Great schedule vs. great record
I've been stirring the pot on RSD regarding seedings for Nationals. It seems there's a bit of a fuss on what matters more: a good record versus mediocre competition or a mediocre record versus good competition. I come down squarely on the side of good competition. I want to reward teams for taking on harder challenges and I think those that have been exposed to Nationals-level competition will do better at Nationals.
Which is more important? Should seedings be a reward for the season so far or a decent guess at the final finishes of the teams?
Which is more important? Should seedings be a reward for the season so far or a decent guess at the final finishes of the teams?
posted by d at 1:30 PM
10 Comments:
Neva,
Most of my discussions with UPA coordinators have been under the framework that seeding is not predictive of final results. It is a ranking of perceived strength, based on past data.
I think most controversy involves lack of data, and, of course, how folks weight the data.
My feeling is that head to head, when you have it, is an important piece of data. My opinion, in terms of the referenced discussion, after taking a brief look at it, is that the head to head wins of Delaware over the 2 NE really push down the seedings for the NE teams.
-G
I guess I'm in the minority here, but to me, a head to head game is more or less just another game. In the case of a team like Delaware, it's significant, because the two most impressive wins on their resume are probably those two head-to-head wins against the New England teams. But it strikes me as odd that it's so much more useful (in a seeding argument) for Delaware to have a regular season win over Dartmouth than, say, Texas.
Tarr -
It doesn't strike me as that odd. Imagine Delaware had beaten Stanford at some tournament. Would we then seed them ahead of Stanford? Of course not.
Think about it by analogy to college football. Imagine that USC lost last year to UCLA. Would that be a sign that UCLA should have been in the national championship game? No.
Discounting outlying results is perfectly reasonable, and thus non-anomalous results are more valuable than anomalous ones.
Simply playing a tough schedule shouldn't be worth much. You need to hold your own against those teams. Suppose A is 24-10 and played a tough schedule and B is 40-2 against a weak schedule. A might be 23-0 against swill teams and 1-10 against good teams, while B is 39-0 and 1-2. Or A could be 16-0 and 8-10, and B 40-0 and 0-2 (losing 15-2 and 15-6). You need to look more closely.
I'd like to see the records for each team broken down by opponent RRI (or UPA Top 25), and maybe an average +/- for close calls, say against 1-10, 11-25, 26-50, and >50. Or you could simply remove every victory against a team ranked below #50 or #35 or whatever the highest ranked team is that had an expected score of closer than 15-10.0 against, say, the #16 team.
But thinking about it some more, I'm not sure that in general this would tell you any more about a team than simply looking at their RRI. But I suppose it could help in a tough case. And it would make the teams' records more directly comparable.
Sam,
No, of course, you don't seed them ABOVE Stanford. But in my opinion, a win over Stanford should help them a LOT more in seeding arguments than a win over Dartmouth. It should motivate moving them up the rankings moreso than the Dartmouth win, because it would be more impressive than the Dartmouth win.
That's my point - that you look at overall season and figure out how well a team has played. If head to head results produce an unambiguous ordering, then they would lead you to the same conclusion that looking at all the games would, anyway.
Jim - It's not a perfect system by any means, but the RRI calculations do attempt to do something like that - it gives less credit for getting pummeled by great teams than the UPA top 25 rankings do. Playing top teams close is worth a lot, though, as it probably should be. That's another one of those old questions: how much difference is there between a 14-13 win and a 13-14 loss.
One last comment, re: Neva's original question, "Should seedings be a reward for the season so far or a decent guess at the final finishes of the teams?"
To a degree, it has to be a little of both. On the club level where teams are more consistent from year to year, I think you can err more toward the latter. The code for this is "considering the previous years' results". On the college level, where predictions are a lot harder, I think you need to focus on the current season a lot more. That said, I don't see a great record against a weak schedule as inherently more deserving of reward than a good record against a hard schedule.
I guess I was trying to say that there is a way to use a team's record by controlling for the schedule. Compare A and B based on how they did against Top 10 teams, against the second tier, and against the third tier. So maybe Delaware's 25-4 gets broken down into 0-0, 2-3, and 23-1, while UCSB's 19-16 becomes 1-13, 4-3, and 14-0. Each of those sub-records becomes directly comparable. UCSB's tough schedule explains their record but also doesn't really give them much credit, since most of those losses weren't that close and they are probably bumped up by 3 close games (1 W, 2 L) against #4 UCLA.
I like Parinella's break-down of a team's record into tiers. Here's how I would break down the records of the four teams in question:
Dartmouth (NE 1) (28-10 overall)
0-7 against the top tier; 7-10 against the middle tier; 21-0 vs. bottom tier
Tufts (NE 2) (36-11 overall)
2-6 against the top tier; 6-4 against the middle tier; 28-1 vs. bottom tier
Swartmore (ME 1) (37-0 overall)
0-0 against the top tier; 7-0 against the middle tier; 30-0 vs. bottom tier
Delaware (ME 2) (27-4 overall)
0-0 against the top tier; 5-3 against the middle tier; 22-1 vs. bottom tier
As you can see, Tufts and Dartmouth both have just as many wins against middle or top tier teams as Swartmore does, but have played more games against that level of competition. Tufts two wins against the top tier should be especially weighted and should bump the NE teams up over the ME ones, in my opinion.
For reference, here's how I broke down their competition...
Top tier teams include:
Colorado
UCLA
UC-Davis
UC-Berkeley
Florida
Georgia
British Columbia
Wisconsin
UNC
Middle tier teams include:
Carleton
Michigan
Washington
UC-Santa Barbara
Cornell
State College
Emory
Tufts
Delaware
Swartmore
Dartmouth
Northwestern
Purdue
MIT
Bottom tier:
Everyone else
Did this before Gwen did it, posting anyway since I took the trouble...
So using Jim's method on Tufts and Swarthmore, not doing anything about close games, and using the UPA Top 25:
Tufts is 35-12.
Top tier (1-10): 1-6
Second tier (11-25): 2-2
Third tier (26-50): 8-4
Bottom tier (50+): 24-0
Swat is 37-0.
Top tier (1-10): 0-0
Second tier (11-25): 0-0
Third tier (26-50): 9-0
Bottom tier (50+): 28-0
Not really sure that this makes seeding any clearer.
Despite all our comments, the ME teams got the nod in the final seedings. My bet is that they end up as the 15th and 16th place teams when its all said and done though.
Who will save the Metro East?
Post a Comment
<< Home