How Ascend the Table measures MLS teams

Source data

American Soccer Analysis is the V1 data source. Final MLS league scores and results are the canonical observations. ASA expected-goals data may add context or support validation, but it never replaces a missing final score and missing xG is never shown as zero.

The model sequence

1. Validate eligible league matches and map both participants to the active 30-team roster.
2. Roll plain Elo and Elo-modified ratings forward in kickoff order.
3. Fit home and away scoring strengths from final goals with explicit shrinkage.
4. Build a Poisson score matrix and apply the Dixon-Coles low-score correction.
5. Validate rankings, probabilities, selections, cutoff, and publication completeness.

Elo and Elo-modified

Elo estimates relative team strength from results and opponent strength. V1 uses a fixed K-factor of 20 and a 65-point home-field adjustment. There is no rating damping.

Elo-modified adds a bounded goal-margin multiplier. It is the primary power-ranking order. Plain Elo remains separately labeled, and neither is blended with a third-party published ranking.

Final 2025 ratings initialize the 2026 preseason state. Ratings then roll forward without an in-season reset.

Poisson and Dixon-Coles probabilities

Poisson estimates how often each scoreline should occur from the two teams’ home and away scoring strengths. Dixon-Coles adjusts the four low-scoring cells—0–0, 1–0, 0–1, and 1–1—before the score matrix is normalized.

These models produce matchup probabilities and strength diagnostics, not another Elo rating. A 55% home-win probability describes the modeled matchup; it is not a guarantee.

Game of the Matchday and five picks

The featured match balances competitiveness, expected attacking activity, and the mean quality of the two teams. The fixed V1 weights are 45%, 30%, and 25%, respectively.

For picks, the model takes the most likely outcome in each fixture and requires at least 40% confidence. Eligible fixtures are sorted by confidence and separation; at most five are selected. If fewer qualify, the publication says so.

Matchday update cadence and freshness

There are two authoritative compute runs per matchday cycle: one pre-matchday run at the first eligible evaluation on T-2, and one post-matchday run after every match is final plus a four-hour settling buffer. Later article stages can reuse a valid pre-matchday artifact; they do not create extra model runs.

Every ranking and team page shows its data cutoff and methodology version. A failed run cannot replace the last valid publication.

Known limitations

• Home advantage and the Dixon-Coles correlation are fixed V1 constants.
• The score model does not include player availability, lineups, or injuries.
• There is no cross-league strength adjustment.
• Schedule irregularities can require an explicit, reviewed cycle override.
• Recent form can diverge from the longer-run rating; neither should be mistaken for standings.