Lies, Damn Lies, and Sausage

Joe Mitchell has been keeping detailed statistics on SoCG acceptance rates for years. Here's a graph (also in PDF), distilled mostly from Joe's numbers, showing the number of papers submitted and accepted to SoCG every year since its inception.

A few points about this graph:

• I counted only full papers, not short communications (1995–1997) or video descriptions (1993–present). There is no record of how many short communications were submitted in 1996, so the record low submission count shown in the graph is actually an overestimate!

• From 1997 until 2001, the SoCG program included seperate "theory" and "applied" tracks. The submission and accetpance rates for the theory track are shown separately. The acceptance rates for theory and applied papers were always roughly equal.

• The huge spike in submissions in 1997—more than twice as many as 1996, and almost twice as many as in 1998—follows the introduction of the applied track. Most of the new 1997 submissions were applied-track papers from outside the usual SoCG community. After a dismal 21% acceptance rate, the the submission pool very quickly swung back to the traditional computational geometers submitting the usual number of papers.

• There was a spike in submissions in 1989 when the conference was first held in Europe, and again in 2000 when the conference was first located in Asia.

• The dip in 1996 mirrors a temporary increase in the number of geometry papers at STOC, which was also part of FCRC that year.

• The number of accepted papers has remained virtually constant for 21 years. The shortest programs were in 1986 (34 papers) and 2002 (35 papers). The largest number of papers was in 2004 (49 papers). The 2004 program included a day of parallel sessions; nobody seemed to object (or notice). With these three exceptions, there have been 41±3 SoCG papers every year. As a result, there is no significant correlation between the number of submissions in year n and the number (or fraction) of papers accepted in year n-1.

• Just in the last three years, there has been a significant increase in submissions that does not seem to be correlated with any external factors—location, cost, previous acceptance rates. The field simply appears to be growing significantly. Is this a reflection of the turn-of-the-century faculty hiring boom? Of the .com crash (more smart people going to grad school instead of industry)? Of CARGO and ITR funding? Your guess is as good as mine.
  

Comments

-- The number of accepted papers has remained virtually constant for 21 years.

So either the field has not grown in 21 years, or somehow we missed the opportunity to grow the conference in proportion with it. Either way, not good.

Observe that the number of acceptances did not go up when the applied track was created. In other words, we invited the neighbors to dinner but we did not place any extra settings at the table.

[JeffE]

I agree. We missed a good opportunity.

Perhaps the biggest reason that the number of acceptances did not go up in 1997 was the choice of location. The conference site in Nice ( http://www-sop.inria.fr/prisme/scg97/Carrefour.html ) had one large conference room), making it impossible to add parallel sessions. One one point during the planning stages, after seeing the huge number of submissions, the organizers toyed with the idea of adding an extra day to the conference. I don't recall why that didn't work out, but I suspect it was primarily logistics. Nice is a very popular place in June; reserving the conference site for an extra day, with only six months advanced warning, may have been simply impossible.

Nobody expected the number of submissions to DOUBLE.

To cope with the flood of submissions, more than 30 ten-page submissions were accepted only as short communications, each with two pages in the proceedings and no talk. Despite the absence of any real alternative, this was not a popular decision with anyone. Communications were not supposed to be a dumping ground for results that weren't good enough to be "real" papers. Short communications were voted out entirely that year.

The question of how to deal with the extra submissions was the hottest topic at the business meeting. Add more days? Add parallel sessions? Shorten some or all of the talks? (Raimund Seidel seriously suggested limiting each talk to five minutes. After all, people don't go to conferences for the talks, but for the social interaction.) In the end, the community voted by a bare majority to expand the conference to four days, to keep talks at 20 minutes, and not to add parallel sessions.

The 1998 and 1999 conferences WERE four days long, but for some reason the program did not expand significantly. In 2000, the program went back down to three days.

For more insight into the discussion, check out the SoCG business meeting minutes from 1996 ( http://compgeom.cs.uiuc.edu/~jeffe/compgeom/files/scg96-minutes.html ), 1997 ( http://compgeom.cs.uiuc.edu/~jeffe/compgeom/files/scg97-minutes.ps.gz ), and 1998 ( http://compgeom.cs.uiuc.edu/~jeffe/compgeom/files/socg98minutes.html ).