Vols Data Lab

Raw margin lies. Beating an FCS team by 40 is not the same as beating Georgia by 3. To weight competition you need to know how good each opponent is — but that depends on how good their opponents were. The schedule graph is circular.

rating(i) = average over games of [ adjusted margin + rating(opponent) ]

We start every team at zero and iterate until nothing moves. That fixed point is the only set of ratings consistent with every result in the league — which is why this needs all 452 FBS games, not just Tennessee's.

• Home field: the home team's margin is reduced by 2.5 points before it counts.
• Blowout cap: margins pass through a smooth cap at 35 points, so a 50-point win counts meaningfully more than a 20-point win but nowhere near twice as much. Running up the score earns very little.
• Non-FBS opponents: pinned at -28, so scheduling three FCS teams never looks like playing three average FBS teams.
• Small samples: ratings regress toward zero by n/(n+1), so nobody is crowned off one September blowout.

Units are points per game against an average FBS team on a neutral field. A +14 team should beat a +3 team by about 11 on neutral turf.

The dashed line is what the rating expected: against an opponent rated -5.3 the margin should be about +14.2. Dots above it are games Tennessee played better than its own rating; dots below are worse.

Expected wins sums each game's win probability. Comparing it with the actual record separates “good” from “fortunate” — a team can win close games it had no business winning, and the gap shows up here.

Volatility is the spread of those gaps. A low number is a team that performs the same every week; a high number is one nobody can predict.

Games against FCS opponents are excluded — an unrated opponent has no expectation to measure against — so the record here can differ from the headline record. Both are shown rather than quietly reconciled.

Schedule D is the average defensive rating of the opponents faced — how hard the slate was. Higher is a better defense.

Weighted D is the same average, but weighted by production. It answers a different question: of the defenses this player actually gained yards against, how good were they?

weighted D = Σ(production × opponent D) / Σ(production)

The gap between them is the story. A positive gap means production skewed toward the tougher defenses. Negative means it was padded against the weaker ones.

We deliberately do not publish a synthetic “adjusted yards” figure. Converting a points-based team rating into an individual yardage projection would require a yards-per-point constant nobody can justify. Every number above is exact arithmetic on real per-game results — no modeling assumptions, nothing to take on faith.

Known limitation: opponent quality here is measured in points, not yards. A defense can prevent points while giving up yardage, so this is a proxy rather than a direct measure. A yards-based rating is wired up and switches on automatically once a CFBD_API_KEY is configured.