h-index

Chris Leonard points to a proposal by Bob Michaelson for a new bibliographic impact metric, called the h-index:

A scientist has index h if h of his/her N papers have at least h citations each, and the other (N-h) papers have fewer than h citations each.

Chris ponders:

In a relatively low-citing field like CS, I wonder whether this approach would also be considered for assessing one’s output?

I found the appellation "low-citing" rather startling (given how many of my papers have more than 50 references), but it's a good question. Fortunately, it's also a fairly easy question to answer, given the propensity of ((most) current) computer scientists to put their papers on the web.

Here are h-factors for a few prominent computational geometers, according to Google Scholar, with name collisions filtered by hand eye. These aren't the only people I checked, but I left anyone with an index below 20 off the list.

• Pankaj Agarwal: 32
• Jon Bentley: 33
• Bernard Chazelle: 34
• Ken Clarkson: 20
• Herbert Edelsbrunner: 37
• David Eppstein: 29
• Paul Erdős: 30 (sic)
• Leo "unsorted" Guibas: 43
• Mark Overmars: 23
• Micha Sharir: 43
• Jack Snoeyink: 21
• Emo Welzl: 23

(For the record, my GSh-factor is 13. Look at my pocket watch. Concentrate. You are getting verrrry sleepy. You want to cite me more. At the count of 3 you will wake up relaxed and energized. 1...2...)

So there aren't any geometers in the range of Donald Knuth's 58 and Bob Tarjan's 61, and even they're nowhere close to physicists Ed Witten's 110. So Chris was right!

It's goes without saying that Google Scholar is heavily skewed toward more recent citations. (Paul Erdős gets only a 30?) Arguably, that bias makes it a better indicator of current, direct impact, but it skews the index from its original intention. On the other hand, what else could I use? Science Citation Index does a horrible job indexing computer science papers. Only David knows his own h-index for sure.

Oh, I can just see those tenure and promotions committees foaming at the mouth. “A number! A single number! Now, finally, we can make rational, objective decisions!” Lord help us.

Update: Piotr Indyk (24) chides me for omitting Ken Clarkson (20) and Jon Bentley (33). Shame on me. Bad, bad, bad.

Comments

[D. Eppstein]

Re: "Only David knows his own h-index for sure."

Actually, I don't think I do. First, because I haven't done a thorough search through SCI or Google for quite a while, so I'm missing a lot of recent cites. Second, because I'm still not really sure what counts as a paper. (Are journal versions and conference versions separate or should they be lumped together?) And third, because these numbers are only useful in comparisons if everyone's is calculated the same way. So I'd rather stick with something repeatable like your by-eye scan of Google Scholar than try to correct my own numbers to make them more accurate.

[anonymous]

A problem with this measure is that it assumes the slope of the citation pyramid is the same for all authors. No offense, but I'd take Steve Cook's h=27 over any of the h=30-40s that you list (or any of those over my measly h=11). It seems that a more sensible measure is the corrected h-index:

h * [# citations of most cited paper]/1000.

Interestingly this gives a score, for papers only, of Knuth=34.974 compared to Cook=33.021 or Tarjan=55.754 which in terms of scholarly papers sounds about right (Knuth's enormous contribution in books & the TeX system have to be measured in a different way).

-- Paul Erdős gets only a 30?

Erdos has always been known for having many rather minor publications. I wouldn't be surprised if even a time-corrected count didn't push him much past 40.