Monday, February 13, 2012

On the advantages of open-source econometrics

I find it surprisingly difficult to encourage my colleagues or my employer to go for open-source software. It appears to be very difficult to overcome the idea that if it is free, it cannot be good. But open-source software is not only good and free, by its very name, it is possible to look at its entrails. This implies that if there is an error, anybody can find it and fix it. With proprietary software, errors are much more difficult to detect, and then one is at the mercy of the publisher to do something about and disseminate a patch. You may think this is not a problem for you, well let's see whether this example hits closer to home.

Talha and Yasemin Yalta have looked at a series of econometrics software bundles and in particular tested for their accuracy. It turns out that fixes for gretl, the open-source one, were rapidly applied. The commercial products, though, took up to five years to correct them. And I know of at least one case where the software publisher refused to correct an error, because "published results would then not replicate." And imagine all the errors we do not see because the code is secret...

PS: another interesting point the authors raise is that commercial publishers do not make old versions of their software available for replication purposes. That can be a serious problem, for example if someone were accused of falsifying results, or just being sloppy.

2 comments:

Anonymous said...

The issue about replicability with older versions is a very serious issue. Is some software library with such versions available?

Economic Logician said...

Very good point. Is such a library available? Is there even a facility that would allow to read old storage formats for digital data?