We need different types of citation: Replicates, Falsifies, DependsOn, Acknowledges …
March 25, 2014
How should scientists, and reporters, discuss work that has failed to replicate? The original Barr and colleagues article remains in the scientific literature; failed replication alone is not grounds for retraction.
He’s right, of course: we certainly don’t want to retract every paper whose conclusions can’t be replicated, for all sorts of reasons: they may subsequently be replicated after all; the paper may contain other useful information even if the experiment in question was flawed; the replication studies themselves probably rely on the original’s Methods section; authors should not be punished for unfortunate outcomes unless they were fraudulently obtained.
What we want is for that Barr et al paper, whenever anyone looks at it, to be displayed with a prominent header that says “The following studies attempted to replicate this finding but failed:”, and a list of references/links. And, for that matter, another header saying that the following other studies did replicate it.
For web-sites to automatically produce that kind of annotation, they need articles that cite the original to include an additional piece of metadata, along with the author/year/title/journal/etc. metadata that identifies the cited paper. That additional ingredient is the citation’s type, which should be one of a small set of defined values.
What values are relevant? I won’t try to come up with an exhaustive list at this point, but obvious ones include:
- Replicates — the current paper replicates work done in the cited paper (and so provides evidence, though not proof, that the cited paper’s conclusion is correct).
- FailsToReplicate – the current paper attempts to replicate work done in the cited paper, but fails (and so provides evidence that the cited paper is mistaken).
- Falsifies — the current paper shows definitely that the cited paper is wrong. This is a stronger statement than FailsToReplicate, and would be used for example when the new work shows conclusively that the experimental protocol of the original was critically flawed.
- DependsOn — the current paper depends on information from the cited paper, such as the phylogeny that it proposes or the vertebral formula that it gives. For these purposes, the cited paper is treated as an authoritative source.
- Acknowledges — the current paper uses ideas proposed in the cited paper, and gives credit to the original.
There are all sorts of practical issues that will impede the adoption of this idea (not least the idiot fact that the citation graph is a trade secret rather than a freely available database), but let’s ignore those for now, and figure out what taxonomy of citation-types we want.