Tennessee Soccer Stats
Newsletter · July 2026

The First ELO Change

TSSE was built in one offseason: eighteen seasons of results, a rating engine, and a research series, all in the weeks since the 25-26 season ended in June. Before the girls kick off in August, that research goes back into the engine. This is the changelog, and a note about what we're keeping under the hood from here on.

Why change anything

The first version of this rating system treated every program the same way twice: at birth and every offseason. A brand-new school started at 1500 plus a small class adjustment, whether it opened in a booming Williamson County suburb or a county that had never fielded a varsity team. And every June, every program's rating slid 25 percent back toward the mean, whether it graduated two players or eleven.

Both of those were reasonable defaults when the engine knew nothing. It knows things now. An offseason of reports established, with numbers, that a program's context predicts where it will settle (The Zip Code Effect), that keeping a core together beats renting one (The Freshman Report), that program experience matters where roster age does not (The Experience Report), and that the coach is the thread that holds the whole thing across graduation cycles (The Coaching Carousel). An engine that ignores its own site's findings is leaving accuracy on the table. So, two changes.

Change one · Smarter starting ratings

When a new program enters the pool, TSSE now hands it a prior instead of a blank slate. The logic is the same one a good scout would use before ever seeing the team play: what class will they play in, what division, and what does the neighborhood around the school look like?

The evidence for that last part is the strongest single finding on this site. Among public programs in the v1 ratings, the wealthiest quartile of neighborhoods carried an average rating 169 points above the poorest quartile in boys and 188 points in girls, and the private-public gap runs another 80 to 90 points on top of class effects. None of that is destiny. Bearden's quarter-century of coaching continuity beats its zip code, and Tri-Cities beats its own SEI every season. Which is exactly why the prior only takes a fraction of the observed gap:

ComponentRough weightWhere it comes from
Base1500Every program, both pools, unchanged.
Class & division≈ −50 to +120Historical cross-class results; the D2-AA privates enter highest, 1A enters lowest. Carried over from v1.
Neighborhood (SEI)up to ≈ ±45About a quarter of the observed quartile gap. Weighted slightly heavier for girls, where the money signal is stronger.
Hard cap±150 totalNo program starts more than 150 from center, ever.

A prior is a starting guess, not a verdict. New programs keep the accelerated update weight over their first 20 games, so a team that outplays its neighborhood gets credit fast. The point of the change is narrower than it might sound: for the first month of a new program's life, the engine now guesses like someone who has read our research instead of someone who hasn't.

Change two · Regression that reads the roster

The offseason is where high school ratings go to die. Every senior class walks at graduation, and a flat 25 percent slide toward the mean was our way of admitting we didn't know who lost what. But we track rosters now, season over season, and the rosters say the flat rate is wrong in both directions.

I paired every team-season roster with the rating change that followed it: 6,141 team-season pairs across twelve years. The share of a roster that graduates predicts the next season's rating change at r = −0.24, which for one offseason variable is a loud signal. Split into fifths, it looks like this:

Next-season rating change by share of roster graduating (6,141 team-seasons)
~9% seniors
+26
~18% seniors
+15
~23% seniors
+1
~30% seniors
−11
~42% seniors
−30
Average ELO change the following season, grouped into quintiles of senior share. Teams that keep their core don't just hold their rating, they grow it; teams that graduate the spine give back a competitive season's worth of points.

The coach matters too, independently. Programs that changed head coaches between seasons behave like an extra three points of regression compared to programs that kept the same voice on the sideline. Put the two together and the spread is stark: a team returning its core under the same coach holds about 93 percent of its distance from the mean; a team that graduates heavy and changes coaches holds about 87. A flat rate splits the difference and gets both teams wrong.

6,141team-season pairs analyzed
−0.24corr: senior share vs next-season change
+26 → −30avg swing, lowest to highest turnover
~2×regression gap, calm vs turbulent offseasons

So the offseason slide is no longer one number. From the 2014-15 season forward, as far back as the roster registry reaches, each program's regression is set from what it actually loses; older offseasons keep the flat rate, since there is no roster to read:

TermRough weightWhat it reads
Base regression≈ 18%The floor case: an ordinary roster, ordinary turnover.
Graduation term≈ ±8 pts of regressionScales with senior share relative to the state median (~24%). Heavy senior classes regress more; returning cores regress less.
Coach change≈ +4 ptsA new head coach between seasons, from the same roster registry behind The Coaching Carousel.
Bounds10% – 35%Hard clamp. No roster data for a season means the old flat rate applies.

One deliberate limit: the regression reads how many players a program loses, not who they are. This site does not track individual minors by name, and the rating system will not either. Team-level counts get us most of the signal with none of the discomfort.

The official curtain

From v2 on, the exact constants are private.

Version 1 of this system was documented down to the update formula, and I think that was right for building trust while the site was new. It has a cost, though. A fully published formula invites schedule-gaming (knowing precisely which opponent moves your rating most), it makes every future tuning decision a public re-litigation, and it hands the whole thing to anyone who would rather rebrand it than link it.

So TSSE is adopting the same posture as every serious rating system, from FIFA's SUM formula to Nate Silver's club ratings: the structure, constraints, and rough weights are public (this page is that disclosure), and the exact constants live behind the curtain. You can always check the output against reality; every rating, every game, and every dataset on this site stays free and open.

Public, forever
  • Boys and girls rated in strictly separate pools
  • Chronological walk of every game; no retroactive edits to played results
  • Goal margin matters, logarithmically, capped so running up the score is worthless past a point; scores capped at 19
  • Updates are zero-sum between the two teams in a match
  • New programs: prior of 1500 + class/division + a bounded SEI term (±150 total cap), then a fast-learning window for their first 20 games
  • Offseason regression bounded between 10% and 35%, set by graduation share and coach continuity
  • Manual overrides exist only for data errors (wrong score, wrong team), never for ratings
  • All ratings, match logs, and the datasets behind them: free, downloadable, CC BY 4.0
Behind the curtain
  • The K-factors (base and fast-window)
  • The exact margin multiplier and its cap
  • The precise class/division and SEI prior constants
  • The regression slope and the coach-change constant
  • The cross-class win-probability anchors
The re-rate

These changes are not a patch stapled onto the end of the list. In July 2026 the engine re-walked every game since 2008 under v2: each program that ever entered the pool entered with its new prior, and every offseason the roster registry can see applied the roster-aware regression. History is re-rated, not rewritten; every result stays exactly as it was played.

The outcome moves like a recalibration, not an earthquake. The v2 list correlates with v1 at 0.999 in both pools, and the average ranked program moved about 11 points in boys and 17 in girls. The programs that moved most are exactly the ones the research said should move:

Moved up under v2
  • Houston girls +64, into the state's top three
  • Baylor girls +63; Webb School girls +64
  • McCallie boys +44; Brentwood boys +29
  • The pattern: strong programs with returning cores and stable sidelines stop being over-regressed every June
Moved down under v2
  • Mostly lower-table programs with heavy annual turnover, by 40 to 70 points
  • Ravenwood passes Franklin for #2 in boys; Collierville joins the girls' top five
  • The pattern: hollowed-out rosters no longer get the benefit of the doubt at the season line
The research behind the changes
Analysis in this newsletter: 6,141 team-season pairs joining the TSSE season snapshots (2014-15 through 2025-26) to the roster registry (team-level grade counts only, no player names) and the coaching registry. Regression figures are empirical year-over-year persistence of rating deviations. Tennessee Soccer Stats is a personal, independent project and is not affiliated with the TSSAA or any school. Questions, corrections, or a methodology argument you want to have? Get in touch.