Friday, November 29, 2019

THE #4 SEEDS: WHO TO PICK FROM A MESS OF TEAMS?

As happens sometimes, the Committee had a mess of teams from which to pick the #4 seeds, with a lot to go on for which ones to pick.

The first group of candidates were the candidates for #3 seeds that didn't get them:

#10 Brown, which met 3 "yes" #4 seed standards and 0 "no."  I discussed Brown in my initial report on the Committee's decisions.  My conclusion was that the Committee probably believed the Ivy League and its members were significantly overrated and that Brown got swept up in that.

#14 Duke, which met 0 "yes" and 0 "no" #4 seed standards.

#18 Washington, with 0 "yes" and 0 "no" #4 seed standards.

#20 Texas Tech, with 0 "yes" and 0 "no" #4 seed standards.

The second group of candidates were teams that my computer didn't have in the #3 seed candidate group:

#13 Rutgers, with 2 "yes" and 1 "no" #4 seed standards.

#21 Texas A&M, with 0 "yes" and 0 "no" #4 seed standards.

A third group, which I'm including because the Committee gave one of them a #4 seed, are those that my computer had just missing a #4 seed:

#22 Penn State, with 0 "yes" and 1 "no" #4 seed standards

#23 Florida, with 0 "yes" and 1 "no" #4 seed standards.

#25 Michigan, with 0 "yes" and 1 "no" #4 seed standards.

Of the teams that had been candidates for #3 seeds, the Committee gave #4 seeds to two of them:  Washington and Texas Tech.  The Committee did not give seeds to two of them:  Duke and Brown.  I covered Brown in my earlier Ivy League report.  I'll come back to Duke in my comments at the end of this report.

Of the other five teams, Texas A&M met 0 "yes" and 0 "no" standards.  I'll come back to it, too, in my comments at the end of this report.

For the remaining four, here are their "yes" and "no" standards:

#13 Rutgers

Here are the "yes" #4 seed standards Rutgers met:

Non-Conference RPI and Common Opponents Score (Standard #44)

The "yes" #4 seed score for this standard is >=48.8539.

Rutgers' Non-Conference RPI was .6780 (#3) and its Common Opponents Score was 2.25 (#11), combining for a score of 49.0300 for this standard.  If Rutgers had not received a #4 seed, there are 6 teams, of 192 seeds over the last 12 years, that were in the "yes" #4 seed category but no longer would be.

Non-Conference RPI Rank and Common Opponents Rank (Standard #54)

The "yes" #4 seed score for this standard is <=42.6

Rutgers' Non-Conference RPI Rank was #3 and its Common Opponents Rank was #11, combining for a score of 42.6 for this standard.  Thus Rutgers was right at the edge of the "yes" score.  If it had not received a #4 seed, there are 3 teams, of 192 seeds, that were in the "yes" #4 seed category but no longer would be.

Here is the "no" #4 seed standard Rutgers met:

Top 50 Results Score and Last 8 Games (Standard #63)

The "no" #4 seed score for this standard is <=-13,056.

Rutgers' Top 50 Results Score was 636 (#34) and its Last 8 Games score was -8, combining for a score of -16,964 for this standard.  Rutgers' poor results were a tie in the 56-100 opponent rank area (v #70 Princeton, away); a tie in the 101-150 rank area (v #138 Nebraska, away); and a loss in the 101-150 rank area (v #111 Maryland, home).  With Rutgers getting a #4 seed, when I change this standard to be consistent with its getting that seed, there are 51 teams, of the 720 Top 60 teams over the last 12 years, that previously were in the "no" #4 seed category that now will move out of that category.

#22 Penn State

Here is the "no" #4 seed standard Penn State met:

RPI Rank and Top 50 Results Score (Standard #28)

The "no" #4 seed score for this standard is <=3,777.

Penn State's RPI Rank was #22 and its Top 50 Results Score was 501 (#41), combining for a score of 3,683 for this standard.  With Penn State getting a #4 seed, there are 7 of the 720 Top 60 teams over the last 12 years that previously were in the "no" #4 seed category that now will move out of that category.

#23 Florida

Here is the "no" #4 seed standard Florida met:

Non-Conference RPI Rank and Common Opponents Score (Standard #53)

The "no" #4 seed score for this standard is <=-0.9000.

Florida's Non-Conference RPI Rank was #25 and its Common Opponents Score was -1.95 (Rank #39.5), combining for a score of -1.100 for this standard.  If Florida had gotten a #4 seed, there are 17 of the 720 Top 60 teams over the last 12 years that previously were in the "no" #4 seed category that would have moved out of that category.

#25 Michigan

Here is the "no" #4 seed standard that Michigan met:

Top 50 Results Score and Last 8 Games (Standard #63)

The "no" #4 seed score for this standard is <=-13,506.

Michigan's Top 50 Results Score was 1,053 (#31) and its Last 8 Games score was -7.  Michigan's poor results were a tie in the #56-100 opponent rank area (tie with #90 Indiana, home); and a loss in the #151-200 rank area (#159 Marquette, away).  If Michigan had gotten a #4 seed, there are 17 of the 720 Top 60 teams over the last 12 years that previously were in the "no" #4 seed cagetory that would have moved out of that category.

Comments About the Committee's Decisions on #4 Seeds

The Committee's historic patterns did not give a clear indication of which teams should get #4 seeds.  Most of the teams' profiles weren't highly persuasive one way or the other.  Here are the Committee's #4 seeds, in order if I assume their order fits with standard bracket placement practices:  Washington (#13), Rutgers (#14), Texas Tech (#15), Penn State (#16).  So, what is the likely thinking behind the Committee's picks?

It seems likely that the Committee first selected Washington and Rutgers.  Washington finished tied with Southern California for 3rd in the Pac 12 conference.  The Pac 12 was the top ranked conference, and Stanford, UCLA, and Southern California already had received a #1 and two #2 seeds, making Washington a strong candidate for at least a #4 seed.  For Rutgers, its #3 Non-Conference RPI Rank could have given the Committee something to hang its hat on, especially combined with Rutgers' results against Common Opponents with other Top 60 teams (#11), even though Rutgers had some poor results.

If that's right, how did the Committee decide on the last two #4 seeds?

For a question like this, the first thing I look at is teams' Top 50 Results scores.  When I looked here, it was clear the Committee did not use teams' good results against Top 50 opponents as a basis for its decision on these seeds.  Here are the Top 50 Results Scores my system assigned the teams competing for the last two spots:

Duke  28,132
Florida  10,131
Texas Tech  3,824
Michigan  1,053
Brown  531
Penn State  501

Since Texas Tech and Penn State got the #4 seeds, this was not the basis for decision.

Based on reviews I've done over the past few years, something else I look at is "conference balancing."  This refers to spreading affirmative decisions around among the top conferences.  When I look at the teams' conference regular season and tournament (for those that have them) finishing positions, here as how my computer placed all of the seeded teams, plus some additional teams that were next in order in the conferences.  The placement is based on the average of the team's regular season and tournament finishing positions.  The conferences are in the order of their average RPI Rank.  For each seeded team, preceding its name I've put the team's seed position based on standard seeding practice.  I've highlighted in yellow which teams won their conference regular season and conference tournament competitions:

#1 Pac 12:

#1 Stanford  #1 = 1
#5 UCLA  #2  = 2
#7 Southern California  #3.5 = 3.5
#13 Washington  #3.5 = 3.5

#2 ACC:

#2 North Carolina  #1 in regular season + #1 in tournament = 1
#3 Virginia  #3 + #2 = 2.5
#4 Florida State  #2 + #3.5 = 2.75

NC State  #5 + #3.5 = 4.25
Louisville  #4 + #6.5 = 5.25
Duke  #6.5 + #6.5 = 6.5

#3 SEC:

#6 South Carolina  #2 + #1 = 1.5
#9 Arkansas  #1 + #2 = 1.5

Florida  #4.5 + #3.5 = 4
Vanderbilt  #4.5 + #3.5 = 4
Texas A&M  #3 + #6.5 = 4.75

#4 Big 12:

#10 Oklahoma State  #1 + #3.5 = 2.25
#15 Texas Tech  #2 + #3.5 = 2.75
#11 Kansas  #5 + #1 = 3

TCU  #6 + #2 = 4

#5 Ivy League:

Brown  #1 = 1

#6 Big 10:

Michigan  #2.5 + #2 = 2.25

#16 Penn State  #4 + #1 = 2.50
#14 Rutgers  #2.5 + #3.5 = 3
#12 Wisconsin  #1 + #6.5 = 3.75

#7 West Coast:

#8 BYU  #1 = 1

From this point of view, for the last two #4 seeds -- Texas Tech and Penn State -- it looks like the Committee was looking at the top conferences, and within them at teams' finishing positions in their conferences.  Plus, within conferences it was giving first preference to teams that finished #1 in the regular season competion and to teams that finished #1 in the conference tournament.  If teams (1) did not finish #1 in the conference regular season or tournament and (2) finished in an average position #4 or lower, those teams did not get seeds.  Thus, using Texas Tech as an example, the decision to seed it suggests the Committee might have decided, "Texas Tech finished #2 in its conference; Duke finished #6.5 in its conference.  As between Texas Tech and Duke, we've got to seed Texas Tech before we seed Duke."  This approach makes sense, seems reasonable to me, and is a possible explanation for all of the #4 seed decisions.

The other thing I think this analysis does is, it makes it virtually certain that the Committee believed the Ivy League was not the #5 conference, notwithstanding what the RPI said.  Indeed, it seems the Committee believed the Ivy League was not the #6 conference, either, but rather was the #7 conference (behind both the Big 10 and the West Coast Conference), at best.

In this light, given the difficulty of distinguishing among teams when the Committee gets to #4 seeds, this seems like a reasonable explanation for what the Committee did and also seems like a reasonable basis for its decisions.


Wednesday, November 27, 2019

THE #3 SEEDS, INCLUDING: DUKE, WASHINGTON, TEXAS TECH, OR WISCONSIN?

In my article on the #2 seeds, my computer had Arkansas, Oklahoma State, and Kansas competing for the last #2 seed, but they didn't get it.  That would put them in line for #3 seeds, and in fact the Committee gave them #3 seeds.

Among the three of them, Arkansas met 3 standards for "yes" #3 seed and 0 "no" standards.  Kansas and Oklahoma State each met 0 "yes" and 0 "no" standards.  This suggested Arkansas should get the top #3 seed and left undefined where to put Kansas and Oklahoma State.  If the Committee's placement of the three teams in the bracket is an indication of how the Committee ranked them, they likewise saw Arkansas as the top #3 seed (#9 seed overall).  Next came Oklahoma State (#10) and then Kansas (#11).  It's worth noting that Arkansas' Top 50 Results Score was 17,641, Oklahoma State's was 3,062, and Kansas' was 2,410, which matched their bracket placement order.

This left the Committee needing to pick the last #3 seed.  According to my computer, the candidates were Duke, Washington, Texas Tech, and Brown, all of whom met 0 "yes" standards for a #3 seed and 0 "no" standards.  The Committee, however, picked Wisconsin, which met 0 "yes" standards and 1 "no."

In my first article on the Committee's bracket decisions, I discussed Brown and the Ivy League, so I won't cover that here.

Wisconsin

As stated, Wisconsin met no "yes" #3 seed standards and 1 "no."  Here is the "no" standard it met:

Conference Standing and Conference RPI (Standard #71)

The "no" #3 seed score for this standard is <=3.1142.

In figuring a team's conference standing, I consider both its regular season standing and its conference tournament finishing position (as the Committee is required to do for at large selections).  In my system, for the conference tournament, the champion finishes in the #1 position, the runner-up in #2, the losing semi-finalists in #3.5 (average of #3 and #4), the quarter-finalists in the #6.5 position (average of #5 through #8), and so on.  I then average the team's regular season standing and its conference tournament position to get its Conference Standing.  Wisconsin was #1 in the Big Ten regular season standings, but exited the conference tournament in the quarter-finals giving it a #6.5 position.  The average of these two positions is #3.75, so in my system that's its Conference Standing.

The Big Ten's Conference RPI was 0.5566.  Combining this with Wisconsin's #3.75 Conference Standing gave it a score for this standard of 3.1053.  This meant it would not get a #3 seed according to the Committee's pattern.  Since Wisconsin did get a #3 seed, I will need to change the "no" #3 seed score for this standard.  When I make the change, there will be 15 teams, of the 720 Top 60 teams over the last 12 years, that previously were in the "no" #3 seed category but that now will not be in that category.  This is a small change.

General Comments

There is no obvious reason, in the data, for the Committee to give Wisconsin the last #3 seed.  The other teams vieing for the seed all had better Top 50 results -- Duke with a score of 28,132, Washington with 4,400, and Texas Tech with 3,824, whereas Wisconsin had only 1,047.

On the other hand, it is possible that the Committee gave little weight to the Big Ten's Conference RPI, which could have made Wisconsin a contender for a #3 seed.  The Committee's doing this would have been consistent both with its not giving Brown a seed (or Yale, Harvard, or Columbia at large selections) and with its giving BYU the last #2 seed -- all reflecting that the Committee was doubtful about the reliability of Conference RPI's and Conference Ranks at least in this area of the conference ratings and rankings.

And, going even further, it is possible that the Committee believed the Big Ten should get at least one position among the top 12 seeds.  If the Committee felt that the Ivy League really was a weaker conference than the Big Ten or the West Coast Conference, this would have made the Big Ten the #5 conference and the West Coast the #6.  All of the better ranked conferences plus the West Coast already had seeds among the top 12 -- the ACC and Pac 12 with 3 each, the SEC and Big 12 with 2 each, and the WCC with 1.  If this is how the Committee looked at it, it could have made sense to them give the Big Ten the last #3 seed, settling on Wisconsin as the best Big Ten candidate.  This is the best explanation I can see for the Committee's decision.




Tuesday, November 26, 2019

THE #2 SEEDS, INCLUDING: ARKANSAS, KANSAS, OKLAHOMA STATE, OR BYU?

For the #2 seeds, according to my computer there were 3 clear teams to get the seeds:

South Carolina met 14 "yes" standards and 1 "no" standard.  I'll cover the "no" standard below.

UCLA met 3 "yes" standards and 0 "no" standards.

Southern California met 3 "yes" standards and 0 "no" standards.

The Committee and my computer agreed, all 3 of these got #2 seeds.

As was the case with the #1 seeds, however, there was not an obvious fourth #2 seed.  According to my computer there were 4 candidates for that spot:

Kansas, meeting 0 "yes" standards and 1 "no."

Oklahoma State, meeting 0 "yes" standards and 2 "no."

Arkansas, meeting 0 "yes" standards and 6 "no."

BYU, meeting 0 "yes" standards and 10 "no."

Doing a detailed review of each team and its "no" standards is tedious, but in this case it proved instructive on what the Committee seemed to think was important, so I'll go through a full review.

But first comes South Carolina's "no" #2 seed standard.

South Carolina

South Carolina met 1 "no" standard for a #2 seed:

Non-Conference RPI Rank (Standard #4)

The "no" #2 seed standard for this is >=35.

South Carolina's Non-Conference RPI Rank was #38.  According to the Committee's pattern, this meant it would not get a #2 seed.  South Carolina also, however, met 14 "yes" standards, so it gave the Committee a profile it hasn't seen over the last 12 years.

When I change the standard to >=39, which I will have to do to accommodate South Carolina's having gotten a #2 seed, there will be 33 teams that historically were in the "no" #2 seed group that now will not be in that group.  That 33 is out of the 720 Top 60 teams over the last 12 years.

One of the things I measure is how important each standard appears to be in the Committee's decisions.  I measure this by counting how many teams for each decision category -- #1 seeds, #2 seeds, ..., at large selections -- have a "yes" or a "no" for that standard.  If no or few teams have a "yes" or "no," it means teams in the decision category have a wide range of scores for the standard.  That, in turn, means that the Committee can't be making much use of that standard when making decisions.  This is the case with teams' Non-Conference RPIs and Ranks.  Ordinarily, the numbers say, they just aren't that important.  (This is true, notwithstanding that Florida State's #1 Non-Conference RPI Rank may have played a part in its getting the fourth #1 seed.)  And, in the case of South
Carolina, the Committee was willing to give it a #2 seed notwithstanding its #38 NCRPI Rank.

Kansas

Kansas met 0 "yes" standards and 1 "no."  Here is its "no" standard:

RPI Rating and Top 50 Results Score (Standard #17)

The "no" #2 seed score for the standard is <=254,481.

Kansas' RPI was .6414 and its Top 50 Results Score was 2,410, giving it a score for this standard of 254,467.  This is just below the "no" standard.  If Kansas had gotten a #2 seed and I were to revise this standard to accommodate that, there would have been only 1 team out of the 760 Top 60 teams over the last 12 years that would have moved out of the "no" #2 seed category.  Thus it would have been a minimal change from the Committee's pattern as to this standard.

Oklahoma State

Oklahoma State met 0 "yes" and 2 "no" standards.  Here are its "no" standards:

RPI Rating and Top 50 Results Score (Standard #17)

This is the same as Kansas' "no" standard.  The "no" #2 seed score for the standard is <=254,481.

Oklahoma State's RPI was .6382 and its Top 50 Results Score was 3,062, giving it a score for this standard of 253,860.  If Oklahoma State had gotten a #2 seed and I were to revise this standard to accommodate that, there would have been 6 of the 760 Top 60 teams over the last 12 years that would have moved out of the "no" #2 seed category.

RPI Rating and Top 50 Results Rank (Standard #18)

The "no" #2 seed score for this standard is <=3.6208.

Oklahoma State's Top 50 Results Rank was #24.  This plus its RPI Rating gave it a score for this standard of 3.6154.  If it had gotten a #2 seed, there would have been 6 teams of 720 over the last 12 years that would have moved out of the "no" 2 seed category.  These 6 probably overlapped with the teams for the previous standard.

Arkansas

Arkansas met 0 "yes" standards and 6 "no" standards.  Here are its "no" standards:

Last 8 Games (Standard #13)

As I explained in my first article in this series about the Committee's decisions, I use this standard to measure a team's poor results.  Although the factor the Committee is instructed to consider is teams' results over the last 8 games (both game results and strength of opponents), I instead consider poor results over the course of the entire season (primarily for computer programming reasons).  The Committee's factor isn't explicitly for poor results, but it already has a number of factors related to good results, so my guess is that the Committee mostly is looking for poor results under this factor.

The "no" #2 seed score for this standard is <=-7.0.

Arkansas' score for this factor was -8.0.  This score came from a tie with a team in the 56-100 RPI Rank range (home v #57 Georgia), a loss to a team in the 56-100 range (home v #88 Oklahoma), and a tie with a team in the 151-200 range (neutral site v #157 Minnesota).

If Arkansas had gotten a #2 seed, there would have been 66 teams of 720 over the last 12 years that would have moved out of the "no" #2 seed category.

RPI Rank and Last 8 Games (Standard #36)

The "no" #2 seed score for this standard is <=-2.3774.

Arkansas' RPI Rank was #6.  Combined with its Last 8 Games Score, its score for this standard was -2.5000.  If Arkansas had gotten a #2 seed, there would have been 5 teams of 720 over the last 12 years that would have moved out of the "no" #2 seed category.

Non-Conference RPI and Last 8 Games (Standard #46)

The "no" #2 seed score for this standard is <=59.8006.

Arkansas' Non-Conference RPI was .6455 (Rank #14).  Combined with its Last 8 Games Score, its score for this standard was 59.1358.  If Arkansas had gotten a #2 seed, there would have been 23 teams of 720 over the last 12 years that would have moved out of the "no" #2 seed category.

Non-Conference RPI Rank and Last 8 Games (Standard #55)

The "no" #2 seed score for this standard is <=-2.3333.

Arkansas' Non-Conference RPI Rank was #14.  Combined with its last 8 Games Score, its score for this standard was -5.7413.  If Arkansas had gotten a #2 seed, there would have been 153 of 720 teams over the last 12 years that would have moved out of the "no" #2 seed category.

Common Opponents Score and Last 8 Games (Standard #90)

The "no" #2 seed score for this standard is <=-1.9167.

Arkansas' Common Opponents Score was 3.86 (Rank #9).  Combined with its Last 8 Games Score, its score for this standard was -2.2143.  If Arkansas had gotten a #2 seed, there would have been 11 of 720 teams over the last 12 years that would have moved out of the "no" #2 seed category.

Common Opponents Rank and Last 8 Games (Standard #91)

The "no" #2 seed score for this standard is <=-2.8000.

Arkansas' Common Opponents Rank was #9.  Combined with its last 8 Games Score, its score for this standard was -4.3333.  If Arkansas had gotten a #2 seed, there would have been 68 of 720 teams over the last 12 years that would have moved out of the "no" #2 seed category.

General Comments About Arkansas

It is possible to interpret the above as suggesting that poor results killed Arkansas' #2 seed chances.  It had a good RPI Rank, a decent Non-Conference RPI Rank, and a good Common Opponents Rank, but they weren't good enough to balance out its poor results.  On the other hand, poor results, by themselves, are a fairly "unused" factor in the Committee's seeding decisions so far as I can tell.

Thus although Arkansas may have been hurt by its poor results, there may have been other factors at play, as discussed at the end of this article.

BYU

BYU met 0 "yes" standards and 10 "no" standards.  Here are the "no" standards it met:

RPI Rating and Top 50 Results Score (Standard #17)

The "no" #2 seed score for this standard is <=254,481.

BYU's RPI Rating was .6350 and its Top 50 Results Score was 4,514, combining to give it a score for this standard of 254,084.  It's worth noting this was a little less than Kansas' score and a little more than Oklahoma State's score.  With BYU getting a #2 seed, there will be 4 of 720 teams over the last 12 years that will move out of the "no" #2 seed category.  This is a very small change.

RPI Rating and Top 50 Results Rank (Standard #18)

The "no" #2 seed score for this standard is <=3.6208.

BYU's Top 50 Results Rank was #18.  Combined with its RPI Rating, this gave it a score of 3.6118 for this standard.  With BYU getting a #2 seed, there will be 7 of 720 teams over the last 12 years that will move out of the "no" #2 seed category.  This is a small change.

RPI Rating and Conference RPI( Standard #20)

The "no" #2 seed score for this standard is <=1.2005.

The West Coast Conference's RPI was .5338 (Rank #7).  Combined with BYU's RPI Rating, this gave BYU a score of 1.1688 for this standard.  With BYU getting a #2 seed, there will be 148 of 720 teams over the last 12 years that will move out of the "no" #2 seed category.

RPI Rating and Conference Rank (Standard #21)

The "no" #2 seed score for this standard is <=3.7363.

The West Coast Conference's Rank was #7.  Combined with BYU's RPI Rating, this gave BYU a score of 3.6356 for this standard.  With BYU getting a #2 seed, there will be 77 of 720 teams over the last 12 years that will move out of the "no" #2 seed category.

RPI Rank and Conference RPI (Standard #31)

The "no" #2 seed score for this standard is <=3.2112.

BYU's RPI Rank was #12.  Combined with the WCC's Conference RPI, this gave BYU a score of 3.0725 for this standard.  With BYU getting a #2 seed, there will be 135 of 720 teams over the last 12 years that will move out of the "no" #2 seed category.

RPI Rank and Conference Rank (Standard #32)

The "no" #2 seed score for this standard is >=22.1.

BYU's score for this standard was 26.7.  With BYU getting a #2 seed, there will be 46 of 720 teams over the last 12 years that will move out of the "no" #2 seed category.

Non-Conference RPI Rating and Conference RPI (Standard #41)

The "no" #2 seed score for this standard is <=1.6318.

BYU's Non-Conference RPI Rating was .6368 (Rank #19).  Combined with the WCC's Conference RPI, this gave BYU a score of 1.5976 for this standard.  With BYU getting a #2 seed, there will be 77 of 720 teams over the last 12 years that will move out of the "no" #2 seed category.

Non-Conference RPI Rank and Conference RPI (Standard #50)

The "no" #2 seed score for this standard is <=3.1842.

BYU's Non-Conference RPI Rank was #19.  Combined with the WCC's Conference RPI, this gave BYU a score of 3.0418 for this standard.  With BYU getting a #2 seed, there will be 26 of 720 teams over the last 12 years that will move out of the "no" #2 seed category.

Conference RPI and Common Opponents Rank (Standard #80)

The "no" #2 seed score for this standard is <=3.1441.

BYU's Common Opponents Rank was #7.  Combined with the WCC's Conference RPI, this gave BYU a score of 3.1320 for this standard.  With BYU getting a #2 seed, there will be 12 of 720 teams over the last 12 years that will move out of the "no" #2 seed category.

Comments on the Choice of BYU for the Last #2 Seed

Looking at the above details, it seems like Kansas would have been first choice for the last #2 seed, followed by Oklahoma State.  For Arkansas, its poor results seem to be a problem.  And for BYU, its conference seems to be a problem.

Notwithstanding this, I think the following considerations are possible and reasonable explanations for the Committee's decision to give the fourth #2 seed to BYU:

1.  The Committee may not have held the West Coast Conference's RPI and RPI Rank against BYU.  It's possible, and maybe likely, that the Committee is feeling cautious about how much emphasis it should place on Conference ratings and ranks, especially in relation to the top mid-major conferences.  Given what happened with the Ivy League this year, as discussed in my first report on the Committee's decisions, this would make sense.  Of BYU's 10 "no" standards, 8 of them involved the WCC's rating and rank.  If the Committee decided not to hold these against BYU, BYU's other two "no" scores were barely less than the "no" #2 seed scores for the standards and would have put BYU, Kansas, and Oklahoma State looking about equal.

2.  During the season, BYU defeated Kansas at Kansas.  Oklahoma State beat Kansas at Kansas in their conference regular season game, but lost to Kansas at a neutral site in their conference tournament game.  The Committee might have seen this is placing BYU ahead of Kansas and Oklahoma State.

3.  Regarding Arkansas, I think the most likely explanation for it not getting a #2 seed is its poor results.  They weren't horrible results, but in the context of what's expected from a #2 seed, they weren't good.

4.  BYU was undefeated, with an 18-0-1 record.  It's tie was against Santa Clara at Santa Clara.  Santa Clara beat almost #1 seed and clear #2 seed UCLA at UCLA.  All rating systems have trouble with undefeated teams, particularly rating systems like the RPI that don't take goal differential into consideration.  It's clear an undefeated team is very good, but rating systems can't tell you whether and if so, by how much, it's even better than the rating systems say.

In this context, the Committee's giving the last #2 seed to BYU seems reasonable.

Monday, November 25, 2019

THE #1 SEEDS, INCLUDING: FLORIDA STATE OR UCLA?

With the quarter-finals coming up, the higher seeds are the home teams.  This is a significant advantage, which puts the Women's Soccer Committee's decisions on the #1 seeds among its most important.

According to my NCAA Tournament bracket formation program, this year three of the #1 seed decisions were easy:

Stanford met 37 "yes" standards for getting a #1 seed and 0 "no" standards.  The Committee's pattern is for a team meeting any 1 of those 37 "yes" standards to get a #1 seed.

North Carolina met 31 "yes" standars and 1 "no" standard.  I'll come back to that "no" standard.

Virginia met 10 "yes" standards and 0 "no" standards.

Thus my computer said those three teams would get #1 seeds, in the order of 1.1 Stanford, 1.2 North Carolina, and 1.3 Virginia -- and they did.

Being curious about how North Carolina met a "no" standard, I checked it out.  They got their "no" on Standard #4, their Adjusted Non-Conference RPI Rank.  The "no" score for that standard is >=25, which means that according to the Committee's pattern, a team with a NCRPI rank of #25 or poorer will not get a #1 seed.  North Carolina's NCRPI rank was #28.

Over the last 12 years, of the 720 Top 60 teams, there have been 33 teams that, under the >=25 standard, would have been assured of "no" #1 seed but that, once I've changed the standard to >=29 to accomodate this year's #1 seed for North Carolina, no longer will have that "no" status.  All of those 33 teams have been outside the #1 to #7 to #14 RPI rank levels, which have been the limits beyond which the Committee has not gone for #1 and #2 seeds respectively.  In other words, the Committee previously has not seen a team with a #28 NCRPI rank that otherwise looked qualified for a #1 seed.  In this context, North Carolina's getting a #1 seed indicates to me that the Committee does not place much weight on a team's NCRPI rank, in the #1 seed selection process.

After Stanford, North Carolina, and Virginia, there was not an equally obvious fourth #1 seed.  Rather, there were two slightly "flawed" possibilities:  Florida State with 1 "yes" standard and 2 "no" standards; and UCLA with 0 "yes" standards and 1 "no" standard.  Here is some detailed information on these two.

Florida State

Florida State met 1 "yes" and 2 "no" standards.

Here is its "yes" standard:

Non-Conference RPI Rank and Conference RPI (Standard # 50)

This standard combines these two factors, each weighted at 50%.  The "yes" #1 seed score for this standard is >=4.3058.

Florida State's Non-Conference RPI Rank was #1 and the ACC's RPI was .5942 (ranked #2), producing a standard score of 4.3273.  If the Committee had not given Florida State a #1 seed, I would have to change the standard score to reflect the Committee's decision.  Of the 48 teams getting #1 seeds over the last 12 years, this changed standard score would have taken only 1 team out of the "assured" #1 seed group.  Thus it would have been a minimal change from the Committee's pattern.  On the other hand, Florida State's score is the 3rd best score for this standard of those 48 teams and also the best score this year.

Here are Florida State's "no" standards:

Top 50 Results Score and Head to Head Score

I'm going to assume for this that readers have read my description, in the immediately previous article, of what these factors are.

The "no" #1 seed score for this standard is <=15,929.

Florida State's Top 50 Results Score was 1,924, which for a #1 seed is not very good -- of the 48 teams getting #1 seeds over the last 12 years, only 2 had a poorer Top 50 Results Score.  Florida State's Head to Head Score was 0.57.  Together they produce a score for this standard of 11,638.  With Florida State getting a #1 seed, I will need to change this standard to accommodate that.  Of the 720 Top 60 teams over the last 12 years, this means that 49 of those teams that previously were assured as "no" #1 seeds now no longer will have that status.

Top 50 Results Rank and Head to Head Score

The "no" #1 seed score for this standard is <=.9503.

Florida State's Top 50 Results Rank was #28, to go with its Head to Head Score of 0.57.  Together, these produced a standard score of .7179.  Since I'll need to change the "no" standard score to accommodate Florida State's #1 seed, this means that 38 of the 720 Top 60 teams over the last 12 years that previously fell within the "no" #1 seed category for this standard now no longer will have that status.

Florida State Comments

Florida State had a winning percentage of 0.7500, which is not great for a #1 seed.  It had only the #28 Top 50 Results Rank, and it had a not outstanding Head to Head Score.  Yet its RPI Rank was #4 and its Non-Conference RPI Rank was #1.  How could that be?

Florida State's very high RPI and Non-Conference RPI Ranks were due to its strength of schedule.  Its non-conference opponents' average RPI rank was 35.  The next team in order was Florida at 46, followed by DePaul at 64.  UCLA was 55 positions farther down at 90.  Looking at conference and non-conference games and opponents' unadjusted RPIs, Florida State played the 2nd toughest schedule (just behind Duke).  UCLA played the 5th toughest.

Thus there's a suggestion, in the Committee's giving Florida State the 4th #1 seed, that the Committee was rewarding Florida State for its having exposed itself to a very tough non-conference schedule.

UCLA

UCLA met no "yes" standards and 1 "no" standard.

Here is its "no" standard:

Non-Conference RPI Rank and Head to Head Score

The "no" #1 seed score for this standard is <=.9818.

UCLA's Non-Conference RPI Rank was #18 and its Head to Head Score was 0.71, producing a score for this standard of .9365.

If the Committee had given UCLA a #1 seed, in adjusting the standard score to accomodate this, 6 of the 720 Top 60 teams over the last 12 years that were in the assured of "no" #1 seed category under this standard would have moved out of that category.

General Comments about the Florida State or UCLA Decision

Given how the Committee placed Florida State and UCLA in the bracket, it's reasonably clear that Florida State was the 4th #1 seed (effectively seeded fourth) and that UCLA was the 1st #2 seed (effectively seeded fifth).  Because Florida State got the #1 seed, it gets home field advantage against UCLA.

Home field is a significant advantage.  In RPI rating terms, it is worth an 0.0150 adjustment in the rating difference between two opponents.  In terms of win likelihoods, this means that:

With Florida State being the home team, based on the RPI rating difference between Florida State and UCLA, as adjusted for home field advantage, Florida State's win/tie/loss likelihoods are 62.8%/13.2%/24.0%.

At a neutral site, Florida State's win/tie/loss likelihoods are 57.8%/13.5%/28.6%.

With UCLA as the home team, Florida State's win/tie/loss likelihoods are 47.4%/15.5%/37.2%.  In this scenario it is more likely the game will be an outright loss or a tie (going to Kicks from the Mark) than that Florida State will outright win.

Thus the Committee's decision to give Florida State the 4th #1 seed, rather than UCLA, was a big one with a potentially significant impact on the outcome of he Florida State v UCLA game.  In my opinion, based on the above analysis it was a reasonable decision.  On the other hand, based on the above analysis, giving UCLA the 4th #1 seed also would have been a reasonable decision.

Saturday, November 23, 2019

WOMEN'S SOCCER COMMITTEE BRACKET DECISIONS - BROWN, HARVARD, YALE, COLUMBIA, AND THE IVY LEAGUE


The Ivy League had the #5 average RPI among the conferences.  Yet Ivy League champion Brown, with a #10 RPI rank, did not get a seed from the Women's Soccer Committee.  The League had 3 other teams among the RPI's Top 50:  Yale at #37, Harvard at #41, and Columbia at #49.  Yet none of those teams got an at large selection.  Why?

I'll start with Brown, then move on to the League, and then come back to Yale, Harvard, and Columbia.

No Seed for Brown

According to my bracket formation program, Brown had factors in its profile that met 3 "yes" standards for getting a #4 seed and 0 "no" standards.  What a "yes" standard means is that over the last 12 years, every team with a profile factor meeting that standard got at least a #4 seed.  Thus Brown's not getting a #4 seed was a break from the Committee's historic patterns.  The question is, "Why?"

(Hereafter, when I refer to the Committee's pattern or use the word "historically," I'm meaning over the last 12 years, 2007 through 2018.)

Here are the "yes" standards Brown met, with some information about each of them:

Conference Standing and Conference RPI (Standard #71)

This standard looks at the team's standing within its conference put together with the conference's RPI.  (For all combination standards, like this one, the two factors are weighted 50% each.)  Under this standard, Brown's #1 conference standing combines with the Ivy League's 0.5629 average RPI to give Brown a score of 3.8707.

Under this standard, the "yes" standard for a #4 seed is a score of 3.8587.  In other words, historically, teams with a score of >=3.8587 always have gotten at least #4 seeds.  Since Brown's 3.8707 is a better score, history says Brown would get a #4 seed.  But it didn't.

How big a change is this from the historic pattern?  A way to measure that is to ask how many teams that historically were "assured" at least a #4 seed by the standard now will not be in the assured group because the standard has to be changed to incorporate Brown not getting a seed.  The answer is, only 1 team, out of 192 seeded teams over the 12-year period.  Thus this is a small change from the Committee's pattern.

Conference RPI and Head to Head Score (Standard #78)

"Head to Head Score" needs an explanation.

One of the factors the Committee is required to consider for at large selections and surely considers for seeds is Head to Head Results.  It's up to the Committee to decide exactly what this means and how to apply it.  Hopefully, it's not simply a matter of, "How did A and B do when they played each other," as that would miss out on the "A beat B," B beat C," and "C beat A" scenario.

The way I come up with a Head to Head Score is to look at every head to head game between Top 60 teams.  Each team in a head to head game gets a positive or negative score based on the game outcome -- win, loss, tie -- and the game location -- home, away, neutral.  I then determine each Top 60 team's average head to head score per head to head game played.  This creates a picture of how a team did in all of its games against Top 60 opponents.

The "yes" score for this combined standard is >=14.3133.

The Ivy League's RPI was .5629.  Brown's Head to Head Score was 1.60 per game (total head to head score of 8, averaged over 5 Top 60 games).  These gave Brown a score for this standard of 14.5461, which indicates Brown would get at least a #4 seed.

Since Brown didn't get at least a #4 seed, how big a change is this from the Committee's pattern?  Out of the 192 seeded teams over the last 12 years, 15 of them ( a little over 1 per year) would have been "assureds" under the 14.3133 standard but won't be once I've revised the standard to incorporate the Brown "no seed" decision.

Head to Head Score and Last 8 Games Score (Standard #88)

Last 8 Games Score needs an explanation.  There are two "secondary" criteria the Committee can consider in at large selections, and almost certainly considers for seeds.  One is results against teams already selected with an RPI rank of 75 or better (for which I use the surrogate of good results -- wins or ties -- against Top 50 teams, since that is what really seems to matter to the Committee).  The other is results over the last 8 games (winning record and strength of opponents).  I believe this second one primarily is looking for poor results towards the end of the season, and as a surrogate I look at poor results -- losses or ties -- against lower-ranked teams over the course of the entire season.

Brown's Last 8 Games Score, in my system, was -2.  This was for a loss against a team ranked in the 56 to 100 range.

The "yes" #4 seed score for this combined standard is >=8.000.  Brown's score was 10.8000, which indicates Brown would get at least a #4 seed.

Since it didn't, how big a change is this?  In the move of this standard from 8.000 to a new score to accomodate Brown's 10.8000 not getting a seed, 24 teams out of the test period's 192 would not have been "assured" a #4 seed -- 2 per year.  This is the biggest of the three changes.

Discussion

Based on the Committee's patterns, Brown would have gotten at least a #4 seed.  And, it would have been a possible #3 seed, for which it met no "yes" standards but also no "no" standards.  But it didn't get seeded.

Is there a likely explanation for this?

One possible explanation could have to do with Brown's good results against Top 50 teams, which seems to be a powerful factor in the Committee's decisions.  I have a scoring system for these results -- wins and ties -- that is very heavily weighted towards good results against very highly ranked teams.  The question the Top 50 Results Score addresses is, "At how high a level have you demonstrated you are able to compete?"

Brown's Top 50 Results Score, in my system, was 531.  Of the 16 seeded teams, 15 of them had a better score than Brown, most of them a lot better.  The only seeded team Brown outscored was Penn State with 501.  Further, other teams in the bracket that didn't get seeded included Virginia Tech (548), NC State (11,664), Texas A&M (1,249), Florida (10,131), and Michigan (1,053).  Indeed, Tennessee with 14,493 and Georgia with 14,490 didn't even get into the Tournament.  Thus the Committee may have concluded that Brown's best results simply weren't good enough for it to be seeded.

There's also another possible explanation when you consider this:  For two of the "yes" standards Brown met, one of the factors was the Ivy League's RPI.  For the other "yes" standard, one of the factors was Head to Head Score in games against Top 60 teams, and three of the five Top 60 teams Brown played were #37 Yale, #41 Harvard, and #49 Columbia (2 wins and 1 tie).  It's possible, especially when combined with Brown's comparatively weak Top 50 Results Score, that the Committee considered the Ivy League's RPI, and its teams' RPI ratings and ranks, to be suspect and that this affected its decision about Brown.

The Ivy League

As the non-conference part of the season moved along, it became apparent that the Ivy League was going to have a very high non-conference winning percentage and that this was going to translate into its ranking as the #5 RPI conference.  This caused the raising of some eyebrows, including mine, since the League's rank as a conference since 2013 had  been 8 (2013), 8, 11, 8, 7, and 9 (2018).  Also, although not unheard of, it is unusual for a conference to move up 4 rank positions in one year.

Last February, I wrote an article, So Your Conference Wants More of Its Teams in the NCAA Tournament: How Much Attention Should Its Teams Pay to the RPI Formula?  Answer: A Lot.  In that article, I described in detail a non-conference scheduling strategy by which a mid-major conference's teams, working together, could make large improvements in their RPI ratings and ranks with no change in the teams' true strengths.  A  big part of the strategy was scheduling non-conference opponents that would allow all of  the conference's teams, and thus the conference as a whole, to have a very high non-conference winning percentage.  This strategy would take advantage of the current structure of the RPI formula.  For an example, I took one of the top mid-major conferences and replaced its 2018 schedule with an alternative schedule implementing the non-conference strategy I described in the article.  That included having the conferences' teams play significantly weaker opponents than they'd actually played and selecting those opponents with a view to their contributions to opponents' strengths of schedule.  I then computed what the conference teams' RPI ratings and ranks would have been with this alternative schedule plugged into the 2018 season.  The alternative schedule dramatically improved all of the conference teams' ratings and ranks and turned the conference into the #1 RPI conference.  Same teams, but a different and significantly weaker schedule.

Coming back to the Ivy League this year, when I saw their high rank as a conference, I immediately looked at the Ivy teams' non-conference schedules to see if they fit the "high non-conference winning percentage" part of the strategy I'd described.  They did.  The result was that the League's overall non-conference winning percentage was .7391, giving it the third highest winning percentage among all the conferences behind only the Pac 12's record .8151 and the ACC's .7789.  The fourth highest, well behind the Ivy League, was the SEC's .6778.

To give an idea of how the Ivy teams did this, one can look at the winning percentages of the non-conference teams they played as compared to those of other conferences.  In this comparison, the League ends up having had the #19 ranked opponents' winning percentages, at .4707.  And, those opponents' schedules weren't strong -- their winning percentage against their opponents ranked the League #10 in opponents' opponents' winning percentage.  Thus the League as a whole achieved very good results against very weak opponents, who themselves played relatively weak opponents.  And due to the RPI's structure, this greatly improved the Ivy League's teams' RPI ratings and ranks.

As an additional check, I looked at the average RPI rank of the League's non-conference opponents, as compared to other conferences' opponents' average RPI ranks.  This had the League in 17th place among the conferences:
Conferences Non Conference Opponents Average ARPI Rank
BigTen 111
BigTwelve 112
ACC 117
SEC 119
BigEast 125
PacTwelve 129
WestCoast 140
BigWest 143
Colonial 143
American 154
MidAmerican 168
MissouriValley 168
BigSky 169
OhioValley 170
AtlanticTen 170
AtlanticSun 174
ConferenceUSA 174
Ivy 179

Simply put, the League's non-conference scheduling, put together with the structure of the RPI, resulted in the League's teams being overrated.

A detailed look at each of the Ivy teams' schedules helps show this:

#10 Brown itself had a decent schedule, with an away tie against #21 Texas A&M and a home win against #16 Hofstra.  After that, its next best result was a home win against #107 Providence.

#37 Yale's best non-conference result was a home win against #104 Fairfield.

#41 Harvard's best non-conference result was a home win against #122 UMass.

#49 Columbia's best non-conference result was a home win against #169 St. Joseph's.

#70 Princeton's best non-conference result was an excellent home tie against #13 Rutgers.  After that, its next best result was a home tie against #69 Villanova.  Princeton's tie with Rutgers and Brown's win against Hofstra and tie with Texas A&M were the League's only positive results against Top 50 teams.

#106 Dartmouth's best non-conference result was an away win over #150 Quinnipiac.

#108 Penn's best non-conference result was a home tie with #104 Fairfield.

#228 Cornell's best result was a home tie with #211 Binghamton.

All of this information says that the RPI, through a combination of its structure and how the League's teams did their non-conference scheduling, overrated the Ivy League and its teams.  And all of this information was available to the Committee.

In this context, I think it's likely that the Committee did not believe the Ivy League really was the #5 ranked conference and that the Committee believed the League's teams were overrated.  The Committee's decisions suggest that this was the case.

There's one other item to consider related to the Ivy League.  It has to do with the number of non-conference games the League's teams play, as compared to the number of conference games they play.

To show why this is important, I took the 2018 season and eliminated all of the non-conference games.  To make the illustration clearer, I also eliminated all of the conference tournament games.  I then computed teams' RPI ratings, including the three elements of their ratings, and their RPI ranks.  This produced the following as the Top 21 teams:
Team Conference RPI Element 1 RPI Element 2 RPI Element 3 RPI Unadjusted URPI Rank
Milwaukee Horizon 1.0000 0.5000 0.5000 0.6250 1
CentralConnecticut Northeast 1.0000 0.5000 0.5000 0.6250 2
Stanford PacTwelve 0.9545 0.5000 0.5000 0.6136 3
TennesseeMartin OhioValley 0.9500 0.5000 0.5000 0.6125 4
Georgetown BigEast 0.9444 0.5000 0.5039 0.6121 5
Samford Southern 0.9444 0.5000 0.5000 0.6111 6
FloridaAtlantic ConferenceUSA 0.8500 0.5389 0.5156 0.6108 7
StLouis AtlanticTen 1.0000 0.4478 0.5265 0.6055 8
NorthCarolinaU ACC 1.0000 0.4500 0.5178 0.6044 9
TexasState SunBelt 0.9000 0.5044 0.5034 0.6031 10
CentralArkansas Southland 0.9091 0.5000 0.5000 0.6023 11
BowlingGreen MidAmerican 0.9091 0.5000 0.5000 0.6023 12
StJosephs AtlanticTen 0.8000 0.5589 0.4909 0.6022 13
NorthTexas ConferenceUSA 0.8500 0.5167 0.5189 0.6006 14
Grambling Southwestern 0.9000 0.5000 0.5000 0.6000 16
Monmouth MetroAtlantic 0.9000 0.5000 0.5000 0.6000 16
Radford BigSouth 0.9000 0.5000 0.5000 0.6000 16
Baylor BigTwelve 0.8889 0.5031 0.4991 0.5985 18
BostonU Patriot 0.8889 0.5000 0.5000 0.5972 19
BYU WestCoast 0.8889 0.5000 0.5000 0.5972 21
SouthFlorida American 0.8889 0.5000 0.5000 0.5972 21

In the table, Element 1 is the team's winning percentage.  Element 2 is its opponents' winning percentage.  And Element 3 is its opponents' opponents' winning percentage.

The teams with 0.5000 for Elements 2 and 3 all are teams whose conferences play full round robins.  For these conferences, the only thing that distinguishes teams from each other is their Elements 1.  And for these teams, every conference game pulls their Elements 2 and 3 towards 0.5000.  The explanation for this is simple: If I'm Team A and I play conference Teams B and C in my conference, the two of them are going to play each other.  One will win and the other will lose, which means they together will contribute a net winning percentage of 0.5000 to my Element 2; or they will tie, making a similar combined contribution of 0.5000.  The same thing will happen to Team B when it plays conference Teams C and D.  And, since I've played Team B, Teams C and D then will be my opponent's opponents and together will contribute a winning percentage of 0.5000 to my opponents' opponents' winning percentage (Element 3).

The teams with Elements 2 and 3 above or below 0.5000 are from conferences that don't play full round robins.  For them, the variations from 0.5000 are strictly a function of which teams they drew in conference play.  And, for the conference as a whole, their Elements 2 and 3 will tend towards 0.5000.

The following table, based on my "conference games only" experiment, shows how this works for conferences as a whole:

Conference Conference ARPI Average
ACC 0.5000
AmericaEast 0.5000
American 0.5000
AtlanticSun 0.4924
AtlanticTen 0.5036
BigEast 0.5028
BigSky 0.5000
BigSouth 0.5000
BigTen 0.4982
BigTwelve 0.5004
BigWest 0.5000
Colonial 0.5000
ConferenceUSA 0.5000
Horizon 0.5000
Ivy 0.5000
MetroAtlantic 0.5000
MidAmerican 0.5000
MissouriValley 0.5000
MountainWest 0.5000
Northeast 0.5000
OhioValley 0.5000
PacTwelve 0.5000
Patriot 0.5000
SEC 0.5000
Southern 0.5000
Southland 0.5000
Southwestern 0.5000
Summit 0.4949
SunBelt 0.5037
WAC 0.5000
WestCoast 0.5000

Withing each conference as a whole, teams' Element 1 winning percentages will balance out at 0.5000.  And their Elements 2 and 3 likewise will balance out very close to 0.5000.

The fundamental lesson from this is that every conference game pulls the conference's teams' Elements 2 and 3 towards 0.5000.

Bccause of this, your proportion of conference and non-conference games matters.  The higher the proportion of conference games, the more your rating gets pulled towards 0.5000.  The lower the proportion, the less your rating gets pulled there.

The Ivy League is a relatively small conference, with a full 7-game round robin and no conference tournament.  In 2018, which should be a fairly representative season, 43.8% of its games were conference games and 56.3% were non-conference.  The Big Ten, on the other hand, is a large conference that plays an 11-game regular conference season and has a fairly large conference tournament.  Altogether, 62.9% of its games were conference games and 37.1% were non-conference.  Thus, slightly oversimplifying, the Ivy League had 43.8% of its games pulling its Elements 2 and 3 towards 0.5000 and the Big Ten had 62.9% of its games doing it.  Thus the Ivy League's small size and lack of a conference tournament gives it a significant edge over the Big Ten in the RPI ratings race because it amplifies the value of the Ivy League's non-conference winning percentage.

When you put the Ivy League's relatively low proportion of conference games together with how the League's teams scheduled their 2019 non-conference opponents, and when you plug that into the RPI formula, it's easy to see how the RPI could have overrated the League's teams for the 2019 season.

Yale, Harvard, and Columbia

With that as context, what might the Committee have been thinking in not giving at large selections to Yale, Harvard, and Columbia?

Yale

#37 Yale met one "yes" standard, but 7 "no" standards.  This means that Yale presented the Committee with a profile it has not seen over the last 12 years.

Here's the "yes" standard Yale met:

ARPI Rating and Top 50 Results Rank (Standard #18)

The "yes" standard for an at large selection here is a score >=3.3704.

Yale's RPI rating was 0.5982 and its Top 50 Results Rank was #30.  Together, these produced a score for this standard of 3.3842, which under the Committee's pattern would have given it an at large selection.

Over the last 12 years, the Committee has made 280 at large selections involving unseeded teams.  Of these teams, there are 15 that were "assured" under this standard going into the season, but that no longer will be assured with the standard revised to reflect Yale's not having gotten an at large selection.  That amounts to a little over 1 team per year.

Here are the "no" standards Yale met:

Top 60 Head to Head Score (Standard #10)

The "no" standard for an at large selection is a score <=-1.7000.  In other words, under the Committee's pattern, a team with a score below that number will not get an at large selection.

Yale's score was -1.7500.  This was based on four games: an away loss to Virginia Tech (#15), a home loss to Harvard (#41), an away loss to Columbia (#49), and a home tie with Brown (#10).

If the Committee had given Yale an at large selection, there is only one team previously assured of a "no" decision under this standard that now would have been removed from that "no" position.  That would have been a minimal change.

Top 60 Head to Head Score and ARPI Rank (Standard #23)

The "no" standard for an at large selection is a score <=-1.5615.  Yale's score was -1.6392.

If the Committee had given Yale an at large selection, there are 5 teams previously assured of a "no" decision that now would have been removed from that position.  This is about a team every other year, a very small change.

Top 60 Head to Head Score and Adjusted Non-Conference RPI Rank (Standard #52)

The "no standard for an at large selection is a score <=-1.6217.  Yale's ANCRPI Rank was 57.  Yale's score for the combined factors was 1.68.

If the Committee had given Yale an at large selection, there is only one team previously assured of a "no" decision that now would have been removed from that position.  That would have been a minimal change.

Top 60 Head to Head Score and Top 50 Results Score (Standard #60)

The "no" standard for an at large selection is a score <=-28,332.  Yale's Top 50 Results Score was 1152.  Yale's score for the standard was -28,598.

If the Committee had given Yale an at large selection, there are only two teams previously assured of a "no" decision that now would have been removed from that position.  That would have been a minimal change.

Top 60 Head to Head Score and Top 50 Results Rank (Standard #67)

The "no" standard for an at large selection is a score <=1.l5921.  Yale's Top 50 Results Rank was #30.  Yale's score for the combined standard was -1.6133.

If the Committee had given Yale an at large selection, there are only two teams previously assured of a "no" decision that now would have been removed from that position.  That would have been a minimal change.

Top 60 Head to Head Score and Common Opponents Score (Standard #86)

The "no" standard for an at large selection is a score <=-11.6100.  Yale's Common Opponents Score was -2.94.  Yale's score for the combined standard was -12.2125.

If the Committee had given Yale an at large selection, there are 8 teams previously assured of a "no" decision that now would have been removed from that position.  That is 2/3 of a team per year and would have been a small change.

Top 60 Head to Head Score and Common Opponents Rank (Standard #87)

The "no" standard for an at large selection is a score <=-1.5928.  Yale's Common Opponents Rank was #51.  Yale's score for the combined standard was -1.6696.

If the Committee had given Yale an at large selection, there are 3 teams previously assured of a "no" decision that now would have been removed from that position.  That would have been a minimal change.

Yale Conclusion

The one "yes" standard that Yale met was a combination standard, half of which relied on its RPI rating.  All of its "no" standards involved its results against other Top 60 teams, with 3 of those 4 results being against Ivy League teams and its only Top 60 non-conference result being a loss.  Whether the Committee gave Yale an at large selection or not, its decision would not have been a major change from past precedent.

It seems likely to me that the Committee felt Yale's RPI rating (and rank) was suspect and that, under that circumstance, it simply didn't have enough good results to merit an at large selection.

Harvard

#41 Harvard met 1 "yes" standard and 2 "no" standards.  This means it had a profile the Committee hasn't seen over the last 12 years.

Here is the "yes" standard Harvard met:

Conference Standing and Conference Rank (Standard #72)

The "yes" standard for an at large selection is <=11.5.  Harvard's Conference Standing was #2 and the Ivy League was the #5 conference.  Harvard's score for the combined standard was 10.6, which historically would have gotten an at large selection.

With the Committee not giving Harvard an at large selection, there are 15 teams (of 280 unseeded at large selections) that in the past were assured under this standard that would not have been assured with the standard adjusted to reflect Harvard's not getting an at large selection this year.  This is a little over 1 team per year.

Here are the "no" standards Harvard met:

Adjusted Non-Conference RPI Rank and Top 50 Score (Standard #47)

The "no" standard for an at large selection is <=833.3.  Harvard's ANCRPI Rank was #93 and its Top 50 Score was 20.  The Top 50 Score came from its win at #37 Yale and its tie at #49 Columbia.  Harvard's score for the combined standard was 773.

If the Committee had given Harvard an at large selection, there are 6 teams that previously were assured of "no" at large selection that would have changed that position, or 1/2 team per year.  This would have been a small change.

Adjusted Non-Conference RPI Rank and Top 50 Score Rank (Standard #48)

The "no" standard for an at large selection is >=259.6.  Harvard's Top 50 Score Rank was #51.  It's score for the combined standard was 276.6.

If the Committee had given Harvard an at large selection, there are 21 teams that previously were assured of "no" at large selection that would have changed that position.  This is just short of 2 teams per year.  This would have been a small to moderate change.

Harvard Conclusion

Harvard's one "yes" standard for an at large selection relied on the Ivy League's #5 rank.  Harvard's non-conference results and its results against Top 50 opponents, on the other hand, worked against it.  And overall, it appears to me that giving Harvard an at large selection would have been a slightly greater change from the Committee's pattern than not giving it a selection.

Further, it again seems likely to me that the Committee considered the Ivy League's rank to be overstated and even Harvard's relatively poor Top 50 Results to be overstated since they were achieved against Ivy League opponents.

Columbia

Columbia, #49, met no "yes" standards and 1 "no" standard.  Here's the "no" standard it met:

Adjusted Non-Conference RPI Rank and Top 50 Results Rank

The "no" standard for an at large selection is >=259.6.  Columbia's ANCRPI Rank was #84 and its Top 50 Results Rank was #52.  Its score for the combined standard was 271.2.

If the Committee had given Columbia an at large selection, 14 teams that previously had been assured of "no" at large selection now would be in a different position.  This amounts to about 1 team per year and would have been a small change.

Overall Conclusion

None of the Committee's decisions on Brown, Yale, Harvard, and Columbia represents a big change from the Committee's patterns.  In addition, it seems reasonably likely that the Committee felt that the Ivy League and its teams were overrated due to how they did their non-conference scheduling.  All in all, I believe the Committee's decisions were reasonable.






Sunday, November 10, 2019

FINAL PROJECTED NCAA TOURNAMENT BRACKET 11.10.2019

Below is my final projected NCAA Tournament bracket for this year.  Below it are a few comments and a brief review of how I produce the bracket.

Here is the key to the numbers in the NCAA Seed or Selection column:

1 = #1 seed (for all of the seeds, I've put them 1.1, 1.2, 1.3, 1.4, 2.1, etc., to indicate the order in which my system places them)

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

5 = unseeded automatic qualifier

6 = unseeded at large selection, with * meaning the team was tied with the 6.5 group but with better results against Top 50 teams, giving it the at large selection

6.5 = Top 60 team tied with last teams "in" but not getting a selection due to poorer results against Top 50 teams

7 = Top 60 team not getting an at large selection, in order starting with the closest to getting a selection

NCAA Seed or Selection Automatic Qualifier ARPI Rank for Formation Team for Formation
1.1 AQ 1 Stanford
1.2 AQ 2 NorthCarolinaU
1.3 0 3 VirginiaU
1.4 0 4 FloridaState
2.1 AQ 5 SouthCarolinaU
2.3 0 8 SouthernCalifornia
2.2 0 7 UCLA
2.4 AQ 9 KansasU
3.1 0 6 ArkansasU
3.2 AQ 12 BYU
3.4 0 11 OklahomaState
3.3 AQ 10 Brown
4.2 0 14 Duke
4.1 0 13 Rutgers
4.3 0 18 WashingtonU
4.4 0 17 WisconsinU
5 AQ 16 Hofstra
5 AQ 22 PennState
5 AQ 50 Milwaukee
5 AQ 24 SouthFlorida
5 AQ 42 Xavier
5 AQ 56 LoyolaChicago
5 AQ 38 StLouis
5 AQ 44 Monmouth
5 AQ 63 SouthAlabama
5 AQ 90 Navy
5 AQ 102 CalStateFullerton
5 AQ 84 BowlingGreen
5 AQ 87 Lamar
5 AQ 119 Seattle
5 AQ 149 CentralConnecticut
5 AQ 47 NorthTexas
5 AQ 76 BoiseState
5 AQ 110 StonyBrook
5 AQ 112 Lipscomb
5 AQ 118 SouthDakotaState
5 AQ 58 Samford
5 AQ 139 Radford
5 AQ 217 Belmont
5 AQ 224 NorthernColorado
5 AQ 246 PrairieViewA&M
6 0 15 VirginiaTech
6 0 30 CaliforniaU
6 0 33 Vanderbilt
6 0 21 TexasA&M
6 0 28 Louisville
6 0 19 NCState
6 0 20 TexasTech
6 0 23 FloridaU
6 0 27 WashingtonState
6 0 36 Clemson
6 0 25 MichiganU
6 0 29 SantaClara
6 0 46 Pepperdine
6 0 26 Memphis
6 0 31 WestVirginiaU
6 0 35 Georgetown
6 0 32 FloridaAtlantic
6 0 34 NotreDame
6 0 40 ArizonaU
6 0 53 TennesseeU
6 0 43 ColoradoU
6* 0 48 TexasU
6* 0 55 TCU
6.5 0 39 AlabamaU
6.5 0 45 IowaU
7 0 57 GeorgiaU
7 0 41 Harvard
7 0 37 Yale
7 0 49 Columbia
7 0 51 UtahU
7 0 54 MississippiU
7 0 52 OregonState
7 0 60 DePaul
7 0 59 Furman

I base this on the factors the NCAA says the Women's Soccer Committee must use in making its at large selections.  For each factor, and for various combinations of factors, I have a scoring system that assigns a score to each team.  I've matched the factor scores with the Committee's decisions over the last 12 years.  I assign a "yes" and a "no" standard to most factors.  For a factor, a "yes" standard means that the Committee always has made a decision in favor of a team with that score or better.  A "no" standard means that the Committee always has made a decision against a team with that score or poorer.  When a team meets both "yes" and "no" scores, it means the team has a profile the Committee hasn't faced over the last 12 years.  After each year, I adjust the factor standards as needed to be sure that every standard is consistent with all of the Committee's past decisions.

Here are a few comments:

Tennessee came in #11 in the SEC regular season competition.  Although Mississippi State last year broke the #11 barrier for getting an at large selection, its rank and rating and Non-Conference rank and rating were so high it made it virtually impossible for the Committee not to give it an at large selection.  Tennessee does not come close to matching the numbers that Mississippi State had last year.  Nevertheless, Tennessee meets 4 "yes" standards and 3 "no" standards.  That gives it 1 more "yes" than "no" standards.  I treat this the same as a team that meets 1 "yes" standard and no "no" standards.  In my system, this year this qualifies Tennessee for an at large selection.  I will not be surprised, however, and I will not criticize the Committee, if Tennessee does not get an at large selection.

Georgia, on the other hand, came in ahead of Tennessee in the SEC regular season competition but meets 2 "yes" and 3 "no" standards.  I treat this the same as a team that meets no "yes" and 1 "no" standards.  In my system, this leaves Georgia without an at large selection.  Since they came in ahead of Tennessee in the SEC regular season competition, however, it will not surprise me and I will not criticize the Committee, if Georgia gets an at large selection.  And, it will not surprise me to see them flipped with Tennessee.

With Tennesee "in" and Georgia "out," there were 2 positions left to fill and 4 teams that meet no "yes" and no "no" standards.  This is not unusual for filling the last at large spots.  The four teams are Texas, TCU, Alabama, and Iowa.  In that situation, I give the empty at large spots to the teams that have the best results (wins or ties) against Top 50 teams, based on a scoring system that is very heavily weighted towards good results against very highly ranked teams.  My scoring system valued Texas' and TCU's Top 50 results much more highly than Alabama's and Iowa's, so Texas and TCU received the selections.

My system leaves the Ivy League with only Brown in the Tournament.  If the Committee gives an at large selection to one or more other Ivy teams -- possibly Harvard, Yale, or Columbia -- I will have more comments to post.