Monday, March 26, 2018

Final Four per the MDS Model (2018)

In the past four seasons since I've been running my model, going into the Final Four I've only picked the champion correctly once: Villanova in 2016 (missed on Gonzaga in 2017, Kentucky in 2015, and Florida in 2014). But good news for Nova fans: I've got Villanova winning it all again, as I did at the beginning of the tournament.

SeedTeam1st RoundChampionshipChampion
Final Four

Monday, March 19, 2018

Sweet 16 per the MDS Model (2018)

A bunch of upsets on the left side of the bracket has left the field with 2 favorites well above the rest: Villanova and Duke. Both team's individual chances are limited by the fact that they likely will have to play each other (but combined they win it almost 37% of the time).

SeedTeam1st RoundElite 8Final FourChampionshipChampion
Sweet 16
9Kansas St.100.00%37.11%17.31%6.13%1.59%
9Florida St.100.00%33.69%13.50%6.34%1.83%
7Texas A&M100.00%37.62%15.57%7.66%2.36%
5West Virginia100.00%33.12%15.25%6.86%3.77%
3Texas Tech100.00%40.98%15.32%6.69%3.59%

Another wrinkle, as illustrated above, is that Duke has the best chance of making the championship game, yet a lower chance of actually winning it. Playing Syracuse in the Sweet 16 greatly elevates their odds of getting there, but Villanova is still the better team, and thus has the stronger chance over all.

Sunday, March 18, 2018

Slow Basketball Teams Aren't Winning Titles

Note a key word in the title of this post: slow teams aren't winning titles, not can't. No one can definitively say whether they CAN'T. But we can determine whether they aren't at an abnormal rate.

On the heels of Virginia's momentous upset loss to UMBC, the first time that a 1 seed has lost to a 16, a hot argument right now is whether teams that play that slow really suffer from an automatic disadvantage due to their pace of play. KenPom weighed in before the tournament on this very subject: 
I'm focusing on the speed part of this angle, so is it really that abnormal that no "ultra-slow" team has won the title? At least since statisticians like KenPom have been tracking pace metrics explicitly?

One of the original ideas underlying basketball analytics is that slowing the game down is an optimal strategy for the underdog, as outlined in Dean Oliver's seminal work, Basketball on Paper:
A third risky strategy-one most people may not have considered-is to slow the pace down, reducing the number of possessions in a game ... This strategy is used when an underdog gets an early lead on a favorite. What it does is limit the better team from taking full advantage of being better. A good team will win out over a bad team if you play long enough. By cutting a game down to fewer possessions, an underdog is limiting how long the favorite has to prove it is better.
(This strategy) fundamentally increase(s) the variability of the difference between points scored and points allowed ... The slow-down strategy actually works by increasing the variability of both offensive and defensive ratings. (Oliver p. 128)
So teams like Virginia play like underdogs even though their favorites, which in turn helps the actual underdogs (since the sport is a zero-sum game). Less possessions = more variability = more of a chance for the underdog.

My first step was to verify that it is in fact the case that slow teams actually aren't winning titles. And they aren't. Since 2002 (when KenPom's data begins), no team in the bottom 10% percentile of adjusted tempo has won it all, and Villanova's championship in 2016 is the closest at the 19.9% percentile.

In light of the above strategy, this presents a chicken-and-the-egg problem. Do teams generally play slow BECAUSE they're worse and realize it? Meaning there just haven't been that many teams that are both good AND slow? 

Using historical futures odds since 2009 (pre NCAA tournament), I determined that there have only been 3 teams each year (on average) that had > 0.1% of winning the championship AND played at a tempo in the bottom 10% percentile. Added together, these good/slow teams have had a 7.97% chance of winning the tournament each season, per the Vegas odds market. That means that there's a 92.03% chance each season that a non-slow team lifts the trophy. Therefore, there's a 24.37% chance that no good/slow team has won it in the past 17 seasons (92.03%^17), which falls well outside any reasonable confidence level to conclude otherwise (that this isn't random and is actually a by product of the style of play). Those teams have probably just been slightly unlucky.

The teams that fall in the bottom 10% this year and are still alive are Michigan and Syracuse (as of when this article was written). So who knows, maybe in 2 weeks one of them will prove us all wrong and finally do it. 

Alternatively, it's not too hard to imagine the counterfactual universe in which all of this is moot and a slow team already won the championship. If Virginia doesn't blow their lead to Syracuse in 2016 and makes the Final Four, they would have had a very good chance of going on to win it all. Even more damning: Wisconsin easily met the criteria in 2015 (7th slowest team in the country), but fell short in the title game by less than two possessions.

Friday, March 16, 2018

Do 12 Seeds Really Upset 5 Seeds More Often Than Expected?

The 12 seed over a 5 seed upset pick is a popular one in bracket pools everywhere. So much so that the value that may lie there is eroded now by its popularity, or so it seems. But do 12 seeds really pull off the upset to an abnormal degree? At least when compared to 11 over 6 and 13 over 4 (relative to their respective seeds)?

If there is a significant difference, the best explanation I've heard (via Mark Titus, @clubtrillion) is that the 12 seed is usually an auto bid mid-major that got hot and won their conference tournament to play their way into the field. Meanwhile, the 5 seed is often a Power 5 team that slumped towards the end of the season and fell down into the spot. Examples of this this year include Ohio St/South Dakota St and Clemson/New Mexico St, where Ohio St finished the regular season 2-3 and Clemson ended the year 3-5 (and their counterparts won their conference tournaments).

I used all first round tournament results since 1985 to determine the trend, and the 12/5 matchup is indeed an outlier:

The 12 over the 5 is actually on par with the 11 over the 6, and very close to the 10 over the 7. Using the linear regression illustrated above by the trendline, we should expect the 12 to beat the 5 only 29.12% of the time. Testing against this null hypothesis generates a z-score of 4.66 over 1,056 games, so we can determine with a very high level of certainty (99.9998%) that this isn't a result of random chance: the 12 seeds do win more often than they're supposed to.