March Madness Manifesto

NCAA Tournament game between Northeastern and Notre Dame at the Consol Energy Center in PittsburghThe NCAA Men’s Basketball Tournament gets underway in earnest on Thursday, beginning an annual tradition of the strangest sporting event found anywhere in America (no, the “First Four” abomination on Tuesday and Wednesday does not count). Consider that college basketball’s championship event draws enormous ratings and interest despite having a regular season that is largely an afterthought for most mainstream sports fans. Meanwhile, the event is trumpeted for providing opportunities to the have-nots, even though history and economics dictate that only a small fraction of the 351 Division I teams have a chance to win the tournament. The beauty of “March Madness” stems not from traditional sources of sports fandom, but rather the frenzied, unpredictable nature of its early rounds, its temporary elevation of the slighted and forgotten, and a brilliant structure for low-stakes gambling that brings out the degenerate in all of us. After filling out a sixty-four team bracket, teams unknown a moment before turn into old friends. The anonymous becomes personal. That is, of course, unless those teams lose or until they reach the next round against a newly favored opponent, at which point they slide back into the forgotten. Loyalty only goes as far as the next round. March Madness is a mercenary event for a success-driven culture.

Given this dynamic, bracket advice is cheap and plentiful. Yet, despite the considerable ink spilled on this subject, it is amazing how consistently terrible this advice often is. Many columnists search through history to show that only teams with particular, obscure traits can win, leaving themselves open to the problem of overfitting. At the extremes, bracket soothsayers tout statistics showing that only teams with blue as a team color have won the National Championship going back more than ten years (except for Louisville in 2013). Other articles focus on picking a certain number of each kind of seed, as though the teams themselves are irrelevant. This line of thinking uses simplicity to overwhelm common sense. If you cannot logically explain why a statistic or factor should affect tournament play, then its correlation to success should not imply causation.

Another example of this deficiency comes from an article that somehow got published in the Washington Post in 2008: “Once you’ve got your national champion, go back and make sure you’ve given that team’s conference the proper respect in earlier rounds. There’s obviously no real link between, say, Duke’s performance in the last round and Clemson’s performance in the first, but I don’t care: I always choose one major conference to ride throughout the tournament. The strategy makes theoretical sense-that there would be some kind of interior consistency in the final results-and it seems to offer major rewards.” It’s entirely unclear how something could have “no link” but also make “theoretical sense.” This is also belied by actual results; in 2011, the University of Connecticut won the national championship, while other members of their Big East conference went 7-10. In contrast to these misleading specifics, other analysts provide platitudes as useless as they are mindless, such as: “Do: Pick a Few Early Upsets” but “Don’t: Go Too Crazy with Your Upset Picks.” These are the bromides of March.

Here, then, is a manifesto setting forth the principles you should follow in filling out your bracket. They are the result of hours of wasted time and were crafted with love.

As a preliminary matter, the most important concept to keep in mind is simple yet revolutionary: your task is not simply to pick the team you believe will win each game, it is to win your pool. If you are seeking to merely avoid embarrassment, then choose all favorites. However, if your goal is to win (or place in) your bracket pool, certain strategies are necessary to improve your chances of differentiating yourself from your competitors. At the same time, you still want to choose teams that can win. A bracket with a 16 seed winning the championship is certainly unique, but it also has no chance of success. Since these two concepts are not always in harmony, the real challenge is to balance them by identifying overvalued and undervalued teams; then just hope you get lucky.

Pay Attention to the Pool’s Size and Scoring Rules

Effective strategies depend on the rules of engagement, and the most important features for these purposes are your pool’s size and scoring rules. Regarding the former, generally the larger the pool, the more risks you must take. The odds are that you will not do significantly better than a “chalk” bracket—one with the higher seeded team winning every game—so in a smaller pool, your choices should more closely conform to that orderly bracket. If there are five people in your pool, then there is no need to make an unlikely championship pick or take many lower seeded teams in the early rounds. Just choose one of the most likely teams to win it all, sprinkle in a few variations, and you’ll give yourself an excellent chance. Most people, however, play in office pools with hundreds of entries. Under this framework, a successful entry will need enough variation from seed expectations to distinguish it from the others who will invariably cover a set of predictable patterns. Likewise, if you are interested in winning a mass competition like ESPN’s bracket challenge, which last year had 11.57 million entries, then you have to choose an off-the-wall champion and/or an incredibly unusual overall bracket. In sum, the key is to increase the wildness of your picks in proportion to the size of your pool.

Another key consideration is the scoring system. The standard scoring rules for most pools is a step ladder framework of 1-2-4-8-16-32, meaning that you receive 1 point for each first-round victory, 2 for each second-round win, and so on, so that the points for a win double each round. Under this method, each round is worth the same amount (32 points) even as the number of games decreases by half each time. Thus, this system places an increasing emphasis on each successive round and great weight on correctly choosing the champion. If playing under these rules, then analyzing your bracket’s optimal differentiation from the norm should focus heavily on the importance of picking a champion. For instance, if you are in an office pool of approximately 300 entries with standard scoring, picking one of the top two or three favorites really limits your chances of winning. At the same time, if the pool isn’t too large, taking a real long-shot to win will be an unnecessary risk.

A different (and frankly, better) scoring system seeks to balance the importance of the various rounds by adding a seed value to a flatter round-value curve. For instance, one system has a 3-6-12-18-24-30 framework for picking a game correctly in each round, along with a “seed bonus.” Under this system, if you correctly picked a No. 2 seed to defeat a No. 15 seed in the first round, then you would receive 5 points (3+2). Likewise, if you correctly chose a 5 seed to win the championship, you would receive 35 points for the championship round (30+5). Here, choosing the champion is still far more important than picking any other individual game. Yet, the number of points at stake is weighted more toward the early rounds, decreasing the fundamental need to do so. This seed-added scoring system shifts the areas where you can vary your bracket to earlier in the tournament, while also decreasing the danger in choosing a longer-shot champion.

In sum, the first step in filling out your bracket is assessing your pool’s size and scoring to determine how much madness you’ll need to predict.

Don’t (Always) Pick the Best Team

As alluded to above, picking the better team in each game is a tactic doomed to failure. Even if you are truly the greatest basketball mind on earth and can uniquely see which team has an advantage in every game according to skill and matchups, following that analysis will not lead to success. Doing so amounts to the philosophy “get everything right.” This is not a good strategy. A single-elimination tournament of amateurs unused to playing under such stakes is bound to deliver tremendous variation. Every NCAA tournament has resulted in some relatively low probability events occurring.

This is not to say that you should just randomly pick among unlikely outcomes. It does mean, though, that because you need to vary your picks to sufficiently differentiate your bracket from the masses of co-workers standing between you and the ultimate prize, choosing some low probability events is a necessity. Since we know that “shocking” upsets are bound to happen, the best strategy is choose true underdogs who are nonetheless more capable than popular perception. Even under standard scoring, choosing an 11-seed with a 40% chance of victory is a better strategy than taking the 6-seed favored to win 60% of the time, even though you are getting worse odds. Likewise, if there is a 1 seed with “only” a 20% chance of advancing to the Final Four, it may be wise to choose a particularly strong 4 seed to beat them in the Sweet Sixteen.

The “Better” Team is not always the Better Team

The NCAA Tournament Selection Committee’s Principles and Procedures state that the committee’s mission is to select the best teams and seed them accordingly. But in reality, their very own guidelines contradict that promise. The committee relies to a large extent on the Ratings Percentage Index (RPI) in seeding the field. Not only is the RPI a deeply flawed metric, but the factors that go into it are related to achievement, not ability. There is a difference. The committee’s emphasis on a team’s “resume” over its “efficiency” seeks to reward and penalize teams for wins and losses, and quality wins and bad losses, rather than just looking at overall performance stripped of luck and circumstance.

Efficiency statistics take this into account, such as those generated by Ken Pomeroy at his Ken Pom blog. Ken Pom judges teams by their offensive and defensive efficiency (measured by points per possession adjusted for schedule strength). Of course, Ken Pom, is not a perfect predictor of performance, either. No one is Nostradamus when subject to the random variation of single-elimination tournaments. But Ken Pom’s efficiency ratings are a better baseline than seeding; comparing the two can often reveal significant value. Following these differences allows bracket variation without sacrificing odds. The selection committee does not heavily rely on efficiency statistics. You should.

Consider Compound Probabilities

The disease of believing that you can analyze a bracket and pick the winning team according to simple logic has a particularly nasty symptom: holding the idea that each round should be picked independently from the others. There is too much randomness in the tournament to pick Team A over Team B and then ignore the possibility that you are wrong when making the next round’s selection between Team A and Team C. In practice, this means applying compound probabilities to your bracket picking. A team with a 51% chance of winning the first round and a 51% chance of winning the second round is not favored to reach the Sweet Sixteen; instead, it should do so just over a quarter of the time. Thus, in a situation with very strong 4 and 5 seeds and a vulnerable 1 seed, you should still account for the 1 seed’s easy road to that game and the toss-up nature of the 4 vs. 5 game. Particularly strong early round challengers should dissuade you from taking a team to advance far.

Coaching Matters

It’s obvious to most people that coaching can make a major difference on the performance of a sports team, especially one made up entirely of amateurs. Less obvious is how this should factor into bracket selections. It can be hazardous to isolate a particular factor as particularly necessary to March achievement. By the time a team has reached the tournament, all of the little factors that lead to team success or failure have been baked in to the cake, so to speak. Yet, there is real reason to believe that certain coaches, even when controlling for team quality, have a particular aptitude or ineptitude for the rigors of tournament coaching. During the regular season, it can be hard to quantify what percentage of a coach’s success can be attributed to actual coaching ability and what should be chalked up to superior recruiting. There is no question that West Virginia’s Bob Huggins, for instance, is an excellent college coach. He recruits good players and those players go on to win games. But in the tournament, he has actually won fewer tournament games than the baseline amount expected, based on the seeds he’s obtained throughout his career. Watching individual games, it can be painfully obvious that certain coaches prepare their teams intelligently to extract every ounce of possibility out of late game situations or to adjust to new defenses, while other coaches and their players look completely lost. The brilliant Brad Stevens, the former Butler coach who is now working wonders with the NBA’s Boston Celtics, always seemed to put his college teams in the best position to succeed. No wonder then that his astronomical Performance Above Seed Expectation (PASE) of 1.510 is the highest of any coach with at least four appearances since the tournament’s format changed in 1985 (meaning that Stevens won a game and a half more per tournament than his seed levels would predict). Examining a coach’s PASE is thus an extremely helpful way of quantifying his ability to improve a team’s tournament odds.

PASE is measured as follows: A coach’s tournament wins minus his expected wins (based on the different seeds he’s had over the years) divided by the years he’s coached in the tournament (PASE=W-Ex. W/Year). Expected wins are simply a historical measure. We now have 31 years of data since the NCAA tournament expanded to 64 teams in 1985. Through last year, each seed has won an average number of games per tournament listed in the following table (remember, there are four of each seed; this table indicates what each one of a particular seed is expected to win):

Chart of wins per NCAA tournament seed

As a simple demonstration, if a coach had been in only one tournament as a 1 seed and won 4 games in that tournament, his PASE would be 0.65 ((4-3.35)/1). Of 2016 tournament coaches with at least four previous appearances in the big dance, the most successful are Michigan State’s Tom Izzo (0.906 PASE), Michigan’s John Beilein (0.740), Arizona’s Sean Miller (0.613), Kentucky’s John Calipari (0.601), Xavier’s Chris Mack (0.490), and Miami’s Jim Larranaga (0.405). The least successful coaches in this year’s tournament with four prior trips are Indiana’s Tom Crean (-0.279), Notre Dame’s Mike Brey (-0.308), Vanderbilt’s Kevin Stallings (-0.343), Temple’s Fran Dunphy (-0.351), Pittsburgh’s Jamie Dixon (-0.466), and Virginia’s Tony Bennett (-0.746).

Avoid Teams Over-Reliant on Home Court

While the commentator mantra that successful tournament teams must show that they can win on the road is usually overblown, there is one way that the RPI-reliant selection committee can often trip themselves up in seeding that is related to road and neutral records. The committee puts an outsized emphasis on “good” wins, defined as beating teams in the RPI top 50. At the same time, holding home court is a huge advantage. While the committee does list road victories in their “team sheets”, those sheets’ “good” win sections do not change based on location, despite the evidence that “[b]eating the 90th-ranked team on the road is about as difficult as beating the 50th-best team on a neutral floor, which is roughly as difficult as beating the 20th-best team on one’s home floor.” This means that certain teams with disproportionate numbers of narrow, home-court wins over excellent teams are particularly susceptible to being grossly over-seeded and unable to beat tournament-level competition away from their own campus’ distinct advantages. This year, Oregon State is particularly weak under this guideline. Their five best wins are the reason they are in the tournament, and all were at home. Meanwhile, they have no road wins against tournament teams. The NCAA tournament is played entirely at neutral sites, and teams that can play up to tournament level competition only at home are at a distinct disadvantage.

There’s a Method to Picking Underdogs

Even while being cognizant of the over-fitting problem described above, it is important to keep in mind particular qualities that increase the likelihood that a substantial underdog can score an upset. As ESPN’s Giant Killers series has repeatedly documented, teams that employ high-risk, high-reward strategies such as forcing turnovers, grabbing offensive rebounds, and successfully shooting three-pointers at high rates are particularly likely to get lucky and exceed their typical performance in a particular game. Conversely, favorites that play a more variable style are more susceptible to having an off night and losing to a lesser team. Look out for high seeds that are overly reliant on these same factors (especially three-pointers). The reason is simple: inferior teams that sometimes play very well and sometimes play very poorly are more likely to beat superior teams than inferior teams that always play at their middling level. A slow pace is also crucial. The fewer the possessions, the fewer opportunities for an underdog’s true quality to manifest itself.

Adjust for the Quality of a Particular Year

A final consideration that is particularly important this year is the overall quality of the seed lines in comparison to their historical average. The adjusted efficiency levels of the top teams, per Ken Pom, vary from year to year, and because these efficiency ratings are zero-sum (how efficiently a team plays in a particular game comes at the expense of another team’s efficiency), weaker top seeds mean stronger lower seeds. Multiplying each of the values in the table above by four will give you a historical baseline for the average number of games each seed line wins in a typical year. For instance, the average 13 seed wins 0.25 games per tournament, which means that on average one 13 seed will win a game per tournament. In years where the top seeds have weaker efficiencies than usual, seed performances are flatter; years with stronger top seeds have starker disparities in success. For instance, last year, five teams had pre-tournament efficiencies of 0.95 or higher (1.00 is the upper bound). That year, three 1 seeds made it to the Final Four and there were remarkably few overall upsets. The upset-filled 2011 and 2014 tournaments, however, had only one and two teams above 0.95 efficiency respectively.

Clicking through Ken Pom’s year-by-year rankings makes it abundantly clear that the strength of the top seeds is remarkably weak this year—weaker even than 2011. Going into this year’s tournament, only Kansas has an efficiency above 0.95, just barely clearing that mark at 0.9503. The 2, 3, and 4 seeds are also historically weak. This has been exacerbated by the disqualification of two of the nation’s best teams this year due to scandal—Louisville and SMU. Not all top seeds are created equally, and in a year where the numbers suggest a more egalitarian field, you should take greater chances.

*            *            *

Using these principles will not guarantee victory. In fact, they likely increase your odds of performing poorly by pushing you to take riskier odds. In order to win your pool, though, separating yourself from your many competitors by taking intelligent risks will best set you up for success. Then, just maybe, you’ll have an entertaining three weeks instead of finding a better way to spend your time


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s