Expected goals plus-minus

As the season ticks on, we're starting to get a bit of data to work with, albeit sometimes in small samples. This season I am not working with a relatively complex player model but am really interested in trying to "play the fixtures" as much as I can. A couple of sources of frustration, therefore, have been as below:

1. Sites which provide the difficulty of fixtures often seem a bit simplistic. For example, the Premier League site shows Wolves as an average "3-rated" opponent, both at home and away yet the reality is more complicated. While at home they have been very solid defensively, with an Expected Goals Conceded (xGC) of just 4.71 in five games, which ranks third best, yet on their travels they have surrendered 7.94 xGC, which is the third worst. Even more confusing is that most sites (though not all) don't distinguish between attacking and defensive fixtures, so facing Leeds at home is presented as an easy "2-rated" fixture when in reality Leeds are indeed a somewhat weak defensive team (14.8 xGC, ranked 16th) but have shown an excellent attacking threat to date (17.1 xG, ranked 2nd).

2. A second issue is that teams have obviously not faced the same opponents to date so teasing out a team's ability from that of their fixtures is difficult. This is particularly true when deciding whether a team is improving or declining (or whether they are staying the same but their opponents strength is the true variable).

Something I want to try and play with is a small part of data I used to embed in my larger model, something we'll call plus-minus (or +/-). What we're basically calculating here is how well a team does against an opponent in comparison to how that team has fared against everyone else. 

So, if in a three game span Chelsea surrender 0.6 xGC, 1.3 xGC and then 1.8 xGC you might conclude they are a little inconsistent and their 1.23 xGC per game would likely leave them in the middle of the pack. You might be able to mentally adjust for the context of their opponents of course, so if they've faced top opponents you give them the benefit of the doubt, but this then becomes a lot things to juggle in your mind. If however, we knew that those three opponents scored, on average, 1.2 xG, 1.5 xG, and 2.5 xG respectively then we can see that Chelsea well out performed expectations and "conceded" 1.5 less expected goals than the league has typically been conceding in these fixtures. This would lead one to conclude Chelsea are an elite defensive side and perhaps worthy of extra investment.

I plan to dig into this data a bit more (and post the attacking equivalent) but for now I just want to highlight Chelsea a little. 

Each bar shows the degree to which Chelsea over or underperformed the xGC that other teams have achieved against that opponent so far this season. Given that this is goals conceded a negative number here is good. The line shows the trend of the season so a downward slant suggests a team continuing to outperform the average, while an upwards slant would suggest a team is getting comparatively worse over time.

Since Mendy (5.2m) arrived, there has been a general acceptance of the narrative that Chelsea's defense has become quite solid, notching a very impressive five clean in eight gameweeks (plus additional success in the Champions League). Chilwell (6.2m) - initially signed by many for his attacking abilities - has been targeted by many, seeing his ownership rise to over 35%, making him the most owned defender in the game. Zouma (5.5m) and James (5.2m) have 21% and 15% ownership numbers respectively so are too gathering some solid attention. Indeed, if you sum the ownership numbers for teams' 'keeper and top three defenders, Chelsea's 80% total far exceeds the likes of Liverpool (48%), Villa (50%), and Leicester (47%). 

I believe this data shows, however, that we are still vastly underrating this defense as of right now. I add that caveat because with COVID, crowded fixtures, and an unproven manager, Chelsea's fortunes could change, but they currently look like just about the surest bet anywhere on the field.

Chelsea's xGC of 7.9 looks very solid, albeit not a million miles away from teams like Man City (9.9), Brighton (10.1), Burnley (10.8), and Villa (10.8). Per the earlier visualization though, this seems to actually underrate Chelsea's performance. A total of 2.2 expected goals came in the single game against Liverpool, and even that was actually 0.6 less than team's are typically surrendering to the Champions (who were full strength at the time, by the way). They have outperformed the typical xGC performance in every single game other than Southampton when they narrowly underperformed the average, and are conceding over half an expected goal per game, less than other teams in the same fixtures.

Compare that to, say, Villa, who have a good overall xGC but have been hit-and-miss conceding higher xG totals that average in games against Fulham, Leeds, and Arsenal, while excelling against Liverpool, Southampton and West Ham. In total they have only conceded about two less expected goals than the average performance against their same opponents, suggesting they have been somewhat benefited by weaker opponents than fantastic defensive play. This runs counter to intuition which says Villa were on a tough run of games, but based on the season so far, we might need to revise our expectations about how hard games against the likes of Leicester and Arsenal really are. 

The really exciting thing about Chelsea is that they also seem to be getting better in recent weeks, with results against Sheff Utd, Newcastle, and Spurs representing three of the best four defensive performances of the season by this +/- metric. Having clinched top spot in their Champions League group, we can probably predict that starters like Chilwell, Zouma, and James are more likely to play in the league too, and their fixtures look quite useful for at least the next six weeks. 

The only downside side is a lack of any truly great individua fixtures, though games against Jimenez-less Wolves (A), West Ham (H), Arsenal (A), and even Villa (H) all look like reasonable bets for a clean sheet. At their respective prices, a double up with Chilwell and James - or for those in need a Martinez replacement, perhaps Chilwell and Mendy - seems like a very solid strategy to get you through the upcoming busy month. 

[EDIT: Some readers might know that I have been on and off with this site for many years, but are currently not super-plugged into the FPL space as other responsibilities have taken over. It's very possible that these posts are talking about "new" data which is actually very old and commonly available, and I just don't know where to look for it, so apologies in advance if I am doing the Uber tech-bro move of proudly announcing my new invention which turns out to just be a bus].