Robust Linear Mixed-Effects Model
Also known as: robust mixed-effects model, robust linear mixed model, robust LMM, Robust Karma Etkiler Modeli
The robust mixed model is a linear mixed-effects model for panel and repeated-measures data that tolerates outliers and heavy-tailed errors. It replaces the usual likelihood with bounded-influence estimating equations, building on the robust restricted maximum likelihood of Richardson and Welsh (1995) and the robustlmm implementation of Koller (2016).
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 the robust mixed model when you have continuous outcomes with a panel or repeated-measures (clustered) structure and you suspect outliers or heavy-tailed, non-normal errors that would distort an ordinary mixed model. It assumes a genuine grouped structure is present and is designed precisely for the case where normality is doubtful. A reasonably large sample is needed — at least about 50 observations — because the variance components and robust scale must be estimated reliably.
Strengths & limitations
- Tolerates outliers and heavy-tailed errors that would bias an ordinary linear mixed model.
- Does not require normally distributed errors, so it suits contaminated or skewed clustered data.
- Respects the panel or repeated-measures structure through random effects while protecting both the fixed effects and the variance components from contamination.
- Needs a fairly large sample (at least about 50 observations); in small samples it can fail to converge.
- When the outlier ratio is very high (above roughly 25%), the variance components can no longer be estimated reliably.
- More computationally demanding and less standard than ordinary mixed models, and typically relies on specialised software.
Frequently asked
How is this different from an ordinary linear mixed model?
The model structure — fixed effects plus cluster-specific random effects — is identical. The difference is in estimation: the robust version replaces the Gaussian likelihood with bounded-influence estimating equations that downweight outliers, so a few extreme points cannot dominate the fixed effects or the variance components.
When should I prefer a permutation test instead?
In small samples (fewer than about 30 observations) the robust mixed model often struggles to converge. A permutation test sidesteps the variance-component estimation and is the safer choice when the sample is that small.
What if more than a quarter of my data are outliers?
With an outlier ratio above roughly 25%, the robust mixed model can no longer estimate the variance components reliably. In that regime bootstrap inference is more trustworthy than relying on the robust fit.
Does it require normally distributed errors?
No. The method is built precisely for heavy-tailed and non-normal errors. It assumes a genuine panel or repeated-measures structure and continuous outcomes, but it does not require the normality that an ordinary mixed model leans on.
Sources
- Koller, M. (2016). robustlmm: An R Package for Robust Estimation of Linear Mixed-Effects Models. Journal of Statistical Software, 75(6), 1-24. DOI: 10.18637/jss.v075.i06 ↗
- Richardson, A. M. & Welsh, A. H. (1995). Robust Restricted Maximum Likelihood in Mixed Linear Models. Biometrics, 51(4), 1429-1439. DOI: 10.2307/2533273 ↗
How to cite this page
ScholarGate. (2026, June 1). Robust Linear Mixed-Effects Model. ScholarGate. https://scholargate.app/en/statistics/robust-mixed-model
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.
- Heteroscedasticity-Robust Standard ErrorsStatistics↔ compare
- OLS RegressionEconometrics↔ compare
- Panel Fixed EffectsEconometrics↔ compare
- Permutation TestStatistics↔ compare
- Robust RegressionStatistics↔ compare