Great Expectations – What Expected Goal Models Tell Us About Fulham and Their Season So Far

This article discusses the statistical performance of Fulham Football Club during the 20/21 Premier League season. It is being written in the run up to Christmas and looks at data up until the end of Gameweek 14 of the season (Fulham have just played Newcastle).

I will look at measures of:

  • Expected Goals (a modelled estimate of the number of goals expected for (xG) or against (xGA) a team based on the number and quality of goals given up)
  • Non Penalty Expected Goals (similar to the measure above but excludes penalties from the analysis of expected goals for (NPxG) or against (NPxGA) a team)
  • Post Shot Expected Goals (a slightly different model which estimates the number of expected goals for (PSxG) or against (PSxGA) based on the quality of the shots taken rather than the chances created)

The article will also looks at how well such statistical models describe the onfield performance of the team and why there could be discrepancies.

Summary

As usual I will start with a synopsis of the post, those who wish to do so can dig into the detail further down the page, but the key takeaways are these:

Fulham have been competitive in almost all games since Gameweek 4…

  • they have conceded 13 goals in 11 games since this time of which 4 have been from the penalty spot
  • they have scored 10 goals in this time, but models would have expected them to have scored between 13 and 14, the shortfall is due to a combination of missed penalties and good goalkeeping (particularly in the Liverpool game).
  • they have averaged a point a game over this run, but if they had scored goals in line with expectations, they could have generated an extra 5 points. Even more points would have been possible if VAR decisions had been favourable.

…this improved performance seems to be due, in part, to a major change in tactical approach…

  • The data shows Fulham having lower possession and playing more direct during this good run,
  • a consequence has been that Fulham have generally had fewer shots, but of much higher quality, resulting in a net increase in expected goals scored

...in addition, Fulham have gone from being a team which massively underperforms model expectations, to one that broadly meets them…

  • At the start of the season, Fulham were conceding much better shots than they should given the chances being created, we noted here that this was a symptom of many recently relegated teams.
  • But this has now stopped being the case and since gameweek 4, opponents shots have been almost exactly as good as predicted by xG models.
  • Further, since gameweek 3, Alphonse Areola has consistently outperformed model expectations with his shot stopping, so Fulham are conceding fewer goals than models expect.
  • However, this overperformance by Areola is more or less offset by Fulham’s offensive underperformance, which is due to a combination of Fulham’s penalty conversion weaknesses and the overperformance of opposing goalkeepers.

…as such, if Fulham can maintain underlying performance, there is potential for a higher points yields going forward, this will be essential to survive this season as the current point accumulation rate is unlikely to be sufficient.

Team Performance

The chart below gives a simple overview of Fulham’s performance through the season to date. It uses 3 game rolling averages to smooth the data and to partly neutralise the effect of opposition difficulty. It shows expected goals scored against expected goals conceded, as well as the average rate of actual points accumulation.

It shows that, over the season, Fulham has gone from having a net negative xG (ie expecting to concede more goals than they score) to a net positive xG, before reverting to a net negative again (with the gap closing towards the end). This reflects generally improving performance both offensively and defensively, with the latter part of the season reflecting that they played Liverpool and Man City in this time (and Man City in particular, put up some big xG numbers against Fulham).

You can see as well that, as the xG gap closed, Fulham began accumulating points. They became competitive in matches from around gameweek 4. Since which time they have been picking up around a point a game on average.

The general improvement in results and performance seems consistent with the arrival of the new defensive personnel (phased in from gameweek 4) and a change in tactical approach, with Scott Parker increasingly favouring a low possession, high speed counterattacking system over the initial possession based approach (which was used throughout the Championship season).

This change in tactical approach is clear in the data too, the chart below shows the trend for lower possession as the season progresses.

In addition, the following chart shows a measure I developed called ‘Total Progressive Ratio’ which measures the proportion of possession actions (passing and dribbling) which are directed towards the oppositions goal. It is essentially a measure of how direct a team is.

The chart shows that as the season has gone on, Fulham have become more direct, and their opponents have become less so. Fulham have transitioned from being a team that gets caught on the break, to the team that does the breaking!

One notable consequence of this tactical shift is a change to the nature of the shots Fulham are taking.  Earlier in the season, I wrote about how Fulham were taking high volumes of low-quality shots. This has now changed as shown in the chart below, Fulham are now, generally, taking fewer shots (blue line is shots for Fulham and orange is against).

But at the same time, the quality of those shots has increased significantly, this is shown in the chart below which measures expected goals per shot (a measure of the probability of a goal from each chance created). Again Fulham are presented by the blue line and opposition shots by the orange.

This shows that the gap in quality between the shots Fulham take and give up has closed over the season.

So to summarise the story so far, models are showing that Fulham have become competitive in matches, firstly as personnel have changed and also as tactics have moved to a more pragmatic approach.

In the next section, I will look at how well Fulham’s performance is captured by these models, which is another matter that has a material impact on the outlook for Fulham.

Model Performance – Goals Against

So far this season, Fulham have conceded 23 goals from 14 matches, an average of 1.64 per game which ranks 15th in the league.

This is a very high rate to concede goals, but performances do appear to be improving as Fulham conceded 10 goals in their first 3 games, so have conceded only 13 goals in the subsequent 11 games. This is an average of 1.18 goals per game. In that run, only 3 teams (Palace, Man City, Everton) have managed more than 1 goal against Fulham and 4 of the goals have been penalties (often fairly dubious VAR jobs).

From an expected goals modelling perspective, Fulham have a total xGA (expected goals against) of 21.7 over the 14 games, which is slightly lower than the actual number of goals conceded (23). Interestingly, the expected goals conceded per game has not particularly changed over the course of the season:

  • In the first 3 games Fulham had an average xGA of 1.7 per game and PSxG of 2.5 (but were actually conceding 3.33 goals per game) – so Fulham were massively underperforming expectations
  • In the last 11 games, Fulham have had an average xGA of 1.5 per game and PSxG of 1.65 (but have actually been conceding 1.18 goals per game)- so Fulham are now overperforming expectations.

In my blog posts earlier in the season, I set out a concern that Fulham were conceding goals because their defence was so open and weak, that the quality of chances given up were actually much worse than they appeared, I cited numerous examples where opposition players took shots from positions where normally they would be under pressure, but Fulham’s defence applied no pressure at all, and I worried that the xG models were overestimating the quality of Fulham’s defence.

The first bullet above is evidence of this effect.

 My blog posts at the time also highlighted that both conceding better shots than expected and letting in more of the shots than expected were characteristics of relegated teams from previous seasons, and Fulham were doing both.

So what has happened, since Gameweek 3, is not that the number or quality of chances given up has particularly decreased (although they have a bit), but that the quality of resulting shots has decreased to a level that matches model expectations, and also our goalkeeper, Alphonse Areola, has been saving more of the shots than one would expect.

This is worth unpacking further, the chart below shows cumulative excess Post-Shot Expected goals – so it measures whether the shots opponents take are better than we would expect given the chances they are creating. As noted above, my hypothesis is that a marker of weak defences is that they give up better shots than you would expect because they apply less pressure to shooters than most models would assume.

I would say that this chart shows Fulham demonstrating this worrying characteristic of giving up better shots than expected in the first 4 gameweeks, but since then, excluding one game (against Everton) the level of post-shot expected goals is a near exact match to the pre-shot expected goal model.

That means that with the exception of the Everton game, opponents are now taking shots of the quality that the models would expect given the chances they are creating. I take this as evidence that Fulham are now performing like a ‘typical premier league defence’ whereas prior to gameweek 5, they were not meeting this standard.

The notable thing here is that this is a fairly good match to the introduction of Anderson and Adarabioyo into the Fulham defence.

The other key development is form of Alphonse Areola. The chart below shows, for each game this season whether Areola conceded more goals than we would expect from ‘an average premier league goalkeeper’ (shown with red bars) based on the Post-Shot xG model or fewer goals than we would expect (green bars).

The Orange line shows the cumulative over or under performance across all the games.

This chart show that after a rocky opening 2 games (where he conceded 7 goals against a PSxG of only 5), Areola has subsequently outperformed expectations in 10 out of 11 games (Everton again being an anomaly). Over that 11 game run, Areola has let in 13 goals, but PSxG models say an average keeper would have conceded 18.3 goals. So he is conceding significantly fewer goals than we might expect.

If we combine the last two charts together, we get a statistical story for the defensive record during the season to date. The chart shows the cumulative difference between expected goals and actual goals during the season (excess goals).

It shows the season in two segments:

1) The first 3 gameweeks of the season, Fulham are conceding many more goals than models predict, opponents are finding it easier to shoot than models predict and Fulham’s goalies are not doing a good job of saving the shots that they face.

2) The subsequent 11 gameweeks, Fulham are conceding fewer goals than predicted by xG models, this reflects that (game against Everton aside), opponents are taking shots of the quality you would expect given their shooting positions and Areola is then saving a much higher proportion of shots than models would predict.

So bad was the opening few games of the season, that the consistent team overperformance against the xG model in the 11 games since has not yet fully reversed the early damage done, but the trend is positive. Fulham are currently beating the xG models.

Model Performance – Goals For

So far this season, Fulham have scored 13 goals in 14 games, an average of 0.93 per game which ranks 16th in the premier league.

Expected Goal modelling says Fulham should have scored 16.6 goals this season (1.18 goals per game), or the 12th best in the league.

Immediately we note that Fulham are scoring fewer goals than expected, indeed there are 3.6 missing actual goals versus the models.

Obviously, part of this is to do with penalties, Fulham have scored 2 out of 5 penalties this season, where xG models say we should have scored 3.7. So penalties account for about half of the missing goals.

Removing penalties from the model (NPxG) shows us Fulham have 11 goals from NPxG of 12.9. This underperformance of 1.9 is difficult to pin down to a single source and is small enough to be just bad luck.

We can get a little closer by looking at the quality of Fulham’s actual shots on a post-shot xG model.

The chart below (which is a little chaotic but more explanation will follow), shows Non- Penalty Expected Goals on both a pre-shot (NPxG – a measure of chance quality) and post-shot (NPPSxG – a measure of shot quality) basis.

It is hard to pull much information out of this, until I show you the net cumulative difference between the pre and post shot model – the chart below does this, and this is measuring the cumulative difference between the quality of chances Fulham create and the quality of shots they take – essentially asking whether Fulham are efficient at converting good chances into good shots.

The chart shows that, over the course of the season, Fulham’s shots have been an exact match to the quality of chances created. In other words, Fulham are performing like a normal premier league team when it comes to shooting (non penalty obviously).

So poor shooting does not explain Fulham’s missing goals. The difference can, however, be largely explained by the performance of opposing goalkeepers, for example, in the Liverpool game, Fulham generated post shot xG of 2.2 but only scored 1 goal, essentially Allison’s goalkeeping performance in that game is enough, along with penalty missing to explain a lot of Fulham’s underperformance against expected goals for the season.

The total overperformance of goalkeepers (cumulative) is shown in the chart below:

So to summarise what we have learned:

  • Fulham have a season xG of 16.6 but have only scored 13 goals
  • This leaves 3.6 missing goals which is explained by
    • 1.7 goals lost due to poor penalty taking
    • 2 goals lost due to better than expected goalkeeping from opponents (of which 1.2 came in the Liverpool game)

Conclusion

To bring this all together, I think the key points so far are:

  1. Fulham’s modelled performance is improving and Fulham are broadly generating xG at the same rate they concede it.
  2. Fulham have moved from fundamentally underperforming modelled defensive performance (due to inadequate defence) to overperforming modelled defensive performance due to exceptional goalkeeping (even despite their poor penalty taking record)
  3. The overperformance above is largely offset by the teams underperformance in terms of scoring goals – this underperformance is due to their penalty taking woes and, to a small extent, good goalkeeping (but this is really just in one game)

These developments have seen Fulham reach a point where they are generally competitive in games, and indeed, if they converted penalties at a typical rate then they would, most likely, have generated a 14 point return from the last 10 games. This would certainly be survival form.

The return could have been higher still, were it not for some questionable penalty decisions against Fulham, and or some inspired goalkeeping by Allisson in the Liverpool match. A return of 18 points from these 10 games would not be beyond what the performances deserved.

Going forward I expect to see Fulham mix things up with the tactical approach, combining the extent to which they are defensive and direct in matches with games where they try to control the ball. If Fulham are competitive in all their games going forward, to the extent that they have been over the last 10, then a higher points yield should be possible.

This is critical because I believe a total approaching 40 points will be required for survival this season, so the current run rate of a point a game is unlikely to be enough. Maintaining the team’s competitiveness is likely to require the continued form of Alphonse Areola and it remains to be seen whether he is able to continue to beat the xG models as he has over the season so far.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: