Attack & Defence Strength Ratings: Parameterizing Football Poisson Models

1. Introduction: The Parameterization Bottleneck of Football Poisson Models

In quantitative association football analytics, the Poisson probability distribution serves as the fundamental workhorse for forecasting match outcomes, calculating fair 1X2 probabilities, establishing Asian Handicap lines, and constructing exact scoreline matrices. Under the standard bivariate Poisson formulation, the number of goals scored by the home side ($X$) and the away side ($Y$) across 90 minutes are modeled as discrete random variables characterized by intensity parameters $\lambda$ and $\mu$:

While the mathematical machinery for converting $\lambda$ and $\mu$ into market probabilities is well understood, recreational modelers and beginner analysts invariably stumble upon the foundational operational bottleneck: where do the expected goal intensities $\lambda$ and $\mu$ actually come from?

Subjective heuristics, media narratives, and unadjusted historical averages fail to account for schedule strength disparities, home field advantage dynamics, and asymmetric defensive efficiency. An attacking team that scores three goals against a relegation candidate cannot be expected to generate the same offensive output when traveling to face the league's stingiest defense.

This technical guide details the exact mathematical methodology for calculating attack and defence ratings from empirical league standings, derives the joint expectancy formulas, works through a step-by-step example with hypothetical team figures, introduces exponential time-decay weighting, and compares raw historical goal statistics against modern Expected Goals ($xG$) frameworks.

2. League Baseline Constants: Dissecting the Macro-Environment

Before evaluating individual clubs, a model must calibrate the macroeconomic scoring baseline of the target competition. Scoring baselines differ between competitions and from season to season, so they must be measured for the league being modelled.

Key League Baseline Parameters

1. Total Matches Played ($N_{\text{matches}}$): For a standard 20-team league operating a double round-robin format, the season comprises $20\times 19=380$ fixtures.
2. Total Goals Scored at Home ($G_{\text{home, total}}$): The aggregate number of goals scored by all home teams across the sampling window.
3. Total Goals Scored Away ($G_{\text{away, total}}$): The aggregate number of goals scored by all visiting teams.
4. Average Home Goals per Game (${\overline{G}}_{\text{home}}$): $${\overline{G}}_{\text{home}}=\frac{G_{\text{home, total}}}{N_{\text{matches}}}$$
5. Average Away Goals per Game (${\overline{G}}_{\text{away}}$): $${\overline{G}}_{\text{away}}=\frac{G_{\text{away, total}}}{N_{\text{matches}}}$$

In the English Premier League across the 2023–24 and 2024–25 seasons (760 matches, football-data.co.uk), the baselines were ${\overline{G}}_{\text{home}}\approx 1.66$ and ${\overline{G}}_{\text{away}}\approx 1.45$, an average of ${\overline{G}}_{\text{total}}\approx 3.11$ goals per game and a home advantage multiplier of $1.66/1.45\approx 1.14$.

3. Formulating Attack Strength Indices ($A_i$)

A team's Attack Strength Index measures its offensive potency relative to the league average across equivalent venue conditions (home or away). It is formally defined as the ratio of goals scored by the club per game to the league's baseline scoring rate:

Home Attack Strength ($A_{\text{home},i}$)

Away Attack Strength ($A_{\text{away},i}$)

An index of $1.00$ represents a perfectly average attacking unit. An attack rating of $A=1.35$ signifies that the club scores 35% more goals than an average team in that venue, whereas $A=0.75$ indicates an offensive unit generating 25% fewer goals than the league norm.

4. Formulating Defence Strength Indices ($D_i$)

Conversely, a team's Defence Strength Index measures its defensive resistance relative to the league baseline. Crucially, lower defence values signify superior defensive performance, as a lower index indicates fewer goals conceded:

Home Defence Strength ($D_{\text{home},i}$)

Away Defence Strength ($D_{\text{away},i}$)

Notice the vital denominator pairing: when evaluating Team $i$'s home defence, we divide by the league average away goals scored (${\overline{G}}_{\text{away}}$), because away teams are the opponents attempting to score at Team $i$'s stadium. Similarly, Team $i$'s away defence is scaled by league average home goals (${\overline{G}}_{\text{home}}$).

5. Synthesizing Expected Goal Intensities ($\lambda$ and $\mu$)

With individual strength ratings computed, forecasting expected goals for an upcoming match between Home Team $H$ and Away Team $A$ becomes an elegant multiplicative synthesis:

This formulation accounts for all three primary drivers of football goal scoring:

1. The venue-specific baseline scoring environment (${\overline{G}}_{\text{home}}$ or ${\overline{G}}_{\text{away}}$).
2. The attacking prowess of the scoring team in that venue ($A_{\text{home},H}$ or $A_{\text{away},A}$).
3. The defensive vulnerability of the defending opponent in that venue ($D_{\text{away},A}$ or $D_{\text{home},H}$).

6. Step-by-Step Worked Example: Arsenal vs. Liverpool

To demonstrate the operational execution of this methodology, let us calculate the Poisson parameters for a marquee Premier League fixture: Arsenal (Home) vs. Liverpool (Away).

Standings Data (Hypothetical Mid-Season Sample)

• League Averages: ${\overline{G}}_{\text{home}}=1.55$ goals/game, ${\overline{G}}_{\text{away}}=1.20$ goals/game.
• Arsenal at Home (10 matches): 26 goals scored ($2.60$/game), 8 goals conceded ($0.80$/game).
• Liverpool Away (10 matches): 22 goals scored ($2.20$/game), 11 goals conceded ($1.10$/game).

Step 1: Compute Arsenal's Home Ratings

Step 2: Compute Liverpool's Away Ratings

Step 3: Calculate Expected Goals ($\lambda$ and $\mu$)

Feeding $\lambda =1.846$ and $\mu =1.467$ into an independent Poisson model gives fair match probabilities of $P(\text{Arsenal Win})\approx 46.6\%$, $P(\text{Draw})\approx 22.6\%$, and $P(\text{Liverpool Win})\approx 30.8\%$. A Dixon-Coles adjustment would move a little probability into low-scoring draws.

7. Advanced Calibration: Time-Decay Weighting ($\xi$)

A fatal flaw of using simple cumulative season averages is the assumption of stationarity: it assumes a team's performance six months ago under an injured key striker carries the same predictive validity as their performance last weekend. In Dixon and Coles' (1997) seminal framework, historical matches are weighted exponentially using a time-decay parameter $\xi$:

Where $\mathrm{\Delta }t$ represents the time elapsed in days or weeks since match $t$. The value of $\xi$ is fitted on past matches. With $\xi =0.002$ per day, a match played two weeks ago carries about 1.4 times the weight of one played six months earlier; with $\xi =0.005$, about 2.3 times.

8. Limitations: Goals vs. Expected Goals ($xG$)

While historical goal data is readily accessible, realized goals are notoriously noisy. Over a 10-match home sample, a team may score 25 goals from only 15.2 Expected Goals ($xG$) due to unsustainable finishing luck or goalkeeping errors. Modern quantitative syndicates substitute raw goal counts with Expected Goals ($xG$) and Expected Goals Against ($xGA$) when computing $A_i$ and $D_i$, because xG varies less from match to match than goals and the ratings settle after fewer games.

9. Home Advantage Is Not Constant

Home advantage changes over time and has to be re-estimated. In the 2,490 Premier League matches from August 2019 to January 2026 (football-data.co.uk), the ratio of home to away goals ranged from 1.01 in 2020–21, when matches were played without spectators, to 1.34 in 2022–23, with an average of 1.18. Crowd effects on referees, travel and tactical trends are possible drivers. In practice, re-estimate the home baseline over a rolling window of recent seasons rather than relying on a long-term average.