In working on my "regular" end-of-season analysis of what the Women's Soccer Committee's NCAA Tournament seeds and at large selection might be, the numbers I am seeing have made something clear: This year, there is nothing "regular" about what the Committee will be seeing. Because of that, this article will include more and different details than what I have provided in earlier years, so you can see what the Committee will be facing and, once you have seen the Committee's actual decisions, you can decide whether the Committee has been "up to" the moment.
First, I'll start with some information on why this season is not "regular."
Proportions of Out-of-Region Games
The following table shows the historic proportions of games that the four regions' teams have played against out-of-region opponents. I place each State's teams in the region in which the State's teams as a group play either the majority or plurality of their games. To see the regions -- Middle, North, South, and West -- and the States within them, go to the RPI: Regional Issues page at the RPI for Division I Women's Soccer website. The data in the below table are from the years 2013 through 2024 (excluding Covid-affected 2020).
The next table breaks down the historical out-of-region percentages by region, including showing the distribution of out-of-region games among the other regions:
This year, likely driven by the changed financial landscape for Division I sports. the out-of-region numbers have declined dramatically, notwithstanding the increased out-of-region travel for teams from conferences with recent major expansions of their geographic footprings:
Comparing this to the first table above, there has been a 28.5% reduction in out-of-region games.
Here are this year's breakdowns for the regions:
Comparing this to the second table above, you can see the reduction in out-of-region travel for each region. There is an 18.3% reduction for teams from the Middle, 29.3% for the North, 31.0% for the South, and 33.5% for the West.
In looking at these reductions, consider that for the NCAA RPI to function as a fully national rating system, there must be a large number of out-of-region games. If there are not enough out-of-region games, then what you are seeing in the NCAA RPI ratings and rankings is how teams within a region compare to each other, but not how teams from a region compare to teams from other regions.
Levels of Parity Within Regions
An indicator of parity within a region is the proportion of in-region games that are ties. The following table shows the historic proportions of ties, by region. The data for the table are from the years 2010 through 2024 (excluding Covid-affected 2020), with all games that were ties at the end of regular time treated as ties (for those years when the rules provided for overtime games).
As you can see from this table, historically the West has had the highest proportion of ties, followed by the Middle and North, with the South having the lowest proportion of ties. In terms of parity, the order is from the West with the greatest parity to the South with the least.
Here is the table for this year:
As you can see, the proportions of in-region ties are higher for all regions. In other words, it appears there has been an increase in parity within the regions. Once again, the West has the greatest parity and the South the least, with the Middle and North switching places from the historic norm.
Diminished Value of Ties Within the RPI Formula
In 2024, the Women's Soccer Committee changed the RPI Formula. This included a change in how the NCAA computes Element 1 of the RPI, which is a team's Winning Percentage (WP). A way to express the formula for Winning Percentage is:
WP = (Wins + X*Ties) /(Wins + Ties + Losses)
Until 2024, in the formula X was 1/2. In 2024, the Committee changed X to 1/3. Thus the value of a tie went from 1/2 of a win to 1/3 of a win.
As a result of this change, the Committee depressed many teams' ratings, since many teams have one or more ties. As a presumably unintended side effect, the change also punished regions with higher levels of parity and thus more ties.
It appears that an effect of these changes has been to change many upper level teams' profiles enough that they now look poorer than they have in the past. This relates to my annual process of considering what seed and at large decisions we can expect the Committee to make, if the Committee follows its historic decision patterns. I'll go through the expected seeds and at large selections below and hopefully you will be able to see what I mean.
#1 SEEDS
Historically the #1 seeds always have come from the Top 7 teams in the NCAA RPI rankings. Thus teams #1 through #7 are the #1 seed candidates.
Here is a table that relates to this year'sd #1 seed selection process:
Here is a detailed description of this table, as an introduction to the process I use and to the other tables I'll show below for the other seed levels and the at large selections.
I have identified 13 individual factors that the NCAA directs the Committee to consider in its at large selection process. For each of those factors, either the NCAA provides a scoring system (for example, for a team's rating, the NCAA specifies the RPI as the scoring system) or I provide my own scoring system. In addition to those individual factors, I also pair each factor with each other factor, with the scoring system weighting each factor in a pair at 50% of that "paired" factor's value. Altogether this results in 105 paired factors plus the 13 individual factors or a total of 118 factors.
By comparing the factor scores for teams to the Committee's seed and at large selection decisions for teams over the years, for each decision the Committee must make -- #1 seed, #2 seed, etc., and at large selection -- I have identified two score standards for each factor. A "yes" standard for a factor means that teams whose scores for that factor have been better than the "yes" standard always have gotten a positive decision from the Committee. A "no" standard means that teams whose scores have been poorer than the "no" standard never have gotten a positive decision. Using #1 seeds and the NCAA RPI Rating factor as an example:
The "yes" factor score is 0.6986. This means that teams with NCAA RPI ratings better than 0.6986 always have gotten #1 seeds.
The "no" factor score is 0.6479, this means that teams with NCAA RPI ratings poorer than 0.6479 never have gotten #1 seeds.
It is important to note that some teams will have NCAA RPI ratings between the "yes" score of 0.6986 and the "no" score of 0.6479. These are possible, but not assured, #1 seeds based on the Committee's historic patterns.
For each required Committee decision, my system evaluates each team in relation to each factor. The above table, in the 1 Seed Status Based on Standards column shows how this year's candidate group fared in the evaluation process.
As you can see if you look at the 1 Seed Total and No 1 Seed Total columns, Stanford and Notre Dame each meet a number of "yes" standards and no "no" standards. This means that based on the Committee's historic patterns, Stanford and Notre Dame are clear #1 seeds.
On the other hand, if you look at Virginia and TCU, each meets at least 1 "yes" standard and at least 1 "no" standard. This means each has a profile the Committee has not seen before (meaning not since 2007). Based on my years of experience looking at numbers like this, I see something in the Virginia and TCU "no" numbers. They are significantly higher than what I would expect to see for a profile the Committee has not seen before.
Further, if you look at Vanderbilt, Michigan State, and Kansas, they meet 0 "yes" standards and a high number of "no" standards. Again based on years of experience, the "no" numbers are far higher than what I would expect to see for teams in the Top 7 of the NCAA RPI rankings.
Rather than seeing 5 of the 7 candidates for #1 seeds having significant numbers of "no" scores, what I would expect to see is at least several of them having no "yes" and no "no" scores. These then would be the candidates for the remaining 2 #1 seed positions.
The bottom line of this is that most of the RPI Top 7 teams' profiles are far poorer than what one should expect based on past history.
This same phenomenon appears throughout the decisions the Committee must make. Because of this, I have concluded that this year, the changes I described above make it unwise to use the "no" factor scores as a basis for seeing if the Committee's decisions are consistent with the Committee's past decision patterns. The "yes" factor scores should be fine, but not the "no" scores. I have shown this in the 1 Seed Status Based on Standards column by identifying all of the teams other than clear #1 Stanford and Notre Dame as #1 seed Candidates. Looking at the table, when I disregard the "no" scores, each of Virginia and TCU is left with at least 1 "yes" score. Because of this, it appears to me that those teams receivinig #1 seeds would be most consistent with the Committee's historic decision patterns.
Unlike this year, in the past, when there have not been enough teams meeting only "yes" standards to fill a decision group, there have been teams that meet no "yes" and no "no" standards. I then apply a tiebreaker to fill out the group. The tiebreaker is the factor, from among all 118 of them, the scores of which historically have been most consistent with the Committee's decisions as to that group. As it turns out for the #1 through #4 seeds, the factor most consistet with the Committee's decisions is teams' NCAA RPI ratings or ranks. So if, for example, Virginia and TCU had met 0 "yes" standards this year, the Committee's historic decision patterns would have suggested picking from the 5 candidates the 2 teams with the best NCAA RPI ranks. In that case, I would have looked at the 1 Seed Status Based on NCAA RPI Rank column, in which I entered "1 Seed" to indicate that those teams would get 1 seeds if needing to use the tiebreaker. In that case, the Committee's historic patterns would have assigned the remaining 2 #1 seed positions to Virginia and Vanderbilt.
In the table, the #1 Seed Status Based on Standards and Tiebreaker Combined column shows the #1 seeds, based on this process, that would be most consistent with the Committee's historic patterns.
#2 Seeds
With the #1 seeds most consistent with the Committee's historic patterns identified, next comes the #2 seeds using the same process as applied to the #2 seed candidate group, but excluding the already identified #1 seeds. The candidate group is teams with NCAA RPI ranks of #13 or better.
Of the candidates, after disregarding the "no" scores, there are 4 teams that have at least 1 "yes" score: Vanderbilt, Florida State, Duke, and Georgetown. So, given the constraints this year, those teams receiving #2 seeds would be most consistent with the Committee's historic patterns.
#3 Seeds
I will go through the remaining seeds in the same fashion as for the #2 seeds.
Kansas, Colorado, LSU, and Louisville as #3 seeds would be most consistent with the Committee's historic patterns.
#4 Seeds
There are three teams with at least 1 "yes" score: Michigan State, Tennessee, and Washington. There still is one #4 seed position to fill. Disregarding the "no" scores leaves all the other teams in the #4 seed candidate group (teams ranked #28 or better) as possibilities. Since the tiebreaker is teams' NCAA RPI ranks, the best ranked remaining team is West Virginia, so it fills the remaining #4 seed position.
#5 Through #8 Seeds
The process for these seeds is the same as above, except that the Committee has done these seeds only for a few years and the data establishing Committee patterns are relatively limited. Based on what the Committee has done for these seeds so far, the following changes appear to best identify the Committee's patterns:
1. The tiebreaker for these seeds is a paired factor rather than simply their NCAA RPI ranks. The paired factor is NCAA RPI Rank and Top 50 Results Score combined (the higher the score, the better); and
2. The Committee's selection of the #5 through #8 seeds is more like the Committee's at large selections than the Committee's #1 through #4 seeds. So the best standards references are to the standards for at large selections.
With that in mind, here are the tables for the #5 through #8 seeds (identified in the column headings as 4.5 through 4.8):
NOTE: In the table, the lack of entries for Fairfield indicate that it played no opponents with NCAA RPI ranks of 60 or better. That puts it out of consideration for an NCAA Tournament seed or at large position.
At Large Positions
The tiebreaker for at large positions is factor pair of NCAA RPI Rank and Top 50 Results Rank combined (the lower the score, the better). The at large candidates are teams ranked #57 or better (that have not been seeded).
After the seeding, there are 8 additional at large positions to fill. In the table, the At Large Status Based on Standards column shows the selection of 6 teams, each of which has at least 1 "yes" score: Penn State, Mississippi State, Ohio State, Clemson, South Carolina, and Illinois. This leaves 2 positions to fill based on the tiebreaker. The teams scoring best on the tiebreaker (lowest score) are California and Utah Valley, so their selection best matches the Committee's historic patterns.
Reminder and Summary
This year, for the reasons described above, I am disregarding the "no" scores. How the Committee will treat the negative aspects of teams' profiles this year remains to be seen. Maybe they will recognize that recent changes are causing teams' profiles to seem poorer than they have been in the past; but maybe they won't.
In the meantime, here is a summary of the above seeds and at large decisions. In the Seed or At Large Selection column, #1 seeds are 1.0, #2s are 2.0, #3s are 3.0, #4s are 4.0, #5s are 4.5, #6s are 4.6, #7s are 4.7, and #8s are 4.8. Unseeded Automatic Qualifiers are 5.0. Unseeded at large selections are 6.0. At large candidates from the Top 57 that do not get at large selections are 7.0. (The order of teams within the different groups does not have any significance.)
No comments:
Post a Comment