By Stephen McGeoch
The discrepancy between the expected and actual goals is certainly remarkable. With 20 games played, Celtic have scored 40 non-penalty goals in the SPFL this season. That’s 2.0 goals per match against an estimate of 2.8 goals. It feels like a lot, especially since it’s from a fairly large - 438 – number of shots.
So, is that just statistical variation, or is there something deeper going on?
We can do a quick test as follows.
The xG model suggests that Celtic should have scored 20 x 2.8 = 56 non-penalty goals. That’s an average probability of scoring per shot of 56 / 438 = 0.13. That would be quite a good rate by historical standards; most teams have an average of about 0.10 ie it usually takes about 10 shots to get a goal.
Celtic’s actual conversion rate is 40 / 438 = 0.09, which is a bit on the low side. We use something called the binomial distribution to test our hypothesis that the underlying rate of scoring is 0.13 per shot and ask what’s the chance of only scoring 40 (or fewer) instead of the expected value of 56 goals?
When you plug in the numbers, you find that the probability that the shortfall is just down to statistical variance is a bit less than 1%.
Something else is going on.
Postecoglou’s team are certainly taking lots of shots: 21.9 per 90. Lennon’s 2019/20 team were taking only 17.5, but with a huge conversion rate of 0.16. His 2020/21 team shot a bit more (17.8). They were memorably less accurate (0.10), and we can all remember Christie reliably picking out the back rows of the empty stands, but still actually a little more productive than this season’s squad so far.
Let’s take a look at who was doing the shooting in the 2019/20 team. The top 6 shooters and their conversion rates are as follows. They accounted for 69% of all shots.
Player Shots Goals per Shot Edouard 100 0.21 Forrest 70 0.14 Christie 52 0.19 McGregor 49 0.14 Ntcham 47 0.09 Griffiths 46 0.20
Now let’s do the same for this season’s team (and bear in mind that this is for only 20 matches).
Player Shots Goals per Shot Turnbull 68 0.07 Abada 50 0.12 Rogic 35 0.09 McGregor 34 0.03 Jota 33 0.15 Kyogo 31 0.26
Postecoglou’s team is producing lots of opportunities for players who aren’t able to convert them. Our most deadly players, Jota and Kyogo, are each taking less than half of the shots of Turnbull.
So why the big discrepancy between expected and actual goals?
Here’s my guess.
The best chances usually fall to the best finishers. So, when we have an opportunity with xG of 0.20, that value is based on the fact that a decent finisher, probably a striker, is most likely to be taking it. I reckon Celtic are making good chances for players who aren’t good finishers.
Turnbull is the most obvious example, but McGregor is getting chances and not taking them, and the next 4 players in volume of shots after Kyogo are Johnston (25), who has been having a nightmare, Ralston (22), who’s actually a good finisher with G/Sh of 0.18, Bitton (21) and Juranovic (20). Most of those guys have little in common with the “benchmark” striker of the model who converts 1 in 5 of his 0.20 xG chances.
Lennon’s successful 2019/20 team created chances for the players who were best able to convert them. That hasn’t happened yet for Postecoglou, but he has lost a very good player in Edouard and he has been hampered by injuries at the same time as creating a new style of play.
I’d expect his team to become more efficient as the squad settles.