Monday, November 14, 2016

NCAA Tournament Bracket: Summary and Thoughts

Looking back over my analyses of the Committee's at large selection decisions, the first thing I note is that my bracket simulation program matched the Committee's decisions with one exception.  Here is what the program came up with:

In clearly:  Iowa State

In but possibly out:  Texas Tech, NC State, Michigan, and Oklahoma State

Out but possibly in:  DePaul, Texas A&M, Loyola Marymount, Virginia Tech

The only difference between this and the Committee's decisions is that the Committee put Texas A&M in and Iowa State out.

I haven't done detailed analyses of NC State and Michigan because no one has raised serious objections about their having received at large selections.

Following my analyses, it still looks to me like Iowa State should have gotten an at large selection, in which case it also looks like Oklahoma State would have been out.  I'd say it looks to me like the Committee erred on that one.

Regarding all the other Committee decisions, based on the data and the history of how the Committee has applied the criteria to the data, I don't see any Committee decisions that were improper.  They could have made different decisions that also would have been proper, but the ones they made were within the discretion allowed them by the rules.

NCAA Tournament Bracket: Loyola Marymount and Virginia Tech "Out" -- Why?

This hopefully will be my last analysis of specific Committee decisions this year, examining the Committee's not giving at large positions to Loyola Marymount and Virginia Tech.  Some have argued that one or both of these teams should have gotten at large positions in the Tournament, neither got one.

Loyola Marymount

Here is Loyola Marymount's profile:

ARPI  0.5744
ARPI Rank  52
ANCRPI  0.5924
ANCRPI Rank  41
Top 60 Results Score  18
Top 60 Results Rank  49.5
Conference Standing  4
Conference ARPI  .5354
Conference ARPI Rank  9
Head to Head Results Score  -0.37
Common Opponents Results Score  -3.63
Common Opponents Results Rank  54
Last 8 Games Score  -12

Loyola Marymount met 0 "yes"; and 6 "no" standards.

"No" standards:

ARPI Rank >=48, and Top 60 Results Score <=172
ARPI Rank >=48, and Top 60 Results Rank >=47
ARPI Rank >=49, and Top 60 Results Rank >=46
ARPI Rank >=51, and Top 60 Results Rank >=41
Top 60 Results Rank >=45, and Last 8 Games Score <=-11
Top 60 Results Rank >=46, and Last 8 Games Score <=-10

Essentially, Loyola Marymount had two things going against it:

  • It did not have any really outstanding results.  Its best wins were against Pepperdine (#38), Long Beach State (#47), Ball State (#51), and San Diego (#71); and its best ties were against Texas Tech (#42) and Tulsa (#84)
  • It had some pretty poor results.  Its worst loss was to St. Mary's (#211) and its worst tie was against UC Santa Barbara (#113).
In both cases, Loyola Marymount was pretty far off from what historically has been expected for an at large selection.  Fundamentally, Loyola Marymount simply hadn't done enough to distinguish itself from the field of at large competitors.

Virginia Tech

Virginia Tech's situation is not as straightforward.  Here is its profile:

ARPI  0.5859
ARPI Rank  39
ANCRPI  0.5900
ANCRPI Rank  45
Top 60 Results Score  2304
Top 60 Results Rank  28
Conference Standing  9
Conference ARPI  .5928
Conference Rank  2
Head to Head Results Score  -1.67
Common Opponent Results Score  -4.68
Common Opponent Results Rank  58
Last 8 Games Score  -10

Virginia Tech met 8 "yes" standards; and 7 "no" standards

"Yes" standards:

ARPI Rank <=39, and Top 60 Results Score >=96
ARPI Rank <=41, and Top 60 Results Score >=384
ARPI Rank <=39, and Top 60 Results Rank <=45
ARPI Rank <=41, and Top 60 Results Rank <=35
ARPI Rank <=39, and Conference ARPI >=.5809
ARPI Rank <=41, and Conference ARPI >=.5839
ARPI Rank <=42, and Conference ARPI >=.5867
ARPI Rank <=41, and Conference Rank <=2

"No" standards:

Head to Head Results Score <=-1.37
ANCRPI >=42, and Conference Standing >=9
Conference Standing >=3.75, and Head to Head Results Score<=-1.33
Conference Standing >=5.25, and Head to Head Results Score <=-1.17
Head to Head Results Score <=-1.33, and Common Opponents Rank >=47
Head to Head Results Score <=-1.22, and Common Opponents Rank >=53
Head to Head Results Score <=-1.00, and Common Opponents Rank >=58

Thus, going for it, Virginia Tech had its ARPI rank, some good results against Top 60 teams, and its conference's strength.  Going against it, it had overall poor head to head results against Top 60 teams (notwithstanding that a couple of its results were good); its poor conference standing; and poor results against common opponents with other Top 60 teams.

Given Virginia Tech's unusual profile, with both strong positives and strong negatives, my best guess is that its #9 position in its conference standings was decisive.  In the context of its mixed record in relation to the Committee's criteria, #9 simply was too poor a conference standing, in comparison to its competitors, even though its conference was ranked #2 (#3 in the ANCRPI).

Opinion

Given the criteria and the Committee's history applying the criteria to teams' profiles, it's hard to see Loyola Marymount getting an at large selection.  As for Virginia Tech, either a "yes" or a "no" decision would be reasonable to me, although it would have been very difficult for the Committee to include the #9 team from a conference, particularly a conference that is not the #1 conference.

Saturday, November 12, 2016

NCAA Tournament Bracket: Texas Tech "In"; Why?

I've already written about possible rationales for the Committee's not giving at large selections to Princeton, Iowa State, and DePaul; and giving selections to Texas A&M and Oklahoma State.  So, what about Texas Tech, which got an at large selection?

Texas Tech

Texas Tech will be a good case for writing about a number of the Committee's criteria.  But first, here is Texas Tech's profile:

ARPI  .5838
ARPI Rank  42
ANCRPI  .6257
ANCRPI Rank  24
Top 60 Results Score  578
Top 60 Results Rank  34
Conference Standing  7.25 (average of regular season standing and conference tournament standing)
Conference ARPI  .6010
Conference Rank  1
Top 60 Head to Head Results  -0.50
Top 60 Common Opponent Results  -7.05
Top 60 Common Opponent Rank  60
Last 8 Games  -7

Texas Tech met 3 "yes" standards for getting an at large selection; and 1 "no" standard.  It thus presented the Committee with a profile the Committee hasn't had to decide on over the last 9 years.

"Yes" standards:

ARPI Rank <=42, and Conference ARPI >=.5867
ARPI Rank <=44, and Conference ARPI Rank =1
Conference ARPI Rank = 1, and Top 60 Head to Head Results >=-0.77

"No" standards:

ANCRPI Rank >=12, and Top 60 Common Opponent Results <=-6.95

It looks, from this analysis, like a major factor for the Committee, in giving Texas Tech an at large selection, was the strength of the Big 12, the #1 ARPI-ranked conference this year.  The strength of the Big 12 alone didn't do the trick, but rather contributed when paired with Texas Tech's ARPI rank and its head to head results against Top 60 teams.  At least, that's what it looks like to me.

Conversely, the Committee appears to have been willing to look past Texas Tech's comparative common opponent results in relation to other Top 60 teams, since those results were the poorest of the Top 60 teams, with no team having Texas Tech's profile on this issue having gotten an at large selection over the last 9 years.

NCAA Criteria.

There has been some discussion about why Texas Tech's results over its last eight games didn't outright disqualify it from an at large selection.  Before getting to that specific question, here's an overall review of the criteria the Committee is required to apply.

The NCAA has three primary criteria the Committee is required to apply.  In capsule form, they are the Adjusted RPI (which has a number of sub-criteria), head to head results, and results against common opponents.  It is left to each Committee member to decide how much weight he or she wants to give to each of these criteria.

For head to head results and results against common opponents, so far as I know the NCAA gives no further elaboration on exactly what those criteria mean.  For the Adjusted RPI criterion, on the other hand, the NCAA does elaborate.  That criterion includes:
  • overall record
  • Division I record
  • overall RPI rank
  • non-conference record and RPI rank
  • conference regular-season record and conference tournament results
If the Committee cannot make a decision based on the primary criteria, then there are two secondary criteria, both of which it must consider:
  • Results against teams already selected to participate in the field (including automatic qualifiers with RPI of 1-75)
  • Late season performance -- defined as the last eight games including conference tournaments (strength and results)
I've often wondered whether the Committee really follows the order of (1) see first if you can make a decision based on the primary criteria and then, only if you can't, (2) add the secondary criteria into the mix.  My experience in trying to do my own selections over the years was that it was almost always not possible to make decisions using only the primary criteria.

There have been some arguments, by DePaul advocates, that DePaul should have received an at large selection rather than Texas Tech.  Those arguments, in part, have focused on Texas Tech's last eight games, suggesting that Texas Tech's results in those games should have disqualified it, thus leaving a space for DePaul.  Looking at DePaul, however, may provide some insight into the Committee process, at least this year.  As my previous analysis of DePaul's profile indicates, it looks like their Non-Conference ARPI rank of #130 was the factor that cost them an at large position.  Since the ANCRPI as a factor applies when the Committee is considering the primary criteria, it is possible that the Committee had decided not to give an at large selection to DePaul before it even considered Texas Tech's last eight games.  If so, then a decision not to give Texas Tech an at large position probably would not have helped DePaul anyway.

In the balance of this post, I'll give some of my thoughts about the head to head results and results against common opponent criteria, as well as about the last eight games criterion.

Head to Head Results.

Using head to head results as a criterion is trickier than it might seem on first glance.  The easiest way to use it is to look to see if there was a head to head game where two teams are in contention with each other for one of the last at large positions.  Using one head to head game as a basis for choosing between two teams, however, leaves a lot to be desired.  This is because there always is the possibility of an A beat B, B beat C, and C beat A scenario, or of a comparable but more complicated scenario.  My conclusion, over the years, has been that this is too simplistic a use of this criterion -- although using it this way is within a Committee member's authority.

Rather than that approach, the approach my system takes for a team is to look at all its head to head results against Top 60 teams.  Essentially, this applies the Head to Head Results criterion to the entire field of potential at large teams.  My system assigns a value to each head to head result:  +2 for a win, -2 for a loss, +1 for an away tie, 0 for a neutral site tie, and -1 for a home tie.  The system then, after tallying a team's points for all of its head to head games, computes the average per game.  To me, this gives a good picture of where a team stands on this criterion against the field of the Top 60 teams.  For Texas Tech, this yielded a Head to Head Results score of -0.50.  Looking then at the standards, over the last 9 years, every team from the #1 ranked conference with a Head to Head Results score of -0.77 or better has gotten an at large selection.  In other words, Texas Tech's head to head results actually are pretty good given that it's from the #1 conference -- which means that a lot of its head to head games have been against very strong teams.  (Compare this to DePaul's Head to Head Results score of -0.60, with DePaul playing in a significantly weaker conference.)

It's important to note that my system for applying this criterion does not distinguish opponents based on their ranks.  A head to head win against a Top 60 opponent has the same value, no matter the opponent's rank.  On the other hand, when I get to the secondary criterion of "results against teams already selected," I do value results in relation the opponent's rank.

Results Against Common Opponents.

How to apply this criterion raises the same questions as the Head to Head Results criterion.  And likewise, rather than simply looking at two teams' results against their common opponents, I do a "Top 60" analysis for each team.  My system compares each Top 60 team to each other Top 60 team, looking at the results those two teams had against common opponents, and assigning values to those results using the 2, 1, 0, -1, -2 point system I described for valuing head to head results.  If I'm evaluating Team A in comparison to Team B, I then add up each teams' cumulative points related to their common opponent results.  This leaves the two teams either tied in their common opponent points, in which case their common opponent score in relation to each other is 0; or one has a net positive score and the other has a net negative score.  I go through the same process for Team A in relation to Team C, then Team D, and so on.  Once I've done this for all Top 60 teams with which Team A had common opponents, I determine Team A's total net common opponent score.  I then divide that score by the number of common opponent games Team A had with other Top 60 teams, to get an average score per common opponent game.  I do this for each Top 60 team, so I end with a table of average common opponent scores for all 60 teams.  (Actually, my computer does all of this.)

Although I'm pretty confident no Committee member does it this way, it provides me with a basis for evaluating Committee decisions, and looking for Committee patterns in relation to the Common Opponent Scores my system has produced.  These patterns then define the Common Opponent standards, which I then apply to determine whether the Committee's current decisions follow the same patterns as their decisions over the last 9 years.

Last Eight Games.

One of the arguments against Texas Tech getting an at large selection has been that it lost 7 of its last 8 games and that the Committee therefore failed to properly apply this criterion.

I have to say that this criterion, to me, is odd.  The criterion doesn't explicitly say what it's looking for -- good results, poor results, ...?  I always have interpreted the other secondary criterion, Results Against Teams Already Selected, as looking for positive results -- good wins and good ties.  And, my experience in applying it that way myself, in attempting to predict what the Committee will do, has suggested to me that's how the Committee uses it.  So, for the Last Eight Games criterion, I've doubted it is looking for good results, since that largely would duplicate what the Results Against Teams Already Selected is looking for.  My alternative has been to treat the Last Eight Games Criterion as looking for poor results.

There was a period of a few years when my system actually looked at teams' last eight games.  When I did that, I found two things:  (2) I couldn't find evidence of the Committee ever making a decision based on the last eight games; and (2) the process was very labor intensive from a programming perspective.  Given those two things, and not wanting to abandon the criterion altogether, I decided to create a surrogate for it:  poor results over the entire season.  That system assigns negative values to poor results, with the amounts of the values related to the rankings of the opponents.  I still find little evidence of the Committee making decisions based on this criterion surrogate, but I do find some; and the surrogate doesn't have the labor intensity demand.  Besides which, if there's a particular question about a team's actual last eight games, we always simply can get the team's record and look at it.

Using my system's surrogate criterion, Texas Tech came out with a -7 score for this criterion.  DePaul, on the other hand, came out with a -12.  In other words, DePaul's poor results were worse than Texas Tech's.  (On the other hand, Texas A&M, which got an at large selection, came out with -17.)

However you look at a team's poor results, an important point is that whether you're looking at the entire system or looking only at a team's last eight games, you need to be looking not only at the team's W-L-T record, but at the opponents the team was playing -- as the criterion says, "strength and results."  This is something the arguments against Texas Tech, due to their 1-7 record over their last 8 games, have not said much about -- the strength of its opponents in those games.

Looking at Texas Tech's last 8 games, they were:


  • W Iowa State #48
  • L Oklahoma State #56
  • L Baylor #65
  • L West Virginia #2
  • L Texas #78
  • L TCU #34
  • W Oklahoma #13
  • L West Virginia #2
If I'm a Committee member, and we've gotten to the secondary criteria (which I assume happened with Texas Tech), I'm asking myself, "Is this record sufficient to exclude Texas Tech from further consideration?"  And, from a disinterested outsider's perspective, I ask myself, "Is it clear that the Committee, by not disqualifying Texas Tech, failed to consider this criterion?"  From my perspective, the Committee could have considered the criterion and simply concluded that given the level of Texas Tech's opponents over the last 8 games, its record wasn't poor enough to disqualify it from getting an at large selection.  In fact, the Committee might have included that this record indicated Texas Tech legitimately was in the at large consideration mix.  This is not to say that the Committee made the right or wrong choice, only that I don't see that their decision on Texas Tech indicates there was something fundamentally wrong with how they did their job.


Wednesday, November 9, 2016

NCAA Tournament Bracket: Did an NCAA Data Error Affect the Committee's Decisions?

One of my big concerns, the last week of the season before the NCAA Tournament, is that the NCAA will end up with a data error that affects, in a significant way, the "Selection" reports the NCAA staff provides to the Women's Soccer Committee.  These reports are the primary resource the Committee has when it is making its seeding and at large selection decisions.  It's a concern to me because the NCAA does not publish these reports as soon as they have entered the final games data into their system.  Rather, they publish them a few days after the Committee has made its Tournament decisions.  Thus there's no way for anyone to identify any data errors and give the staff a chance to correct the reports before the Committee has made its decisions, or at least tip the Committee off as to what the error is and of what its effects are.

On reviewing the NCAA's just-released reports -- two days after the Committee made its decisions -- it turns out there was such an error.

The error itself may seem inconsequential, on first glance.  A SWAC semi-final tournament game, between Howard and Arkansas Pine Bluff, got reported into the NCAA's system as a 3-1 win by Arkansas Pine Bluff.  In fact, the game was a 0-0 tie, with Arkansas Pine Bluff advancing on Kicks From the Mark.  As a result of the error, the NCAA's data show Arkansas Pine Bluff with a record of 9-7-1 (0.5588), rather than the correct 8-7-2 (0.5294).  They show Howard with a record of 13-5-2 (0.7000), rather than the correct 13-4-3 (0.7250).  What this means, in RPI terms, is that Arkansas Pine Bluff's opponents got more strength of schedule credit than they should have; and Howard's opponents got less.  And this effect rippled through the two teams' opponents' opponents' ratings.

So, whom did these teams play, outside their SWAC games?


  • Arkansas Pine Bluff played Oral Roberts, Central Arkansas, UALR, and, Oklahoma State.
  • Howard played Cleveland State, Radford, Mt. St. Mary's , Longwood, George Mason, Princeton, Robert Morris, VMI, and Navy.
And what effect did the data error have on the ARPI ratings and rankings of key teams?  What follows shows, in the middle, an ARPI rank.  On the left is the team the NCAA, with its data error, gave that rank.  On the right is the team the correct data would have given that rank.

Notre Dame  8  Connecticut
Connecticut  9  Notre Dame

The Committee gave Notre Dame a #2 seed and Connecticut no seed.  Question:  With the correct data, would the Committee have done the seeding differently?  Specifically, would it have seeded Connecticut and/or given Notre Dame a lower seed.

Oklahoma  12  Virginia
Virginia  13  Oklahoma  12

The Committee gave Virginia a #3 seed and Oklahoma no seed.  The correct rankings would not have changed this.

Florida State  15  Duke
Duke  16  Auburn
Auburn  17  Florida State

The Committee gave Florida State and Duke #3 seeds and Auburn a #4 seed.  Question:  With the correct data, would the Committee have done the seeding differently?  Specifically, would it have seeded Auburn ahead of Florida State?

Marquette  24  Ohio State
Ohio State  25  Marquette

This ranking error is inconsequential.  Both were at large selections.

Wisconsin  35  Kent State
Kent State  36  Wisconsin

This ranking error is inconsequential.  Wisconsin got an at large selection and Kent State was an Automatic Qualifier.

Missouri  43  South Alabama
South Alabama  44  Missouri

This ranking error is inconsequential.  Missouri got an at large selection and South Alabama was an Automatic Qualifier.

Loyola Marymount  51  Ball State
Ball State  52  Loyola Marymount

This ranking error is inconsequential.  Neither team got an at large selection.

DePaul  53  UCF
Northeastern 54  DePaul
UCF  55 Oklahoma State
Oklahoma State  56  Northeastern

Northeastern was an Automatic Qualifier.  DePaul and UCF did not get at large selections.  Oklahoma got an at large selection.  In my opinion, it's very unlikely that UCF's moving up two positions would have made a difference.  Thus it's most likely that this ranking error was inconsequential.

Thus overall, the data error may have had an effect on seeds.  We'll never know.  It seems very unlikely it had an effect on at large selections.  Nevertheless, sooner or later, an error like this will have an effect.  The NCAA has had data errors in the past, at one of least might have been significant, affecting #1 seeds.  It would be wise for the NCAA to develop a system for vetting its final data.  Otherwise, sooner or later the NCAA is going to be in a very embarrassing position.

Tuesday, November 8, 2016

2016 NCAA Tournament Bracket: Texas A&M and Oklahoma State "In" and DePaul and Iowa State "Out" -- Why?

I promised I would take a look at this question.

In the bracket simulation my computer produced, it had Iowa State and Oklahoma State "in" and DePaul and Texas A&M "out."  Here's how they came out in terms of meeting "in" and "out" standards.  (If you don't know what I mean by "standards," read the second preceding post about Princeton where I give an explanation.)  Here's how these four teams came out on the standards in general:

Iowa State:  met 4 "yes" standards; and no "no" standards

Oklahoma State:  met 5 "yes" standards; and 5 "no" standards

DePaul:  met 2 "yes" standards; and 11 "no" standards

Texas A&M:  met no "yes" standards; and 3 "no" standards

Based on this, the obvious computer choice was Iowa State "in" and Texas A&M "out."  This left a decision between Oklahoma State and DePaul, when looking at these four teams.  Both of these teams, with some "yes" and some "no" standards, presented profiles the Committee hasn't seen over the last 9 years.  My approach with teams like this is that the team with best "yes" to "no" comparison gets the selection, so as between Oklahoma State and DePaul, the decision went to Oklahoma State.   But, what underlies those numbers?

Here are the teams' "profiles."  For a description of what the non-obvious factor titles are referring to, again go to the second preceding post about Princeton.


Factor Iowa State Oklahoma State DePaul Texas A&M
ARPI 0.5789 0.5709 0.5721 0.5702
ARPI Rank 48 56 54 57
ANCRPI 0.5899 0.5897 0.5156 0.6085
ANCRPI Rank 46 47 130 29
Top 60 Results Score 548 144 7200 365
Top 60 Results Rank 35 41 23 40
Conference Standing 6 6 2.5 6.75
Conference ARPI 0.6010 0.6010 0.5492 0.5754
Conference ARPI Rank 1 1 6 4
Top 60 Head to Head Results Score -1.00 -0.60 -0.60 -0.71
Common Opponent Results Score -3.68 -4.70 -3.32 -4.35
Common Opponent Results Rank 55 59 51 57
Last 8 Games (Season Poor Results) Score -8 -8 -12 -17

With those team profiles in mind, here are the "yes" and "no" standards the teams met:

Iowa State.  4 "yes" standards; and no "no" standards:

"Yes" standards:

Conference Standing <=6, together with Conference ARPI >=.5868

Conference Standing <=6.25, together with Conference ARPI >=.5930

Conference Standing <=6.75, together with Conference ARPI >=.5971

Conference Standing <= 6.75, together with Conference ARPI rank = 1

You'll note that the first three "yes" standards involve the same two factors.  What this means is that Iowa State met this paired factor "yes" standard by a significant margin.

Essentially, over the last 9 years, every team that had Iowa State's conference standing, conference ARPI, and conference rank profile has gotten an at large position in the Tournament.  Since Iowa State meets no "no" standards, the Committee's decision not to give Iowa State an at large selection was a break from past precedent.  It's possible that the Committee considered the size of the Big 12 conference -- 9 teams -- and felt that a 6th place finish was not sufficient to justify an at large selection, even if a 6.75 place finish would be sufficient for a team from a larger conference.  But see Oklahoma State.

And, it's worth noting that during Big 12 conference play, Iowa State beat Oklahoma State @ Oklahoma State.

Oklahoma State.  5 "yes" standards; and 5 "no" standards:

"Yes" standards:

Conference Standing <=6, together with Conference ARPI >=.5868

Conference Standing <=6.25, together with Conference ARPI >=.5930

Conference Standing <=6.75, together with Conference ARPI >=.5971

Conference Standing <= 6.75, together with Conference ARPI rank = 1

Conference ARPI rank = 1, together with Head to Head Results score >=-0.77

The first four of these match the four "yes" standards for Iowa State.  Iowa State and Oklahoma State were tied for 6th in the conference standings.  (For conference standings, my system averages the regular season standings with the conference tournament finishing position standings.)  With the two teams being tied, the Committee conceivably could have decided it would treat one team as the 5th in the 9-team Big 12 and the other as the 6th (they split the 5 and 6 positions in the regular season and went out in the Big 12 tournament quarterfinals).  And, whereas they may have felt the 6th place team was too far down in the ranks of a 9 teams conference for an at large selection, they might also have felt the 5th place team was high enough.

On the last of the "yes" standards, Oklahoma State's head to head results against Top 60 teams were better than Iowa State's.  The difference, however, was only slight:  Oklahoma State was 2-6-2 and Iowa State was 2-6-1.  This could have given the Committee a very slight basis for preferring Oklahoma State over Iowa State.

"No" standards:

ARPI Rank >=55, ANCRPI Rank >=31

ARPI Rank >=48, Top 50 Results score <=172

ARPI Rank >=55, Top 50 Results score <=380

ARPI Rank >=51, Top 50 Results Rank >=41

ARPI Rank >=55, Head to Head Results score <=-0.43

All of these standards indicate that given Oklahoma State's ARPI rank of #56, its team profile (apart from conference standing, conference rank, and Head to Head Results score combined with conference rank) historically would have been insufficient for an at large selection.

DePaul.  2 "yes" standards; and 11 "no" standards:

"Yes" standards:

Conference Standing <=2.5, Conference ARPI >=.5486

Conference Standing <=2.75, Conference Rank <=6

It's important to note that DePaul played #7 ARPI ranked and NCAA Tournament #2 seed Georgetown twice, winning the first time and tieing the second.  Before last year, the value of these results under the Top 60 Results factor might have been enough to secure DePaul an at large position.  Last year, however, the Committee's decisions on some teams meant that I had to eliminate some of the standards for "yes" decisions based on Top 60 Results.  In any event, based on the current standards (consistent with all decisions over the last 9 years), DePaul's results over Georgetown were not sufficient to meet any Top 60 Results standard for a "yes" at large selection decision.

"No" standards:

ANCRPI Rank >=129

ARPI Rank >=47, ANCRPI Rank >=115

ARPI Rank >=48, ANCRPI Rank >=87

ARPI Rank >=50, ANCRPI Rank >=78

ARPI Rank >=54, ANCRPI Rank >=76

ANCRPI Rank >=105, Top 60 Results Rank >=20

ANCRPI Rank >=102, Conference Standing >=2.5

ANCRPI Rank >=113, Conference Standing >=2.25

ANCRPI Rank >=102, Conference Rank >=6

ANCRPI Rank >=105, Conference Rank >=5

ANCRPI Rank >=91, Head to Head Results Score <=-0.40

With DePaul's Non-Conference ARPI Rank of 130, DePaul did not come close to having an ANCRPI Rank sufficient to meet the historic standards for an at large selection.  What this suggests, pretty clearly to me, is that from the Committee's perspective DePaul was done in by its poor non-conference results, notwithstanding its good results with Georgetown and its standing within the Big East.

Texas A&M.  0 "yes" standards; and 3 "no standards":

"No" standards:

ARPI <=0.5704

ARPI Rank >=55, Top 60 Results Score <=380

ARPI Rank >=55, Head to Head Results Score <=-0.43

It's important to note here that Texas A&M was close to avoiding these "no" standards, with its ARPI of 0.5702, its Top 60 Results Score of 365, and its Head to Head Results Score of -0.71.  Since Texas A&M got an at large selection, what this means is that once the dust clears I'll need to adjust my system's standards so that the above 0.5704 is reduced below 0.5702; the 380 is reduced below 365; and the -0.43 is reduced below -0.71.  Based on my experience, none of these is a significant adjustment.  Thus another way to look at Texas A&M would be as a team that meets no "yes" standards and likewise meets no "no" standards.  If that were the case, I would have had them in the same boat as Oklahoma State, with its 5 "yes" standards and 5 "no" standards (i.e., "yes" and "no" balancing out).  And, in that circumstance, one could consider that Texas A&M's 6.75 (top half) finish in the SEC with its 14 teams was a better finish than Oklahoma State's 6 (bottom half) finish in the Big 12 with its 9 teams.  This would be particularly true if the Committee relied on the Non-Conference Adjusted RPI in evaluating conference strength:

1  Pac 12  0.6077
2  Big 12  0.6051
3  ACC  0.6026
4  SEC  0.6005

This is part of the information the Committee receives.  They well might have referred to it, as the NCAA staff has suggested that the rationale for having the ANCRPI is that it gives a better picture of conference strength.  Based on my experience, these four average conference ANCRPIs are quite close, enough so to say that they indicate the four conferences are pretty much equal in average strength.  The Committee certainly may have felt that way.

My Opinion.  Based on all of the above, I think it's pretty clear that the Committee left DePaul "out" because of its very poor Adjusted Non-Conference RPI rank, which was far below the ANCRPI ranks of teams getting at large selections over the last 9 years.  People may disagree on that decision, but I believe it was a decision well within the Committee's discretion.

I also believe the Committee's decision on Texas A&M is defensible, with the Committee concluding (1) although it had no significant positives, it also had no significant negatives and (2) its finishing position in the SEC was better than the finishing positions of Iowa State and Oklahoma State in the Big 12, in relation to conference size.

As between Iowa State and Oklahoma State, however, I find it hard to find a good Committee rationale for selecting Oklahoma State.  Yes, the two were very close, but Iowa State looks like the choice more consistent with the criteria, to me.  I wouldn't say, however, that this was a major error by the Committee, rather a minor error and not something to be outraged about (unless you're Iowa State).

Monday, November 7, 2016

2016 NCAA Tournament Bracket: Princeton Part 2

There's another way to look at Princeton as an at large candidate, showing that sometimes it doesn't have to be as complicated as the ordinary process makes it.  Simply look at the ranks of teams Princeton beat, tied, and lost to.  Here's looking at it that way:

Wins:  Teams ranked 77, 86, 132, 133, 154, 165, 180, 222, 225, 288

Ties:  Teams ranked 46, 130, 138

Losses:  Teams ranked 2, 41, 66, 136

Do these suggest that Princeton is a #31 team?  Or do they suggest Princeton is somewhere in the range from 45 (where a tie was their best result) to 140 (where a loss was their worst result)?  Sometimes, it may help to use a simple "smell" test.

2016 NCAA Tournament Bracket: Princeton

One of the questions the Women's Soccer Committee probably is dealing with is what to do with Princeton:  Do they get an at large position in the Tournament or not?

Here is Princeton's profile:

ARPI: 0.5948

ARPI Rank:  #31

ANCRPI:  0.6442

Non-Conference ARPI Rank:  #18

Results Against Top 60 Teams Rank:  55

Standing Within the Ivy League:  5

Ivy League Average ARPI:  .5376

Ivy League ARPI Rank:  8

Head to Head Results Against Top 50 Teams Score and Rank:  -1.67/56

Common Opponent (With Other Top 60 Teams) Results Score and Rank:  -0.18/29

Last 8 Games Score:  -12

Several of these aspects of the profile are ones I've developed as surrogates for some of the NCAA's at large selection criteria.  I developed them in order to be able to create a computer program that will make objective decisions rather than relying on my own or others' subjective decisions.

  • Results Against Top 60 Teams assigns values to wins and ties against Top 60 teams, with those teams broken down into subgroups by ARPI rank.  The higher the opponent's rank, the higher the score for a win or tie against the opponent.  The points awarded are on a geometric scale, so good results against very highly ranked teams are very valuable.  Apart from the ARPI, I have found this profile aspect to be the most consistent with the Committee's decisions.
  • Head to Head Results assigns values to wins, ties, and losses in head to head games against Top 60 teams.  It then averages each team's scores for all head to head games based on the number of those games, to give an average score per head to head game.  This factor, unlike the Results Against Top 60 Teams, does not distinguish among the Top 60 teams based on their ranks.  A positive score is good and a negative score is bad.
  • Common Opponent Results identifies each common opponent a team had with another Top 60 team, compares the two Top 60 teams' results against that opponent, and assigns scores based on the comparison to the two Top 60 teams' results.  It then averages each team's scores for all of its common opponent games based on the number of those games, to give an average score per common opponent game.  A positive score is good and a negative score is bad.
  • Last 8 Games is a surrogate for an NCAA criterion.  The NCAA criterion looks at a team's results over the last 8 games.  I've found no evidence that the Committee actually uses that criterion, and I'm not sure what it is looking for, but my best guess is it's looking for poor results.  Based on that guess, as a surrogate my system looks at poor results over the course of the season, assigning negative values based on how poorly ranked the team is against which the poor result occurred.
When my system evaluates teams, it compares the aspects of their profiles, either as individual factors or as paired factors (such as ARPI rank combined with Top 60 Results rank), to the patterns of the Committee's decisions over the last 9 years.  The patterns are a set of "standards":
  • A "yes" standard means that over the last 9 years, a team that met that standard always got a "yes" decision from the Committee either for a particular seed or for an at large selection.
  • A "no" standard means that over the last 9 years, a team that met that standard always got a "no" decision from the Committee either for a particular seed or for an at large selection.
If a team this year meets both "yes" and "no" standards, what that means is the Committee has not met a team with this one's profile, over the last 9 years.  Princeton is in that position this year.

In terms of "yes" standards, the most obvious one for Princeton is ARPI Rank:  an ARPI rank of #34 or better is a "yes" standard.

Here are other "yes" standards that Princeton meets:
  • ARPI Rank <=35 and ANCRPI >=0.6168
  • ARPI Rank <=43 and ANCRPI Rank <=22
  • ARPI Rank <=37 and Conference ARPI >=.5322
  • ANCRPI Rank <=22 and Conference Standing <=6
  • ANCRPI Rank <=19 and Conference Standing <=5.75
  • ANCRPI Rank <=19 and Top 60 Common Opponent Score >=-2.21
  • ANCRPI Rank <=22 and Top 60 Common Opponent Score >=-2.10
It's important to note that all of Princeton's "yes" standards are based on either the ARPI or the ANCRPI.

In terms of "no" standards, here are the ones Princeton meets:
  • Top 60 Head to Head Results Score <=-1.37
  • Top 60 Results Rank >=54.5 and Conference Standing >2
  • Top 60 Results Rank>=45 and Last 8 Games Score <=-11
  • Top 60 Results Rank >=46 and Last 8 Games Score <=-10
  • Top 60 Results Rank >=51 and Last 8 Games Score <=-8
  • Top 60 Results Rank >=54 and Last 8 Games Score <=-5
  • Conference Standing >=3.75 and Top 60 Head to Head Results Score <=1.33
  • Top 60 Head to Head Results Score <=-1.25 and Last 8 Games Score <=-4
Based on this evaluation, the Committee could go either way with Princeton.  They have the RPI in their favor.  Except for that, however, their profile is against them.

This raises the question of how the Committee members look at the RPI.  Princeton is an example of a problem the RPI has.  When one calculates the Strength-of-Schedule portion of the RPI (effectively weighted at 50% of the RPI itself), the formula gives an effective weight of 80% to the opposing team's winning percentage and only 20% to its opponents' winning percentages.  In other words, what Team A contributes to its opponents' strength of schedule is mostly Team A's winning percentage.  A result of this is that, if one compares a team's ARPI rank to its rank in terms of what it contributes to an opponent's strength of schedule, those two ranks can be quite different.  I ran some calculations this morning, and the ranks of Princeton's opponents in terms of what they contribute to its strength of schedule are roughly 9 positions better, on average, than their ARPI ranks.  In other words, Princeton is a beneficiary of this flaw in the RPI formula.  I ran some other very crude calculations and came up with an RPI-flaw-corrected rating of #38.  I haven't done this for the ANCRPI, but presumably Princeton's ANCRPI rank likewise would be poorer if flaw-corrected.  It's thus possible that with corrections Princeton would not meet any "yes" standards.  Whether the Committee members understand this RPI flaw, and if they do are willing to take it into account, is something I don't know.

Given all of this, in my Bracket Simulation, I had to make a choice of what I think the Committee will do about the previously unseen profile that Princeton presents.  My choice was that the Committee will not give Princeton an at large selection.  I most certainly, however, could be wrong.

Sunday, November 6, 2016

2016 NCAA Tournament Bracket Simulation: End of Season -- FINAL -- Simulation

1 - #1 seed
2 = #2 seed
3 = #3 seed
4 = #4 seed

5 = unseeded automatic qualifier

6 = unseeded at large selection

7 = not getting at large selection, but next in line

Simulation isn't clear:

* = might do better than simulation says, based on numbers
** = might do poorer than simulation says, based on numbers

NCAA Seed or Selection Automatic Qualifier ARPI Rank Team
1 AQ 1 Stanford
1 AQ 2 WestVirginiaU
1 ** 3 SouthCarolinaU
1 ** 6 SouthernCalifornia
2 AQ* 4 FloridaU
2 5 NorthCarolinaU
2 AQ 7 Georgetown
2 9 NotreDame
3 ** 11 UCLA
3 ** 12 VirginiaU
3 ** 14 Clemson
3 AQ** 19 MinnesotaU
4 AQ 8 ConnecticutU
4 * 10 BYU
4 13 OklahomaU
4 AQ 17 FloridaState
5 AQ 33 Samford
5 AQ 35 KentState
5 AQ 38 Pepperdine
5 AQ 41 Harvard
5 AQ 43 SouthAlabama
5 AQ 46 Bucknell
5 AQ 47 LongBeachState
5 AQ 55 Northeastern
5 AQ 59 UNLV
5 AQ 72 FloridaGulfCoast
5 AQ 85 Charlotte
5 AQ 86 Monmouth
5 AQ 92 Liberty
5 AQ 93 Seattle
5 AQ 104 EasternWashington
5 AQ 120 Albany
5 AQ 121 NorthernKentucky
5 AQ 139 IllinoisState
5 AQ 141 Dayton
5 AQ 166 SouthDakotaState
5 AQ 174 SIUEdwardsville
5 AQ 214 StFrancis
5 AQ 220 HoustonBaptist
5 AQ 273 AlabamaState
6 15 Duke
6 16 Auburn
6 18 ArkansasU
6 20 NebraskaU
6 21 UtahU
6 22 Rutgers
6 23 KansasU
6 24 OhioState
6 25 Marquette
6 26 NorthwesternU
6 27 PennState
6 28 Memphis
6 29 ColoradoU
6 30 SMU
6 32 CaliforniaU
6 34 TCU
6 36 WisconsinU
6 37 SantaClara
6 ** 42 TexasTech
6 44 MissouriU
6 ** 45 NCState
6 48 IowaState
6 ** 49 MichiganU
6 ** 56 OklahomaState
7 * 31 Princeton
7 * 39 VirginiaTech
7 * 52 LoyolaMarymount
7 * 53 UCF
7 * 54 DePaul
7 * 57 TexasA&M
40 StJosephs
50 Rice
51 BallState

Saturday, November 5, 2016

Reports Summary: Games Through Friday, November 4

The post below this one is my updated NCAA Tournament bracket simulation, now based on games through Friday, November 4.  Here are a number of notes:

Seeds:  The #1 and #2 seeds are clearer, the #3 and #4 seeds not so clear.  There are unseeded teams that could receive seeds, the most likely of which is Florida State.  Connecticut also is a possibility for a seed, as are Duke, Rutgers, and Nebraska.  Of the seeded teams, the most likely to be bumped from seeding are Minnesota, Auburn, Arkansas, and Virginia.  This all is based on the simulation having North Carolina, Minnesota, and Florida winning their conference tournaments.

At Large "In":  There are some "iffy" teams getting at large selections.  Among -- but not all of -- them are:

#29 Princeton is relying almost exclusively on its ARPI and ANCRPI ranks.  Its best result is a tie against #48 Bucknell.  At #29, based on past history, it should be a lock for an at large selection, but based on a close review of its actual games, it looks undeserving of an at large selection.  Assuming Princeton wins its game tonight against Penn, Princeton will test the Committee's allegiance to the RPI.  If Princeton loses, it may make the Committee's job easier.

#45 NC State is relying mostly on some very good results, but also has a bunch of negatives.  Here, the test will be of the value the Committee assigns to very good results, especially in comparison to NC State's not-so-good position in the ACC.

#55 Oklahoma State has both positive and negative attributes.

#42 and #44 Texas Tech and Missouri have positive attributes, with Texas Tech also having a negative, but they are just on the right side of the border.

At Large "Out":  There are some other teams that might get at large selections.  Among -- but not all of -- them are:

#40 Virginia Tech has some positive attributes, but a poor position in the ACC.  Although it is not unheard of for the #9 team in a conference to get an at large selection, it is very difficult no matter how strong the conference.  And this year, the ACC is not the top conference according to the ARPI and the ANCRPI.

#39 and #51 Loyola Marymount and Michigan are relatively strong potential candidates for at large selections.  They don't have any strong positive attributes, but they don't have any strong negatives either.

#57 DePaul has a couple of positive attributes, but a bunch of negatives.

And, if one wants to drop down in the rankings to areas from which the Committee has not picked an at large selection over the last decade, #58 Texas A&M and #63 Baylor might receive some Committee consideration although each has negatives that probably are too strong.




2016 NCAA Tournament Bracket Simulation: Update Including Games Through Friday, November 4

1 = #1 seed
2 = #2 seed
3 = #3 seed
4 = #4 seed

5 = unseeded Automatic Qualifiers

6 = unseeded at large selections

7 = teams just missing at large selections

NCAA Seed or Selection Automatic Qualifier ARPI Rank Team
1 AQ 1 Stanford
1 AQ 2 WestVirginiaU
1 3 SouthCarolinaU
1 6 SouthernCalifornia
2 AQ 4 NorthCarolinaU
2 AQ 5 FloridaU
2 AQ 7 Georgetown
2 8 NotreDame
3 12 UCLA
3 13 Clemson
3 14 VirginiaU
3 17 ArkansasU
4 10 OklahomaU
4 11 BYU
4 15 Auburn
4 AQ 21 MinnesotaU
5 AQ 9 ConnecticutU
5 AQ 31 StJosephs
5 AQ 35 Samford
5 AQ 37 KentState
5 AQ 41 Pepperdine
5 AQ 43 Harvard
5 AQ 46 SouthAlabama
5 AQ 47 LongBeachState
5 AQ 48 Bucknell
5 AQ 54 UNLV
5 AQ 56 Northeastern
5 AQ 73 FloridaGulfCoast
5 AQ 82 EasternKentucky
5 AQ 84 Charlotte
5 AQ 87 Monmouth
5 AQ 91 HighPoint
5 AQ 93 Seattle
5 AQ 98 EasternWashington
5 AQ 103 StephenFAustin
5 AQ 106 Milwaukee
5 AQ 121 Hartford
5 AQ 137 IllinoisState
5 AQ 146 OralRoberts
5 AQ 223 StFrancis
5 AQ 231 ArkansasPineBluff
6 16 Duke
6 18 FloridaState
6 19 NebraskaU
6 20 Rutgers
6 22 UtahU
6 23 KansasU
6 24 Marquette
6 25 OhioState
6 26 NorthwesternU
6 27 PennState
6 28 Memphis
6 29 Princeton
6 30 ColoradoU
6 32 SMU
6 33 CaliforniaU
6 34 TCU
6 36 WisconsinU
6 38 SantaClara
6 42 TexasTech
6 44 MissouriU
6 45 NCState
6 49 IowaState
6 55 OklahomaState
7 39 LoyolaMarymount
7 40 VirginiaTech
7 51 MichiganU
7 57 DePaul
50 BallState
52 Rice
53 UCF
58 TexasA&M
59 CentralMichigan
60 Providence
61 NorthTexas
62 BostonCollege
63 Baylor