Friday, April 30, 2010

Random seeding gripe

In the ACBL seeding system, points for winning or placing in a major event decay arithmetically. So, you get 11-n points for winning the Spingold n years ago. Suppose "Bob" won in 2000 and 2001, and "Zia" won in 2003. As of 2004, Bob gets more points than Zia, 15 to 10, which seems fair. But when we get to 2010, suddenly Zia's win is worth more than Bob's two, 4 to 3. To put it differently, it looks crazy that a win 9 years ago is twice as good as a win 10 years ago, while wins initially decay only mildly, 10% a year.

I suggest an exponential decay factor, maybe .9. This would avoid these odd reversals. Given my (lack of) record in major events, my gripe has no self-interest component.

4 comments:

Kenny Z said...

You're probably right, but I think you overstate your case when you say the current system calls a win 9 years ago "twice as good" as a win 10 years ago. In that case, arithmetic calculation would seem more reasonable to me. Yes, 2 seeding points are twice as many as 1, but it is the raw number, not the factor that matters in this case. A win 9 years ago is worth just one seeding point more than a win 10 years ago. A win 2 years ago is worth only 1.5 times more than a win 4 years ago, but in practical terms that difference is more substantial than the difference between 9 and 10 years ago. The raw number "3" matters more than the factor 1.5.

Jonathan Weinstein said...

Right, I guess to put it differently than "twice as good" I would just say that winning both Vanderbilt&Spingold 10 years ago is now equal to winning just one 9 years ago. If we all think this seems wrong, some change should be made, and the simplest one is to just use a multiplicative decay factor.

Anonymous said...

Good point Jon. I am a big fan of exponential decay too.

Re: self interest. My best guess is that your future record will make this discussion relevant for you.


Regards,
Alex

Jeff said...

If they start fixing things, they could also include the MP formula -- arithmetic to 10,000 and then nothing. The USBF logarithmic method is much better.

I have had little luck in getting the ACBL to look at technical issues.