Sunday, March 2, 2014

Prediction Recap (2/24)

#1 Syracuse @ Maryland:  Syracuse by 1
#1 Syracuse @ #14 Virginia:  Virginia by 2

Syracuse is barely favored at mediocre Maryland, which could lead to the spectacle of having the #1 team lose four straight games. 

Syracuse pulls out a 2 point win at Maryland, so we don't get the spectacle of four straight losses, but they do lose to Virginia by 19 (!) points.
 
Virginia Tech @ #5 Duke:  Duke by 31

VT loses by "only" 18.
 
#20 Michigan @ Purdue:  Michigan by 3

A good chance for Purdue to steal a meaningless but satisfying upset.

Purdue loses by 1 point in OT -- but as Feinstein would say, that's what bad teams do.

#7 Cincinnati @ #21 Connecticut:  UConn by 6

UConn by ... 6.
 
#8 Kansas @ Oklahoma State: OKSt by 5

The PM may be the last believer in OK State (although they had a good win this week against Texas Tech).  If the PM is to be believed, they have a good chance to beat Kansas next Saturday.

Yes indeed, you mockers!  OKState wins by 7.
 
#11 Louisville @ #22 Memphis: Louisville by 3

A good illustration of how powerful the HCA is in college basketball. 

Memphis wins by 6 in a game that Louisville led by 8 points or so with just a few minutes to go.  Louisville is now 1-4 against ranked teams this season, which can't be a good sign.
 
#15 Iowa @ Minnesota: Iowa by 4
#11 Creighton @ Xavier:  Creighton by 3.5
#17 Kansas State @ Iowa State:  KSU by 4
#19 Texas @ Oklahoma: Oklahoma by 12
#25 Gonzaga @ St. Mary's:  SMU by 2

We could easily see all the ranked teams lose. 

Five ranked teams losing in the same weekend?  Are you crazy?

Iowa lost, Creighton lost,  Kansas State lost, and Texas lost.  (Gonzaga won fairly easily.) 
Crazy like a fox.  :-)

4 comments:

  1. I think most predictive systems have OK St as a top 20 team (I have them top 15, Sagarin as #13). Vegas also had them favored by 2 against Kansas. It's just the descriptive systems like AP Poll, RPI, etc. that have them (deservedly) lower (I have them as a 9-seed I believe).

    "Louisville is now 1-4 against ranked teams this season, which can't be a good sign." Have you tested this? Would be interesting to see if it is true, and if so, may point to doing something like weighting tougher games more. I think KenPom this year is doing something like this, weighting games against similar-rated teams more than others.

    ReplyDelete
  2. Apparently OK State isn't a lock to make the Tournament, which would be shocking.

    To the extent that I've tested weighting games based upon opponent's strength it hasn't added anything. That isn't too surprising, as most rating systems already explicitly use the opponent's strength. I think at some point I had the "winning percentage against the AP Top 25" and it wasn't valuable, but I don't currently get that info.

    I exchanged some emails with KenPom about exactly what he is doing, and from what I understood, he's weighting expected results much more than unexpected results. So if you're expected to win a game by 8 and you win by 6, that game gets counted heavily. If you lose by 14 (or win by 25), the game would count less. I'll have to be honest and say I don't understand that reasoning at all -- it's essentially algorithmic confirmation bias. It seems to me that (if anything) you'd want to weight the unexpected results more. Still, I've learned not to dismiss an idea until I test it, but I haven't had a chance to look at this approach. Maybe in the off-season.

    ReplyDelete
  3. Yeah, this goes back to what we discussed last year, about "deserving" (aka "resume" aka "descriptive") versus "best" (aka "true strength" aka "predictive"). I'm firmly in the former camp, but others disagree. Every professional league out there, for instance, uses "deserving" as their criteria.

    I have never found weighting to be of any benefit, whether it be recency weighting, opponent weighting, removing garbage time, or whatever. Any signal that is there usually gets drowned out by the decreasing sample.

    ReplyDelete
  4. I've had the same experience as you with weighting. And KenPom explicitly said that it improved the predictive performance of his rating (caveated for only February / March games, I think he said). So all in all a puzzling result from him.

    ReplyDelete