Frank Windmeijer talks with
ScienceWatch.com and answers a few questions about
this month's Emerging Research Front Paper in the field of
Economics & Business.
Article: A finite sample correction for the variance of
linear efficient two-step GMM estimators
Authors: Windmeijer, F
Journal: J ECONOMETRICS, 126 (1): 25-51, MAY 2005
Addresses: Inst Fiscal Studies, Ctr Microdata Methods &
Practice, 7 Ridgmt St, London WC1E 7AE, England.
Inst Fiscal Studies, Ctr Microdata Methods & Practice,
London WC1E 7AE, England.
Why do you think your paper is highly
The estimation method for which my standard error correction works
particularly well is very popular, especially when using panel, or
longitudinal, data. As the usual standard errors are often much too small
when the number of observations are moderate, the correction has been
adopted quite quickly into mainstream statistical packages and is now being
used in applied research, leading to much more reliable inference.
Does it describe a new discovery, methodology, or
synthesis of knowledge?
"It was while working with them on a review book
chapter on these methods that I discovered where the
problem of the downward bias in the variance estimate came
It describes the development of the methodology used to account for
estimation of nuisance parameters that do not affect the estimate of the
variance in large samples, but have an impact when the sample size is small
or moderately large.
The problem was quite well-known, but the paper develops the method to take
account of the extra variation due to the estimation of these nuisance
parameters. The result is that the corrected standard errors are in general
larger and much more reliable in finite samples, but also still correct
when the sample size is large.
Would you summarize the significance of your paper
in layman's terms?
As for any statistical analysis, obtaining correct confidence intervals is
important, for example, to assess the direction, if any, of a causal
Standard methods often rely on large sample results to inform the
reliability of results in observed data. Often these approximations are
reliable, but not in the case of the particular estimation method
considered here, leading often to confidence intervals that are too small,
i.e., leading to a false sense of precision.
The standard error correction described in the paper corrects for this
bias, inducing much more reliable confidence intervals on which to base,
for example, policy decisions.
How did you become involved in this research and
were any particular problems encountered along the way?
During my time at University College, London, and the Institute for Fiscal
Studies, I had the privilege of working with Professors
Richard Blundell, and
Stephen Bond, who were pioneers in the development
of this estimation method for dynamic panel data models.
It was while working with them on a review book chapter on these methods
that I discovered where the problem of the downward bias in the variance
estimate came from. It then took a bit more research to actually be able to
correct for this bias in a systematic way, although once I isolated the
cause, this was not too onerous.
Where do you see your research leading in the
In recent work with Whitney Newey, a Professor of Economics at the
Massachusetts Institute of Technology, we have developed a similar
correction for a more complicated estimation method for the same type of
As these estimators are less widely used, I now have a Ph.D. student
looking at providing software code for wider dissemination, and to assess
the properties inherent in a wide area of applications.
Do you foresee any social or political
implications for your research?
Not directly, but indirectly, as the applied research in this field will be
based on better inference.
Frank Windmeijer, Ph.D.
Professor of Econometrics
Department of Economics
School of Economics
Finance and Management
University of Bristol
Centre for Microdata Methods and Practice
Institute for Fiscal Studies
KEYWORDS: GENERALIZED METHOD OF MOMENTS; VARIANCE CORRECTION; PANEL DATA;
PANEL-DATA MODELS; OF-MOMENTS ESTIMATORS; GENERALIZED-METHOD; INFERENCE;
INSTRUMENTS; RESTRICTIONS; TESTS; WEAK.