Form Score — Understanding The New Football Analytics Metric That Measures A Team’s Form
Note: This article is based on football analyst Sezer Unar’s introductory article to the metric form score. All credit of the metric goes to him.
When it comes to evaluating a team’s chances ahead of the next game, we’re usually presented with their recent form over the past 5 games.
The visuals on TV depict their recent trend in performances, either W, D, or L.
Does this give a complete picture of a team’s form? Maybe not the best.
We’re already well aware of rolling averages of xG which are popular in football analytics circles.
This is an old picture below (from the platform The Full Time Whistle Co. where I’m a co-founder) which compares the xG vs GF(goals scored) of teams in the Premier League in the first half of the 2022–23 season.
In case you want to understand xG in more detail, we’ve covered xG in detail in this post on TFTWC along with its application and variance in xG, which you can read for a firm understanding of what the xG metric is.
Rolling xG and xGA, which are metrics that compare a team’s xG and xGA with a moving average over the course of the season, are pretty common in football analytics circles.
This is an example, comparing Arsenal’s xG & xGA over Mikel Arteta’s tenure as Gunners boss.
Coming back,
If a team has won their last 5 games, they should be in the best form right?
However, Sezer Unar, the creator of the new metric — “form score” argues that it doesn’t.
Unar suggests that analysing a team’s xG and xGA separately and calculating their form based on those metrics over a 7-game period i.e their previous 7 games, gives a much more well-rounded & whole picture of a team’s form.
He writes in his post:
When looking at the league table, I thought it would be nice to see a rating of the teams’ form, rather than the last N matches won or lost, so that the table becomes colorful, more entertaining, and informative.
First of all, I distinguish between a team’s offensive and defensive form. It’s great to score a lot of goals, to score 3 xG per game, but it’s equally negative if you’re conceding 3 xG per game. That’s why I don’t take the xG difference into account, it makes the most sense to calculate them separately.
It’s obvious to everyone that looking at xG based on a single match is complete nonsense. I might be too optimistic by saying, everyone. So it’s better to look at the last few games. And the last few games? Inspired by David Sumpter’s article here, I set the number at 7.
It is also important to note that the result 7 games ago cannot have the same weight as the result one game ago. At the end of the day, we are assessing the form of the team. A match almost a month ago gives us only a glimpse, but a very limited glimpse. It’s better not to treat the light from far away the same as the light from near. However…
It’s not a good idea to put too much emphasis on the last game. Maybe the team got a red card, or maybe they played against a very strong team. One match is one match.
So a match count of less than 7 doesn’t give us a healthy picture of form, while equal weighting makes this metric unhealthy. Moreover, “more matches” helps to somehow minimize the impact of “extreme” matches.
For the R-based function code, you can refer Sezer Unar’s article as I’m more proficient in Python and don’t know R very well.
However, according to Sezer’s calculations, the weight of each of the 7 games is as follows:
This picture depicts the impact each of the previous 7 games has on the next fixture. The immediate last game has an impact of 19.2% while the first game in the form series has an impact of 10.2%.
So 52% of a team’s form comes from the last 3 matches and the remaining from the last 4 matches before them.
Sezer further writes:
There is not an extreme difference between this weighting and equal weighting.
But is this enough? At the end of the day we get an average of xG for and xG against, but what does that mean? For example, is a team with a weighted xG of 1.25 over the last 7 matches good or bad? Shouldn’t there be a benchmark?
I had team statistics for the English Premier League, La Liga, and Bundesliga for the last 5 seasons, so I applied this process to those leagues and calculated the z-score corresponding to the xG values I obtained.
I scaled the values between -2 and 2 to -5 and 5 and finalized our metric.
The reason I’m doing this is to make the final score more understandable.
So a team with a -5 form score is in terrible form, while a team with a 5 form graph is going through a great period.
So in general, form score is a rating from -5 to +5 calculating a team’s form based on xG and xGA.
Here are Sezer’s results of his “form score” calculations for the Premier League on the date 16th March 2023.
Comparing ‘Form Score’ to Results
Form score can be an interesting and exciting new metric to determine a team’s defensive and offensive form.
But how does form score relate to and actually play a part in predicting results?
Sezer’s form score chart ahead of gameweek 28 was as follows:
Some games got postponed, however, the results from matchday 28 were:
Where Form Score Was Right
In accordance to the form score, Newcastle beat Nottingham Forest 2–1, with Newcastle being more in form in their form score. Both Brentford & Leicester who were nearly equal in form score ended up drawing 1–1, Chelsea v Everton also played out a 2–2 draw with similar form scores & Everton’s defensive form score resulted in 2 goals conceded and Arsenal thumped Crystal Palace 4–1 as expected by their form score.
Where Form Score Was Wrong
Southampton v Tottenham ended in a 3–3 draw despite Spurs’ better defensive form score. Leeds also emphatically beat Wolves 4–2, however, there was a red card involved in that game. While Aston Villa beat Bournemouth 3–0 despite both teams having similar form scores.
Form Score’s Accuracy
So the accuracy of form score is about 66.67% from this set of fixtures, not taking into consideration the Leeds vs Wolves game because of the red card as it was an “extreme game”.
That means that form score does have an impact going into fixtures. But of course, doesn’t tell the whole story, because a team might turnaround a poor run with a surprise result, as the second leg of Chelsea vs Borussia Dortmund in the UEFA Champions League 2022/23 panned out. Dortmund were on a 15-game unbeaten streak while Chelsea had extremely poor form going into the gam.e
Of course it would be better to test form score’s accuracy over the course of an entire season (but I don’t have access to Stats Bomb data for the moment).
Conclusion
Form score does seem like a useful metric, and as it grows and evolves, using further stringent parameters, it could be a viable form guide and result predictor.
Form score does provide an interesting insight into a team’s form & the way they might play, and has some intriguing future applications such as players’ form, as Sezer Unar writes,
Potentiality
This type of method can also be used to measure players’ form. While it is relatively logical to measure a team’s form through xG, it would be quite illogical for players. However, if you already have a player-based per-match rating system in place, a weighted moving average can provide a useful perspective. As a fan, there have been many times when I’ve been like “this player has been awful in recent weeks, why is he playing?”. At least we can express this feeling in numbers.
Summarising, form score seems like a more detailed way to evaluate a team’s form as compared to just listing W,D,L. And separating it into offensive and defensive form is a really great way to bifurcate and analyse both aspects of a team’s performance in their form.
It’s only early days for the metric form score, however, its evolution as time progresses and the further applications it could have can be compelling and insightful pieces of data in the football analytics spectrum.
You can read Sezer Unar’s original article introducing “form score” here.
You can also check out the platform The Full Time Whistle Co. where I’m a co-founder, for more in-depth football content.
Thank you for reading.
