多因素研究设计方差分析的实施及交互效应考察

The implementation of variance analysis and analysis of interactive effect in multi-factor research design

多因素研究设计在科学探索中应用广泛,可同时探索多个因素对结局指标的影响。在对这类资料进行统计分析时,交互效应是不应轻易回避的问题;对交互效应考察后方决定后续进行主效应分析或单独效应分析。但不少研究者会忽略对交互效应的考察或未根据其检验结果正确选择分析方法。本文基于SPSS 20.0和R 3.6.1统计软件,对析因设计和重复测量设计这两种常见的多因素设计资料进行数据模拟与分析,阐释考察交互效应的重要意义及相应的分析要点,指出忽略交互效应的统计学显著性可能导致的后果,包括:导致检验效能降低,容易得出错误结论;导致损失原始资料中的宝贵信息,或导致分析结果丧失实际意义等。建议在对研究数据进行分析时,首先判断是否存在交互效应,然后正确选择主效应分析或单独效应分析,以避免得出片面的、错误的结论。

Multi-factor research design is widely applied in scientific research. It can simultaneously explore the effects of multiple factors on outcome indicators. The consideration of the interactive effects of different factors is a critical issue when analyzing this type of data. The analytic strategy for main effects or simple effects depends on the significance of the interactive effect. However, many researchers tend to skip the analysis on interactive effects, or wrongly select statistical analysis method because of ignoring the test result. In this study, SPSS 20.0 and R 3.6.1 statistical software were used to simulate and illustrate how to analyze data from two most popular multi-factor design data--factorial design and repeated measurement design. The significance of evaluating interactive effect and corresponding key point analysis was explained. The possible consequences of ignoring the statistical significance of interactive effects were indicated, that include leading to low inspection efficiency, prone to draw wrong conclusions, loss of valuable information in the original data, or loss of practical significance of the analytic results. It is suggested that in the analysis of research data, we should first judge whether there are interactive effects, and then correctly choose main effect analysis or single effect analysis to avoid one-sided and wrong conclusions.