NCAA Tournament 2013: Computers, like humans, spit out flawed predictions, too

[CDC84]

http://aol.sportingnews.com/ncaa-basketball/story/2013-03-20/ncaa-tournament-2013-president-obama-final-four-picks-kenpom-vs-rpi

[Ekrub]

Count me as one who views Kenpom as superior to RPI. Margin of victory (within reason as sometimes final scores don't indicate how close/big of a blowout a game is) Kenpom isn't an going to ever be perfect but in the way of predicting games, IMO, he has the best system. It's a tool, not the entire sport. There is a reason we play the game and I'm sure Ken Pomeroy is happy that his predictions aren't perfect.

RPI has too many flaws and too many ways that it can be "gamed"

[gamagin]

Ok, THIS is disrespect !

http://espn.go.com/video/clip?id=9074010

[Once and Future Zag]

Count me as one who views Kenpom as superior to RPI. Margin of victory (within reason as sometimes final scores don't indicate how close/big of a blowout a game is) Kenpom isn't an going to ever be perfect but in the way of predicting games, IMO, he has the best system. It's a tool, not the entire sport. There is a reason we play the game and I'm sure Ken Pomeroy is happy that his predictions aren't perfect.

RPI has too many flaws and too many ways that it can be "gamed"

RPI has 2 data points per game W/L and Location

KenPom's systems treats (essentially) every possession as a source of data points.

Hard to say that former one isn't going to be less precise a metric than the latter.

[NumberCruncher]

Count me as one who views Kenpom as superior to RPI. Margin of victory (within reason as sometimes final scores don't indicate how close/big of a blowout a game is) Kenpom isn't an going to ever be perfect but in the way of predicting games, IMO, he has the best system. It's a tool, not the entire sport. There is a reason we play the game and I'm sure Ken Pomeroy is happy that his predictions aren't perfect.

RPI has too many flaws and too many ways that it can be "gamed"

Me too.

All methods of predicting games are flawed, some more than others. One reason efficiency ratings are flawed is that there is no way of separating out garbage time stats, which are virtually meaningless. I could come up with many other reasons. Still, it's far better than RPI.

[CDC84]

Florida is just an odd team. There will always be a team or two out there that just doesn't "make sense" no matter what system you use.

The Gators haven't beaten a decent opponent outside the state of Florida. I've seen nearly everyone of their road losses, and it always comes down to their guards hucking up bad shots in clutch situations (especially Kenny Boynton) and some incredibly low basketball IQ. They go into vapor lock mode.

I have always had some issues with margin of victory because a lot of coaches, once they establish a nice lead (say, 17 points) in the 2nd half will actually do the math on how many possessions are left in the game and work the shot clock to death even if it leads to a missed field goal. Mark Few did this in a number of games this season. Teams that could've been blown out by 30 were beaten by 15. Granted, they were still beaten by 15, which is a pretty good margin. It's smart basketball.

[VinnyZag]

Pomeroy posted an alternative point of view on his blog, which is not behind his paywall.

Decoursey seems to be the main anti-stats guy on Twitter right now. Watching him argue with what he derisively calls "Pomeroy people" has been amusing today.

[rennis]

They go into vapor lock mode.

Did you invent that? Great metaphor!

[caduceus]

The thing about these stats to remember is that they only go so far. It's a statistical model designed to capture a 40 minute basketball game into a few numbers. Are they useful? Sure. They can be quite informative about the overall strength of a team, and can point out important details that might otherwise be obscured by things like the pace of a game.

Many people confuse a prediction with a statistical probability. Seth Davis makes a prediction. Just because Pomeroy's log5 model gives Florida and Louisville the best chance at winning it all doesn't mean at all that it's predicting those two to be in the National Championship game. If it ends up being Kansas and the Zags (or even Montana vs. Harvard), it doesn't at all mean that the probabilities are wrong. Every team in the tournament has a non-zero chance of winning it all, whatever the number is.

Statistical models are not crystal balls, and will never capture all the variables involved that make up a ball game. But they can be a very useful tool in addition to other means like "the eye test."

[caduceus]

I'll add this here, since it was also on KenPom's blog. Pretty interesting.

The pundits are really going out on a limb this year. Way to take a chance.


The tournament is wide open
by Ken Pomeroy on Tuesday, March 19, 2013

We’ve heard that a lot from all of the experts all season. Just three days ago, this column promised us there was massive parity in the game, with quotes from the top coaches. Jim Boeheim says there would be ten different winners if we played the tourney ten times!

So which wacky teams did the experts pick to win it all? Well, for starters, not a single person picked Duke (despite Goodman’s piece proclaiming them the favorites two weeks ago). Just three of the 34 experts picked a non-one seed and in each case it was a two-seed. And 27 of the 34 picked Louisville. Here’s a complete rundown of picks posted at ESPN.com, CBSSports.com, USAToday.com, SI.com, and Yahoo.com.

Alaa Abdelnaby: Louisville
Greg Anthony: Louisville
Nicole Auerbach: Louisville
Jay Bilas: Louisville
Jeff Borzello: Louisville
Mateen Cleaves: Louisville
Seth Davis: Louisville
Dennis Dodd: Louisville
Gregg Doyel: Ohio State
Jeff Eisenberg: Louisville
Brad Evans: Louisville
Pat Forde: Louisville
John Gasaway: Louisville
Pete Gillen: Miami
Scott Gleeson: Kansas
Jeff Goodman: Louisville
Doug Gottlieb: Louisville
Seth Greenberg: Louisville
Mike Lopresti: Louisville
Joe Lunardi: Louisville
Stewart Mandel: Louisville
Matt Norlander: Louisville
Jerry Palm: Louisville
Gary Parrish: Louisville
Digger Phelps: Miami
Eric Prisbell: Louisville
Jon Rothstein: Louisville
Wally Szczerbiak: Gonzaga
Pete Thamel: Louisville
Peter Tiernan: Louisville
Dick Vitale: Louisville
Dan Wetzel: Indiana
Jay Williams: Ohio State
Luke Winn: Louisville