Cluster-Robust Standard Errors
Cluster-Robust (Clustered) Standard Errors · Also known as: clustered standard errors, cluster-robust inference, clustered variance estimator, Küme Robust Standart Hatalar
Cluster-robust standard errors correct the variance of regression coefficients when observations are correlated within clusters such as schools, hospitals, or regions. The clustered sandwich estimator grew out of Liang & Zeger's (1986) generalized estimating equations and was synthesized for applied work by Cameron & Miller (2015), delivering valid inference when ordinary standard errors would be too small.
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 cluster-robust standard errors whenever observations are naturally grouped and may be correlated inside those groups — panel, longitudinal, or cross-sectional designs with a defined cluster structure — and the outcome is continuous or binary. The approach assumes the cluster structure is correctly specified, clusters are mutually independent, and there are enough of them (at least about 20, with a minimum sample of roughly 50). It is not appropriate with very few clusters, where the asymptotic correction breaks down.
Strengths & limitations
- Produces valid inference under arbitrary correlation within clusters, without modelling that correlation explicitly.
- Leaves the coefficient estimates untouched — only the standard errors, t-tests, and confidence intervals are corrected.
- Applies broadly across panel, longitudinal, and cross-sectional data with a defined grouping.
- Relies on asymptotics in the number of clusters; with fewer than about 20 clusters the standard errors become unreliable.
- Requires a correctly specified cluster structure and independence between clusters.
- With very few clusters (below about 10) the cluster correction is essentially meaningless.
Frequently asked
Do cluster-robust standard errors change my coefficients?
No. The point estimates from the underlying regression stay exactly the same. Only the variance — and therefore the standard errors, t-statistics, and confidence intervals — is corrected for within-cluster correlation.
How many clusters do I need?
At least about 20 clusters are recommended for the asymptotic correction to be trustworthy. With fewer than 20 the standard errors become unreliable and a wild cluster bootstrap is preferable; with fewer than about 10 the correction is essentially meaningless and a permutation test is the better choice.
At what level should I cluster?
Cluster at the level where correlation is plausibly induced — typically the sampling or treatment-assignment unit, such as the school, hospital, or region. The cluster structure must be correctly specified and clusters should be independent of one another.
How is this different from heteroscedasticity-robust standard errors?
Heteroscedasticity-robust (Huber-White) standard errors allow each observation its own error variance but still treat observations as independent. Cluster-robust standard errors go further, permitting arbitrary correlation among all observations within the same cluster.
Sources
- Liang, K. Y. & Zeger, S. L. (1986). Longitudinal Data Analysis Using Generalized Linear Models. Biometrika, 73(1), 13-22. DOI: 10.1093/biomet/73.1.13 ↗
- Cameron, A. C. & Miller, D. L. (2015). A Practitioner's Guide to Cluster-Robust Inference. Journal of Human Resources, 50(2), 317-372. DOI: 10.3368/jhr.50.2.317 ↗
How to cite this page
ScholarGate. (2026, June 1). Cluster-Robust (Clustered) Standard Errors. ScholarGate. https://scholargate.app/en/statistics/cluster-robust-se
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.
- OLS RegressionEconometrics↔ compare
- Panel Fixed EffectsEconometrics↔ compare
- Permutation TestStatistics↔ compare
- Wild BootstrapStatistics↔ compare