多元回归分析中的交互作用问题——以语言阈限假设检验为例

Interaction in Multiple Regression Analysis as Exemplified by Testing the Linguistic Threshold Hypothesis

前期研究将二语水平(和母语阅读能力)由连续型变量转化为有序的类别变量,依据类别变量的水平采用皮尔逊相关分析、不含交互作用项的多元回归分析和方差分析的方法检验语言阈限假设。本文认为,语言阈限假设检验的核心是作为连续型变量的二语水平和母语阅读能力在预测二语阅读能力时是否存在交互作用,最恰当的统计分析方法是含交互作用项的多元回归分析。通过交互作用诊断、回归模型比较和简单斜率分析验证了语言阈限假设,为后期研究提供了统计分析上的启示。

In an attempt to test the linguistic threshold hypothesis, previous studies transformed L2 proficiency （and L1 reading ability） from a continuous variable into an ordinal categorical one, and.based on the ordinal levels, they conducted Pearson correlation analyses, multiple regression analyses with an interaction term excluded as well as ANOVAs. This arti- cle argues that the core of testing the linguistic threshold hypothesis lies in testing the interaction between L2 proficiency and L1 reading ability on L2 reading ability and that the most proper statistical method is a multiple regression analysis with an interaction term included. Interaction diagnostics, a regression model comparison and an analysis of simple slopes join together to confirm the linguistic threshold hypothesis, providing insight into statistical analysis for further studies.