In the preceding post, I explained how I produce simulated end-of-season ratings and ranks. In this post, I will explain how I use those ratings and ranks and other data from the simulated season in order to identify teams likely to be candidates for NCAA Tournament seeds and at large selections.
The NCAA annually, in its Pre-Championship Manual, identifies the factors the Women’s Soccer Committee is to consider when seeding and making at large selections for the NCAA Tournament. Based on the Manual and studies of game results data as compared to the Committee’s decisions since 2007, I have identified 15 key factors for the decisions:
RPI rating
RPI rank
Non-conference RPI rating
Non-conference RPI rank
Value of good results (wins or ties) against Top 50 opponents
Rank of good results (wins or ties) against Top 50 opponents
Value of head-to-head results (wins, ties, or losses) against Top 60 opponents
Rank of head-to-head results (wins, ties, or losses) against Top 60 opponents
Value of common opponent results (wins, ties, or losses) compared to the results of other Top 60 teams that have played the same opponents
Rank of common opponent results (wins, ties, or losses) compared to the results of other Top 60 teams that have played the same opponents
Conference standing (average of regular season conference standing and finishing position in conference tournament)
Conference RPI
Conference RPI rank
Value of poor results
Poor results rank
In addition to these factors, I have created another group of paired factors: I pair each individual factor with each other factor, with each factor weighted at 50%, which results in 103 additional factors or a total of 118 factors both individual and paired.
For each of the individual factors that does not have an NCAA or conference-created scoring system, I have created my own scoring system. You can find the details of my scoring systems in an NCAA Tournament Decisions resource I created for coaches.
By comparing the NCAA Tournament seed and at large selection decisions over the years to the factor scores, for each type of decision -- a particular seed level or an at large selection -- I have identified certain factor scores that (1) always have gotten teams "yes" decisions and (2) always have gotten teams "no" decisions. For example, for #1 seeds, the team with the #1 RPI rank always has gotten a #1 seed (a yes decision) and teams with RPI ranks of #8 or poorer never have gotten #1 seeds (a no decision). Most of the 118 factors have both yes and no standards for each type of decision.
At the end of the season, I match teams’ factor scores with the yes and no factor standards. For each type of decision, this allows me to see how many yes and no factor standards a team meets. If a team meets only yes standards for a particular decision, then the team should get a yes decision. If it meets only no standards, then it should get a no decision. If it meets no yes and no no standards, then it is a candidate for a yes decision but also might get a no decision. If it meets both yes and no standards, then it has a profile the Committee has not seen before and likewise might get either decision.
Most often, at the end of the season, for a particular Committee decision some teams will meet some yes standards and no no standards. If there are not enough teams only meeting yes standards to fill the decision quota (such as four #1 seeds), then ordinarily the remaining teams to fill the quota will come from teams that meet no yes and no no standards. For the choice among those teams, based on comparing past Committee decisions to teams’ factor scores, I have identified which factor(s) matches best with the Committee choices. Using at large selections as an example, if I first give at large selections to teams that meet one or more yes standards for at large selection and no no standards, my selections on average match the Committee selections historically for all but about 2 at large selections per year. If I then consider teams that meet no yes and no no standards, using the factor that best matches the Committee choices from 0 yes - 0 no teams, I narrow the gap to matching the Committee decisions for all but 1 at large selection per year. As a matter of interest, in this process, the factor that best matches the at large selections is the paired factor of RPI rank and Top 50 results rank.
Using my pre-season simulated end-of-season ratings and ranks and related data, given the excess number of ties and the treatment of all games with a win likelihood of 50% or more as wins, teams’ data are too exagerrated for me to use the process I just have described to do a reasonable pre-season simulated NCAA Tournament bracket. I can, however, show how teams fare in the factor standards yes and no system. So, for what it is worth, here is how teams fare for each of the NCAA Tournament decisions (excluding #5-8 seeds, initiated last year):
At Large Selections (showing Top 80 teams)
#4 Seeds (showing Top 40 teams)
No comments:
Post a Comment