It goes without saying that it has been a difficult two weeks for the Michigan State community. Tuesday night, the Spartans' men's basketball team returned to the Breslin Center for the first time since the tragic shooting on Feb. 13 that claimed the lives of Brian Fraser, Arielle Anderson and Alexandria Verner.
For at least a few hours, the Spartans put a smile on the faces of countless Spartans fans in East Lansing and around the world. Michigan State posted perhaps the most impressive win of the season to date as the Spartans defeated the No. 17 Indiana Hoosiers, 80-65.
As the regular season winds down, the Green and White appear to be trending up once again. The post-season is right around the corner. March beckons. What might that post-season look like? Let's dig into the numbers.
Quick Big Ten Tournament Update
Earlier this week, I provided a detailed look at Michigan State prospects for the rest of the regular season. The Spartans' win over the Hoosiers. My simulations suggest that there is about a 20% chance that the Spartans will run the table leading up to the Big Ten Tournament and a 66% chance that the Spartans lose no more than one game.
It currently appears that Michigan State is not going to be able to reschedule the game against Minnesota, which is not necessarily a bad thing. If the Spartans can win the final three games of the Big Ten season, my simulations suggest that a top four seed in the Big Ten Tournament is virtually guaranteed (99% chance), with the No. 3 seed being most likely.
If the Spartan do lose one of the final three games (which the math says is the most likely outcome), the No. 5 seed (38% odds) is now the most likely result, but the No. 6 seed is also very possible (34%).
Bracketology Metrics
As March approaches, however, it is "the big dance" that is still forefront in the minds of Spartans fans. Around this time of year, it is time to get serious about one of my favorite branches of science: bracketology.
I have been fascinated with the structure of the NCAA basketball tournament since I was a child. I still have a laminated folder that contains hand-written brackets from the late 1980s and 1990s. More recently I have started to use the various mathematical tools at my disposal to make predictions about the bracket, create trial backets, and thoroughly analyze the actual bracket once it is released on Selection Sunday.
It would be fair to say that I am more than a little obsessed with March Madness. When it comes to the mathematical tools and metrics used to create the NCAA Tournament brackets, there are two categories worth mentioning.
First, there are predictive metrics. These are metrics designed to predict the outcome of future games. In other words, predictive metrics can be used to forecast point spreads, which directly correlate to victory probabilities. Tempo-free efficiency metrics such as those published by Kenpon and Bartorvic fall into this category.
Second, there are results-based metrics. These metrics are generally designed to evaluate the quality of wins and losses in games that have already taken place. The NCAA used to use a metric called the rating percentage index, or "RPI" that was based on a simple formula employing winning percentages.
Recently, a set of more mathematically sound metrics have been developed that are usually referred to as "strength of resume" or "strength of record" metrics. ESPN tabulates their own version of the "SoR" as does Bartorvic. Michigan State associate athletic director Kevin Pauga developed his own results-based metric called the KPI.
Finally, there are metrics which attempt to blend together the elements of both predictive metric and results-based metrics into a single value. A few years back the NCAA introduced the NET rating system that utilizes elements of both.
The NCAA Tournament Selection Committee uses a combination of the metrics described above, but the NET metric is the one that tends to grab the most headlines. "Bracketologists" will reference "quad one" or "quad two" wins and losses, which are strictly based on the NET rankings, to compare teams.
Over the years, I have developed my own method and strategy for evaluating teams as we head into March. For a predictive metric, I simply rely on Kenpom's efficiency margin rankings. For a results-based metric, I have developed my own strength-of-record calculation.
My "Dr. G&W-SoR" calculates the expected number of wins that an average Top 25 team (with a Kenpom efficiency margin of 19.00) would have for each team's actual schedule, using Kenpom data to estimate win probabilities. Incidentally, this is exactly the way that I calculate strength of schedule. I then compare the actual number of wins each team earned compared to the expected win total for the reference team with the same schedule.
This year, I also developed a single blended metric that combines Kenpom efficiency and my SoR metric into a single value. I used the final NCAA Tournament seed lines from the previous two tournaments to tune the blended metric to best match the committee's actual seed list.
I can now use my blended NCAA Tournament metric to predict the seed list and generate sample brackets.
Best Case/Worst Case for Michigan State
Using my new tournament metric, I can easily explore different possible scenarios for the remainder of the current Michigan State season. For example, what are the best and worst cases that we can imagine? The first question to ask is if Michigan State is safely in the NCAA Tournament now?
The worst possible scenario for the Spartans would be if the Spartans were to lose all the remaining regular season games. Michigan State would then have a final record of 17-13 entering the NCAA Tournament.
With this record, I estimate that the Spartan's would enter the Big Ten Tournament as the No. 10 seed. For the sake of argument, I will assume that Michigan State would face, and lose to, No. 7 Iowa in the first game to finish at 17-14.
Would the Spartans still hear their names called on Selection Sunday with such a dreadful finish? My Tournament metric suggests that in this scenario, that Michigan State would just barely sneak into the Tournament as a member of the First Four. Also note that if somehow the Minnesota game were to be rescheduled and the Spartans were to lose that game, the Spartans would slide into the NIT in this scenario.
Fortunately, my calculations suggest that there is only about a 3% chance that Michigan State will lose out. Even so, I think that one more win is needed for the Spartans to feel completely safe on Selection Sunday.
On a more positive note, what is the best possible scenario for the Spartans? First, Michigan State would need to win all three regular season games to finish at 20-10. This would then most likely result in the No. 3 seed in the Big Ten Tournament. Let's then assume that the Spartans would beat teams with strong NET ratings such as Indiana, Maryland, and then Purdue to claim the Big Ten Tournament crown.
In this scenario, my algorithm would place the Spartans as the best team on the No. 5 seed line. If the committee is feeling generous, it is possible at a 23-10 Michigan State team could slide up to the No. 4 seed line. That is likely the best-case scenario.
The Bracket Matrix
But what is the most likely place for the Spartans to land on Selection Sunday? A middle-of-the-road scenario would be one where Michigan State loses on the road to Iowa, beats Nebraska on the road, and then beats Ohio State in the season finale to finish the regular season at 19-11.
This would likely result in a No. 5 seed for Michigan State in the Big Ten Tournament. For this exercise, I will assume that the Spartans beat No. 12 seed Nebraska and then loses to No. 4 seed Indiana in the quarterfinals.
In this scenario, my metric would place Michigan State on the No. 8 seed line, which incidentally is exactly where the Spartans are ranked now in my system (No. 30 overall).
But how does the prediction of my algorithm compare to other public brackets?
Fortunately, the bracket matrix website collects data from over 100 NCAA Tournament brackets available on the internet. Figure 1 below shows the distribution of the current seed predictions for Michigan State.
 |
| Figure 1: Distribution of Michigan State's projected seeds, as of Feb. 22 and as reported by the bracket matrix website |
As we can see, the consensus of the online brackets agrees with my formula's current prediction of a No. 8 seed for the Spartans. Michigan State is a No. 8 seed in 45% of the public brackets, a No. 7 seed in just over 30% of the brackets, and a No. 9 seed in 19% of the brackets. Less than 5% of the brackets have the Green and White as a No. 6 or a No. 10 seed.
If this were to come to pass, some of Michigan State's possible first round opponents include (in order of predicted difficultly) Arkansas, Auburn, Texas A&M, Providence, Nevada, Kentucky, Duke, Oklahoma State, Florida Atlantic, North Carolina State, Pittsburgh, and Missouri.
The Spartans currently project to be favored against all of those teams on a neutral floor except the first three on the list.
Just for fun, here is my algorithms raw seed line prediction as of Feb 23. The teams shaded in yellow are the "last four in" and the teams shown in orange are the "last four out." After each team name I have included the current NET, Kenpom, and my SoR rankings.
 |
| Table 1: Dr. Green and White's current NCAA Tournament seed list as of Feb 23. |
Note that this is an unoptimized list. If I were to use this list to create a bracket, I would move some of the teams around based on applying "the eye test" to their resumes. I would also place teams into regions considering the potential for rematches (such as Michigan State versus Illinois in a No. 8 versus No. 9 game, which will not happen), and the location of each game. The committee places a large emphasis on reducing travel distance when possible.
While there are a few notable differences, my raw metric compares well to the current brackets from ESPN and the average seeds shown on the bracket matrix website.
Comments
Post a Comment