Tuesday, April 15, 2008

Deadlines (tax, conference, ...)

For the first time, I had to rush on the last day to get taxes out in time. It was a rather intense few hours, and I couldn't help but think that most of this would have been impossible in the pre-electronic era: I was able to download missing documents, get forms online, and even download tax software, all in a matter of a few clicks.

I'm also somewhat ashamed to admit that I rather enjoyed the adrenaline rush of getting to the post office with 10 minutes to spare and sending everything off. It reminded of conference submission deadlines (and I imagine that in the days before electronic submission, people literally had to rush to the post office to send copies of a submission in).

But then I got to wondering. In this hyper-competitive era, with submissions increasing, and acceptance rates, is it slowly becoming less likely that papers submitted at the last minute can compete with more polished submissions ?

Now I like data as much as anyone else, and was wondering what kinds of statistics we could glean from information already available. For example, every year PC chairs trot out statistics about acceptance rate as a function of when papers were submitted. Because of the almost hyper-exponential rate of submissions as the deadline approaches, these numbers are necesarily unreliable, but to the extent that one believes in such statistics, there's always a kind of maximum point somewhat away from the actual deadline.

I don't know if this tells me what I need to know though: does a drift of the maximum point away from the deadline prove my point ? Lacking any actual information about which papers are more "polished" i.e maybe re-submissions, or papers prepared earlier, one has to work off the assumption that papers that are ready earlier are submitted earlier, and I'm not willing to buy that argument without more data.

So maybe I should rely on anecdotal evidence. How about it, readers ? In your personal experience (I emphasize this point: I want to distinguish your opinions from what you've encountered in your own work), do you think that it's become less likely that a last-minute paper gets into a conference, especially the top ones ? Feel free to generalize this question to other deadline-based submission fora (even proposals, if you wish).

p.s A blog post without links ! Scandalous...


  1. You're taking this Tax deadline way too seriously. If you file in the next couple of days (16th or 17th), nothing bad will happen. I've done it many times, and they've never rejected my tax returns or said they need to charge me interest for the two days delay.

  2. Anecdotal observation:
    I have been on PCs of conferences which allocate paper numbers in some increasing order. My experience has been that those with the lowest paper numbers and the highest paper numbers are easy rejects. I don't have any data mapping this to times of submission, but I'm sure some PC chairs could oblige (indeed, I've occasionally seen some ad hoc analysis of probability of acceptance conditioned on time of arrival presented at business meetings).


  3. Paper submission numbers are biased by time zone among other things. Generally, I will do the first submission of a paper when there is a convenient break in the last day rush (say while a co-author is reading a draft) since it is a bit time-consuming to go through the submission server but I am typically comfortable doing this any time up to an hour or two before the deadline.

    I have records as PC chair from Complexity 08. Maybe because the deadline was 6pm Pacific time there were not that many new submissions in the last two hours (13/87), but here are the stats.

    Our acceptance rate was about 37%.
    In reverse order of submission from the end it is RRRBRAABARARA where B involves a major bug later noticed by the authors.

    Now the state of polish of a paper might be more related to how shortly before the deadline the last revision of a paper was sent. Here is that two hour reverse order sequence ARARRARAABARRRRAABARR.
    Of course many of these represent multiple revisions. If one did it by individual event then there would be even more A's.


Disqus for The Geomblog