语言学实验研究中基于项目的方差分析——兼为一篇被指虚构统计的论文辩护

当一个语言学实验同时有被试样本和项目样本时,基于被试和基于项目做两次方差分析检验同一个效应,是比较科学的做法.这样的研究在外语界不仅少见,而且有的还被批评为虚构统计.基于项目的方差分析,讨论其适用范围,给出具体做法和SPSS操作步骤.为某篇被批评的外语实证研究论文做方法上的辩护,指出了其中每个效应出现两个F检验的来由.

Analysis of variance by items in linguistic experiments

When a linguistic experiment consists of both subject samples and item samples, it is a better way to apply analysis of variance （ANOVA） by subjects and by items respectively to test the same effect. However, this kind of approach is not commonly used in foreign language empirical researches in China. Some researches adopting this approach were even criticized as ＂fabricate statistic results＂. The present paper first provides a brief introduction to ANOVA by items. Then it discusses some related issues and provides operational steps in SPSS. Finally, it makes a defense against the strong criticism to a certain paper in which each effect was tested by two F-tests. Two F-tests from ANOVA by subjects and by items in that paper are appropriate whereas the critic is completely ignorant about the ANOVA by items.