{
  "name": "Documented default tuning constants and breakdown figures for robust regression in R, Python statsmodels and scikit-learn",
  "description": "A cross-software comparison of the tuning constants, scale estimators and breakdown figures stated in the official documentation of R MASS::rlm (7.3-65), Python statsmodels RLM (0.15.0) and scikit-learn 1.1.14 for Huber, bisquare, Hampel, MM, Theil-Sen estimators. Compiled by reading each manual page; no single page lists all rows together.",
  "page": "https://casrai.org/guides/robust-regression",
  "publisher": "CASRAI",
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "dateModified": "2026-11-18T08:20:00Z",
  "temporalCoverage": null,
  "spatialCoverage": null,
  "columns": [
    {
      "name": "Estimator or setting",
      "unit": "",
      "description": ""
    },
    {
      "name": "Software",
      "unit": "",
      "description": ""
    },
    {
      "name": "Parameter",
      "unit": "",
      "description": ""
    },
    {
      "name": "Documented value",
      "unit": "",
      "description": ""
    },
    {
      "name": "What the documentation says",
      "unit": "",
      "description": ""
    }
  ],
  "rows": [
    {
      "Estimator or setting": "Huber M-estimator",
      "Software": "R MASS::rlm",
      "Parameter": "psi.huber k",
      "Documented value": "1.345",
      "What the documentation says": "Default for the Huber psi; the Huber estimator has a unique solution"
    },
    {
      "Estimator or setting": "Tukey bisquare M-estimator",
      "Software": "R MASS::rlm",
      "Parameter": "psi.bisquare c",
      "Documented value": "4.685",
      "What the documentation says": "Can have multiple local minima, so the starting point matters"
    },
    {
      "Estimator or setting": "Hampel M-estimator",
      "Software": "R MASS::rlm",
      "Parameter": "psi.hampel a, b, c",
      "Documented value": "2, 4, 8",
      "What the documentation says": "Can have multiple local minima, so the starting point matters"
    },
    {
      "Estimator or setting": "MM-estimator",
      "Software": "R MASS::rlm",
      "Parameter": "k0 of the initial S-estimator",
      "Documented value": "1.548",
      "What the documentation says": "Tukey's biweight initialised by an S-estimator; case weights not supported"
    },
    {
      "Estimator or setting": "Scale estimate",
      "Software": "R MASS::rlm",
      "Parameter": "scale.est",
      "Documented value": "MAD (default)",
      "What the documentation says": "Rescaled median absolute deviation of residuals; Huber proposal 2 is the alternative"
    },
    {
      "Estimator or setting": "Iteration control",
      "Software": "R MASS::rlm",
      "Parameter": "maxit, acc",
      "Documented value": "20, 1e-4",
      "What the documentation says": "Defaults for the maximum iterations and convergence tolerance"
    },
    {
      "Estimator or setting": "Huber M-estimator",
      "Software": "Python statsmodels",
      "Parameter": "HuberT t",
      "Documented value": "1.345",
      "What the documentation says": "Default for the HuberT norm used by RLM"
    },
    {
      "Estimator or setting": "Huber regressor",
      "Software": "Python scikit-learn",
      "Parameter": "epsilon",
      "Documented value": "1.35",
      "What the documentation says": "Advised for 95% statistical efficiency; outliers are downweighted, not discarded"
    },
    {
      "Estimator or setting": "Theil-Sen estimator",
      "Software": "Python scikit-learn",
      "Parameter": "Breakdown point (simple linear regression)",
      "Documented value": "about 29.3%",
      "What the documentation says": "Robustness decreases as the number of predictors grows"
    }
  ],
  "sources": [
    {
      "title": "R MASS::rlm manual page (7.3-65)",
      "url": "https://stat.ethz.ch/R-manual/R-devel/library/MASS/html/rlm.html",
      "accessed": "2026-11-15"
    },
    {
      "title": "statsmodels HuberT norm",
      "url": "https://www.statsmodels.org/stable/generated/statsmodels.robust.norms.HuberT.html",
      "accessed": "2026-11-15"
    },
    {
      "title": "statsmodels Robust Linear Models (0.15.0)",
      "url": "https://www.statsmodels.org/stable/rlm.html",
      "accessed": "2026-11-15"
    },
    {
      "title": "scikit-learn User Guide 1.1.14 Robustness regression",
      "url": "https://scikit-learn.org/stable/modules/linear_model.html",
      "accessed": "2026-11-15"
    }
  ]
}