此，最好使用正确的方法，特别是对于从样本量较小的样本得出的方程，以及对于预测变量得分极端的个体。为便于使用正确的方法，提供了一个计算机程序用于临床实践。
