Citing a paper with multiple versions

Question

In my area (computer science) it's common for a paper to be posted on arxiv, then appear in a conference a few months later, and finally appear in a journal several years later. This raises the question: which version should I cite?

The general practice I've seen is to cite the latest version. However, I do not agree with this general practice for several reasons:

• The chronology becomes unclear. e.g. I might say "Bobson [Bob14] builds on the seminal work of Adamson [Ada15]" It looks like the chronology is backwards, but the correct chronology is that [Ada15] appeared on arxiv in 2013 and then appeared in a journal in 2015.
• The most recent version is often not the best version to read. For example, the conference version of a paper is usually heavily abridged (due to a page limit) and the journal version is hidden behind a paywall and strangely formatted. Thus the arxiv version is usually the best version to read.
• If I am citing a specific part of the article (e.g. "Theorem 4.2 on page 23") that only appears in the arxiv version or has a different number in the journal version, then it doesn't make sense to cite the "wrong" version.
• It means I have to keep updating my citations as the paper moves through different versions. And this leads to inconsistent citations.

The only advantage I see to citing the most recent version is that it indicates how "good" the paper is, as indicated by the prestige of the conference or journal.

I am often tempted to cite the arxiv version in these cases, as it is gives the best chronology and is usually the most readable version. (I can justify it by saying that this is the version I actually read.)

Is going against this general practice acceptable? Might I annoy someone by not acknowledging that their paper appeared in a very prestigious venue? Or would someone be unhappy that I don't have the venue information in my references?

How should I cite articles that have multiple versions? In particular, how do I communicate (i) the correct chronology, (ii) the version with the right information, and (iii) the conference/journal that accepted it?

One solution is to cite all the versions, but I think that is impractical and confusing - and I have never seen it done.

Answer 1

In math we have the same issues. Here is what I do, most of which is standard:

1. If a paper has been published, always cite the published version.
2. If you want to point out the chronology, in particular because it is not clear from publication dates, do it in the text. People understand that publication dates do not align perfectly with when the papers were originally written. (My papers don't appear in the order I wrote them.) If you refer to arXiv versions, it is easy to see the chronology, at least up to the level of month/year, from arXiv ids.
3. If something is accepted to a certain journal/conference proceedings, you can say "Journal XXX, to appear" in the references. If you want to say something was announced at a certain conference, you can say so in the text.
4. Make references (e.g., theorem numbers) to the version you cite. I personally like to include version numbers in arXiv citations for this reason (e.g., arXiv:1601.0000v37). If you want to refer to something that appeared only in a previous version, I would cite the published paper if published and say in the text something like: This result appeared as Thm A.B.C of an earlier version (arXiv:1601.0000v36) of [citation]" possibly with some remark about why it's not in the current version. (Hopefully, if you're using it, it hasn't been removed because there was an error in the proof.) But if the paper has not been published, you can either do a similar thing or just put a citation to the version you want to refer to. However, it's probably best to point out in such situations that the result doesn't appear in later versions so the reader doesn't accidentally check the wrong version.

Answer 2

I would cite only one, and that would be the

(a)latest (b) peer reviewed item (c) from its first place of publication.

Answer 3

The hierarchy is usually 1) the journal paper (if exists, usually the most recent), 2) the conference paper (if exists), 3) non-peer-reviewed sources, including arXiv.

Never cite arXiv if there is journal version out there, since in many fields citations are valued, and citations are formally collected only for "reputable" sources. SCI index for example, counts only journal citations of journal publications. So, citing arXiv instead of journal is really bad.

Another, equally important reason is, that arXiv papers are not subject to peer review and basically anyone can "publish" anything. Consider this: in arXiv paper you find some interested claim you plan to rely on. You don't find this claim in the journal paper. Perhaps you don't find it, because reviewers made the authors remove it, since it was not substantiated by results, and citing the arXiv paper because it says something that the journal paper does not, would be extremely foolish. Conference papers from reputable conferences are usually peer reviewed too, but authors cannot fix as many things as in journal paper, so they come as the second choice.

So, there is reason that peer review exists, and you really should rely on reviewed versions of the paper.

Answer 4

I have the same problem and cannot offer you a solution for which I claim it to be the correct one. However here are some thoughts:

• For work citing other work like in your example "Bobson [Bob14] builds on the seminal work of Adamson [Ada15]" I would use the version [Bob14] references since this is the one [Bob14] builds on.

• I usually prefer a peer-reviewed publication to arXiv.

• Typically I stick to the first peer-reviewed publication I used and only "update" if a later version contains a result that I need.