To RPI, or Not to RPI? That Is the Question

Every May, college baseball turns into a sport of closing arguments. Teams have spent three months building a résumé, but in the final days before the NCAA Tournament field is announced, the conversation becomes less about what happened and more about how it should be described. A win becomes a Quadrant 1 win. A loss becomes something to be explained away. A road trip in March becomes evidence of ambition. A home weekend against the wrong opponent becomes the kind of thing a team drags around long after everyone has forgotten the score.

This is the strange little courtroom RPI built.

Nobody loves RPI in a pure sense. It is too blunt for that. It does not know why a midweek game was played with the fourth arm available, or why a northern team had to spend the first month of the season on the road, or why one conference schedule creates opportunities another league never sees. It can be awkward and unforgiving, and in the right argument, almost everyone can find a reason to dislike it.

But the committee has not treated RPI like something it dislikes. For more than a decade, it has treated it like the language of the room.

That distinction matters this year because there are more ranking systems in the conversation, and more ways for people to frame the same team differently. One model may make a team look safe. Another may make the same team look vulnerable. That does not make the debate useless, but it does make the historical question more important. The tournament is not built by the ranking system people wish mattered most. It is built by a committee with habits, and those habits leave a paper trail.

So we went back through the modern RPI era to see how strong that trail really is.

The starting point is 2013, which is not an attempt to make the sample prettier. RPI changed in a meaningful way that year, so folding in older seasons would make the comparison less useful. There was also no tournament in 2020, which leaves us with 12 selection cycles. The important part of the exercise was using the RPI as it existed at the time of selection, not after the tournament had already started changing the numbers. If the question is what the committee saw, then Selection Monday is the only useful snapshot.

The formula itself is not hard to explain, even if the downstream effects can get strange. RPI is built from three pieces: a team's modified winning percentage, its opponents' winning percentage, and its opponents' opponents' winning percentage. A team gets only a quarter of the formula from its own record. Half comes from the quality of the teams it played, and the last quarter comes from the quality of the schedules those opponents played.

That is where the system starts to become less like a ranking and more like an ecosystem. A schedule is not just a list of games. It is a series of attachments. Play a team that wins a lot, and it helps. Play a team that wins a lot against a good schedule, and it helps more. Play a team that loses a lot against a bad schedule, and the damage can linger.

The modified part matters too. A road win is worth more than a home win. A home loss hurts more than a road loss. Neutral games sit in the middle. In theory, the idea is reasonable enough. Baseball is hard on the road, and the formula rewards teams that prove they can win away from home. In practice, it means scheduling becomes its own form of roster management. You are not just deciding who you can beat. You are deciding whose entire season you are willing to attach to your own.

That still does not make RPI perfect. It only explains why it has carried so much force.

The first chart is the one that should slow the entire debate down. Across the last 12 tournament cycles, the mean absolute deviation between RPI rank and regional host seed is 3.07 spots. In a sport with more than 300 Division I teams and only 16 regional hosts, that is not loose correlation. That is not the committee glancing at RPI before doing something entirely different. That is the committee almost tracing it.

There are years where the line gets a little messier. There are individual teams that get moved up or down in ways that still frustrate people years later. But the broad pattern is not subtle. RPI has not been background noise. It has been the outline.

That is why 2022 looks so loud on the chart. The mean deviation that year was 5.88, almost double the 12-year average. Texas A&M, Texas, Oklahoma State, Maryland, Georgia Southern, Southern Miss, Florida, and North Carolina were all part of a bracket that felt less like a slight adjustment and more like the committee was rewriting the top of the page. You can defend pieces of it. You can argue strength of schedule mattered, conference context mattered, and the committee was not obligated to follow a formula. All of that can be true.

The point is that everyone noticed.

When the committee moves meaningfully away from RPI, the season does not blend into history. It becomes one of the years people cite later because it looks different from the rest of the sample.

Two debates, not one

That is the larger context for the two questions being debated now. There is the question of who belongs in the field, and there is the question of who deserves to host. Those are related, but they are not the same. Blending them together is how the conversation gets sloppy.

Mercer belongs in the first conversation. Southern Cal belongs in the second.

The field case: Mercer (2026). RPI #28 · 44–14 · Q1+Q2 9–13. Not trying to look like a regional host. The question is whether the profile fits the historical shape of non-host entrants from multi-bid conferences.

The host case: Southern California (2026). RPI #8 · 42–14 · Q1+Q2 9–11. Not asking whether it belongs in the 64. Asking whether it belongs in the 16, the room where RPI has been most predictive.

Mercer will not be a regional host. The question is whether Mercer looks like the kind of team that has historically made the NCAA Tournament from a multi-bid league. Once the debate is framed that way, the profile looks a lot less fragile than the public conversation around it.

Mercer sits at RPI 28 with a 44–14 record and a 9–13 mark against Q1 and Q2 opponents. The top end of the résumé is not perfect. A 9–13 Q1+Q2 record is not the kind of thing a team puts on a billboard. But bubble résumés are almost never clean. That is why they end up in the argument at all.

The better question is whether Mercer's flaws are large enough to pull it outside the historical shape of the field. When plotted against non-host tournament entrants from multi-bid conferences over the last 12 tournaments, Mercer does not look like a team being stretched into the bracket. It looks like a team that fits the neighborhood.

That is the part worth emphasizing. There is a difference between having something to criticize and having enough to exclude. Mercer has areas a committee could discuss, but the overall profile sits comfortably within the kind of team that has made this tournament before. If the committee is using the same historical language, Mercer should not be treated like a team hoping for a miracle. It should be treated like a team that would require a real explanation to leave out.

That does not mean RPI alone gets Mercer in. It means the historical comparison does not support casually pushing Mercer out.

The host line

Once that field question is separated, the host discussion becomes cleaner. Southern Cal is not asking whether it belongs somewhere in the 64. Southern Cal is asking whether it belongs among the 16. That is a much harder room to enter, but it is also the room where RPI has been most predictive.

The RPI top-16 has been a remarkably strong map of the host line. Across the 12-year window, the committee included an average of 13.8 RPI top-16 teams as regional hosts each year. That is roughly 86 percent. The committee may argue about the order. It may slide one team down, pull another team up, reward a particular conference profile, or punish a team whose résumé has the wrong kind of weakness. But most years, the host pool still looks very much like the RPI top-16.

That is not a small point. If the committee usually includes nearly 14 of the RPI top-16 as hosts, then being well inside that group has historically meant something. It does not guarantee a host spot, but it puts the burden of explanation on the move away from it.

Again, 2022 is the year that refuses to behave. Only 11 of the RPI top-16 hosted that season. Wake Forest, Vanderbilt, DBU, Georgia, and Notre Dame were all inside the RPI top-16 and did not host. Georgia Tech was 18th and also missed. That was not normal committee movement. That was a different kind of year.

The reach and snub charts show why.

The reach chart captures the teams outside the RPI top-16 that were still given host spots. There are 27 of them across the sample, and a few are impossible to ignore. Texas A&M in 2022 went from RPI 23 to the number 5 national seed. Louisiana Tech in 2021 went from RPI 33 to a host. Cal State Fullerton in 2015 went from RPI 29 to the 14 host.

Those examples matter because they show the committee does have room to move. It can elevate a team when it believes the résumé deserves it. It can look beyond the number and decide the number is missing something. But the chart also shows that these are not everyday moves. They stand out because they are large, and in many cases because the surrounding résumé needed to carry the argument.

The same is true on the other side.

The snub chart is where 2022 becomes almost unavoidable. Wake Forest at RPI 8 did not host. Vanderbilt at 10 did not host. DBU at 14 did not host. Georgia at 15 did not host. Notre Dame at 16 did not host. That is the kind of season that can make people nervous in any future year because it proves the committee can break from the board if it wants to.

But proving something can happen is not the same as proving it usually happens.

There have been other high RPI teams left out of hosting. Rice was 9th in 2014. Fairfield was 9th in 2021, though that season came with its own scheduling distortions. Wake Forest was 10th in 2024. Alabama was 10th in 2025. These are real examples, and any honest RPI argument has to make room for them.

These cases are memorable because they are exceptions, not because they are the ordinary rhythm of the committee.

Testing USC against the host history

That is where Southern Cal enters the article in a much cleaner way. The Trojans are not being compared to Mercer. They are not being used as another generic example of RPI mattering. They are being tested against the host history.

Southern Cal sits at RPI 8 with a 42–14 record and a 9–11 mark against Q1 and Q2 opponents. The Q1+Q2 percentage is not dominant. That is the obvious counterargument, and it should not be ignored. But the visual comparison to the last 12 years of regional hosts is hard to dismiss. USC is not sitting in some empty corner of the chart where historical hosts do not live. It is sitting in a place that has usually belonged to hosts.

That is why Indiana State in 2023 is such a useful comparison. Indiana State was 9th in RPI at the time of selection, went 10–10 against Q1 and Q2 opponents, and had only one Q1 win. That profile was not spotless, but it was good enough to host because the overall body of work and the RPI position carried significant weight.

If Southern Cal is sitting at RPI 8 with a comparable broad profile, the question becomes fairly direct. Is the committee applying the same standard, or is this the year the standard changes?

The verdict

That is the real tension underneath the entire RPI debate. It is not whether someone can find a ranking system that likes Mercer less or Southern Cal less. Of course someone can. That is how ranking debates work. The question is whether those systems reflect the way the committee has actually behaved.

History says the committee has treated RPI as the primary framework, especially at the host line. It has made adjustments, but most of the time those adjustments have happened around the edges of the RPI picture, not in defiance of it. A 3.07 mean absolute deviation is not a committee wandering away from the formula. It is a committee living inside the formula and occasionally rearranging the furniture.

So the conclusion is not that RPI should end the conversation. That would be too simple, and it would ignore the very real judgment the committee has always had. The conclusion is that RPI has earned historical weight because the committee has given it historical weight.

For Mercer, that means the field case should be viewed through the lens of similar tournament entrants, not as a team that belongs on any first four teams out lists.

For Southern Cal, that means the host case should be viewed through a history in which RPI 8 teams have usually been treated like hosts, even when the résumé had imperfections.

Maybe this year is different. Maybe the committee places more weight on another system. Maybe 2026 becomes one of those seasons that stands out later, the way 2022 stands out now.

But if that happens, it should be understood as a change, not as business as usual. The committee has spent the last 12 tournaments telling us what language it speaks. The accent changes from year to year, and sometimes the sentence gets a little strange, but the language has been the same.

Notes & methodology: RPI was recomputed weekly Monday-by-Monday using the same modified-winning-percentage formula in effect since 2013 (1.3 weight for road wins, 0.7 for home wins, with mirrored loss weights). Quadrants follow the standard split: Q1 = home opp. RPI ≤30, neutral ≤50, away ≤75. Q2 = home ≤75, neutral ≤100, away ≤135. Records reflect Selection Monday snapshots, not end-of-tournament numbers. The 12-cycle window excludes 2020, when the tournament was cancelled. The "multi-bid league" backdrop in Figure 02 is the set of non-host entrants from conferences that placed two or more teams in the field that year, limited to those with an RPI of 60 or better (323 team-seasons); conference-tournament auto-bid winners whose RPI fell outside that range are excluded so the comparison reflects the at-large field. Conference assignments use 2026 alignment throughout.