Cross-Classified Multilevel Models in Education
Also known as: Cross-Classified Random Effects Models, CCREM, Cross-Classified Multilevel Modeling, Multiple Membership Cross-Classified Models
Cross-classified multilevel models extend hierarchical linear modeling to situations where units belong to two or more groupings that do not nest neatly inside one another. In education, students are often classified by both school and neighborhood, or by primary and secondary school across time — classifications that cut across each other rather than form a clean hierarchy. These models assign a random effect to each classification simultaneously, partitioning variance among them and yielding correct inferences where a purely nested model would be misspecified.
Key highlights
- Correctly models non-nested grouping structures that purely hierarchical models cannot represent.
- Partitions variance among crossed contexts (e.g., school vs. neighborhood), answering which matters more.
- Avoids the bias from forcing crossed data into a single hierarchy or omitting a classification.
- Extends naturally to multiple-membership structures for mobile students and multiple teachers.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use cross-classified multilevel models whenever the data's grouping structure is genuinely non-hierarchical — students belonging to both schools and neighborhoods, mobile students spanning multiple schools, examinees crossed with raters, or pupils nested in current schools but also in prior schools. They are essential when ignoring a crossed classification would misattribute its variance to another level or to error. They require enough units in each classification and more computation than nested models, and like all observational multilevel models they describe variance and associations rather than establishing causal effects of any one context.
Strengths & limitations
- Correctly models non-nested grouping structures that purely hierarchical models cannot represent.
- Partitions variance among crossed contexts (e.g., school vs. neighborhood), answering which matters more.
- Avoids the bias from forcing crossed data into a single hierarchy or omitting a classification.
- Extends naturally to multiple-membership structures for mobile students and multiple teachers.
- Computationally heavier than nested models; large cross-classifications can be slow or hard to converge.
- Reliable variance estimation needs adequate numbers of units in each classification.
- Specifying and interpreting several crossed random effects is more complex and error-prone.
- Remains observational: variance attributed to a context is not a causal effect of that context.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
What is the difference between nested and cross-classified structures?
In a nested structure each lower-level unit belongs to exactly one higher-level unit — every student is in one classroom, every classroom in one school — forming a strict hierarchy. In a cross-classified structure, units are simultaneously grouped by two or more classifications that cut across each other, so a school contains students from many neighborhoods and a neighborhood sends students to many schools. Nested models assign one random effect per hierarchy level; cross-classified models assign a separate random effect for each crossed classification.
When do I need a multiple-membership rather than a cross-classified model?
Multiple membership applies when a single unit belongs to more than one group within the same classification — a student who attended several schools, or who is taught by several teachers, over the period studied. Instead of one school effect, the model uses a weighted combination of the relevant school effects. Cross-classification concerns membership in different classifications (school and neighborhood). Real data often needs both at once, giving a combined multiple-membership cross-classified (MMCC) model.
Why not just pick the most important grouping and use a nested model?
Because doing so misallocates variance and can bias estimates. If you model only schools and ignore crossed neighborhoods, neighborhood variation gets absorbed into the school effects or the residual, overstating how much schools matter and giving incorrect standard errors. When two contexts genuinely cross, only a model with both random effects can correctly separate their contributions; choosing one and dropping the other is a specification error, not a simplification.
Sources
- 1.Goldstein, H. (2011). Multilevel Statistical Models (4th ed.). Wiley.ISBN 9780470748657
- 2.Raudenbush, S. W. (1993). A crossed random effects model for unbalanced data with applications in cross-sectional and longitudinal research. Journal of Educational Statistics, 18(4), 321–349.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 22). Cross-Classified Multilevel Models in Education. ScholarGate. https://scholargate.app/education/cross-classified-multilevel-education