Robust Regression
Zaman, Asad, Peter J. Rousseeuw, and Mehmet Orhan. "Econometric applications of high-breakdown robust regression techniques." Economics Letters 71.1 (2001): 1-8.
The SSRN version (pre-publication pdf) inserts a few lines which were accidentally omitted from the published version.
A Zaman, PJ Rousseeuw, M Orhan - Economics Letters, 2001 - Elsevier
A literature search shows that robust regression techniques are rarely used in applied
econometrics. We present a technique based on Rousseeuw and Van Zomeren [Journal of
the American Statistical Association, 85 (1990) 633–639] that removes many of the ...
Cited by 92 Related articles All 14 versions Cite
This is one of my most popular papers, and has been cited by 92 authors. The published version is available here
1. Introduction
Since it is well-established that even high quality data tend to contain outliers, and it is a rare
economic data set that qualifies as ‘high-quality’, one would expect far greater reliance on robust
regression techniques than is actually observed. Searching through ECONLIT for the term ‘robust
regression’ led to a remarkably low total of 14 papers published in different journals most of
which apply robust regression techniques to real economic data sets. We believe that this reluctance
to use robust regression techniques may be due to the following factors.
The belief that large sample sizes make robust techniques unnecessary — with enough data,
one can find the truth.
The belief that outliers can be detected simply by eye, or by looking for unusual OLS
residuals, or by sensitivity analysis, obviating the need for a robust analysis.
Existence of several ‘robust regression’ techniques with little guidance available as to which
is appropriate.
Unfamiliarity with interpretation of results from a robust analysis.
Unawareness of gains available from robust analysis in real data sets.
In this paper we hope to remove some of these barriers and facilitate more routine use of a particular
set of robust regression procedures, based on Rousseeuw and Van Zomeren (1990) and modern
algorithms.