July 4, 2013, 11:23 a.m. ET
The Dismal Science's Crusading Voice
Economist Bruce McCullough Continues a Largely Solitary Push to
Require Replication in Research
By  BRENDA CRONIN

Bruce McCullough, an economist with a specialty in statistical
software, wants to inject some more rigor into the dismal science.

In the physical sciences, it is common to replicate most findings,
checking important results by having others attempt to repeat the
exercise working from the same data. In some cases, it's a matter of
life and death. Replication is seldom done in economics, however, as
the incentives often aren't as clear.

Mr. McCullough, a professor of statistics at Drexel University's LeBow
College of Business, is a longtime advocate for changing that. He says
replication should be done before publication and not after.

"No one should take publication in a peer-reviewed journal as the
final word on any topic," he says. "Rather, it's the first word." Mr.
McCullough, who recently demonstrated a replication in his office on
Drexel's Philadelphia campus, has been an enthusiast since he began
teasing out bugs in econometric software decades ago.

The push for replication got a high-profile boost recently, when a
graduate student at the University of Massachusetts Amherst turned up
a spreadsheet error in an influential 2010 paper on government debt.
The authors, who had made their data available online, acknowledged
the mistake but said it didn't change their conclusion.

[snip]

Mr. McCullough's finding prompted the AER editor in 2004, Ben
Bernanke, to step up enforcement of the journal's replication policy.
The change, spelled out in an editorial statement by Mr. Bernanke, now
Federal Reserve chairman, required authors to provide electronic
versions of their data and code—the computer calculations to run on
the data—so their work could be replicated.

But with the AER policy on replication being the exception among most
economic journals, the 53-year-old Mr. McCullough presses on.

"To rely on the results of a paper when the results cannot even in
principle be verified assumes that the authors are infallible," Mr.
McCullough says. "And we're all human. So, simply because we're all
human, any article should have its data and code made available…so
other people can check that no mistakes were made."


[Full]
http://online.wsj.com/article/SB10001424127887324436104578581620428594356.html
_______________________________________________
pen-l mailing list
[email protected]
https://lists.csuchico.edu/mailman/listinfo/pen-l

Reply via email to