Yes, in general we prefer conferences to journals. It’s a little more complicated than “journals don’t matter.” In each subfield of CS (systems, AI, graphics, etc), there are several conferences and several journals. Generally, the best conferences and even the second-tier conferences, are more prestigious than even the top-tier journals. The best journals are better than low-tier conferences. And peer-reviewed glossy magazines, like Communications of the ACM or IEEE Computer, are high-visibility and well respected.
Bear in mind that the conference and journal models aren’t so different in their final output. In each area of CS that I’m familiar with, the papers that appear in “Proceedings of the ACM Symposium on Quintessence” or whatever the conference is called are written in the same style and have comparable length etc to the papers at the ACM Transactions on Quintessence [journal]. So the real difference is in the reviewing process, not in the writing style.
Neither the journal or conference model is great. The problem with journals is that they’re incredibly slow—for one of my articles, it took a year and a half from submission to publication. They’re also annoyingly bureaucratic—they have fussy typesetting and layout rules that generally don’t improve the quality of the final work.
The impression people have is that it’s rare for a paper to be rejected outright—mostly even a relatively weak paper gets “here are the major flaws, fix and resubmit.” And then it becomes siege warfare where the author fixes a few flaws and sends it back for re-review and at some point either the author gets fed up and quit, or the reviewers get fed up and say “okay fine, publish it.”
The problem with conferences is that they have to make an up-or-down decision on each submitted paper. The timeline means there’s no way for them to say “this paper has serious but correctable flaws and we will publish it if you fix it.” You can always resubmit to a different conference, but it sometimes happens that conference A says “too much X”, and and conference B says “not enough X”—there’s no continuity in reviewers. (In contrast, journal reviewers stay with the paper from submission through the “revise and resubmit” cycle.) The advantage of conferences is that the author gets an answer quickly and the paper, if published, gets attention quickly.
There’s been some motion towards hybrid publication systems. There’s a highly respected conference on Very Large Databases (VLDB). For the last few years, the model has been that you submit your paper to the VLDB Journal. The journal has a fixed reviewing cycle of a few months, fixing the “slow reviews” process. Authors of published papers get to give a talk at the VLDB conference, fixing the visibility problem. OOPSLA, a prominent programming language conference, has switched to two-round reviewing, which helps fix the “up or down and no continuity” problem.
As to how the problem got going—Some part is that this is path dependent and the problem feeds on itself. Once journals become low status, people stop reading them. In science, visibility is the coin of the realm, and a conference talk at the leading venue is therefore much better than a journal article nobody reads. And once journals are low-status, the leading people don’t want to review for them, and so you get lower-quality reviews. Conversely, once a conference becomes highly selective and prestigious, people send papers there because they want the stamp of approval of “the best people send their papers to ACM-Quintessence and only the top 20% get in.”
I don’t have a great sense of why CS has this model and no other field does. I have one guess: Most papers are rejected for being boring / incremental, not for being wrong or having weak evidence for an interesting claim. Mostly computers are easy to experiment on. The machines are pretty deterministic, it’s easy to share code and data, and most things we care about aren’t very expensive.
The conference model is pretty good for deciding if something is interesting. Whereas a journal has an editor and a couple reviewers, the conference model is to assemble a program committee of 15-30 people, and they discuss and vote on each paper. So that gets you a bigger sample of expert opinion on whether something is interesting and valuable.
My impression is that in physics, experiments are easier to get wrong and more important to replicate, and therefore the reviewers are there to check the details, not just give a high-level analysis of whether the paper solves an interesting and important problem with a solution with nontrivial elements that can be reused elsewhere.
I don’t love the CS publication model, but my sense is that nobody likes the publication process. Pretty much all the jokes and complaints about peer review in physics or biology feel relevant to me. So I think it’s just a sign of a healthy scientific field that having a highly-respected publication means having it scrutinized by skeptical experts who don’t pull their punches.
Yes, in general we prefer conferences to journals. It’s a little more complicated than “journals don’t matter.” In each subfield of CS (systems, AI, graphics, etc), there are several conferences and several journals. Generally, the best conferences and even the second-tier conferences, are more prestigious than even the top-tier journals. The best journals are better than low-tier conferences. And peer-reviewed glossy magazines, like Communications of the ACM or IEEE Computer, are high-visibility and well respected.
Bear in mind that the conference and journal models aren’t so different in their final output. In each area of CS that I’m familiar with, the papers that appear in “Proceedings of the ACM Symposium on Quintessence” or whatever the conference is called are written in the same style and have comparable length etc to the papers at the ACM Transactions on Quintessence [journal]. So the real difference is in the reviewing process, not in the writing style.
Neither the journal or conference model is great. The problem with journals is that they’re incredibly slow—for one of my articles, it took a year and a half from submission to publication. They’re also annoyingly bureaucratic—they have fussy typesetting and layout rules that generally don’t improve the quality of the final work.
The impression people have is that it’s rare for a paper to be rejected outright—mostly even a relatively weak paper gets “here are the major flaws, fix and resubmit.” And then it becomes siege warfare where the author fixes a few flaws and sends it back for re-review and at some point either the author gets fed up and quit, or the reviewers get fed up and say “okay fine, publish it.”
The problem with conferences is that they have to make an up-or-down decision on each submitted paper. The timeline means there’s no way for them to say “this paper has serious but correctable flaws and we will publish it if you fix it.” You can always resubmit to a different conference, but it sometimes happens that conference A says “too much X”, and and conference B says “not enough X”—there’s no continuity in reviewers. (In contrast, journal reviewers stay with the paper from submission through the “revise and resubmit” cycle.) The advantage of conferences is that the author gets an answer quickly and the paper, if published, gets attention quickly.
There’s been some motion towards hybrid publication systems. There’s a highly respected conference on Very Large Databases (VLDB). For the last few years, the model has been that you submit your paper to the VLDB Journal. The journal has a fixed reviewing cycle of a few months, fixing the “slow reviews” process. Authors of published papers get to give a talk at the VLDB conference, fixing the visibility problem. OOPSLA, a prominent programming language conference, has switched to two-round reviewing, which helps fix the “up or down and no continuity” problem.
As to how the problem got going—Some part is that this is path dependent and the problem feeds on itself. Once journals become low status, people stop reading them. In science, visibility is the coin of the realm, and a conference talk at the leading venue is therefore much better than a journal article nobody reads. And once journals are low-status, the leading people don’t want to review for them, and so you get lower-quality reviews. Conversely, once a conference becomes highly selective and prestigious, people send papers there because they want the stamp of approval of “the best people send their papers to ACM-Quintessence and only the top 20% get in.”
I don’t have a great sense of why CS has this model and no other field does. I have one guess: Most papers are rejected for being boring / incremental, not for being wrong or having weak evidence for an interesting claim. Mostly computers are easy to experiment on. The machines are pretty deterministic, it’s easy to share code and data, and most things we care about aren’t very expensive.
The conference model is pretty good for deciding if something is interesting. Whereas a journal has an editor and a couple reviewers, the conference model is to assemble a program committee of 15-30 people, and they discuss and vote on each paper. So that gets you a bigger sample of expert opinion on whether something is interesting and valuable.
My impression is that in physics, experiments are easier to get wrong and more important to replicate, and therefore the reviewers are there to check the details, not just give a high-level analysis of whether the paper solves an interesting and important problem with a solution with nontrivial elements that can be reused elsewhere.
I don’t love the CS publication model, but my sense is that nobody likes the publication process. Pretty much all the jokes and complaints about peer review in physics or biology feel relevant to me. So I think it’s just a sign of a healthy scientific field that having a highly-respected publication means having it scrutinized by skeptical experts who don’t pull their punches.