Expected Canaries and the xG Crystal Ball

Written by: Christian Wulff (@ahellofabeating)

International breaks are a good opportunity to pause momentarily, analyse the current status of the Premiership table, then stick your neck out and do some projections for the rest of the season.

 The reason why we at Modern Fitba and the football analytics community in general put so much emphasis on Expected Goals (xG) and use it for our analysis is twofold: Over a smaller samples of games we consider it a better measurement of a team’s underlying performances than goals and actual results. This again is based on the proven assumption that these underlying performances (measured in xG) are a more accurate indicator of future results.

 Simply put; taking the xG values of each team after their 12 or 13 games up until now and projecting that over the full season is likely to be a more accurate prediction than simply calculating points per game so far and multiply with 38.

 There are caveats of course. xG isn’t a perfect indicator of future performances (if it was, I’d be rich), it is simply a better one than goals scored. It also includes an assumption that the quality of chances created and conceded by a team over the first third of the season will be closely replicated over the last two thirds. But of course we know there will be exceptions where a team’s performances start improving or declining significantly from this point on.  Again, it comes down to the degree of accuracy: it is more likely that the xG values of a team will stay the same over the remainder of the season than that their points or goals per game will.

 For teams (and fans), xG might best be seen as their canary down the mine: giving a warning before a threat is apparent or reassurance that everything is well.  

 If results have been good but the xG values a lot worse, beware: all is actually not well and improvement in performance is needed in order for current results to continue.

 Results not been great but xG a lot better? Keep the faith; continue to do what you’re doing and results will highly likely improve.

 Results and xG match up? Congratulations / commiserations: your team is actually as good/terrible/meh as the league table indicates.

 (Note: in terms of projecting what the league table might look like at the end of this season, the Modern Fitba community is currently working on a couple of different models. This is part of why we have this website; publishing work and models that are work in process, being open about its strengths and weaknesses and then reflect on what works and how it can improve. Both models rely heavily on xG data, and you should really read both Jason’s Projected League Table (updated on Twitter yesterday) and Jamie Kilday’s Expected Points. Also check out Seth Dobson’s  work on Expected Win Percentage in Scottish Football which looks at shots taken (quantity of chances) rather than xG (quantity and quality of chances combined)

 The current xG state of affairs in the Premiership is as follows:

xG Table.PNG

It might be tight at the top of the table point wise, but there is absolutely nothing in the xG data that suggest there will even be a hint of a genuine title race this season.  Celtic is simply producing and restricting chances at such a rate that if sustained over a whole season they will be clear winners again.

It’s the same at the bottom of the table. The underlying performances of the bottom three teams have been so much worse than Motherwell, that it be a huge surprise if any of them climb above 10th place.

While they’ve had a very disappointing season so far in terms of results, Motherwell should be absolutely fine, even on current form.  While St. Johnstone is a huge 12 points ahead of them (with a game more), the difference in their actual performances haven’t been anywhere as big – there is simply not as much between these two teams as the table currently suggest.

This leads us to perhaps the most interesting team when it comes to the xG numbers. The early season ‘crisis’ at Aberdeen has officially been declared over with three straight league wins, including over Killie and Hibs, and a League Cup Final to come next week.

Let this be Aberdeen’s official warning: that canary does not seem well. The Don’s mini-revival seems to be built on shaky ground: In the last three league games they’ve scored six goals but only produced chances with a combined xG vale of 1.85. They’ve conceded no goals other than a penalty against Kilmarnock, but their non-penalty xG against was 2.19. Disregarding the penalty, the gap between Aberdeen’s goal difference and their xG difference over the last three games is over 6.

In fact, overall this season the quality of the chances Aberdeen have created is lower than the chances they’ve conceded. And that is excluding the four penalties against them. Beside the Dons, only the four bottom teams in the league have such a negative xG difference.

Now, it is entirely possible that this string of good result will spur Aberdeen on to a consistently better performance level. Any additions in the next transfer window or a tactical changes worked on throughout the winter break might also help.

But let’s be clear: unless Aberdeen improve - or the teams around them deteriorate significantly - their current level of performance will see them struggle to get into the top 6.

In terms of the rest, the distance from Hibs and Kilmarnock up to Rangers and Hearts isn’t huge, but there is a gap there and especially Rangers have been increasing it lately.

The last point is on Livingston: they fully deserve their surprising 7th place in the table so far with their xG difference also being the 7th best this season.

 

Beyond xG difference

When looking at the above, even if you accept that xG is a better projector of future performances it would be fair to point out the fact that the league isn’t primarily decided by goals difference, but by points. It could be a slight danger that overall xG numbers can be padded out (especially in smaller samples) through some very good single game performances.

But what if we had simply awarded the points in each game this season based on the quality of the chances created and conceded, rather than goals for and against?

I can assure you that not even Modern Fitba is actually advocating such a rule change, but it could help judge the overall quality of a team so far set within the context of having to go through 12 or 13 individual battles 90 minute battles against another team.

A team could be rampant over a couple of games, maybe scoring 3 or 4 xG more than their opponent. But if they in the other 10 games have had a chance quality slightly below their opponent, their performances as a whole might be slightly overstated if you look purely at the total xG difference.

For this exercise, I’ve award a win to any team that had an xG of 0.50 or more above their opponent, and a draw if the difference is less. This is admittedly an arbitrary way of deciding what should constitute ‘an xG draw’. The logic is that through rounding up there would be a full goal between the teams (that’s usually how you win games) and it by using this threshold we get 19 ‘xG’ draws so far (25% of games) – quite comparable to the 14 actual draws this season (19%).

(Note: I’ve also included xG from penalties in this exercise We usually exclude xG from penalties (as in the table above) in our analysis because ‘creating’ penalties are usually a lot less repeatable than creating other chances, and hence a less reliable predictor for future chance creation.  For example; if you created on average of 10 chances and was awarded 3 penalties in the first 10 games, you are a lot more likely to keep creating around 10 chances a game than you are getting 3 penalties in the next 10 matches. But since we are trying to determining the point allocation in single games based on xG, penalties – with a fixed xG of 0.80 – are too big a factor to exclude in this context).

Applying the above criteria to the Premiership so far, we end up with the following table:

Points Table.PNG


In terms of positioning, there are no significant changes. Celtic’s dominance is now reflected in points; 7 down to Hearts (with a game less) and 8 to Rangers. Giving points based on xG instead of goals would have seen Hibs with five more, and put them in 4th instead of 8th.

 There is still a gap below Rangers and Hearts, although Hibs do have game in hand on their city rivals and Killie.

Again, Aberdeen is the most interesting: while they were 8th in xG difference, attributing game points based on the xG in each match would have put them in 9th place, behind Motherwell and with the same amount of games.

 Again, let’s be clear: I’m not saying that this means Aberdeen will finish 8th or 9th, only that if their underlying performances continue at this rate throughout the season, they will struggle heavily to break the top 6.

 There is also a faint hope for Dundee; while they are six points adrift of Accies in the league table their xG difference is slightly better and they are equal on points if these were decided by xG.

 And I can hear a lot of people muttering that they don’t need stats to tell them St. Mirren are in major, major trouble and they’ll be right, you don’t. But the stats are confirming it, nevertheless.

 

So that is my xG crystal ball projections:

  • No genuine title race.

  • The bottom three will be Dundee, Accies and St. Mirren (in no particular order).

  • Aberdeen’s revival looks shaky unless their performances improve significantly.

 Save this article to your favourites; it might be a great way of ridiculing me at the end of the season.