The Shape of Code

I am a sporadic reader of In the Pipeline, a blog covering drug discovery and the pharma industry, subjects about which I have no real interest but the author is a no nonsense guy whose writing I enjoy reading. A topic that regularly crops up is retraction of a published paper (i.e., effectively saying “ignore that paper we published way back when”). Reasons for retraction include a serious mistake, plagiarism of somebody else’s work or outright fabrication of data.

Retraction of papers published in software engineering journals is rare, why is that? I don’t think software engineering researchers are more/less honest than researchers in other fields. I could not find any entries on Retraction Watch.

Plagiarism certainly occurs and every now and again a paper is retracted for this reason.

Corrections to previously published papers certainly occur on a regular basis, but I don’t recall seeing a retraction because of a serious error (but then I rarely get to gossip around the coffee table in university departments and am not that well up on such goings on).

Researchers are certainly not above using the subset of a benchmark that shines the most favorable light on their work, or simply performing misleading comparisons. Researchers who do such things are seem more as an embarrassment than a threat to academic integrity, they are certainly not in the same league as those who fabricate data

Fabrication of data in software engineering? I’m sure it goes on, but unless the people responsible own up I think it is unlikely to be detected (unless the claims are truely over the top). There is no culture of replication in software engineering or of building on other peoples’ work (everybody is into doing their own thing); two very serious problems, but not the topic of this discussion.

In fact software engineering is the ideal discipline for an academic fraudster: replication is very rare, everyone doing their own thing, a culture of poor/nonexistent record keeping and experimental data is rarely kept past the replacement of the machine on which it sits (I am regularly told this when I email authors asking for a copy of their raw data for my book). Even in disciplines whose characteristics are at the other end of the culture scale, it can take a long time for fraud to be uncovered.

From time to time authors I contact tell me that the numbers appearing in the published paper are incorrect; often there is an offer of the correct numbers and sometimes a vague recollection of what they might be. Sometimes authors don’t reply to my email, is the data fake or is talking to me not worth their time (I have received replies to this effect)?

Am I worried about fraud in software engineering research? No, incorrect data in published work is more likely to occur because of clerical mistakes, laziness or incompetence.