Multilevel SEM

Multilevel structural equation modeling extends SEM to clustered data by separating the covariance structure into within-cluster and between-cluster parts and fitting latent variable models at each level.

다층 구조방정식모형은 공분산 구조를 집단 내와 집단 간 부분으로 나누고 각 수준에서 잠재변수 모형을 추정하여 구조방정식모형을 군집 자료로 확장한 방법입니다.

How do I fit a structural equation model when respondents are nested in groups such as schools or companies?

응답자가 학교나 회사 같은 집단에 속해 있을 때 구조방정식모형은 어떻게 추정하나요?

Multilevel SEM fits measurement and structural models separately at the individual (within) and group (between) levels, so latent relationships are not confounded by clustering. It fits survey items collected from people nested in organizations, schools, or regions when a sufficient number of clusters is available.

When to use Multilevel SEM

• Employees nested in companies or students nested in schools answering multi-item scales
• Questions about whether construct relationships differ within versus between groups
• Latent constructs with meaningful variance at the cluster level

When to use something else

• Only observed variables with a single outcome: use a multilevel (mixed-effects) regression
• Few clusters: use single-level SEM with cluster dummies (fixed effects) or small-sample-corrected cluster-robust standard errors, and avoid between-level modeling

Assumptions and requirements

• Sufficient number of clusters for between-level estimation
• Non-trivial intraclass correlations for the indicators
• Approximate multivariate normality, or a robust estimator
• Correctly specified models at both levels

What the analysis reports

• Intraclass correlations of indicators
• Within- and between-level loadings and path coefficients
• Chi-square, CFI, TLI, and RMSEA
• Level-specific SRMR

How to report it

The two-level model fit the data well, χ²([df]) = [value], CFI = [value], RMSEA = [value], SRMR-within = [value], SRMR-between = [value].

Related methods

• Structural Equation Modeling (SEM)
• Linear Mixed-Effects Model
• Multilevel CFA

Run Multilevel SEM on Quorant