Model clustered, nested, and repeated data without pretending observations are independent
Multilevel and mixed-effects models can represent dependence across students, schools, sites, people, teams, or repeated observations when the design and sample support them.
Start with the source of dependence
The first decision is not random intercept versus random slope. It is which observations share information, at what level, and how the research question crosses those levels.
- Map units and levels
- Distinguish nesting from crossing
- Identify within- and between-unit questions
Specify variation the design can support
Random effects, time trends, covariance structures, and cross-level terms add meaning only when supported by the design and data. Singular fits and unstable variance estimates are evidence to investigate.
- Compare defensible specifications
- Check residual and influence patterns
- Use sensitivity analysis for key choices
Interpret at the correct level
A within-person change is not automatically a between-person difference. Centering, coding, and interaction choices determine the meaning of coefficients and should be documented.
- Coefficient interpretation map
- Variance and uncertainty summary
- Prediction or marginal-effect display
Review my multilevel design
Bring the research question, design, or existing output. A first review can identify the next statistical decision and the information needed to scope it.