The effect of different effect size measures on mediation analysis performance

Clinical trials primarily focus on assessing the impact of interventions or exposures on desired outcomes. However, it is equally important to conduct research that delves into the underlying mechanisms through which these interventions influence outcomes. This aspect of clinical research, which aims to understand the causal pathways between the intervention and the outcome, often goes overlooked. Mediation analysis methods provide a valuable approach to investigate and unravel these mechanisms by examining the role of intermediate variables. In mediation analysis, the development of various effect size measures has greatly contributed to our understanding of the strength, direction, and practical significance of relationships between variables, particularly in the context of mediating variables. Despite the increasing popularity of mediation analysis, there remains a limited focus on effect size measurements specific to this analysis. The commonly used mediation ratio, although prevalent, exhibits limitations in its performance, particularly in small sample sizes, and may introduce bias into the results. According to the simulation results, it was seen that the mediation ratio measures were far from the expected value and the bias was high in cases where the sample size was below 500. To address these limitations and enhance the validity of inferences, it is essential to consider alternative effect size measures. One such measure that has demonstrated favorable performance, even in small sample sizes and with minimal bias, is the proposed R2 regression explanation coefficient. Incorporating this measure alongside the mediation ratio provides a more comprehensive and reliable evaluation of the mediation model. Therefore, it is recommended that future studies exploring mediation models embrace a broader perspective by considering multiple effect size measures. It has been observed that the proposed R2 regression explanation coefficient, which is recommended to be used in addition to the mediation ratio, performs well even in small sample sizes such as 100 and 300 and has minimum bias. For this reason, considering alternative measures together without focusing on a single effect size measure in studies in which the mediation model is established will make the research findings more reliable.