Picking the Perfect Bracket, First Round

While there are only two teams with an AdjEM over 30, the middle of this tournament is stacked.


Now that the bracket has been revealed, let’s apply the past analysis and pick the perfect bracket.  However, this is probably the worst year to try to do so.  A record number of teams have an adjusted efficiency margin (AdjEM) of 23 or higher (17), 22 or higher (24), and 20 or higher (26).  While there are only two teams with an AdjEM over 30, the middle of this tournament is stacked.

Let’s start with everyone’s favorite matchup: the 5 v. 12 seed.  Each of the four matchups are mid-major underdogs versus power favorites, which you may recall means a 28% success rate.  So we need to find and pick just one 5-12 upset this year.  While I was ready to pen UNC Wilmington into the second round, their draw with Virginia is brutal.  While the game is set to be played in Florida, a known hotspot for 5/12 seed upsets, over 800 miles away from the Virginia campus, a win by the Seahawks would be a record-setting upset.  No 12 seed has ever defeated a 5 seed with more than a 15 point differential between defenses.  UNC-W has an 18 point differential in defense.  Iowa St. is better than Nevada in both offense and defense so we go chalk there.

I see good arguments for both Princeton and Middle Tennessee.  Princeton has a location advantage, with Notre Dame traveling 440 miles and into a huge winter storm.  Princeton has a better defense but there is a significant 9 point difference in offense.  However, this would not be a unique upset. Nine 12 seed upsets had a better defense and over an eight point difference on offense.  So, it can be done if Notre Dame starts cold.  Middle Tennessee, on the other hand, has a better offense than the favorite, which over half of 12 seed upsets had.  And the difference on defense is only four points.  Working against the Blue Raiders is the location: Minnesota only has to travel 334 miles.  Only one other 5 seed lost closer to home since 2002.  Middle Tennessee also has the advantage of winning last year against a harder opponent with a weaker team.  Let’s side with Kermit Davis Jr. and see which new job he ends up at next year.

Moving on to possibly bigger upsets, we have 13, 14, and 15 seeds that’ll need to be unbalanced offense vs. defense against other unbalanced teams with an AdjEM over 5.5, over rank 105, in a relatively good conference, and with a defense ranking better than 145.  The defense qualifier removes Troy, Jacksonville St., Northern Kentucky, and Florida Gulf Coast.  The AdjEM qualifier takes care of North Dakota, Kent St., Winthrop (despite temptation to overlook this qualifier), and Iona.  Vermont is balanced playing a balanced Purdue (let alone that it is a terrible matchup considering that Vermont beats people inside, which Purdue has an incredible counter for) and Bucknell is balanced as well.

That leaves us with New Mexico St. and East Tennessee St.  Even though Bucknell and Vermont are better teams than NM St., they don’t meet the unbalanced profile that has historically succeeded.  Unfortunately for NM St., their opponent may be too balanced as Baylor only has a 8 rank difference between offense and defense.  On the other side of E Tenn St. is… an unbalanced Florida!  All qualifications are met for this matchup.  Just under half of all tournaments in the kenpom era have had either zero or one 13, 14, or 15 seed upsets.  So I feel confident in sticking with only one major upset, especially considering how good the top 26 teams in this tournament are.

On to 11 seeds.  6-11 games have proven tough to predict in the past.  Most 6 seed losers also had a loss over to a team ranked over 100 in kenpom on their resume.  Maryland meets that criteria (loss to 107th ranked Nebraska at home) and will be under-ranked against Xavier.  Easy choice there.  Creighton’s worst loss was to a team ranked 68.  After Maurice Watson’s injury, the Bluejays could be a trendy pick against a preseason top 25 Rhode Island team.  But Creighton has still played well and I don’t like Rhode Island’s weak 3 point and free throw shooting.  USC and Providence will have a nice game in the First Four but either of them winning would be one of the worst 11 seeds to win.  Kansas St. and Wake Forest are very intriguing options.  However, Wake Forest would have the worst defense of any 11 seed victor in the past decade and Cincinnati would set a new standard for a 6 seed losing in the first round with their worst regular season loss to a team ranked 67.

Now for 7-10 and 8-9 matchups, we should err on the side of the better team in adjusted efficiency margin (71% correct).  Look for AdjEM upsets in 7-10 matchups where the difference in rank is less than 8 and there is a match with the favorite on offense or defense.  8-9 upsets can be a bit more dramatic but are usually because the underdog is better in offense or defense.

Saint Mary’s and Wichita St. are easy favorites.  South Carolina is all defense and Marquette is underranked and all offense.  We’ll go with the AdjEM favorite but 10 seed Marquette.  The last 7-10 is Michigan versus Oklahoma St.  Here, we see an underdog matching the favorite in offense and only three ranks separating the teams.  We’ll go with the upset Oklahoma St., which will also provide us with a tournament champion losing to go with a recent trend and an overall AdjEM upset.

The 8-9 games feature two matchups with defensive minded teams (Vanderbilt-Northwestern, and Miami-Michigan St.).  We’ll look to the other two matchups for upsets.  Seton Hall played well in a tough Big East while Arkansas thrived in a sub-par SEC.  More importantly, Seton Hall is a tough, defensive team against an opposite focus, offensive Arkansas.  Lastly is Wisconsin versus Virginia Tech.  Wisconsin is ranked 21 spots better and only three 8-9, 7-10 matchups included a victory by an underdog worse than 20 spots.  However, the Badgers struggle defending the three and the Hokies shoot 40.3%, 9th best in the nation from three.

That gives us nine underdog wins by seed and by AdjEM.  It may be a good idea to consider some variance with Bucknell, New Mexico St., Princeton, Kansas St., and Rhode Island.  Picking all of them would equal 14 upsets by AdjEM and would match last year’s total, although that was two higher upsets than any other year.  Nine would be on the low end but is tied for the most likely number of upsets by AdjEM.


Winning the Coin Flip Matchups

Selecting the better team in adjusted efficiency margin for 7-10 and 8-9 matchups would have been the correct choice 7 out of 10 times since 2010.

The 7-10 and 8-9 matchups seem like they should be the hardest matchups to pick.  However, using predictive statistical analysis can turn a 50-50 coin flip into an easier, informed decision.

For instance, merely selecting the better team in adjusted efficiency margin for 7-10 and 8-9 matchups would have been the correct choice 71% of the time since 2010.  Of the teams that did not have such an advantage but won, nearly half were ranked within 5 spots of the favorite, and only three teams were beyond 20 spots.  Interestingly, 8-9 matchups showed a wide variance of upsets based on adjusted efficiency margin rank, ranging from a four to 36 rank difference, while 7-10 matchups averaged only a 5.6 average difference and only one team ranked more than eight away.  In fact, 8-9 upsets by adjusted efficiency rank tend to be because one team is better than the other in either offense or defense.  Whereas in 7-10 upsets by adjusted efficiency rank, the upsets occur when the underdog can nearly match the favorite in either offense or defense.

Furthermore, there have been two to three adjusted efficiency rank upsets in all but one year since 2010.  Like we said earlier, choosing the top ranked team is only 71% correct.  Picking the few teams to buck the trend is crucial to selecting a perfect bracket, hopefully now made easier knowing what type of matchup leads to an upset.  These trends will need to be monitored going forward.

A couple other notes about the first round before we move on.  Most upsets overall in the first round will come from teams with an adjusted efficiency margin over 10, usually about rank 70 or better.  However, there are usually about two upsets that come from even unlikelier teams.  All mid-major upsets have had a top 100 victory according to kenpom’s post-tournament stats except for Stephen F. Austin’s wins.  Usually, about 19-20 of the top 25 kenpom teams survive the first round.  It is also very unlikely that a conference tournament champion from a power conference does not win their first game; however, this has happened in four of the last six tournaments.  Finally, first round upsets typically  favor either offense or defense based teams

Next up, we’ll look at the second round in full as we narrow our bracket to 16.

Finding a March Madness Miracle

Dunk City highlights the excitement of an increase in 13, 14, and 15 seed upsets in the last couple of years.

In the last decade of tournaments, there has been at least one upset from a 13, 14, or 15 seed.  Of the 24 such upsets since 2002, half have occurred in the last five years.  What is especially remarkable is how consistent several data points are (based on kenpomeroy‘s pre-tournament adjusted efficiency stats) for these March Madness miracles.

All but two 2, 3, and 4 seeds that were upset had less than 14 rank difference between their adjusted offensive and defensive efficiency.  Raising that number to a 25 rank difference still leaves us with 17 of 24 (71%) favorites who have a significant difference in how they play depending on whether they have the ball or not.

Similarly, only one underdog in these matchups had less than an 18 rank differential (2014 Mercer) between offense and defense.  Only one underdog had a negative adjusted efficiency margin and defensive rank above 145 (2012 Norfolk St. for both).  On offense, only three 13 or 14 seed upsets had an adjusted offensive efficiency rank above 135.  Only five 13, 14, or 15 seeds had an adjusted efficiency margin of 5.5 or less and a rank worse than 105 .  Three of those teams were 15 seeds.

Also, only two teams came from conferences with an overall adjusted efficiency margin worse than negative ten, three teams total worse than negative 8.5, and five total worse than negative 6.

Our parameters for a miracle upset:

  • 2, 3, or 4 seed favorite with a 14-25+ rank difference in adjusted offensive efficiency and adjusted defensive efficiency.
  • 13, 14, or 15 seed underdog with a 18+ rank difference in adjusted offensive efficiency and adjusted defensive efficiency.
    • Positive adjusted efficiency margin overall, better than 5.5, and rank, better than 105.
    • Adjusted defensive efficiency should not be beyond rank 145.
    • Conference affiliation matters to a minor extent, but should be no worse overall than negative 6.

If the Shoe Fits…

11 seed upsets are becoming more frequent and are enjoying greater success.

While 12-5 upset picks are chic first round affairs, picking an 11 seed to make it beyond the second round is starting to become a trend.  George Mason kicked off the party with a Final Four run to remember in 2006, followed by VCU’s improbable 2011 campaign, and Dayton’s 2014 Elite 8 appearance that elevated Archie Miller to “coach rumored for every job opening” status.  Still, only 19 have made it to the Sweet 16 and 46 out of 128 (35.9%) have won a first round game.  However, eight 11 seeds have made it to the Sweet 16 since 2010, one more appearance than 6 seeds, and more than half (15 of 28) won first round games.  So 11 seed upsets are becoming more frequent and are enjoying greater success.

So let’s analyze the 11 seed in greater detail and try to figure out which teams are thriving in this position.  Since 2010, seven power conference teams won from the 11 slot while eight mid-majors won.  It should be noted that most mid-major winners were teams that found past success in the tournament (Gonzaga, VCU, Dayton, Wichita St., and Northern Iowa).  Yet, of the eight 11 seeds that advanced to the Sweet 16, five were power teams.

There is very little consistency  when looking at advanced stats.  Offensive-focused teams (like Marquette 2011 and Gonzaga 2016) are just as likely as defensive teams (see Old Dominion 2010 and Wichita St. 2016) to win.  And against opponents, there is a lot of noise with no clear trends in a particular type of 6 seed that is susceptible to upsets.  Unless you look at the schedule.

Of the 15 11-6 upsets since 2010, all but three occurred against a 6 seed with a bad loss on their resume.  These bad losses include those against a team ranked post-tourney in kenpom over 100 in adjusted efficiency margin. All three that did not have such a loss did lose against teams ranked post-tourney at 69, 73, and 80.  One of those was an SMU team that should have won on a blown call at the buzzer, an Ohio St. shocked by pesky, in-state combatant, Dayton, and Arizona team destroyed by a grossly under-seeded Wichita St.  12 of 15 upsets with a trend is good news for our perfect bracket efforts, and we can add in some x-factor to account for the others.

Finally, there is something else unique about the 11 seed in recent years.  The First Four has provided seven 11 seeds with a warm-up game to get tournament anxiety out of the way.  More than half (VCU, Tennessee, Dayton, and Wichita St.) won their first round game.  While the sample size is small, the early returns require careful consideration of an 11 seed fresh off a First Four victory.

Although their matchups with 6 seeds seem unpredictable and there is admittedly some “gut” feeling that needs to be applied, 11 seeds can and have found their way with 6 seeds who were victims of bad losses in the regular season.

As to the second round, six 11 seeds have defeated 3 seeds and two 11 seeds defeated 14 seeds en route to the Sweet 16 since 2010.  Of the victories against 3 seeds, Gonzaga and Washington were underrated and started the tournament with a better adjusted efficiency margin rank than their second round opponents; and Dayton, NC State, Marquette, and VCU were  offensive teams playing defense-first 3’s.

Going further to the Sweet 16, an 11 seed has only won six times ever.  Since 1990, an 11 seed victory has only occurred because of other upsets that left them playing a 7 or 10 seed.  A perfect bracket will likely not include an 11 seed to the elite 8 considering how unlikely and uncommon it is.  But if other factors lead one to choosing between an 11 and a 7/10, then maybe it is time for Cinderella to put on her slippers again.

When It Pays to Root for the Underdog

Last year, picking Little Rock over Purdue was a better bet than a choosing a clearly inferior Maryland squad to be upset.

The infamous 12 seed has a beaten a 5 seed in 27 of 31 years since the field expanded to 64 teams.  Over the last eight years, 12 seeds are an impressive 21-10-1 against the spread while also going 17-15 straight-up.  Despite the unlikely yet consistent success of 12 seeds, people still refuse to pick them to win.  This makes choosing a 12 seed a valuable proposition when trying to get ahead of others in your bracket pool.  But what makes these 5 seeds so vulnerable?

“The five-seed is where you lose natural geographic-area protection, because only the top-four line receives that protection,” explained Greg Shaheen, former senior vice president of the NCAA and organizer of the NCAA Division I men’s basketball championship. “Once you go to the five-[seed], you could be playing those games anywhere.”

So, has location actually mattered?  Of the 26 upsets since 2002, 16 involved 5 seeds that had to travel over 1,000 miles from campus to the first round location.  Lowering the mileage to 800 miles adds 3 more upsets.  Only one upset involved a 5 seed playing within 350 miles of their homes.  Notably, of the last 19 12-5 upsets, 11 occurred in the Mountain or Pacific time zones, and five from the state of Florida.

However, there have been a few 5 seed teams that traveled west and beat a team with regional home-court advantage.  In 2003, Wisconsin and UCONN won in Spokane versus Utah-based teams; in 2004, Syracuse beat BYU in Denver; in 2011, Kansas State defeated Utah State in Tucson; and last year, Maryland beat North Dakota State in Spokane.

The evidence may be surprising but highlights the importance of selection committee seeding when determining the top 16 teams.  This year, there are first round games in Sacramento, Salt Lake City, and Orlando.  Keep an eye on any 5-12 matchups in those locations.

The other commonly accepted tenet of a 12-5 upset is that it usually involves a mid-major over a power conference foe.  This is true – this type of upset has happened the most (10 times) since 2002.  But that is probably because it is the most common matchup between a 5 and a 12.  When looking at all of the possible chances (35), the fact is that 12 seed mid-majors upset 5 seed power teams only 28.6% of the time.

Compare that to other types of 5-12 matchups.  In the eight power vs. power 5-12 games since 2002, six were upsets.  In more recent years (2009-2016), underdog power teams have been matched up with mid-major favorites; the power teams were successful in three out of the four attempts.  That’s 75% for both types and something to watch going forward.

But we can’t stop there.  This is a perfect bracket we are talking about, after all.  Let’s dig deeper and look into Ken Pomeroy’s statistics since 2002 on these upsets to notice any trends.  Because we are trying to prognosticate what will happen, we will use the stats as they existed before each tournament began.

First, let’s see if the best 5 seeds are immune.  Looking at the adjusted Efficiency Margin rank (EM), we find that 5 seeds ranked in the top 20 account for 12 of the 26 victims.  So any 5 seed can be beat.  How about the other end, are there 12 seeds so bad that they never win?  Only five 12 seeds with an EM rank over 60 have won and none over 95.  That’s our floor then.  What about the gap in EM between the two teams?  Only half of the upsets came from teams ranked within 30 spots of the 5 seed.  The overall rank in EM does not seem to give us much information.

How about offensive versus defensive teams?  A 12 seed with a better adjusted defensive efficiency accounts for just six upsets – the same number of upsets won by a 12 seed that gave up 5-10 more points per 100 possessions and those won by a 12 seed that gave up 10 or more points per 100 possessions.  Looking at adjusted offensive efficiency provides better results.  Nearly half of the upsets (12) involved an underdog that was better on offense than the favorite.  Looking at both stats combined, we find that an underdog that was either better at offense or defense than the favorite won a remarkable 18 times.  Of the other 8 upsets where the underdog was not better in either offense or defense, the remaining 12 seeds were within 5 points per 100 possessions in at least one category and three within a point.  In sum, of the last 26 12-5 upsets, 21 (80.1%) involved a 12 seed that was better or within a point per 100 possessions in either adjusted offensive or defensive efficiency.

Based on the above analysis, here are our theories that give the best chance of a 12-5 upset (in order of what I think is most compelling):

  1. 12s seed with better or nearly equivalent adjusted offensive or defensive efficiency are more likely to take down a 5 seed.
  2. Upsets should have 12 seeds with an EM ranked 60 or better.
  3. An underdog team from a power conference is likely to win.
  4. The 5 seed is playing more than 350 miles away from their campus.
  5. The game is in a Mountain/West region or Florida.

There is still some wiggle room here but applying all of these to last year’s tournament would have shown Maryland and Indiana to be secure because their opponents were over EM rank 60 (80 and 105, respectively), Baylor at risk due to Yale’s superior defensive efficiency and EM ranked 50, and Purdue at risk playing in Denver against Little Rock with an EM ranked 59.  As stated above, any 5 can be beat.  But the worse the 12, the less likely the upset.  Little Rock still probably should not have won that game but this analysis shows why picking Purdue was riskier than a choosing a clearly inferior Maryland squad.