Peter Huber

Peter J. Huber (1934-2018) was a Swiss statistician who played the central role in formalizing robust statistics as a coherent mathematical discipline. His 1964 paper on M-estimators introduced a generalization of maximum likelihood estimation that could be made robust to outliers by choosing an appropriate loss function, providing the theoretical foundation for an entire field.

Contributions to Robust Statistics

Huber's work was driven by a practical observation: classical statistical methods, optimized for the normal distribution, were catastrophically vulnerable to even small deviations from normality. His M-estimator framework parameterized the tradeoff between efficiency at the normal distribution and robustness to contamination, allowing statisticians to choose estimators that were neither blindly trusting nor wastefully cautious.

Beyond M-estimators, Huber made foundational contributions to robust regression, robust covariance estimation, and the theory of minimax robustness. His 1981 book Robust Statistics became the definitive reference for the field.

The Philosophical Stance

Huber was not merely a mathematician solving technical problems. He was a methodological critic who recognized that statistical practice had become dangerously dependent on idealized assumptions. His work on robust statistics was an argument for statistical methods that acknowledge the possibility of model failure, rather than methods that perform beautifully in a fantasy world and collapse in the real one.