Thursday, April 09, 2009

Arbitrators and risk avoidance

The Arbitration Committee exists to settle Wikipedia's toughest disputes.  But do arbitrators get elected on the basis of a successful dispute resolution record, or because they don't step on too many toes?  Last night YellowMonkey, a former arbitrator, argued the latter.  It was the sort of statement that looks like the end of a long period of holding one's tongue, by someone who knows a lot, and it's very intriguing.  

What's especially interesting is how YellowMonkey's contention correlates to a pair of data tables posted at about the same time by Paul August, another former arbitrator.  The tables give the percentage of proposed arbitration cases accepted for consideration, broken down by arbitrator, for 2008 and 2009.  Paul calls these 'Activity tables', but two facts jumped out at me that appeared to support YellowMonkey's contention.

The only arbitrator who has accepted more than half the proposed cases this year is Coren, who received a one year term after getting the lowest percentage support of any arbitration candidate that received an appointment.  If Jimbo hadn't decided to expand the Committee Coren wouldn't be on it.  And the only arbitrator who accepted less than twenty percent of cases is Newyorkbrad, who was elected on a landslide in 2007 with the highest percentage support of any candidate.  Coren has voted to accept 54% of proposed cases so far; Newyorkbrad has voted to accept 18%.  And Newyorkbrad has a three year term so he'll still be a sitting arbitrator next year after Coren's term has ended.  I pulled up the election data on the whole Committee and sorted it by percent cases accepted in 2009:

  1. Coren: 54% accepted, 64% election support in 2008
  2. Cool Hand Luke: 46% accepted, 74%election support in 2008
  3. FayssalF: 41% accepted, 76% election support in 2007
  4. FloNight: 41% accepted, 84% election support in 2006
  5. Rlevse: 38% accepted, 73% election support in 2008
  6. Wizardman: 36% accepted, 66% election support in 2008
  7. Roger Davies: 35% accepted, 80% election support in 2008
  8. Kirill Lokshin: 33% accepted, 98% election support in 2006
  9. Vassyana: 33% accepted, 68% election support in 2008
  10. Casliber: 31% accepted, 92% election support in 2008
  11. Jayvdb: 28% accepted, 68% election support in 2008
  12. Risker: 26% accepted, 87% election support in 2008
  13. Sam Blacketer: 26% accepted, 74% election support in 2007
  14. Stephen Bain: 26% accepted, 67% election support in 2007
  15. Carcharoth: 20% accepted, 67% election support in 2008
  16. Newyorkbrad: 18% accepted, 97% election support in 2007
  17. Average: 33% accepted, 77% election support 

Obviously this doesn't hold up in a completely linear fashion, and the relatively small number of cases proposed in the first quarter of 2009 might limit the value of this data.  It's worth noting that Newyorkbrad's acceptance ratio is nearly unchanged from last year when it was 20%.  

What would make an interesting followup would be to track voting behavior on proposed decisions.  Break down remedy types into two basic categores: hard remedies and soft remedies.  Hard remedies are bans, desysoppings, and other specific restrictions directed at individual editors.  Soft remedies would include warnings, cautions, encouragements and other nonspecific actions.  In particular, broad topic paroles and page paroles that empower arbitration enforcement to impose restrictions belong among the soft remedies because discretionary sanctions are a form of risk avoidance: if a specific action gets applied later on, then any retaliation would likely target the administrator who used discretionary powers, rather than the arbitrators.

One thing needs to be stated explicitly: none of this implies that arbitrators ought to be accepting every case proposal or implementing hard sanctions on everybody.  A hanging judge is as bad as an absent one.  Many disputes come to RFAR prematurely and most editors who get named as parties do not need hard remedies.  The task we elect arbitrators to do is the difficult work of deciding which cases and which editors do need it, and the worry I share with YellowMonkey is that our selection process preferentially promotes arbitrators who are least likely to do so when necessary.  

1 comment:

Gregory Kohs said...

The correlation coefficient for this data series is -0.278. While that's something, it's statistically weak.

Nowhere near the -0.65 found between peak systolic velocity and common carotid artery (distal), for example.