Robust Power Analysis
Robust Statistical Power Analysis · Also known as: power analysis under non-normality, distribution-free power analysis, robust sample-size determination, contamination-robust power
Robust power analysis computes the statistical power or required sample size for hypothesis tests that use robust estimators — such as trimmed means or Winsorized variances — instead of ordinary means and standard deviations. It protects against inflated or deflated power estimates that arise when data contain outliers, heavy tails, or skewness that violate classical normality assumptions.
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 power analysis whenever you plan to analyze data with a robust test (Yuen's trimmed-mean test, Wilcoxon rank tests, bootstrap procedures) or when prior experience or pilot data suggest heavy tails, skewness, or frequent outliers. It is especially important in small-to-medium samples, where a single outlier has an outsized influence on classical variance estimates. Do not use it when data are clearly normal and free of contamination — classical power formulas are simpler and fully efficient in that case. Also avoid it when you have not yet decided which robust test will be applied: the power formula must match the intended test.
Strengths & limitations
- Accurately estimates power when data deviate from normality, preventing underpowered studies.
- Directly coupled to robust tests, ensuring the planned analysis and the sample-size formula are consistent.
- Reduces the risk of inflated type II error caused by outlier-driven variance inflation.
- Applicable to a wide range of robust procedures including trimmed-mean tests, rank tests, and bootstrap tests.
- Provides more realistic sample-size estimates for applied fields where clean data are uncommon.
- Requires specification of a trim proportion or specific robust estimator before data collection, which demands prior knowledge.
- Closed-form power formulas exist only for some robust tests; others require simulation, increasing computational demands.
- Less familiar than classical power analysis; reviewers and IRBs may request justification for the approach.
- Modest efficiency cost under normality: robust estimators are less efficient than classical ones when assumptions hold, so slightly larger n may be needed.
Frequently asked
How is robust power analysis different from classical power analysis?
Classical power analysis uses the mean and standard deviation as location and scale parameters and assumes normality. Robust power analysis substitutes trimmed means and Winsorized variances, matching the estimators actually used in the planned robust test and producing accurate power estimates when data are non-normal or contaminated.
Do I need to choose a trim proportion before running the study?
Yes. The trim proportion determines both the robust estimator and the power formula. A 20% trim is the most common default and performs well across a broad range of distributions, but it should be chosen before data collection and reported in the study protocol.
Can I perform robust power analysis by simulation?
Yes, and for many robust tests simulation is the most reliable approach. You specify the assumed data distribution (including contamination), trim proportion, and effect size, then simulate thousands of datasets and count the proportion of tests that reject H0. The WRS2 package in R provides functions for this.
Will I need a larger sample size than with a classical approach?
Under true normality, slightly larger samples may be needed because robust estimators are not fully efficient in ideal conditions. However, when data are non-normal, classical formulas can underestimate the required sample, leading to genuinely underpowered studies — robust power analysis prevents that more serious error.
What if I am unsure whether my data will be non-normal?
If prior data or subject-matter knowledge suggest any possibility of heavy tails or outliers, robust power analysis is the safer choice. At worst you recruit a few extra participants; at best you avoid a fatally underpowered study.
Sources
- Luh, W.-M., & Guo, J.-H. (2010). Approximate sample size formulas for the two-sample trimmed mean test with unequal variances. British Journal of Mathematical and Statistical Psychology, 63(1), 83–100. link ↗
- Wilcox, R. R. (2012). Introduction to Robust Estimation and Hypothesis Testing (3rd ed.). Academic Press. ISBN: 978-0123869838
How to cite this page
ScholarGate. (2026, June 3). Robust Statistical Power Analysis. ScholarGate. https://scholargate.app/en/statistics/robust-power-analysis
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.
- Effect size analysisStatistics↔ compare
- Power analysisStatistics↔ compare
- Robust independent samples t-testStatistics↔ compare
- Robust one-way ANOVAStatistics↔ compare