Meaningful Mediation Analysis: Plausible Causal Inference and Informative Communication
针对统计调解分析中从调解变量到结果变量的相关性推断和结果沟通两个薄弱环节，提出六项条件、甜蜜点分析和四项沟通要素，并通过文献回顾、数据重分析和模拟验证建议。
Abstract Statistical mediation analysis has become the technique of choice in consumer research to make causal inferences about the influence of a treatment on an outcome via one or more mediators. This tutorial aims to strengthen two weak links that impede statistical mediation analysis from reaching its full potential. The first weak link is the path from mediator to outcome, which is a correlation. Six conditions are described that this correlation needs to meet in order to make plausible causal inferences: directionality, reliability, unconfoundedness, distinctiveness, power, and mediation. Recommendations are made to increase the plausibility of causal inferences based on statistical mediation analysis. Sweetspot analysis is proposed to establish whether an observed mediator-outcome correlation falls within the region of statistically meaningful correlations. The second weak link is the communication of mediation results. Four components of informative communication of mediation analysis are described: effect decomposition, effect size, difference testing, and data sharing. Recommendations are made to improve the communication of mediation analysis. A review of 166 recently published mediation analyses in the Journal of Consumer Research, a reanalysis of two published datasets, and Monte Carlo simulations support the conclusions and recommendations.