NCAA Tournament: Predicting the Bracket, At Large Selections and Seeds

Updated April 2026

The purpose of this page is to show the Women's Soccer Committee's patterns in its NCAA Tournament at large selection and seeding decisions.  The patterns are based on the Committee's decisions, as related to the data the NCAA gives to the Committee, over the years from 2007 to the present.

THE FACTORS THE COMMITTEE CONSIDERS

As set out on the NCAA Tournament: Selection, Seeding, and Bracketing Criteria page, when making at large selections the Committee uses the following "criteria," set out in the NCAA DI Women's Soccer Championship Prechampionship Manual.  The Committee also considers these criteria when seeding, but for seeding also can consider other criteria if it wants to.

"Selection Criteria

"....  Won-lost record

"Strength of schedule; and

"Eligibility and availability of student athletes for NCAA Championships.

"In addition, the Women's Soccer Committee [may consider] the following criteria in the selection of at-large teams for the socccer championship (not necessarily in priority order):

"Primary Criteria

"* Results of the adjusted Rating Percentage Index (RPI);

"Ties and losses against teams ranked below 150 in the adjusted RPI;

"Wins and ties against teams ranked in the top 25 of the adjusted RPI;

"Wins and ties against teams ranked in the top 50 of the adjusted RPI;

"* Results versus common opponents; and

"* Head-to-head competition.

"Secondary Criteria

"If the evaluation of the primary criteria does not result in a decision, the secondary criteria will be reviewed.  All the criteria listed will be evaluated.

"* Results versus teams already selected to participate in the field (including automatic qualifiers with RPI of 1-75);

"*Results against teams ranked 51-100 in the adjusted RPI;

"* Late season performance -- defined as the last eight games including conference tournaments (strength and results); and

"* Results of the Kevin Pauga Index (KPI).

"Recommendations are provided by regional advisory committees for consideration by the Women's Soccer Committee.  Coaches' polls and/or any other outside polls or rankings are not used as a criterion by the Women's Soccer Committee for selection purposes."  [Underlining added.]

Based on these "criteria," on comparisons of the Committee's decisions to the data the NCAA provides to the Committee, and on the practicalities of computer programming, I have developed a list of factors to use in seeking patterns in the Committee's decisions.  To start, the list includes the following single factors.  In Appendix A below, there are explanations of how I compute values for those of these factors for which the NCAA itself or the conferences do not have their own computation systems.

1 NCAA RPI Rating

2 NCAA RPI Rank

3 NCAA NonConference RPI Rating

4 NCAA NonConference RPI Rank

5 Top 50 Results Score

6 Top 50 Results Rank

7 Head to Head v Top 60 Score

8 Head to Head v Top 60 Rank

9 Common Opponents in Relation to Top 60 Score

10 Common Opponents in Relation to Top 60 Rank

11 Poor Results Score

12 Poor Results Rank

13 Conference Combined Standing

14 Conference RPI Rating

15 Conference RPI Rank

In addition to these single factors, I pair each single factor with each other single factor with each weighted at 50%, resulting in an additional 103 paired factors, to give a total of 118 factors.  I do this as a way of representing how the different Committee members might look at the data the NCAA provides to the Committee.

WHICH FACTORS ARE MOST CONSISTENT WITH THE COMMITTEE'S DECISIONS?

What if the Committee were to use only one factor to make its at large selection decisions or its decisions at each seed level?  How well would those decisions match the Committee's actual decisions?  The answers to these questions gives some insight into how important each factor is in the Committee's decision process.

#1 Seeds

This table shows, for NCAA Tournament #1 seeds, what the match rate would have been for the Committee's actual decisions if the Committee simply had made its decisions using one factor.  It shows the factors that have the best match rate, in order.  For example, if the Committee over the years had awarded #1 seeds simply using teams' NCAA RPI ranks -- i.e., had awarded #1 seeds to the teams with NCAA RPI ranks of #1 through #4 -- the #1 seeds would have matched the Committee's #1 seeds 83.3% of the time.  In other words, the numbers say that #1 seeds based only on NCAA RPI ranks would have matched the Committee's actual #1 seeds for a little over 3 #1 seed positions per year.

#1 to #2 Seeds

This table is for #1 to #2 seeds as a group.  Again using NCAA RPI rank as an example, it shows that if the Committee simply had used teams' NCAA RPI ranks to identify the teams that would get either #1 or #2 seeds, they would have matched the Committee's actual #1 to #2 seed group for an average of slightly over 7 out of the 8 positions per year.

#1 to #3 Seeds

This table is for #1 to #3 seeds as a group.  Continuing to use NCAA RPI rank as an example, it shows that if the Committee simply had used teams' NCAA RPI ranks to identify the teams that would be in the #1 to #3 seed group, they would have matched the Committee's actual group for a little over 10 out of the 12 positions per year.

#1 to #4 Seeds

This table is for #1 to #4 seeds as a group.  Continuing to use NCAA RPI rank as an example, it shows that if the Committee simply had used teams' NCAA RPI ranks to identify the teams that would be in the #1 to #4 seed group, they would have matched the Committee's actual group for a little over 14 of the 16 positions per year.

#1 to #8 Seeds

This table is for #1 to #8 seeds as a group.  The Committee has awarded #5 through #8 seeds only since 2022, so the data for identifying Committee patterns is relatively limited.  Because of this, I am limiting what I'm showing to only #1 through #8 as a group, rather than also showing #1 through #5, #1 through #6, and #1 through #7.  One should take the numbers shown by this table as painting only a rough picture of the Committee's patterns.

Nevertheless, continuing to use NCAA RPI rank as an example, it shows that if the Committee simply had used teams' NCAA RPI ranks to identify the teams that would be in the #1 to #8 seed group, they would have matched the Committee's actual group for 28.5 of the 32 positions per year.

At Large Selections

This table is for at large selections.  Using the "best scoring" NCAA RPI Rating and NCAA RPI Rank paired factor, it shows that if the Committee simply had used this factor to make at large selections, the selections would have matched the Committee's actual selections for all but 2 1/3 selections per year (out of 33 or 34 selections per year).

USING THE FACTOR STANDARDS METHOD AS A BASIS FOR DECIONS: WHAT IS IT AND HOW DOES IT WORK?

As the above discussion shows, simply using teams' NCAA RPI ratings and ranks to make at large selections and award seeds for the NCAA Tournament would come close to matching the Committee's actual decisions.  There is a refinement to simply using teams' NCAA RPI ratings and ranks, however, that comes even closer to matching the Committee's actual decisions.  The refinement is the Factor Standards Method.

The Factor Standards Method starts with the establishment of "yes" and "no" standards for almost all of the 118 factors, for each Committee decision.  Using #1 seeds as the Committee decision and teams' NCAA RPI ranks as the Factor, as an example:

Teams with NCAA RPI ranks of #1 or better always have gotten a #1 seed.  Thus the #1 seed "yes" Factor Standard for NCAA RPI Rank is <2 .  (I use <2, rather than 1, for programming purposes.)

Teams with NCAA RPI ranks of #8 or poorer never have gotten a #1 seed.  Thus the #1 seed "no" Factor Standard for NCAA RPI Rank is >7.

Appendix B, below, has tables that show all of the factor standards for at large selections and each of the seed levels.

Each factor standard comes from a comparison of the Committee's historic decisions to teams' factor scores.  With 18 years of decisions and factor scores in the data base, the factor standards are relatively stable (except for the #5 to #8 seeds for which there only are four years' history).  Nevertheless, each year I add that year's Committee decisions and teams' factor scores to the data base to see if there are Committee decisions that require "yes" or "no" standard adjustments.  Once I've made any needed adjustments, the updated standards now are consistent with all past Committee decisions including those in the most recent year.

In any year, for a decision that the Committee must make, a team can be in one of three groups:

1. It can meet one or more "yes" standards for the decision;

2. It can meet one or more "no" standards for the decision;

3. It can meet no "yes" and no "no" standards for the decision.

Because I adjust the factor standards yearly to be consistent with all past Committee decisions, there are no teams in the historic data base that meet both "yes" and "no" standards for a particular decision.  In a future year, however, before I have updated the standards to include that year, there can be teams that meet both "yes" and "no" standards.  When that occurs, it means that either those teams have profiles that the Committee has not seen in the past or the array of teams competing for the particular decision presents a situation the Committee has not seen in the past.

Using the factor standards and teams' factor scores, it is possible, for any particular Committee decision, to see which teams met "yes" standards and thus got a "yes" decision from the Committee, which teams met "no" standards and thus got a "no" decision, and which teams met no "yes" and no "no" standards which I sometimes call 0/0 teams.  Unless enough teams met "yes" standards to fill the particular group, then ordinarily the 0/0 teams are "bubble" teams for the particular decision, meaning those are the teams from which the Committee made the selections to fill the group.  I say "ordinarily" because occasionally the number of 0/0 teams matches the number of spaces the Committee has remaining to fill after selection of the "yes" teams.  In the case of such a match, the bubble teams by default fill the remaining spaces without the Committee having to choose from among the bubble teams.

Assuming, as typically is the case, that the number of bubble teams exceeds the number of spaces the Committee has to fill, the next step for the Factor Standards Method is to fill those spaces.  The method does that by using one factor as the basis for filling the spaces.  The factor it uses is the factor that, if used historically to fill those spaces, would have come closest to matching the Committee's actual decisions over the years.  See Appendix C for a list of the factors that come closest, and are next in order, to matching the Committee's actual decisions, for the Committee's seed and at large selection decisions.

The following table shows the results of using the Factor Standards Method (and also the results of using the single factor method described earlier in this page):

Using #1 seeds at the top of the table as an example, the Number Picked Per Year column shows that the Committee picks 4 #1 seeds per year.   The Most Effective Single Factor column shows that there are four single factors that, if each by itself were used to pick the #1 seeds, would pick the most teams that would have matched the Committee's decisions.  The If Picked Only by Most Effective Single Factor, % of Actual  Decisions Matched by Picks column shows that each Most Effective Single Factor matches 83.3% of the Committee's actual decisions over the years.

The Factor Standards Method, % of Actual Decisions Matched by RPI Only "In" Standards Picks shows the proportion of the Committee's actual #1 seeds that use of the NCAA RPI Rating, NCAA RPI Rank, and NCAA RPI Rating and NCAA RPI Rank (paired factor) "yes" factor standards, as a group, picks as #1 seeds, in this case 38.9%.

The Factor Standards Method, % of Actual Decisions Matched by Other "In" Standards Picks shows the proportion of the Committee's actual #1 seeds that use of the all of the other "yes" factor standards, as a group, picks as #1 seeds, in this case 31.9%.

When looking at the Factor Standards Method, % of Actual Decisions Matched by Other "In" Standards Picks, it is important to be aware that all of the factors, but one, are influenced by the NCAA RPI either directly or indirectly.  Influenced directly are the NCAA ANCRPI rating, NCAA ANCRPI rank, conference NCAA RPI rating, and conference NCAA RPI rank single factors plus all paired factors that include as at least one of those factors or one of NCAA RPI rating or NCAA RPI rank as a factor.  Those influenced indirectly are the Top 50 results, head-to-head results, common opponent results, and poor results factors.  For the top 50 results factors, the results that matter are results against NCAA RPI ranked Top 50 opponents.  For the head-to-head results factors, the results that matter are results in games between teams that are NCAA Tournament candidates, which are at most the Top 57 teams in the NCAA RPI rankings (I use the Top 60 for this factor).  For common opponent results factors, the results that matter are the results in games where two teams that are NCAA Tournament candidates -- at most the Top 57 teams in the NCAA RPI rankings (I use the Top 60 for this factor) -- have a common opponent.  For the poor results factors, the results that matter are results in games against teams with "poor" NCAA RPI ranks.  The one factor that is an exception is the Conference Standing factor.  That factor by itself, however, cannot make any "yes" decisions because standing within a conference by itself never is going to assure a team of a seed or at large selection.  Rather, the strength of the conference also is required when considering standing within the conference.  The bottom line of this is that the Factor Standards Method, % of Actual Decisions Matched by Other "In" Standards Picks incorporates the NCAA RPI, one way or another, in all of its picks.

The Factor Standards Method, % of Actual Decisions Matched by Default Picks Due to Number of Bubble Teams Equaling Number of Open Positions shows the proportion of the Committee's actual #1 seeds automatically filled because the number of bubble teams equals the number of open positions, in this case 9.7%.

The Factor Standards Method, % of Actual Decisions Matched by Tiebreaker Picks says, OK, we still have positions to fill, so we'll use a single factor as a tiebreaker to fill those positions.  If we do that, the column shows the proportion of the Committee's actual decisions that get correctly filled using the tiebreaker.  The Most Effective Factor as Tiebreaker Using Factor Standards Method column shows which factor (or factors), used by itself, comes closest to matching the Committee's actual decisions over time.  In the case of #1 seeds, there are four factors that do equally well and better than any other factors: NCAA RPI Rating, NCAA RPI Rank, NCAA RPI Rating and NCAA RPI Rank (paired factor), and NCAA RPI Rating and NCAA ANCRPI Rank (paired factor).  As the Matched by Tiebreaker Picks column shows, the Committee's actual #1 seeds are matched by each of these tiebreakers 16.7% of the time.

The Factor Standards Method, Total % of Actual Decisions Matched by Picks shows the total of the preceding four columns percentages, in this case 97.2%.  This means that the Committee's actual #1 seeds are matched by the Factor Standards Methods picks 97.2% of the time.  See Appendix C for a table that shows what the total match percentages would be using the best-performing factors as tiebreakers.

The last two columns translate this into practical numbers we're used to dealing with.  The Factor Standards Method, Number of Picks Matching Actual Decisions, Per Year column shows that of the 4 #1 seeds each year, the Factor Standards Method on average picks 3.89 correctly.  The Factor Standards Method, Number of Picks Not Matching Actual Decisions, Per Year column shows the Factor Standards Method on average misses 0.11 #1 seed picks per year.  In other words, the Factor Standards Method correctly picks the #1 seeds almost all of the time, missing roughly 1 every 10 years.

As you go down the table through the seed groupings and at large selections, you can see that the Factor Standards Method comes very close to matching the Committee's actual decisions at every decision level.  Thus, whatever method the individual Committee members are conscious of using when making their decisions and whatever method the Committee as a whole thinks it uses, the Factor Standards Method gives an excellent picture of the Committee "Group Mind's" decision patterns.  

For a compilation of the factor standards that are the "most powerful" as single factors and also in awarding "yes" and "no" decisions and as tiebreakers, for seeds and at large selections combined, see Appendix D below.

You might wonder whether there is a pattern when the Committee decisions are different than what the Factor Standards Method would decide.  It is important to remember that the differences from what the Factor Standards Method would decide all are when the Committee must choose from among 0/0 bubble teams, each of which has nothing outstandingly positive in its profile such that it meets a "yes" standard and nothing outstandingly negative such that it meets a "no" standard.  In this context, I've done an analysis of all the differences in relation to teams' conferences, teams' regions, and whether teams are from Power conferences or from Other conferences.  I've also looked at this separately for differences related to seeds and differences related to at large selections.  Although there are exceptions, I see the following Committee patterns for the differences:

Conferences:  Overall where there are differences, the Committee decisions favor stronger conferences  A big exception, however, is the Big Ten, whose teams come out on the negative side of the Committee's decisions more often than on the positive side.  Overall, this could suggest a Committee bias in favor of stronger conferences.  On the other hand, as shown on the RPI: Measuring the Correlation Between Teams' Ratings and Their Performance page, the NCAA RPI discriminates against stronger conferences and in favor of weaker ones.  So this could suggest that the Committee believes there is discrimination and tries to correct for it.  The Big Ten, on the other hand, is an unusual situation and the data suggest that the Committee for some reason may have a bias against the Big Ten, which is a possibility I have entertained in the past.

Regions:  Overall where there are differences, the Committee decisions favor teams from the South and West and disfavor teams from the Middle and, especially, the North.  The NCAA RPI discriminates especially against teams from the West and, to a lesser extent teams from the Middle, and in favor of teams from the North and South.  The decisions don't match this discrimination pattern.

Power and Other Conferences:  Where there are differences, the Committee decisions clearly favor the Power conferences.

At Large Selections and Seeds, Separately:  For conferences, for regions, and for Power versus Other, the Committee's favoritism overall is greater in relation to seeds than it is for at large selections.  This may reflect that the Committee is explicitly tied to the NCAA-designated criteria for at large selections -- and thus, for example, is strongly tied to the RPI ratings and ranks for at large selections -- and has greater freedom for seeds.

Differences Where Two Teams Are from the Same Conference:  There are 11 cases in which the differences are between two teams from the same conference.  In 9 of those 11 cases, a decision in favor of the team with a poorer NCAA RPI can be explained by a look at where the teams fit within the conference standings, how they did in the conference tournament, and how they did in head to head results against each other.

Appendix E below has a series of tables with the detailed data that shows the above patterns.

MY CONCLUSION

With all of the above as context, my judgment is that teams' NCAA RPI ratings and ranks are by far the strongest driver of the Committee's NCAA Tournament seeding and at large selection decisios.  They are close to, but not quite, determinative.

This makes the NCAA RPI's defects, as shown on the RPI: Measuring the Correlation Between Teams' Ratings and Their Performance and the RPI: Modified RPI? pages, an especially significant problem.

APPENDIX A: DESCRIPTIONS OF SINGLE FACTORS

The values for the NCAA RPI Rating, NCAA RPI Rank, NCAA ANCRPI Rating, NCAA ANCRPI Rank, Conference NCAA RPI Rating, Conference NCAA RPI Rank, and Conference Standing single factors all are based on NCAA formulas or conference formulas.  For the other single factors, the values are based on formulas I have created as described in this Appendix.

Head to Head v Top 60 Opponents Score

The NCAA requires the Committee to consider head-to-head results in making its at large selections. To have a scoring system for this factor, I look at all head-to-head results in games where both opponents are ranked #60 or better.  I look only at these games because as a matter of practical necessity, the Committee must limit itself in the number of games it looks at and since 2007, no team ranked poorer than #57 has received an at large selection.

For a team with a head-to-head v Top 60 game, it receives the following points based on the game result:

Win Away        4

Win Neutral    3

Win Home       2

Tie Away         1

Tie Neutral      0

Tie Home        -1

Loss Away       -2

Loss Neutral   -3

Loss Home      -4

For the team, I compute its score for each of its head-to-head v Top 60 games and then add all of its scores together.  I then divide the total by its number of head-to-head v Top 60 games, to get an average score per head-to-head v Top 60 game.  This is its Head to Head v Top 60 Opponents Score.

This scoring system has some weaknesses.  It does not take into account the rank of an opponent, other than that it is in the Top 60.  Also, it ignores Top 60 teams that have no Top 60 opponents (which happens occasionally though not often).  Also, for teams that have very few Top 60 opponents, the limited data can produce suspect results.

This scoring system is a lot more complicated than simply looking at two teams competing for an at large selection to see if they have a head to head result.  It takes into account the possibility of A beats B, B beats C, and C beats A scenarios and of even more complicated scenarios.  In effect, it considers and scores in a single system the entire complex of head to head results in games involving two Top 60 teams.

Top 50 Results Score

This looks at a team’s results against Top 50 opponents, so is somewhat like the Head to Head v Top 60 Opponents Score factor.  However, it considers only "good" results – wins and ties.  Further, it considers opponent ranks.  Here is its scoring system:

I created this scoring system based on observations of the Committee’s decisions over many years.  As you can see, it is very heavily weighted towards good results against very highly ranked teams.

A way to look at the Head to Head v Top 60 Opponents Score as compared to the Top 50 Results Score is this:

The Head to Head factor gives a rough overview of where a team fits as a competitor with other teams in consideration for at large selections; and

The Top 50 Results factor looks in detail at the level at which a team under consideration has shown it can compete successfully: If a team has a win or a tie against an opponent at the top of the RPI rankings, it will have a very high score, showing that it can compete at the very highest level.

Common Opponents with Other Top 60 Teams Score

As with the Head to Head v Top 60 Opponents Score system, this looks at the entire complex of games involving Top 60 teams with common opponents.  Here is how the system works:

1.  For each Top 60 team, I identify all games where its opponent is one in common with another Top 60 team.

2.  For each of those games, I assign the team a score using the same scoring system as for the Head to Head v Top 60 Opponents Score factor.

3.  When any two Top 60 teams have a common opponent, I determine the difference between their step 2 scores in relation to that opponent.  If the two scores as I have determined them are the same, the difference will be 0.  Otherwise, one team will have a positive score and one a negative.

4.  For a Top 60 team, I then add up all its scores as determined in step 3 to get a total score in common opponent games.

5.  I then determine the number of games the Top 60 team had against an opponent that also played one or more other Top 60 teams.

6.  I divide the total score from common opponent games (step 4) by the total games against an opponent that also played other Top 60 teams (step 5) to get an average score per common opponent.  This is the team’s Common Opponents with Other Top 60 Teams Score.

Poor Results Score

I score poor results as follows:

Opponent ranked 56-100:        Tie -1; Loss -2

Opponent ranked 101-150:      Tie -3; Loss -4

Opponent ranked 151-200:      Tie -5; Loss -6

Opponent ranked below 200:  Tie -7; Loss -8

For this factor, I simply add up all the scores to get the Poor Results Score.  Thus a score of 0 means the team had no poor results.

Conference Combined Standing

This combines team regular season standing and conference tournament standing within a single factor.

For conference tournament standing, here is how I set a team’s standing:

Champion:                     1

Runner up:                     2

Losing Semifinalist:      3.5 (the average of the 3-4 positions)

Etc., for tournaments with pre-semifinals games, depending on how many teams play in those games

For teams that do not qualify for a conference tournament, their conference tournament standing is the same as their conference regular season standing.

For the Conference Combined Standing, I use the average of a team’s regular season standing and its conference tournament standing.

It is possible Committee members consider one of regular season standings or conference tournament results to be more important.  If so, the scoring system for this factor does not take that into account.

APPENDIX B: FACTOR STANDARDS

1 to 4 Seeds and At Large Selections

5 (4.5) to 8 (4.8) Seeds

APPENDIX C: MOST POWERFUL TIEBREAKERS, FOR EACH COMMITTEE DECISION

#1 Seeds

#1 to #2 Seeds

#1 to #3 Seeds

#1 to #4 Seeds

#1 to #8 Seeds

At Large Selections

APPENDIX D: MOST POWERFUL FACTORS

The following table is a resource for considering which factors are the most "powerful" in the Committee's decision process.  There is an explanation below the table.

This table shows two sets of data: (1) on the left, in the Necessary or Equally Powerful columns, it shows information based on using the Factor Standards Method, and (2) on the right (except for the farthest right column), in the Single Factor Method Most Powerful columns, it shows information based using a single factor as the basis for decisions.

In the Factor Standards Method columns on the left, a cell with a "1" means that the factor in that row affects the decision in that column, as described below.  In the Single Factor Method columns on the right, a cell with a "1" means that the factor in that row is among the factors with the best "match rate" with the Committee's actual decisions.

The first blue column shows the total number of Committee decisions affected by the factor in that row, as described more below.  The second blue column shows the total number of Committee decisions for which the factor in that row is among the factors with the best "match rate" with the Committee's actual decisions.  The grey column on the right is the sum of the numbers in the blue columns.

In the Factor Standards Method columns, there is color coding:

The NCAA Tournament Rank cells are gold highlighted and assigned 1s because of NCAA Tournament Ranks' importance in deciding which teams do not get seeds or at large selections, as well as their importance in other Committee decisions.  Teams' ranked #8 and poorer do not get #1 seeds, #14 and poorer do not get #1 to #2 seeds, #24 and poorer do not get #1 to #3 seeds, #28 and poorer do not get #1 to #4 seeds, #50 and poorer do not get #1 to #8 seeds, and #58 and poorer do not get at large selections.

Green highlighting in a cell means that there is at least one case where a team meets this factor's "yes" standard but does not meet any other "yes" standard.  Thus this factor is needed in order to identify all teams that meet at least one "yes" standard.

Salmon highlighting in a cell means that this factor is not needed when indentifying teams that meet "yes" factors, but there is at least one case where a team meets this factor's "no" standard but does not meet any other "no" standard.  Thus this factor is needed in order to identify all teams that meet at least one "no" standard.

Grey highlighting is for factors that are not needed for identifying all teams that meet at least one "yes" or one "no" standard, but the number of teams that meet their "yes" standards is in the same range as the gold and green highlighted teams.

The listed factors with "1s" and color coded green or salmon in the Factor Standards Method columns are sufficient, as a group, to identify all teams that meet "yes" or "no" standards and thus all that are necessary for the Factor Standards Method to work.

APPENDIX E: PATTERNS TO COMMITTEE'S "OVERRIDES" OF FACTOR STANDARDS METHOD TIEBREAKER DECISIONS

The following table shows all the cases in which the Committee overrode the Factor Standards Method's tiebreaker decisions.  It shows both the teams that were "in" due to the Committee's decisions and the teams that were "out."  It is important to remember that the tables cover 18 years of Committee decisions, so that the number of differences per year is relatively small.

The following table draws on the above table to show the Committee patterns by conference.  Note in particular that the Committee decisions suggest it believes, at least for the bubble teams, that the RPI (as a tiebreaker) underrated the SEC, ACC, old Pac Twelve, and West Coast conferences and overrated the Big East, Big Ten, Ivy, and American conferences:

The next two tables break the conferences down into at large for the first table and seeds for the second table

At Large

Seeds

The next tables are similar to the ones for conferences, but are four the four regional playing pools:

The next table is similar but is for Power as compared to Other conferences and combines the three tables into one: