Information Technology Reference
In-Depth Information
is widely practiced, and is consistent with the expection in many countries that the
outcomes of publicly funded research be made freely available.
Misrepresentation
Misrepresentation occurs when a paper does not accurately reflect the outcomes that
were observed or the contributions of previous research. When presenting results,
researchers are expected to ensure that they are accurate, describe any experimental
issues or limitations that could have affected the outcome, provide enough detail to
enable reproduction or verification, be fair in description of other work, report nega-
tive as well as positive results, not state falsehoods, and take the effort to ensure that
statements are complete and accurate. However, an honest mistake is not misconduct.
In its clearest form, misrepresentation is fraud: the making of claims that are
outright false. Another form of misrepresentation is when authors imply that they
have high confidence in their results when in fact the experiments were preliminary
or were limited in some way. For example, reported running times may be based on
a small number of runs with high variance, or there may be uncertainties about the
quality of the implementation. Even more dubious are cases where the efficiency
of a method being tested is based on some parameters, and the reported times are
those achieved by tuning the parameters to the input data. Failure to report relevant
unsuccessful experiments is explicitly condemned in the academic codes of conduct.
That is, there is a grey area between work that is fraudulent and work that is
sloppy. Choosing a poor system as a baseline might just be lazy. Badly implement-
ing a baseline—and thus exaggerating the benefits of a new approach—without
verifying that the baseline works as well as was previously reported, begins to look
more like deception.
Other forms of misrepresentation concern interpretation of past work. Abehaviour
that is far too common—and, arguably, is fraud—is to understate other people's work.
It can be tempting for authors to exaggerate the significance and originality of their
results, and to diminish the status of previous results in the field, to increase the
likelihood of their work being published. If you would be uncomfortable defending
what you have written about other people's work, then your text should probably be
changed.
The issue of misrepresentation arises with online publication. When an author
discovers an error in an online paper it is all to easy to correct it silently, with no
explicit indication that the paper has changed. Modifications to online papers should
always be made explicit, by use of a version number and date of publication; and the
original version should continue to be available, as others may have referred to it.
Online archives provide versioning functionality; but simply placing an updated
version of a published paper on an archive does not adequately advertise that the
paper has been changed and what the changes are. Retrospective alteration of a
document is not something that should be done lightly.
