How the model works
From MLS results to rankings and matchup probabilities
Ascend’s model (the ATT Model) keeps two questions separate: how strong a team looks across the season, and how a specific matchup is likely to unfold. Reader pages call its ATT Rating “Rating” and its ATT Forecast “Match predictions.”
1. What the model is trying to answer
ATT Rating asks which MLS teams are strongest right now based on the results accumulated across the season. Match probabilities ask a different question: given the two teams and the venue, how often should the home win, draw, and away win occur?
Neither answer is a guarantee. The value is a consistent baseline that makes movement, disagreement, and uncertainty easier to see from one matchday to the next.
2. What goes into the rankings
Completed MLS scores and results move each team’s rating. The size and direction of the change depend on the opponent: beating a highly rated team means more than beating a much lower-rated one, while losing as a strong favorite costs more.
The calculation also gives the home team a fixed venue advantage before the result is evaluated. ATT Rating adds a restrained score-margin adjustment, so a multi-goal result can carry more information without allowing one match to overwhelm the season.
3. How match probabilities are produced
The scoring model builds separate home and away attack and defense rates from completed goals. Those rates are compared with league scoring averages and lightly pulled toward the league average when a sample is small.
For each fixture, the two teams’ attack and defense profiles produce a home and away expected-goals rate. These are model-generated scoring expectations, not observed expected-goals data. A Poisson score grid turns those rates into possible scorelines and then into home-win, draw, and away-win probabilities.
Dixon-Coles makes a small correction to 0–0, 1–0, 0–1, and 1–1, where independent Poisson estimates can be least realistic. The adjusted score grid is normalized before the three result probabilities are published.
4. How ATT Rating changes after results
ATT Rating is continuous: each completed result starts from the ratings produced by all earlier eligible results. It does not restart because a source matchday arrives late or appears out of calendar order.
A result moves both teams in opposite directions. Expected results usually produce smaller changes; surprises and larger winning margins can produce larger ones. The updated value becomes the starting point for future rankings and matchup probabilities.
5. Rating vs form vs standings
ATT Rating is the rolling team-strength measure used to order the Power Rankings. Form is the short record of recent matches. Standings record points earned under league competition rules. They can disagree without one of them being wrong.
A club can be in excellent five-match form while its longer rating remains cautious, or rate strongly before its table position catches up. The site labels these signals separately so one is not presented as another.
6. When the site updates
Each eligible matchday cycle has one authoritative pre-matchday calculation and one authoritative post-matchday calculation. The post-matchday update waits until every included match is final and a four-hour settling period has passed.
Official completed makeup matches are included. When the source matchday is out of chronological order, the site labels the edition a Schedule-adjusted MLS update, keeps the source matchday as context, and carries the season rating forward without pretending the model rewound.
7. Technical details for readers who want them
- Teams begin the historical replay at 1500; final 2025 ratings initialize 2026.
- Plain Elo and ATT Rating use a fixed K-factor of 20 and a 65-point home-field adjustment.
- ATT Rating uses a capped logarithmic goal-margin multiplier and no in-season damping.
- Poisson home/away rates use five league-average pseudo-matches for shrinkage and clamp expected scoring rates between 0.15 and 4.50.
- Dixon-Coles uses a fixed correlation of −0.08 for the four low-score cells.
- Power Rankings sort by ATT Rating, then plain Elo, season goal difference, and stable team identity.
- A featured matchup balances probability balance, attacking activity, and team quality. Highlighted picks must clear a 40% confidence floor and never exceed five.
8. Data/source note
Match data is supplied by American Soccer Analysis. Final goals and results are the observations that update Elo and ATT Rating. Source expected-goals data is supplemental context and validation data; it never replaces a missing final score, and a missing value is never displayed as zero.