样品(材料)
数学
调解
计算机科学
计量经济学
人工智能
背景(考古学)
匹配(统计)
方案(数学)
理论(学习稳定性)
心理学
标识
DOI:10.1080/10705511.2026.2679039
摘要
Multilevel structural equation modeling (MSEM) with maximum likelihood (ML) estimation can produce biased results in small samples, particularly those with missing data. While alternative modeling approaches exist, MSEM remains essential for designs involving multilevel mediation with upper-level outcomes. In such cases, selecting an appropriate estimation strategy is critical. This study addresses these challenges through a Monte Carlo simulation comparing ML and Bayesian estimators in a 1–2–1 mediation within MSEM under varied conditions. Results indicate that, when estimating multilevel mediation in MSEM with incomplete small samples, estimator selection should be guided by data characteristics (such as effect size, sample size, and the extent of missingness), and the availability of prior information, rather than assuming either ML or Bayesian methods to be universally superior. This study contributes to methodological literature by providing systematic simulation evidence and practical, scenario-based recommendations for selecting estimation strategies in multilevel mediation contexts with incomplete small samples.
科研通智能强力驱动
Strongly Powered by AbleSci AI