SEC Versus Big Ten, Through Everyone Else

The SEC is better than the Big Ten, but not at the top. Their best teams are even; the SEC wins on depth.

On the night of September 12, #1 Ohio State played at #4 Texas in Austin. Texas won, 24 to 23.

Vegas had Texas as a 2.5-point favorite. The rating this essay uses had Ohio State by 2. It picked the loser.

Two other games linked the SEC and the Big Ten that Saturday. Mississippi State won at Minnesota, 38 to 13, and Michigan beat #11 Oklahoma, 17 to 10. The two leagues haven't met again all regular season.

Is the SEC better than the Big Ten? Fans and writers argue it every fall. The two leagues rarely play each other, so settling it through their direct matchups rests on thin evidence. The index this essay develops runs through everyone else instead. The two leagues played 66 games against other leagues. We join those and every other FBS game this season into one network.

Let's start small, inside one league, with a big upset from October 2026.

Is one shared opponent enough?

Auburn versus Georgia. The Deep South's Oldest Rivalry. Before they met on October 17, the two SEC teams had one opponent in common. Both beat Vanderbilt at home, Georgia by 24 and Auburn by 6. Through Vanderbilt, Georgia looks 18 points better.

One opponent is not much to go on, so we turn to the Massey rating, which counts every game.1 It gives each team one number, the margin it would beat an average FBS team by on a neutral field. Each team starts at last season’s number. Every game it plays then moves that number toward how the team is playing now.

For data scientists: This is Kenneth Massey's least-squares rating of scoring margins, with David Harville's carry-over as a Bayesian ridge prior, refit every week.

Before kickoff, the Massey rating had Georgia 19.8 points better, close to Vanderbilt's 18. Add home field and a small adjustment for big gaps, and Georgia was a 21-point favorite with a 90% chance to win.2 Table 1 shows each step. Vegas had Georgia by 19.5.

Auburn won, 27 to 24.

Georgia turned the ball over three times. The last came with the score tied and 58 seconds left, a fumble recovered at Auburn's 44. Auburn moved into range, and Alex McPherson kicked a 34-yard field goal as time ran out.

One shared opponent is not enough. You need all the games. Even then, the Massey rating gave Auburn about one chance in ten, and Auburn took it.

Should an upset count?

A Georgia fan might call that loss a fluke, one that shouldn’t count against Georgia. Famous upsets cut both ways. When Michigan beat Ohio State in 2024, keeping the game made four of the next five forecasts worse. Ohio State went on to win the national title. When Vanderbilt beat Alabama that season, keeping the game made later forecasts better. Table 2 sets both beside Auburn’s win.

So we tested 33 ways of giving less weight to the games the forecasts missed by the most.3 None forecast later games measurably better than keeping every game.

Since no way of discounting upsets forecast better, the Massey rating keeps every game, Auburn's win over Georgia included.

Who links the SEC to the Big Ten?

Texas against Ohio State was one of only three games between the SEC and the Big Ten all regular season. Three games are far too few to compare the leagues' 34 teams. The opponents the two leagues share fill the gap. Let’s consider them one at a time.

Western Kentucky, of Conference USA, lost at Georgia by 50 and at Indiana by 38. Through the Hilltoppers, Georgia is 12 points better than Indiana. This path alone favors the SEC.

Kent State, of the Mid-American, lost at South Carolina by 57 and at Ohio State by 56. Through the Golden Flashes, the two are a point apart. This path alone calls it a tossup, leaning SEC.

UTEP, of the Mountain West, lost at Oklahoma by 51 and at Michigan by 35. Through the Miners, Oklahoma is 16 points better. This path alone favors the SEC by more than two touchdowns. Then Michigan beat Oklahoma by 7.

Each is a path, a chain of games from one league to the other (Figure 1). Any one game can turn on a fumble, a missed kick, or a penalty. A comparison built on two games carries that luck with it. Play them again, and the comparison might favor the other team. The Massey rating uses every path at once, and leans hardest where many short paths agree.

What comes out is a league gap: How many points the average team of one league would beat the other's by. That gap, with an honest margin of error around it, is the Richardson Index, the measure we developed to answer this question.4

For data scientists: The Richardson Index sits on top of Massey's ratings. It adds league gaps with a linking error in their margin of error, each game's share of a comparison, and tiers inside each league. The network view of Massey's method is Bozzo and Franceschet's.

How much does each game count?

A casual fan would settle which league is better with head-to-head games. The SEC and the Big Ten played just three. How much of the answer do those three decide? Only 14% of it.5 Every game this season carries a share of the evidence, by how much it reveals about the two leagues. The 66 games the two leagues played against outside opponents carry 71% of the evidence (Figure 2). Games between other leagues carry 13%, and games inside the SEC or the Big Ten carry under 3%.

On the same Saturday Texas beat Ohio State, Mississippi State won at Minnesota by 25. Both teams sit near the bottom of their leagues. Through week 11, the Richardson Index puts the SEC 4.88 points ahead of the Big Ten. Take away the Mississippi State game, and that lead falls to 4.25. Take away Texas over Ohio State, and it moves by a hundredth of a point. Texas and Ohio State played the bigger game, but a one-point finish says little. A 25-point win by a team with few other ties to the other league says a lot.

How wrong could the league gap be?

How sure can anyone be of that 4.88-point lead? Start with who played whom. Two leagues joined by many games can be compared closely. Two joined by only a handful cannot, because one odd result can move the whole comparison, as Mississippi State's win at Minnesota did. From who played whom alone, the SEC and Big Ten gap in the Richardson Index has a margin of error of about 6 points (Figure 3). For the ACC and the Pac-12, who never met this season, it is about 7.

Last season's ratings cut that margin of error. The Richardson Index starts every team at last season's rating, then adds this season's games.6 So the SEC leads the Big Ten by 4.88 points, with a margin of error of about 4.

For data scientists: The schedule's margin of error comes from the effective resistance between the two leagues, scaled to points. The Richardson Index's 90% margin of error has two parts: The ratings' own uncertainty and a linking error. The linking error comes from a jackknife, which drops each game in turn and measures how far the league gap moves. It covers 69 games, 66 against other leagues and 3 between the SEC and the Big Ten.7

So which league is better?

That margin of error still leaves the SEC ahead. The lead could be as small as 0.7 points or as large as 9.0, and both ends favor the SEC. On a neutral field, an average SEC team would beat an average Big Ten team about 64% of the time. Five other ways of measuring the lead also put the SEC ahead, by 3 to 5.5 points a game.8

A Big Ten fan might complain that starting every team at last season's rating hands the SEC a head start. It does the opposite. Last season ended close, with the SEC only 2 points ahead. This season's games alone put the SEC about 8 points ahead.9 The Richardson Index blends the two, so it lands between them, at 4.88.

Since 2001, the Big Ten has led the SEC once, in 2005, by 2.8 points. The SEC has led every other season (Figure 4). Each season, take the best league and measure how far it was ahead of the second-best. The ten biggest of those leads since 2001 all belong to the SEC. This season's SEC lead is the biggest of them all.

Where does the SEC's lead come from?

Since the playoff began in 2014, the SEC has won six national titles, the Big Ten four, and the ACC the other two. Big Ten fans can brag that the last three went to them: Michigan, Ohio State, and Indiana. SEC fans love to talk about how tough every week in their league is. More than half of the SEC's teams are in the playoff committee's latest top 25, but only a third of the Big Ten's. To see whose claim holds up, we compared the leagues level by level.

So we split each league into tiers: Groups of teams whose ratings sit too close together to say which team is better. This season each league splits into four. In Figure 5, choose a tier to compare it with the other league's tiers, or switch to the national ladder of all 35 tiers.

The top tiers are two teams each, and Georgia and Alabama lead Ohio State and Indiana by 4.2 points. That is less than the margin of error, so statistically, neither top tier is better than the other. Below the top tiers, the SEC's lead grows: 5.7 points in the second tier, 6.7 in the third, and 10.2 in the bottom. The last two leads are bigger than their margins of error.

For data scientists: The tiers come from complete linkage on each league's ratings. No two teams in a tier differ at the 95% level. A credible set finds almost the same top and bottom tiers.10

SEC fans are right about the gauntlet: Even the SEC's bottom tier is level with the Big Ten's third tier. It ranks 12th of 35 nationally. The two leagues' top tiers are even. The SEC's lead comes from its depth.

What should fans watch for next?

The next SEC and Big Ten meeting may come in the playoff. In simulations of the rest of the season, an SEC against Big Ten playoff game turns up 92% of the time.11 Three bowls also pair the two leagues, and the Richardson Index can forecast them before the teams are known. Each bowl usually takes teams of the same rank from each league. At those ranks, the SEC has a 68% chance in the ReliaQuest Bowl and 63% in the Citrus Bowl. The Music City Bowl is a coin flip, leaning slightly toward the Big Ten. Bowl games are harder to read, though. Players skip them for the draft or the transfer portal, and some teams play under interim coaches.

Before any of that, Clean, Old-Fashioned Hate. On November 28, Georgia hosts Georgia Tech. The two teams have no opponent in common, so the forecast runs through the leagues. Georgia rates first in the SEC and Tech 13th of 17 in the ACC. The Richardson Index puts the SEC 11.7 points above the ACC, the biggest gap between the two leagues in this game since 1983 (Table 3, Figure 6). As of November 17, Vegas had not posted a line. Tech last beat Georgia in 2016, and Georgia has won the eight meetings since.

For my part, I hope the rating's pick is wrong. Georgia Tech is my alma mater and where I first learned data science. The Massey rating has Georgia as a 32-point favorite, leaving Tech with just a 2% shot at winning. So you're telling me there's a chance.


How This Was Made

Data

Every FBS game from CollegeFootballData, regular season and bowls, 2001 to 2026, with the rivalry history back to 1946. This season runs through week 11, every game through November 15, from a snapshot taken November 17, 2026.

The rating

Each game is one line of a least-squares fit (Massey, 1997). For home team \(i\) and away team \(j\), with home margin \(m_g\) and home field \(h\):

$$m_g = r_i - r_j + h + \varepsilon_g$$

Each rating has a normal prior centered on the team's full rating from last season, with a spread of 8 points (Harville, 1980). That makes the fit a Bayesian ridge regression, refit every week. The prior is worth about three games of evidence, and home field starts at 2.5 points. The prior soaks up offseason change, the transfer portal included; from week 7 a typical team's own games outweigh it. We chose this engine against three other rating methods by forecasting real games. On 200 simulated seasons with known answers, played on real schedules, it tracked the planted truth at a correlation of 0.93.

The forecast

A team's place \(p\) is its rating minus its league's average, and the league gap \(L\) is the difference between two league averages. With \(H\) the home-field term and \(s = +1\) for the home side, \(-1\) for the away side:

$$\hat{y} = 0.866\,\left(\Delta p + L + H\right) + 0.94\,s$$

Both constants were fitted on every FBS game from 1950 to 2025, each margin from the home side. The fit used 48,061 games, each scored with point-in-time ratings from earlier games only. The tables' adjustment is \(\hat{y}\) minus the plain sum. The win chance is a normal curve of past forecast misses, with a spread of 15.87 points. Each game's spread moves a little with how sure its ratings are. Table 4 walks the Texas and Ohio State forecast the same way, from Ohio State's side.

Upsets

A big upset is a win by a team the rating gave under 20%. The 33 discounting versions came in nine families. Among them: Margin caps, heavy-tailed fits, and rules that drop or shrink a game beyond a noise floor. Each was tuned on 2003 to 2014 and scored on 2015 to 2025, with whole seasons resampled. Matched on the size of the miss, an upset teaches the rating 1.2 times as much as a favorite's win (90% range, 0.8 to 1.7). For about four in ten big upsets, leaving the game out would have improved the later forecasts, the same as for ordinary games. Across 576 big upsets from 2003 to 2025, 56% of the 6,664 later forecasts came closer without the upset. The Brier score still favored keeping them: When an upset helps, it helps a lot. Every discounting family's best version came within 0.003 points a game of keeping every game, held out.

The Richardson Index

It is built on Massey's team ratings and adds four things. The first is the gap between any two leagues' average ratings. The second is a 90% margin of error on that gap, including a linking error. The third is each game's share of a comparison, and the fourth, tiers inside each league. The network view comes from Bozzo and Franceschet, who read Massey's method as a measure of each team's place in the network of games.

The schedule's margin of error

From this season's regular-season games alone, with no starting ratings, the standard deviation of a league gap is

$$\sigma_{\text{sched}} = 15\,\sqrt{R_{\text{SEC, B1G}}}$$

where \(R\) is the effective resistance between the two leagues in the network of games. That gives 3.79 points, a 90% margin of error of 6.2. Treat every game as a 1-ohm wire and run one unit of current from the SEC's teams to the Big Ten's. A game's share of the comparison is the power its wire uses; the shares sum to 1. Leaving one game out is computed exactly with a rank-one update, checked against full refits.

The margin of error

The Richardson Index's margin of error adds the ratings' own uncertainty and a linking error. The linking error is a jackknife over the 69 games the two leagues played outside their leagues. The idea comes from how test makers link two exams:

$$\sigma^2 = \sigma^2_{\text{rating}} + \sigma^2_{\text{link}} = 2.13^2 + 1.36^2 = 2.53^2$$

So through week 11, the Richardson Index puts the SEC 4.88 points ahead, with a 90% margin of error of 4.2. On seasons with planted answers, these margins of error held the truth 91% to 95% of the time. Mississippi State over Minnesota makes up 20% of the linking error; without it the lead falls to 4.25. Without the three direct games it is 4.49 (0.32 to 8.66).

The baseline start

Starting every team at zero, the 2026 gap is 8.2 (0.4 to 15.9). The carried start put the SEC only 2.0 ahead. Through week 11 the Richardson Index still leans 54% on it, so it lands between the start and this season's games alone.

Other checks

With bowls left out, our gap is 5.5, CollegeFootballData's Elo 3.0, and the same rating fitted to Vegas spreads 4.1. Read at the same 16 places in each league, the gap is 4.7. With the Big Ten's two worst teams dropped and every SEC team kept, it is 3.0.

The tiers

Complete linkage across the whole league, chosen from four ways of splitting by how well each recovered planted leagues with known tiers. The recovery score is modest: 0.47 on the adjusted Rand index, where 1 is exact and 0 is chance. A league with no real tiers still tends to come out in about four tiers. The credible set adds three teams to the tiers it finds. It puts Texas in the SEC's top tier, Oregon in the Big Ten's, and Minnesota in the Big Ten's bottom tier. Tiers are found and compared with the same season's ratings, so read the tier leads as a description of the season.

Elo and Vegas

Ours, five Elo versions, and the Vegas line were scored on the same 4,679 games, 2019 to 2025 (Figure 7). The average miss of the forecast margin was 12.84 points for ours and 12.21 for the Vegas line. CollegeFootballData's Elo missed by 12.97, the other four by 13.18 to 13.79. Ours, that Elo, and the Vegas-fitted rating agree on the better league in all 14 seasons since 2013. Across all league pairs they agree 92% of the time.

The playoff

We played out the rest of the season 20,000 times from the ratings through week 11, with this season's rules for the field. A stand-in for the committee ranks teams by rating minus 3 points a loss. It matched the committee's real final top 25 closely in 2024 and 2025. An SEC against Big Ten final comes up 17% of the time. At least one SEC against Big Ten playoff game comes up 92% of the time. The SEC wins the title in 47% of runs, as often as Notre Dame, the Big Ten, and the ACC together.

Every pair of leagues

The Big Ten and the SEC are one of 45 pairs this season. Figure 8 gives the same comparison for every pair, every season since 2001.

The Machine

This series' pick'em model has its picks for the rest of the season on its page. A weekly scoreboard sets it against Vegas, the Crowd, and the AP poll.

Hardware

An M5 MacBook Pro with 32 GB of memory.

Help

Claude helped write the code and edit the essay.

References

  • Massey, "Statistical Models Applied to the Rating of Sports Teams," honors thesis, Bluefield College, 1997.
  • Harville, "Predictions for National Football League Games via Linear-Model Methodology," JASA 1980.
  • Bozzo and Franceschet, "The Massey's Method for Sport Rating: A Network Science Perspective," arXiv 2017.
  • Kolen and Brennan, Test Equating, Scaling, and Linking, 2004.