Beyond shots:

A new approach to quantifying

scoring opportunities

Dan Altman

@NYAsoccer
#OptaProforum
Beyond shots – Dan Altman

Why quantify scoring opportunities?

• Basing player ratings on points is difficult

It’s hard to tell which goals and plays were pivotal

Shapley values offer one possibility, but they depend on hypotheticals

• Basing player ratings on goals is difficult, too

Goals can be idiosyncratic, and many players aren’t directly involved in them

• We need less noisy metrics for the performance of players all over the pitch

Metrics for individual players should sum to a team metric correlated with results

Metrics for players should be consistent from year to year and give good incentives

Metrics for teams should pick up predictive power early in the season
Beyond shots – Dan Altman

The status quo: expected goals based on shots

• Why use shots?

Shots are more common than goals, and creating shots seems to be a more
consistent part of performance than scoring goals

• How are expected goals calculated?

Each shot has an estimated chance of becoming a goal based on historical

averages in the same location and situation

• What can we use expected goals for?

Breaking down expected goals into contributions by individual players results in

player ratings that sum to a team’s expected goals and are persistent over time

Early in the season, a team’s expected goals are a better predictor of its final
position in the table than many other metrics, including shot ratios
Beyond shots – Dan Altman

The status quo: expected goals based on shots

Ratings give credits

for contributions to
own team’s expected
goals and demerits
for responsibility for
opposing team’s
expected goals

Expected goal
difference was
correlated at 0.83
with final positions in
the table in 2012-13
Beyond shots – Dan Altman

Is there an alternative?

• A shot is just one kind of situation on the pitch

The estimated chance of scoring a goal within a given time period can be attached
to any class of situations that can be measured consistently over the season

• Player ratings can be based on situations as well

Just as for shots, contributions to situations can be divided up among players, and
arbitrary rules of thumb for dividing credit can be avoided

• Looking behind the shots may yield more useful information

Simply measuring the chances of scoring from shots says nothing about how the
shots were created

If certain situations usually lead to high-quality shots and goals, they may be a
better basis for ratings
Beyond shots – Dan Altman

An early example: danger zone entries

When the ball enters

this “danger zone”,
what is the chance of
scoring before it
leaves or is lost?

Expected goal
difference was
correlated at 0.83
with final positions

Top 10 players with

2,000+ minutes:
Suárez, Carroll, Van
Persie, Agüero,
Lukaku, Silva, Bale,
Snodgrass, Benteke,
Lambert  
Beyond shots – Dan Altman

Further development: a five-zone system

Each zone contains

all zones closer to
goal

Players receive
credit for advancing
the ball between
zones based on
differences in
.011 .028 .063 .141 .325   expected goals

Expected goal
difference from all
zone entries was
correlated at 0.82
with final positions
Beyond shots – Dan Altman

Further development: a five-zone system

• These ratings use simple zones for offensive production only

Could assign demerits to defenders using same system as for shot-based ratings

A better solution would be to use tracking data to assign demerits

Zones could also be designed to conform to common attacking channels

• But zones that cover more of the field allow the rating of more players

Top 10 attackers for offensive production per minute in 2012-13: Suárez, Silva,
Cazorla, Hazard, Mata, Tévez, Agüero, Carroll, Bale, Lukaku

Top 10 central midfielders for offensive production per minute in 2012-13: de

Guzmán, Touré, Gerrard, Davis, Osman, Dembélé, Ki, Cabaye, Lampard, Noble

Top 10 fullbacks for offensive production per minute in 2012-13: Baines, Johnson,
Coleman, Rangel, Gibbs, Enrique, Azpilicueta, Walker, Evra, Zabaleta
Beyond shots – Dan Altman

Can we do better?

• Tracking data allow more complex definitions of situations leading to goals

TRACAB data from the Premier League track each player, the ball, and referees
25 times per second on a pitch with roughly 1,000,000 possible locations

These data can be synched with Opta f24 files to merge positional information and
description of events in the match

• Preliminary work suggests algorithms can identify important situations

In a test, a prototype algorithm correctly spotted every 2-on-1 and 3-on-2

advantage in front of goal during several matches

Over a season, each situation can be assigned a probability of leading to a goal

based on the actual outcomes listed in the f24 files

Credits for creating situations can be split between the runners, passers, decoys,
and so on in any combination; demerits for allowing them go to the defenders
Beyond shots – Dan Altman

Can we do better?

Criteria were two

players behind the
penultimate defender
and in the middle of
the pitch, of whom
one had possession

Manual checks for

corners, celebrations

Of two 2-on-1
situations in this
game, one led to a
goal; one player (#7)
received the pass in
one and was the
runner in the other
Beyond shots – Dan Altman

Going beyond: the pass before the through ball

From the yellow

FB area, the passing
lanes to and
AM between all the other
CB players are visible

In mathematics, this
ST area is called the
GK
CM kernel of a star-
shaped space
CM ST

CB
Do moves preceded
by passes into the
AM kernel of attacking
players lead to better
opportunities to
FB score?
Beyond shots – Dan Altman

How might all of this be useful?

• Algorithms can quickly identify players who are influential on the pitch

Even the tracking algorithm runs in a few minutes, versus watching hours of video

• The same algorithm can be used for leagues around the world with f24 data

Players in known leagues can be used as reference and compared to players in

other leagues, e.g. “Find me the Eden Hazard of Brazil’s Serie A”

• Individual player ratings sum to team ratings that are correlated with results

This means players who improve their ratings will also improve the team’s results;
it sets good incentives and enhances the predictability of results and finances

• Team ratings pick up information quickly on likely final positions in the table

An early read can be helpful for planning transfer spending and future finances
Thank you
Dan Altman
@NYAsoccer
#OptaProForum