Multilevel Exploratory Factor Analysis (ML-EFA)
Multilevel Exploratory Factor Analysis · Also known as: ML-EFA, multilevel factor analysis, two-level exploratory factor analysis, hierarchical exploratory factor analysis
Multilevel exploratory factor analysis uncovers latent factor structures simultaneously at two or more levels of a data hierarchy — for example, both within individuals and between groups — without imposing a fixed structure in advance. It is essential whenever survey or test items are collected from respondents nested inside classrooms, organisations, or clinics.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use multilevel EFA whenever items are collected from respondents nested in identifiable groups and the ICCs are non-negligible — organisational surveys, classroom-based assessments, multi-site clinical studies, and experience-sampling data are prime examples. It is also the right choice when you suspect the construct structure may differ across levels. Do not use it when the data are not hierarchically structured, when group sizes are very small (fewer than roughly 10 per group) or the number of groups is small (fewer than roughly 30), or when the ICCs are uniformly near zero, in which case standard EFA is sufficient.
Strengths & limitations
- Correctly partitions item variance into within-group and between-group sources, preventing confounded factor solutions.
- Allows the factor structure to differ across levels, revealing level-specific dimensionality that a flat EFA would obscure.
- Provides ICC estimates per item, supporting decisions about whether a multilevel approach is necessary.
- Integrates naturally into multilevel SEM workflows, where the EFA solution can later be tested with a confirmatory multilevel CFA.
- Handles designs common in educational, organisational, and clinical research where single-level assumptions are routinely violated.
- Requires a sufficient number of groups (typically at least 30) and adequate within-group sample sizes to estimate the between-level covariance matrix reliably.
- Model complexity increases substantially: two separate factor structures must be interpreted, rotated, and justified.
- Between-level factor solutions are often based on far fewer units (groups) than within-level solutions, leading to unstable estimates when the number of groups is moderate.
- Software support is more limited than for standard EFA; Mplus is the most capable platform, with R packages (lavaan, Mplus-via-MplusAutomation) as alternatives.
Frequently asked
How does multilevel EFA differ from simply running EFA separately in each group?
Running EFA separately in each group pools within-group and between-group variance within each group's analysis, and it cannot estimate a stable between-group covariance matrix. Multilevel EFA explicitly separates the two sources of variance across all groups simultaneously, yielding level-specific factor solutions that are not confounded.
What ICC threshold justifies a multilevel approach?
There is no hard cutoff, but ICCs above approximately 0.05 are commonly treated as meaningful. Even modest ICCs can produce notably biased standard-EFA solutions if the cluster sizes are large, because the total between-cluster variance accumulates quickly. When all ICCs are below 0.02 and group sizes are small, a single-level EFA is usually adequate.
Can the number of factors differ at the within and between levels?
Yes — and they often do. The between-level structure reflects variation among group means, which is a different phenomenon from individual variation within groups. A construct that appears multidimensional within groups may be essentially unidimensional between groups, or additional group-level dimensions may emerge that have no within-level counterpart.
What software runs multilevel EFA?
Mplus provides the most complete implementation with flexible rotation options and fit indices at each level. The R package lavaan supports multilevel CFA but has limited exploratory functionality; MplusAutomation allows Mplus to be driven from R. Specialized multilevel EFA is not available in standard SPSS or SAS without additional modules.
Should I follow multilevel EFA with multilevel CFA?
Yes, the standard workflow mirrors what is done for single-level factor analysis: use multilevel EFA on a calibration sample to discover the level-specific structures, then cross-validate those structures with multilevel CFA on an independent sample. This two-step process guards against capitalising on chance features of the exploratory solution.
Sources
- Muthén, B. O. (1994). Multilevel covariance structure analysis. Sociological Methods & Research, 22(3), 376–398. DOI: 10.1177/0049124194022003006 ↗
- Ryu, E. & West, S. G. (2009). Level-specific evaluation of model fit in multilevel structural equation modeling. Structural Equation Modeling: A Multidisciplinary Journal, 16(4), 583–601. DOI: 10.1080/10705510903203466 ↗
How to cite this page
ScholarGate. (2026, June 3). Multilevel Exploratory Factor Analysis. ScholarGate. https://scholargate.app/en/psychometrics/multilevel-exploratory-factor-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Bifactor ModelPsychometrics↔ compare
- Confirmatory factor analysisPsychometrics↔ compare
- EFAStatistics↔ compare