At the start of this season, we unveiled an expected goals model to be used on the SPL (I refuse to call it the SPFL), with updates in August and November. Unfortunately, due to work commitments I have struggled to find the time to post updates.
However, last week I was invited to take part on my friend Grant's weekly Scottish football show Campbell's Footballs. We spoke about the battles at the top & bottom of the tables in both the Scottish and English top flights, as well as the international break. This episode can be viewed here.
Expected vs. actual performance
Our model gives an idea of how many points a team should have earned based on the chances they created. Comparing this to their actual performance shows which sides are under-performing (below the diagonal line) and which are over-performing (above the diagonal line).
Below the line is where the notable teams lie. Ross County currently lie bottom, with Dundee just two spaces above them. However, according to our model, both teams should be above Partick, Hamilton and St. Johnstone. Similarly, Motherwell are under-performing, and should really be in the top half of the table.
We will now break down attacking and defending individually to identify why these teams aren't performing as well as we'd expect.
Reality vs. expectations
Our model also specifically tells us how many goals we would expect each team to score/concede in each game. Over a season, we can compare this number to the amount that were actually scored to get an idea of what is causing the deviations from the model.
So for attack we divide the number of goals scored by a team by how many we would have expected. For example, if a team were expected to score 8 goals, but scored 10, then we would give them a score of 10/8 = 1.25. In other words, they have scored 25% more goals than we would have expected. We repeat this with goals conceded to get the following graph:
In attack, the most remarkable team here is Hamilton, who are wildly over-performing (a whopping 27.6% increase on expectation). At the other end of the scale, Dundee's troubles are clearly caused by their huge under-performance up-front, with 20% fewer goals scored than they should have managed.
In defence the best performers are Celtic, whilst Motherwell and Ross County are conceding more goals than they should. This is a harder category to judge the meaning of, because it can be a combination of luck (great chances happen to fall to poor finishers) or goalkeeping.
The remaining seven teams all are within 10% of what we would expect in both attack and defence.
Repeating the analysis from November, we can see how each team are performing in attack and defence specifically.
Starting with attack, we can compare the number of chances each team has created (x-axis) with the average quality of each chance (y-axis).
What is interesting, though, is the different ways each team has racked up the goals. Celtic's high goal tally has come from accumulating loads of chances - an average of 16.8/game. Rangers, meanwhile, have created far fewer chances (almost 100 fewer), but they are striking in how good the quality of their chances are - an average Rangers chance is 27.8% more likely to result in a goal than an average Celtic chance. And because of this, Rangers are in fact predicted to score more goals - although in reality they have only managed to score 2 more with Celtic having a game in hand. But really, this is very encouraging for Rangers, and a lesson for sides to prioritise quality over quantity.
At the other end, Partick stand out for their complete lack of threat. Whilst they have a decent average xG, they are creating significantly fewer than the rest of the league. Coupled with their under-performing finishing (see previous section), then it really spells danger for the Jags.
Now let's repeat the same graph but with chances conceded:
Rangers are interesting again, as they are conceding high-quality chances, albeit not many. Hamilton are also poor at stopping high-quality chances. However, the side with the most predicted goals against them is - you guessed it - Partick. Honestly, how they are not bottom of the table is a mystery.
As we earlier pointed out, Ross County are likely bottom because they are conceding far more goals than expected. This graph suggests that whilst they are limiting the quality of chances, they are conceding a lot of them - maybe they should work on trying to limit chances, regardless of where they are from? Perhaps they are allowing too many shots from range.
Below is a plot of the sides' expected points over the course of the season, cut off at 28 games so that all teams have played the same number:
At the bottom Partick have started to improve, although our previous graphs indicate that they are still clearly the league's poorest side. But now, it appears as if they are roughly at the level of Hamilton. Were Hamilton not over-performing up front, then they would be in serious danger of automatic relegation. I certainly wouldn't rule it out, but the smart money has to say Partick. Ross County, meanwhile, have performed consistently all season, and may just need luck to break their way.
Motherwell early-on looked to be in contention for a European spot, but now lie 8th after a slow decline (albeit they are still fifth in the expected table). This appears to largely coincide with their sale of striker Lewis Moult to Preston North End in the transfer window.
For each match our model gives a probability of victory for each side. So which matches have had the least likely outcomes? Here are the top three:
3rd: Hearts 0-0 Ross County
A placid 0-0 draw midway through the season is an odd one to bring up, but as the shot maps below shot, Hearts should have won strongly. In terms of xG, Hearts won this game 3.18-0.38, which gave them a probability of victory of 92.7%. The chances of a draw were only 5.18%.
In terms of impact, this is relatively low - Hearts are clearly mid-table destined, whilst Ross County may have stolen a point here, but their under-performance (highlighted elsewhere) has seen them sink to the bottom of the table.
This match marked Hamilton's first win at Ibrox for 91 years, as David Templeton and Darren Lyon struck for a surprise victory. In fact, this was even more surprising when you see how many big chances Rangers created. A terrible day at the office for Alfredo Morelos, whose changes contributed to an overall Rangers xG of 4.29, versus 1.49 for Hamilton. Overall, this meant a Hamilton win was only 2.85% likely to happen.
We've already shown that Hamilton are over-performing, and this is a big moment for them, as they get 3 points when 0 seemed inevitable. Without these three points they would be within striking distance of the relegation zone, whilst Rangers would be looking much more secure in second right now.
Two late goals from the home side saw them come back to win 2-1 over Dundee. Whilst Partick created decent chances during the game, Dundee had numerous high-quality chances in-and-around the 6-yard box, and really should have won comfortably. In terms of xG Dundee won 4.54-1.31, but instead left with nothing. This match really highlights the failings of Dundee's strikers, with strikers Sofien Moussa and AJ Leitch-Smith alone expected to score 3.77 goals in this match (Leitch-Smith did score Dundee's sole goal). This is further compounded by the fact Leitch-Smith missed a penalty (which for some reason hasn't appeared on the graph below). Overall, Partick only had a 1.68% chance of winning here.
Had Partick lost this match as they ought to have, they would currently be bottom, and there really is every chance that they could survive thanks to these three points. Had Dundee won, they would be nine points clear of the bottom two, and looking far more likely to survive.
So this very comprehensive look at the SPL will hopefully give you an idea of the stories behind the results this season.
With only a handful of games to go before the end of the season, I am going to stick my neck on the line and make a few predictions (since Grant has already made me do so!):
- Celtic will win the title (obviously), whilst Rangers will ultimately be quite comfortable in second.
- Aberdeen's current points tally will see them stave off Hibernian for third. But unless Motherwell claim a shock Scottish Cup win, both sides will be in Europe next year.
- It's too late for Motherwell to finish in the top half of the table, but I expect them to win the bottom-half of the league.
- Partick are just playing too badly to avoid relegation, I think they are still the most likely side to finish bottom.
- Ross County have quite a task to get out of the bottom two, I'm not sure they'll manage it, because their nearest safe side is Dundee, who are also under-performing (and are therefore just as likely to turn their fortunes around).
So let's see how well these predictions age!