Psychosomatic Medicine and Psychotherapy, University Hospital Freiburg, Freiburg, Germany 79104.
Psychother Res. 2009 Jul;19(4-5):482-92. doi: 10.1080/10503300902905939.
In explorative regression studies, linear models are often applied without questioning the linearity of the relations between the predictor variables and the dependent variable, or linear relations are taken as an approximation. In this study, the method of regression with optimal scaling transformations is demonstrated. This method does not require predefined nonlinear functions and results in easy-to-interpret transformations that will show the form of the relations. The method is illustrated using data from a German multicenter project on the indication criteria for inpatient or day clinic psychotherapy treatment. The indication criteria to include in the regression model were selected with the Lasso, which is a tool for predictor selection that overcomes the disadvantages of stepwise regression methods. The resulting prediction model indicates that treatment status is (approximately) linearly related to some criteria and nonlinearly related to others.
在探索性回归研究中，通常应用线性模型，而不质疑预测变量与因变量之间的关系的线性，或者将线性关系作为一种近似。在这项研究中，展示了具有最佳标度变换的回归方法。该方法不需要预定义的非线性函数，并产生易于解释的变换，将显示关系的形式。该方法使用德国多中心项目的住院或日间诊所心理治疗指征的数据集进行说明。使用 Lasso 选择要纳入回归模型的指征，Lasso 是一种用于预测因子选择的工具，可以克服逐步回归方法的缺点。得出的预测模型表明，治疗状况与某些标准呈（近似）线性关系，与其他标准呈非线性关系。
