Skip to main content
Springer Nature Link
Log in

R-estimation in linear models: algorithms, complexity, challenges

  • Original Paper
  • Published:
Computational Statistics Aims and scope Submit manuscript

Abstract

The main objective of this paper is to discuss selected computational aspects of robust estimation in the linear model with the emphasis on R-estimators. We focus on numerical algorithms and computational efficiency rather than on statistical properties. In addition, we formulate some algorithmic properties that a “good” method for R-estimators is expected to satisfy and show how to satisfy them using the currently available algorithms. We illustrate both good and bad properties of the existing algorithms. We propose two-stage methods to minimize the effect of the bad properties. Finally we justify a challenge for new approaches based on interior-point methods in optimization.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Subscribe and save

Springer+ Basic
$34.99 /Month
  • Get 10 units per month
  • Download Article/Chapter or eBook
  • 1 Unit = 1 Article or 1 Chapter
  • Cancel anytime
Subscribe now

Buy Now

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6

Similar content being viewed by others

Notes

  1. Usually, \(\varphi :[0,1]\rightarrow \mathbb {R}\) is assumed to be a non-decreasing, square-integrable, and bounded function on (0, 1), standardized so that \(\int _0^1\,\varphi (t)\,dt=0\) and \(\int _0^1\,\varphi ^2(t)\,dt=1\). Moreover, some authors require that \(\varphi (t)\) is odd, an assumption that is not necessary in general but is needed, for example, when estimating the intercept parameter based on signed rank scores. Another common assumption is that the scores generated by \(\varphi (t)\) sum up to zero as required in Jaeckel (1972) to ensure the translation equivariance of the resulting estimator.

  2. Jaeckel in a seminal paper (Jaeckel 1972) interpreted (8) as a measure of dispersion of residuals and advocated its use instead of the residual variance on which the classical least squares method is based. Parallel to that, Jurečková (1969) and Koul (1971) suggested other versions of R-estimates. It is shown in Jurečková and Sen (1996), that all the three approaches are asymptotically equivalent in probability and have the same asymptotic properties.

  3. Let \(o_i(\varvec{x})\) denote the value of the \(i^{th}\) smallest coordinate in \(\varvec{x}=\big (x_1,\ldots ,x_n\big )^{\top }\). Putting \(x_{(i)}=o_i(\varvec{x})\), we get \(x_{(1)}\le x_{(2)}\le \ldots \le x_{(n)}\). For such \(\varvec{x}\) that all its coordinates are different from each other, let \(q_i(\varvec{x})\) denote the number of \(x's\le x_i\), that is, the rank of \(x_i\) in \(x_{(1)}\le x_{(2)}\le \ldots \le x_{(n)}\). If \(\varvec{Z}=\big (Z_1,\ldots ,Z_n\big )^{\top }\) is a random vector, the statistic \(Z_{(i)}=o_i(\varvec{Z})\) is called the \(i^{th}\) order statistic and the statistic \(R_i=q_{i}(\varvec{Z})\) is called the rank of \(Z_i\). It is evident that, under our assumptions, \(\varvec{R}\) is a random permutation of \((1,\ldots ,n)^{\top }\). Recall that the ranks are “well defined” only if the probability of coincidence of any pair of coordinates equals to 0. In the case of ties, the corresponding mathematical theory is much more complicated. For details of dealing with ties, see, e.g., monograph (Hájek and Šidák 1967) and the follow-up literature. For our purposes, the vector of ranks will always be a permutations of \(\{1,\ldots ,n\}\), and in the case of ties the ranks will be attributed according to the “first-in first-used” rule.

  4. Note that if the gradient of \(\mathcal {D}_{\varphi }\big (\varvec{\beta }\big )\) exists, it is equivalent to the regression rank test statistic \(S(\varvec{\beta })\) of Jurečková (1969); while if the gradient of \(\mathcal {D}_{\varphi }\big (\varvec{\beta }\big )\) does not exist, then \(S(\varvec{\beta })\) exists but is multivalued.

  5. It is worth noting that information about R add-on packages that provide newer, faster, and/or more efficient algorithms covering robustification of statistical methods is regularly updated in Maechler (2020).

  6. Observe that the line search produces a point \(\varvec{\beta }_s\) in a nonsmooth point of Jaeckel’s dispersion \(\mathcal {D}_{\varphi }\). It can be useful to add a small random perturbation of \(\varvec{\beta }_s\).

  7. Osborne’s method can be run from a vertex of the arrangement (10) only. Switching to that method requires an intermediate step, similar to what is known as “crossover” in optimization. To find Osborne’s starting vertex having \(\varvec{\beta }_s\) in hand from Stage 1, we compute the permutation Q such that \(r_{Q_1}(\varvec{\beta }_s) \le r_{Q_2}(\varvec{\beta }_s) \le \cdots \le r_{Q_n}(\varvec{\beta }_s)\) and solve the linear programming problem \(\min _{\varvec{\beta } \in \mathbb {R}^p}\left\{ \sum _{i=1}^n a(i) (y_{Q_{i}} - \varvec{x}_{Q_i}^{\top }\varvec{\beta })\ |\ y_{Q_{1}} - \varvec{x}_{Q_1}^{\top }\varvec{\beta } \le y_{Q_{2}} - \varvec{x}_{Q_2}^{\top }\varvec{\beta } \le \cdots \le y_{Q_{n}} - \varvec{x}_{Q_n}^{\top }\varvec{\beta }\right\} \).

References

  • Agresti A (2018) Statistical methods for the social sciences, 5th edn. Pearson Education Limited, London

    MATH  Google Scholar 

  • Antoch J, Ekblom H (1995) Recursive robust regression—computational aspects and comparison. Comput Stat Data Anal 19:115–128

    MATH  Google Scholar 

  • Antoch J, Ekblom H (2003) Selected algorithms for robust regression estimators. In: Dutter R et al (eds) Developments in robust statistics. Physica-Verlag, Heidelberg, pp 32–48

    MATH  Google Scholar 

  • Antoch J, Janssen P (1989) Nonparametric regression \(M\)-quantiles. Stat Probab Lett 8:355–362

    MathSciNet  MATH  Google Scholar 

  • Bertsekas D (2009) Convex optimization theory. Athena Scientific, Nashua

    MATH  Google Scholar 

  • Bertsekas D (2015) Convex optimization algorithms. Athena Scientific, Nashua

    MATH  Google Scholar 

  • Bilgic Y, Susmann H (2013) rlme: an \(R\) package for rank-based estimation and prediction in random effects nested models. R J 5:71–79

    MATH  Google Scholar 

  • Björck Å (1996) Numerical methods for least squares problems. SIAM, Philadelphia

    MATH  Google Scholar 

  • Björck Å (2015) Numerical methods in matrix computation. Springer, Heidelberg

    MATH  Google Scholar 

  • Cassart D, Hallin M, Paindaveine D (2010) On the estimation of cross-information quantities in rank-based inference. In: Antoch J, Hušková M, Sen P (eds) Nonparametrics and robustness in modern statistical inference and time series analysis: a Festschrift in honor of Prof. Jana Jurečková, Vol. 7 of IMS Collections, Institute of Mathematical Statistics, pp 35–45

  • Černý M et al (2022) A class of optimization problems motivated by rank estimators in robust regression. Optimization 71:2241–2271

    MathSciNet  MATH  Google Scholar 

  • Černý M, Hladík M, Rada M (2021) Walks on hyperplane arrangements and optimization of piecewise linear functions. Technical report. arXiv:1912.12750

  • Černý M, Rada M, Sokol O (2020) Rank-estimators for robust regression: approximate algorithms, exact algorithms and two-stage methods. In: Huynh V et al (eds) Integrated uncertainty in knowledge modelling and decision making, volume 12482 of lecture notes in computer science. Springer, New York, pp 163–173. https://doi.org/10.1007/978-3-030-62509-2_14

  • Cheng K, Hettmansperger T (1983) Weighted least squares rank regression. Commun Stat A Theory Methods 12:1069–1086

    MATH  Google Scholar 

  • Crimin K, Abebe A, McKean J (2008) Robust general linear models and graphics via a user interface (Web RGLM). J Mod Appl Stat Methods 7:318–330

    MATH  Google Scholar 

  • Draper D (1988) Rank-based robust analysis of linear models. I. Exposition and review (with discussion). Stat Sci 3:239–271

    MATH  Google Scholar 

  • Dutta S, Datta S (2018) Rank-based inference for covariate and group effects in clustered data in presence of informative intra-cluster group size. Stat Med 37:4807–4822

    MathSciNet  MATH  Google Scholar 

  • Dutter R, Huber P (1981) Numerical methods for the non-linear robust regression problem. Commun Stat Simul Comput 13:79–114

    MATH  Google Scholar 

  • Gao F, Han L (2012) Implementing the Nelder–Mead simplex algorithm with adaptive parameters. Comput Optim Appl 51:259–277

    MathSciNet  MATH  Google Scholar 

  • George K, Osborne M (1990) The efficient computation of linear rank statistics. J Stat Comput Simul 35:227–237

    MathSciNet  MATH  Google Scholar 

  • Golub G, van Loan C (1996) Matrix computations, 3rd edn

  • Grötschel M, Lovász L, Schrijver A (1993) Geometric algorithms and combinatorial optimization. Springer, Heidelberg

    MATH  Google Scholar 

  • Hájek J, Šidák Z (1967) The theory of rank tests. Academia, Prague

    MATH  Google Scholar 

  • Hallin M et al (2013) One-step \(R\)-estimation in linear models with stable errors. J Econom 172:195–204

    MathSciNet  MATH  Google Scholar 

  • Hallin M, Mehta C (2015) \(R\)-estimation for asymmetric independent component analysis. J Am Stat Assoc 110:218–232

    MathSciNet  MATH  Google Scholar 

  • Hallin M, La Vecchia D (2017) \(R\)-estimation in semiparametric dynamic location-scale models. J Econom 196:233–247

    MathSciNet  MATH  Google Scholar 

  • Hallin M, Oja H, Paindaveine D (2006) Semiparametrically efficient rank-based inference for shape II. Optimal \(R\)-estimation of shape. Ann Stat 34:2757–2789

    MathSciNet  MATH  Google Scholar 

  • Hallin M, Paindaveine D, Verdebout T (2014) Efficient \(R\)-estimation of principal and common principal components. J Am Stat Assoc 109:1071–1083

    MathSciNet  MATH  Google Scholar 

  • Hardy G (1988) Inequalities, 2nd edn. Cambridge University Press, Cambridge

    MATH  Google Scholar 

  • Hettmansperger T (1984) Statistical inference based on ranks. Wiley, New York

    MATH  Google Scholar 

  • Hettmansperger T, McKean J (2011) Statistical inference based on ranks, 2nd edn. CRC Press, Boca Raton

    MATH  Google Scholar 

  • Hladík M, Černý M, Antoch J (2020) EIV regression with bounded errors in data: total least squares with Chebyshev norm. Stat Pap 61:279–301

    MathSciNet  MATH  Google Scholar 

  • Jaeckel L (1972) Estimating regression coefficients by minimizing the dispersion of the residuals. Ann Math Stat 43:1449–1458

    MathSciNet  MATH  Google Scholar 

  • Jurečková J (1969) Asymptotic linearity of a rank statistic in regression. Ann Math Stat 40:1449–1458

    MathSciNet  MATH  Google Scholar 

  • Jurečková J (2016) Averaged extreme regression quantiles. Extremes 19:41–49

    MathSciNet  MATH  Google Scholar 

  • Jurečková J et al (2016) Behavior of \(R\)-estimators under measurement errors. Bernoulli 22:1093–1112

    MathSciNet  MATH  Google Scholar 

  • Jurečková J, Sen P (1996) Robust statistical procedures. Wiley, New York

    MATH  Google Scholar 

  • Jurečková P, Sen J, Picek J (2012) Methodology in robust and nonparametric statistics. Chapman and Hall/CRC, Boca Raton

    MATH  Google Scholar 

  • Jurečková J, Picek J, Schindler M (2019) Robust statistical methods with R. Chapman and Hall/CRC, Boca Raton

    MATH  Google Scholar 

  • Kloke J, Mckean J (2012) RFIT: rank-based estimation for linear models. R J 4:57–64

    MATH  Google Scholar 

  • Kloke J, Mckean J, Mushfiqur R (2009) Cluster correlated errors. J Am Stat Assoc 104:384–390

    MATH  Google Scholar 

  • Koenker R (2005) Quantile regression. Oxford University Press, Oxford

    MATH  Google Scholar 

  • Koller M (2016) robustlmm: an \(R\)-package for robust estimation of linear mixed-effects models. J Stat Softw 75:1–24

    MATH  Google Scholar 

  • Koul H (1971) Asymptotic behavior of class of confidence regions in multiple linear regression. Ann Math Stat 42:42–57

    MATH  Google Scholar 

  • Koul H, Sievers G, McKean J (1987) An estimator of the scale parameter for the rank analysis of linear models under general score functions. Scand J Stat 14:131–141

    MathSciNet  MATH  Google Scholar 

  • Le Cam L (1986) Asymptotic methods in statistical decision theory. Springer, New York

    MATH  Google Scholar 

  • Letellier T et al (2000) Statistical analysis of mitochondrial pathologies in childhood: identification of deficiencies using principal component analysis. Lab Investig 80:1019–1030

    MATH  Google Scholar 

  • Louhichi S, Miura R, Volný D (2017) On the asymptotic normality of the \(R\)-estimators of the slope parameters of simple linear regression models with associated errors. Statistics 51:167–187

    MathSciNet  MATH  Google Scholar 

  • Madsen K, Nielsen H (1990) Finite algorithms for robust linear regression. BIT 30:682–699

    MathSciNet  MATH  Google Scholar 

  • Maechler M (2020) CRAN task view: robust statistical methods. Technical report. https://cran.r-project.org/web/views/Robust.html/

  • Marazzi A (1992) Algorithms, routines and S-functions for robust statistics, Wadsworth, Pacific Grove

  • Mathworks, Robustfit, Technical report. https://www.mathworks.com/help/stats/robustfit.html/ (2023)

  • McKean J, Hettmansperger T (1978) A robust analysis of the general linear model based on one-step \(R\)-estimates. Biometrika 65:571–579

    MathSciNet  MATH  Google Scholar 

  • McKean J, Sievers G (1989) Rank scores suitable for the analysis of linear models under asymmetric error distributions. Technometrics 31:207–218

    MathSciNet  MATH  Google Scholar 

  • Nelder J, Mead R (1965) A simplex method or function minimization. Comput J 7:308–313

    MathSciNet  MATH  Google Scholar 

  • O’Leary D (1990) Robust regression computation using iteratively reweighted least squares. SIAM J Matrix Anal Appl 11:466–480

    MathSciNet  MATH  Google Scholar 

  • Osborne M (1982) A finite algorithm for the rank regression problem. J Appl Probab 19:241–252

    MathSciNet  MATH  Google Scholar 

  • Osborne M (1985) Finite algorithms in optimization and data analysis. Wiley, New York

    MATH  Google Scholar 

  • Osborne M (2001) Simplicial algorithms for minimizing polyhedral functions. Cambridge University Press, Cambridge

    MATH  Google Scholar 

  • Press W et al (2007) Numerical recipes: the art of scientific computing, 3rd edn. Cambridge University Press, New York

    MATH  Google Scholar 

  • Rao C (1993) Computational statistics. Handbook of statistics 9, North Holland, Amsterdam

  • Rockafeller T (1970) Convex analysis. Princeton University Press, Princeton

    MATH  Google Scholar 

  • Roos C, Terlaky T, Vial J (2005) Interior point methods for linear optimization. Springer, New York

    MATH  Google Scholar 

  • Saleh A, Picek J, Kalina J (2012) \(R\)-estimation of the parameters of a multiple regression model with measurement errors. Metrika 75:311–328

    MathSciNet  MATH  Google Scholar 

  • Schlossmacher E (1973) An iterative technique for absolute deviations curve fiting. J Am Stat Assoc 68:857–859

    MATH  Google Scholar 

  • Sen P, Singer J, Pedroso de Lima A (2010) From finite sample to asymptotic methods in statistics. Cambridge University Press, New York

    MATH  Google Scholar 

  • Shano D, Rocke D (1986) Numerical methods for robust regression. Linear models. SIAM J Sci Stat Comput 7:86–97

    MathSciNet  MATH  Google Scholar 

  • Sievers J, Abebe A (2004) Rank estimation of regression coefficients using iterated reweighted least squares. J Stat Comput Simul 74:821–831

    MathSciNet  MATH  Google Scholar 

  • Singer S, Singer S (2004) Efficient implementation of the Nelder–Mead search algorithm. Appl Numer Anal Comput Math 1:524–534

    MathSciNet  MATH  Google Scholar 

  • van Eeden C, Kraft C (1972) Linearized rank estimates and signed-rank estimates for the general linear hypothesis. Ann Math Stat 43:42–57

    MathSciNet  MATH  Google Scholar 

  • Wolke R (1992) Iteratively reweighted least squares. Algorithms, convergence and numerical comparison. BIT 32:506–524

    MathSciNet  MATH  Google Scholar 

  • Wolke R, Schwetlik H (1988) Iteratively reweighted least squares. Algorithms, convergence and numerical comparisons. SIAM J Sci Stat Comput 9:907–921

    MathSciNet  MATH  Google Scholar 

  • Yu C, Yao W (2017) Robust linear regression: a review and comparison. Commun Stat Simul Comput 46:6261–6282

    MathSciNet  MATH  Google Scholar 

Download references

Acknowledgements

We thank M. Hallin, M. Rada, and other colleagues for a thorough personal discussion and interesting remarks. The authors received valuable input from an associate editor. The work of J. Antoch and M. Černý was supported by the Czech Science Foundation under Grant 22-19353S. The work of R. Miura was supported by JSPS KAKENHI Grant Number 17H02508.

Author information

Authors and Affiliations

  1. Department of Probability Theory and Mathematical Statistics, Faculty of Mathematics and Physics, Charles University, Sokolovska 83, 18675, Prague, Czech Republic

    Jaromír Antoch

  2. Department of Econometrics, Faculty of Informatics and Statistics, Prague University of Economics and Business, Winston Churchill Square 1938/4, 13067, Prague, Czech Republic

    Jaromír Antoch & Michal Černý

  3. Faculty of Economics, Tohoku University, Tohoku University Campus, Kawauchi, Aoba Ward, Sendai, 980-8576, Japan

    Ryozo Miura

Authors
  1. Jaromír Antoch
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Michal Černý
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Ryozo Miura
    View author publications

    You can also search for this author inPubMed Google Scholar

Corresponding author

Correspondence to Michal Černý.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Antoch, J., Černý, M. & Miura, R. R-estimation in linear models: algorithms, complexity, challenges. Comput Stat 40, 405–439 (2025). https://doi.org/10.1007/s00180-024-01495-0

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00180-024-01495-0

Keywords