Robust Weighted Least Squares (Robust WLS)
Robust Weighted Least Squares · Also known as: robust weighted least squares, RWLS, heteroscedasticity-robust WLS, outlier-robust weighted regression
Robust WLS combines weighted least squares — which corrects for known or estimated heteroscedasticity — with robust M-estimation that down-weights influential outliers. The result is a regression estimator that is simultaneously efficient under non-constant error variance and resistant to observations that would otherwise distort coefficient estimates.
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 Robust WLS when your data exhibit both heteroscedasticity (non-constant error variance) and a meaningful proportion of outliers or high-leverage observations — conditions that render OLS inefficient and standard WLS non-robust. It is well-suited to economic, financial, or survey data where extreme values arise naturally. Do not use it when the error distribution is genuinely symmetric and light-tailed with constant variance (OLS is simpler and equally efficient), or when the proportion of outliers exceeds the breakdown point of the chosen M-estimator (roughly 50% for bisquare). If heteroscedasticity alone is the concern without notable outliers, standard WLS or OLS with robust standard errors is sufficient.
Strengths & limitations
- Simultaneously handles heteroscedasticity through observation weighting and protects against outliers through a robust loss function.
- The IRLS algorithm converges to a consistent, asymptotically normal estimator under mild regularity conditions.
- Flexible choice of loss functions (Huber, bisquare, Hampel) lets practitioners tune the trade-off between robustness and efficiency.
- Produces valid sandwich standard errors that remain reliable even under mild model mis-specification.
- More efficient than robust OLS alone when a variance structure can be specified or estimated.
- Requires specifying or estimating the initial heteroscedasticity weight structure; misspecification reduces efficiency.
- Convergence of IRLS is not always guaranteed for highly non-linear weight functions or extreme outlier patterns.
- Interpretation is less straightforward than OLS because coefficients correspond to a down-weighted sample rather than the full dataset.
- Breakdown point of the robustness component depends on the chosen loss function and can be lower than the analyst expects.
- Computationally more intensive than standard WLS or OLS, especially for large datasets.
Frequently asked
What is the difference between Robust WLS and WLS with robust standard errors?
Standard WLS with robust (sandwich) standard errors corrects the inference for heteroscedasticity not captured by the weight matrix but leaves the point estimates unchanged and sensitive to outliers. Robust WLS additionally modifies the point estimates themselves by down-weighting outlying observations through a robust loss function, so both the estimates and the inference are protected.
Which robust loss function should I choose — Huber or bisquare?
Huber weights are a compromise: they behave like OLS for small residuals and like least-absolute-deviations for large ones, giving moderate protection with high efficiency. Bisquare (Tukey) weights completely zero out observations beyond a threshold, offering stronger protection but lower efficiency when outliers are actually informative. Start with Huber for general use; switch to bisquare only if you are confident that extreme observations are truly anomalous.
How do I know if my data need Robust WLS rather than just robust standard errors on OLS?
If diagnostic plots show both non-constant residual variance (heteroscedasticity) and influential outliers (high Cook's D or hat values), Robust WLS is warranted. If only heteroscedasticity is present without influential outliers, WLS or OLS with robust standard errors is simpler and adequate.
Does Robust WLS require any distributional assumptions?
No strong distributional assumption is required; Robust WLS is consistent and asymptotically normal under mild regularity conditions regardless of the exact error distribution. This makes it particularly attractive when the error distribution is unknown or heavy-tailed.
Can Robust WLS be applied to panel data?
Yes. In panel settings, the weights can reflect both cross-sectional variance differences and the robust down-weighting of outlying observations or time periods. However, the weight construction becomes more complex, and dedicated panel-robust estimators (Robust panel WLS or panel GLS with robust adjustments) are often preferred for balanced panels.
Sources
- Huber, P. J. (1981). Robust Statistics. Wiley. ISBN: 978-0471418054
- Greene, W. H. (2018). Econometric Analysis (8th ed.). Pearson. ISBN: 978-0134461366
How to cite this page
ScholarGate. (2026, June 3). Robust Weighted Least Squares. ScholarGate. https://scholargate.app/en/econometrics/robust-wls
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
- Quantile RegressionEconometrics↔ compare
- Robust GLSEconometrics↔ compare
- Robust OLSEconometrics↔ compare
- Weighted Least SquaresStatistics↔ compare