Biometrika Trust

The Effect of an Outlier on L-Estimators of Location in Symmetric Distributions

Author(s): H. A. David and J. K. Ghosh
Source: Biometrika, Vol. 72, No. 1 (Apr., 1985), pp. 216-218
Published by: Oxford University Press on behalf of Biometrika Trust
Stable URL: https://www.jstor.org/stable/2336355
Accessed: 24-01-2024 16:05 +00:00

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Biometrika Trust, Oxford University Press are collaborating with JSTOR to digitize,
preserve and extend access to Biometrika

This content downloaded from 158.182.10.114 on Wed, 24 Jan 2024 16:05:35 +00:00
All use subject to https://about.jstor.org/terms
 216 Miscellanea

Biometrika (1985), 72, 1, pp. 216-8

Printed in Great Britain

The effect of an outlier on L-estimators of location

in symmetric distributions

BY H. A. DAVID

Statistical Laboratory, Iowa State University, Ames, Iowa 50011, U.S.A.

AND J. K. GHOSH
Indian Statistical Institute, Calcutta, India

SUMMARY

The effect is studied of an outlier which has the same symmetric distribution as the
other observations except for a change in location and a possible increase in scale. We
show that the median is the most bias-resistant estimator, in the class of L-statistics with
symmetric nonnegative coefficients that add up to one, for a class of distributions which
includes the normal, double-exponential and logistic distributions.

Some key words: Bias; Convexity; Inequality; Order statistic; Robustness.

Let Z1, ..., Zn be independent variates with finite expectations. Suppose that one of
Z's, we do not know which, represents an outlier. The outlier has distribution function
G(x), the other variates have distribution function F(x) which is symmetric abou
have density f (x).
It is desired to estimate ,u, the mean of the target population, in the presence of the
unidentified outlier. Let the order statistics formed from the Z's be denoted by
Zl:n < ... < Zn:n As estimators we shall consider in this note the class {Ln}
functions of order statistics, so-called L-estimators,

n n

Ln= aiZin, ai = 1, ai =an+1-Oi > (for alli) (1)

i= i=l

We are concerned with the

of L-estimators, see Rosenberger & Gasko (1983).
We begin by considering the subclass {IMrn} of basic estimators
Mr,n= (Znr+i n?Zrn) (r = [4n]+1,...,n),
which includes the median. Clearly, Ln can be written as a linear function of the Mr,n,
with nonnegative coefficients. It will be shown that under certain conditions E(Mr,n) is an
increasing function of r. From this follows in particular a formal proof of the intuitively
appealing result that the median is the most bias-resistant estimator in the class {Ln} of
(1).
Let br,n = E(Zr+l:n-Zr:n). Then 2E(Mr+1,n-Mr,n) = 6r,n n-r,n- We have, compare
David & Groeneveld (1982), taking pi = 0 without loss of generality,

br n n-rn= () J {1 -G(x) -G( -x)}Fr- 1(x) {1 -F(X)}n- 1-r{F(x) -r/n} dx. (2)

This expression, without the factor 1-G(x)-G(-x), has been studied by David &
Groeneveld (1982). From the argument there given, Case 2, it follows that a sufficient

This content downloaded from 158.182.10.114 on Wed, 24 Jan 2024 16:05:35 +00:00
All use subject to https://about.jstor.org/terms
 Miscellanea 217

condition ensuring brn -

R(x) = { 1-G(x)-G(-x)}/f (x) (3)
be increasing in x for x > 0.
Of special interest is the situation G(x) = F{(x-A)/1} for A > 0, a > 0 when

R(x) = r(x, t) dt,

where

r(x,t) = {f( )+f( )} (x) (4)

Example 1. If f (x) = (27) -e-2X then

r(x, t) = 2 exp {2-t2/' ? 2x2(1_- l/2)} cosh (txV2).

Thus r(x, t), and hence R(x), is an increasing function of x for x > 0 and a > 1.

The same result is easily shown to hold for the double exponential f (x) = e- Ix 1. Note
that, in the special case A = oo, stronger results are possible (David, 1985).
For o = 1, the location-shift case, there is an interesting connection with hypothesis
testing. It is well known (Lehmann, 1959, p. 330) that f (x- 0) is a monotone likelihood
ratio family if and only if f = - log f is convex. Assume this is so and also that f has a
derivative, i'. We continue to take f to be symmetric, so that / is symmetric. It can be
shown that, under these assumptions,

r(x, t) = {f (x + t) +f (x-t)}/f (x),

being an increasing function of x for x ) 0 and t > 0, is equivalent to the concavity of

Y'(x) for x > 0. It is also easy to show that, for a > 1, the concavity of *'(x) continues to
imply the increasing character of

r(x, t) {f (x+t) +f x-t (x)

Example 2. For the logistic distribution, with density f(x) = e-x(I +e-x)-2, *(x) is
convex and *'(x) is concave for x ) 0, so that (3) is increasing in x for x > 0, with

G(x)=L1+e{ (x-)}1 ( x 1).

For the uniform distribution, (3) is not applicable but some direct arguments are
possible. The median is no more bias-robust than any other Ln in (1) not involving the
extremes.

ACKNOWLEDGEMENT

This work was supported by a grant from the U.S. Army Research Office.

REFERENCES
DAVID, H. A. (1985). Order statistics under non-standard conditions. In Biostatistics: Statistics in
Biomedical, Public Health and Environmental Sciences, Ed. P. K. Sen. Amsterdam: North-Holland.
To appear.
DAVID, H. A. & GROENEVELD, R. A. (1982). Measures of local variation in a distribution: Expected length of
spacings and variances of order statistics. Biometrika 69, 227-32.

This content downloaded from 158.182.10.114 on Wed, 24 Jan 2024 16:05:35 +00:00
All use subject to https://about.jstor.org/terms
 218 Miscellanea
LEHMANN, E. L. (1959). Testing Statistical Hypotheses. New York: Wiley.
ROSENBERGER, J. L. & GASKO, M. (1983). Comparing location estimators: Trimmed means, medians, and
trimean. In Understanding Robust and Exploratory Data Analysis, Ed. D. C. Hoaglin, F. Mosteller and J.
W. Tukey, pp. 297-338. New York: Wiley.

[Received August 1984. Revised October 1984]

This content downloaded from 158.182.10.114 on Wed, 24 Jan 2024 16:05:35 +00:00
All use subject to https://about.jstor.org/terms