If you played the 2007 NCAA Tournament five times, you’re probably getting Florida over Ohio State in the finals four times.
— Dan Wolken (@DanWolken) April 6, 2014
Now what are the actual odds of any one team making the title game? Well, you'll get slightly different numbers from Pomeroy or Sagarin or Nate Silver's model at 538, but in general the best team in the nation will have something around a 25% chance of making the title game. This means that the most likely title game match-up has maybe a 6-7% chance of happening. To put that in perspective, a 6-7% chance of something happening is about the same as the odds of a 14 point underdog winning a college basketball game. Or approximately the odds of Rhode Island or Boston College beating Kentucky or Connecticut on a neutral floor.But of course, those are the best teams. What about UConn and Kentucky? Well, Pomeroy put the odds of those two playing in the title game at 0.04%, or close to 2800-to-1. So that's led to the typical complaining that this is proof that advanced stats are useless in March Madness. It's all about just having the best players, man! Just will your way to win!
This reminds me of a snarky joke you'll hear from statisticians if you prod them. Knowing that the odds of winning the jackpot in a major lottery (MegaMillions, Powerball, etc) is on the order of 100,000,000-to-1, they'll joke that nobody has ever won the lottery. After all, for anybody to have won would have required overcoming impossible odds.
But of course, the fact that somebody won the lottery doesn't change the fact that the odds to win were 100 million-to-one. No matter how low of a probability some event is to happen, there is some finite number of times over which that event is likely to happen. If you buy 100 million lottery tickets, you are likely to win the lottery. It's just that your expected financial return will be negative, so it's a bad investment idea. This is basically the same gaffe that creationists make when they say that evolution is unlikely because the odds of beneficial mutations happening are so low: over enough billions of generations, evolution isn't only likely, it's inevitable.
So this brings me back to the Tournament. The fact that a low probability event happened doesn't a priori mean that it wasn't a low probability event. The fact that these two teams made the title game is not data, it is an anecdote, which is quite different. The only way to provide evidence that the computer models are bad is to look at a large number of games to see if the probabilities don't add up in a reasonable way... some sort of Bayesian approach would probably do. And of course, this has been done. The computer ratings are very accurate. 80% Pomeroy favorites win just about 80% of the time. Of course, about 20% of the time they lose, too. As they should. In fact, if 80% favorites didn't get upset ~20% of the time, that would be proof that the computer models are bad. Though you'll never see it framed that way in the media.
You can actually see the math in the model just in Kentucky's run alone. Their last four games have all come down to effectively the final possession. Treating those four games as a coinflip says they had a 1-in-16 chance of winning all four games (~6% chance). That's already pretty close to the 1.8% chance Pomeroy had them pre-tourney. But of course, the odds of this run are even flukier than that. After all, Kansas State shot 5-for-21 behind the arc in a game they lost by 7. Louisville made four shots outside the paint all game. Kentucky's poor three-point shooters hit 7-for-11 behind the arc against Michigan. And Kentucky's late game shooting percentages have been absurdly fluky, such as hitting their final five shots against Michigan (3-for-3 behind the arc over that stretch). Aaron Harrison's threes to beat Wisconsin and Michigan were at least as difficult as Fred Van Vleet's three to beat them in the Round of 32 (and Van Vleet is a better three-point shooter than Harrison). Overall, Kentucky is outscoring opponents 53-33 in the final five minutes of their last four games. You can argue that's "clutch", but remember when Tyler Ennis had incredible clutch stats and this was justification for why Syracuse was winning every close game and was the best team in the nation? Well after starting 8-0 in games decided by 6 points or less, Syracuse lost 5 of their final 6 games decided by six points or less. Don't overreact to small sample size "clutch" stats. They're ephemeral and random.
And that's all before we get into the fact that the reffing in their Final Four and Elite 8 games was NBA-style (no charges called, and any minor contact in the lane was a quick whistle), and thus very much to Kentucky's benefit and to the detriment of teams like Michigan and Wisconsin that rely on good defensive positioning to draw charges and to have very good defensive FTRates. If you play this Tourney over 50 times, how many times does Kentucky make the title game? Probably once. But the fact that a 1-in-50 team made a title game does not invalidate the model any more than a dude winning the lottery invalidates the idea that buying lotto tickets is bad investment advice.
Has UConn's run been as fluky as Kentucky? I guess it depends what you mean. UConn's wins certainly haven't been as fluky. They clearly outplayed both Michigan State and Florida over the last two rounds. There were no fluke shooting percentage oddities in their favor, and neither game came down to the coin flip of a one-possession game. Of course, the game against St. Joe's in Round of 64 went to overtime, so they were just one play from being eliminated right there. And UConn's level of play has been extremely abnormal in the sense that they haven't been playing this well all season long. They happen to have played arguably their two best games of the entire season in the Elite 8 and Final Four. So UConn has been more dominant than Kentucky and played better over their five tourney games. But again, I don't think this is evidence that if we played this tourney over 50 times that UConn would get here more than once. I don't think there was any evidence heading into the tourney that they'd play this well.
In a sense, it's a bit frustrating that we're not getting to see the best teams in the nation play here. The last two seasons our national champion was a team that was arguably the best in the nation. If Florida was playing Louisville or Arizona here, I think we'd feel the same way. Or even if we had a team like Kansas or Michigan State or Wisconsin that was playing like a borderline Top 5 team down the stretch. Instead we're back to a title game like 2011 where we had two good-but-not-great teams that just happened to be playing their best ball and getting some lucky bounces all at the right time.
Although I guess we need games like this from time to time. If the best team won every year it would be boring, no? Let's talk about the individual match-ups:
Final Four ATS: 2-0-0
2014 Tournament ATS: 39-23-4
2013 Tournament ATS: 36-30-1
2012 Tournament ATS: 30-35-2
2011 Tournament ATS: 40-26-1
2010 Tournament ATS: 35-25-3
UConn (+2.5) over Kentucky: As I said above, I think UConn has been playing better than Kentucky, particularly over the last two or three games. But certainly Kentucky is playing like a Top Ten team as well. You can't say either team is so much better than the other that they're an obvious favorite. The computers basically have this game as a toss-up, maybe giving Kentucky a tiny edge. Throw in a couple of points for Kentucky being the "public" team getting all the media hype, and you're at the 2.5 point spread. Just under the "fading the public" logic, you want to take UConn here. But what about the individual match-ups?
With Kentucky, reffing is always a concern. If you get refs willing to call charges while not calling so many anticipatory blocking fouls, their offense can get into a bog really quickly. They are not a good passing team or shooting team. If they get the refs from the Michigan or Wisconsin games, however, they could get to the line a lot against a UConn front line that does commit fouls (7th in the AAC in defensive FTRate). Defensive rebounding was also a problem for UConn during the regular season, though it's been a lot better over the last few weeks. Basically, what you have to ask yourself here is do you believe in UConn's crazy good level of performance over the last few weeks. If you do, Kentucky is going to struggle to score. If you don't, and think UConn is going to regress to their regular season form, Kentucky is going to torch UConn for something like 1.3 PPP.
What about the UConn offense? They're a jump shooting team, which plays into Kentucky's hands a bit. The Wildcats are good at denying three-point opportunities to their opponents (they led the SEC in defensive 3PA%). But without Willie Cauley-Stein, Kentucky's bigs are really not particularly good on screens. Alex Poythress is the only Kentucky big I'd trust guarding a UConn backcourt player on a high screen, and he's been invisible so far in this NCAA Tournament. UConn is going to try to isolate Napier and Boatright off of switches, which they'll use to force the Kentucky defense to rotate (which it isn't particularly good at), to either open up DeAndre Daniels underneath, or a shooter behind the arc (of course, Daniels can shoot threes also).
If neither team is shooting well, the edge clearly goes to Kentucky. They'll likely end up with a FTA advantage and they should win the rebounding battle. If this is a game of volleyball around the rim, that plays to their advantage. But UConn has just been playing so incredibly well lately and it's not due to fluky shooting, so there's no obvious reason to expect a regression to the mean. Without Kentucky having a clear match-up advantage overall, I think we have to go back to the view that (in my opinion) UConn has been playing better than Kentucky in the tourney so far, and the fact that the Vegas line is out of whack due to all the "public" love for Kentucky. Those two things tell me that the smart bet is to ride UConn.
No comments:
Post a Comment