[R] the joy of spreadsheets (off-topic)

[Gabor Grothendieck]

On Tue, Apr 16, 2013 at 1:25 PM, Sarah Goslee <sarah.goslee at gmail.com> wrote:
> Given that we occasionally run into problems with comparing Excel
> results to R results, and other spreadsheet-induced errors, I thought
> this might be of interest.
>
> http://www.nextnewdeal.net/rortybomb/researchers-finally-replicated-reinhart-rogoff-and-there-are-serious-problems
>
> The punchline:
>
> "If this error turns out to be an actual mistake Reinhart-Rogoff made,
> well, all I can hope is that future historians note that one of the
> core empirical points providing the intellectual foundation for the
> global move to austerity in the early 2010s was based on someone
> accidentally not updating a row formula in Excel."
>
> Ouch.
>
> (Note: I know nothing about the site, the author of the article, or
> the study in question. I was pointed to it by someone else. But if
> true: highly problematic.)
>

Herndon, Ash and Pollin (HAP), the authors of the critique, found that
in the highest debt category the Excel error in Rienhart and Rogoff
(RR)  was -0.3 percent points compared to a total error (from that
plus RR's other 2 mistakes) of -2.3 percentage points.  See Figure 1
of HAP. Thus aside from the dubiousness of attributing the coding
error in Excel to Excel itself it was not the main source of the
discrepancy.

Also even if one backs out all three errors that they found, the key
conclusion that GDP growth is declining with debt still occurs (but to
a lesser extent) as pointed out by RR in an initial responding email
reported by Bloomberg News.

The key takeaway here is really unrelated to Excel but rather is that
until data and analyses are shared or made public so that the analysis
can be reproduced one cannot have any real confidence in research
results.

RR
http://www.nber.org/papers/w15639.pdf

HAP
http://www.peri.umass.edu/fileadmin/pdf/working_papers/working_papers_301-350/WP322.pdf

Bloomberg News
http://www.bloomberg.com/news/2013-04-16/reinhart-rogoff-paper-cited-by-ryan-faulted-for-serious-errors-.html

--
Statistics & Software Consulting
GKX Group, GKX Associates Inc.
tel: 1-877-GKX-GROUP
email: ggrothendieck at gmail.com