Wednesday, December 01, 2004

PC discussions

Sariel riffs on Lance's discussion of program committees, and Oded Goldreich's somewhat controversial idea of having PC members contact authors for more information.

I have to pick a bone with a few things he said though (emphasis mine):
IMHO, the PC member should be show caution in contacting the authors, and use the input from the authors with care, but rejecting a paper just because it contains some minor technical error which prevents the PC from fairly judging its quality, seems to me to be an excuse to reject papers without reviewing them carefully. As OG mentioned, this additional input (a correction, etc) probably would have a negative impact on the evaluation of the paper anyway. In any case, I think the PC member should be trusted in using his/her/it common sense and clearcut rules should be avoided.
I am not sure I agree with this. People even in the relatively homogenous theory community have widely divergent views on matters (for example, OG's comments on PCs), and to rely on as ephemeral a notion as "common sense" is asking for confusion and ambiguity when some clearcut rules could make things so much simpler.

The tacit rule we have now: "Do not contact authors" is quite clearcut. It might prevent some paper from getting a fair shake, but it is fair (or at least equally unfair). Changing this to allow for PC contact with authors only introduces even more variability into the review process; maybe I feel conscientious and contact authors, whereas others do not. Maybe I believe honestly that a paper written by a friend is really good and needs to be interpreted "correctly", and contact the friend for a clarification I can use. There are myriad ways in which, applying what I believe to be good sense and fairness, I can reduce the overall fairness of the review process as seen by authors in general.

Ken Clarkson suggested over at Sariel's place that any contact with the authors should be made via the committee as a whole (or via the chair). This is more reasonable, and in fact is how most contact happens right now (when authors withdraw papers, or make bug reports), but to weaken the current proscription on author contact would open up a can of worms that no amount of trusting in good sense can prevent.

The other comment I take exception to is:
In particular, most papers (even if incorrect and junk) are harmless even if accepted to a very good conference.
We are trained in theory to think about resources (time/space/randomness/what have you). I claim that a slot in a good conference is also a resource, and with limits on this resource come hard decisions as to how to use it effectively. This is a zero sum game: an incorrect/junk paper that gets into a conference harms the correct/non-junk papers that didn't. Sariel goes on to to describe a more egregious kind of accepted paper, which I agree causes more grief in the long term, but in the long term we are all dead anyway ! Not making proper local choices (i.e what gets accepted to this conference) is harmful, and the harmful effects of this add up over time.

No comments:

Post a Comment

Disqus for The Geomblog