Misery, misery, misery! Think of it as disappointment brought forward, and therefore sooner processed and gotten over. The inevitable play-off glorious failure to Italy averted. Celtic’s Scotland players now get an extra break in the middle of the Champions League! To soothe any pain further, I consider the overall attacking contributions thus far from the Champions.
Due to low data volumes (having not played enough minutes), the following are discounted: Ajer, Sviatchenko, Boyata, Miller, Kouassi, Benyu, Edouard, Dembele.
Top CATs
I have been developing aggregated metrics to highlight overall performance. Once such is CAT score – Celtic Attacking Threat. This is explained in detail in Top CATs, but as a reminder combines:
- Non-Penalty Goals
- Assists
- Shots on target
- Possessions in Box
- Key Passes
This provides an overall assessment of key actions threatening the opposition goal. The scores based on averages per 90m by this measure are:
At the risk of displaying self-satisfied smugness, the CAT score distribution seems to make sense. The out-and-out attackers (Sinclair and Griffiths) lead, with a 3-point drop to the attacking midfielders and then a further 2-point drop to the defenders and defensive midfielders. Variances within bands can therefore provide insight into form.
Sinclair and Griffiths are clear leaders. Sinclair tops, slightly surprising given his wider left sided role. Dembele, on limited data (not included here), has an average of 12.000 illustrating again his propensity to operate in the box compared to Griffiths.
Then we have the attacking midfielders with Forrest leading the way. Given his style is usually to stay wide, like Hayes, this may appear surprising. Remember he played centrally during the Great Summer Striker Shortage of ‘17.
The CAT scores of Armstrong and Ntcham respectively highlight again that it is not necessarily a case of either/or as they play in different areas of the pitch and do different jobs.
Roberts hasn’t got started yet and has played few minutes (377) and against trickier opposition.
Tierney understandably leads the defenders. Brown’s attacking threat is increasingly limited.
Scoring Contribution
Another aggregated metric, albeit a simpler one, has two variants. Firstly, Scoring Contribution strips out the “threat” element and focusses on pure delivery. It aggregates Non- Penalty Goals (NPGs) and Assists to provide an overall attacking productivity score.
(A reminder: I only count NPGs for analysis purposes as adding penalties artificially inflates penalty takers’ numbers. A penalty has around a 78% chance of success irrespective of the taker. It is a specific skill with its own metrics.)
Based on overall totals, the leaders are:
Whilst, more predictively, if we consider actual minutes on the pitch, the per 90m averages:
Griffiths continues to be remarkably productive given how little he is in possession (that gives me an idea!). Put another way, he manages a non-penalty goal or assist every 79 minutes. McGregor is having a highly productive season, and averages the same every 89 minutes. Whilst Sinclair leads the total productivity table, having played more minutes, he takes 98 minutes to rack up a NPG or assist.
And Roberts has already racked up 1 goal and 2 assists in 377 minutes.
The second variant of Scoring Contribution measures Expected Goals and Expected Assists. Expected Scoring Contribution (xSC) is therefore more indicative of long term performance. Players can benefit from “hot streaks” informed by luck over the short term. Shots let in by fumbling ‘keepers, deflected efforts going in (Ntcham!) or not (Hayes!), prosaic passes leading to improbably long shots providing assists, and so on and so forth.
What probabilities suggest players should have produced can be seen:
Finally, the gap between Actual vs Expected highlights those whose streaks in the short term may be difficult to sustain (largest positive differences) and those whose luck may be about to turn (largest negative differences) and those that are performing around the mean (can be expected to continue that level of productivity):
The Expected Goals model I use was developed using data from the American League (MLS). Celtic players do generally seem to be outperforming it – that is, scoring more goals than the model would suggest. If you were to use a model based on Scottish league games only (as many fine folks out there do) I’d expect Celtic players to outperform that also given the quality gap between Celtic players and those from other clubs (respectfully). Until there is a unified model, this will potentially be the case. Which is why Expected Goals and Assists should be considered just another data point to build up a picture, rather than any Holy Grail of football analytics.
I said finally, but as you were reading the above I went off and calculated the number of possessions per scoring contribution. So, finally, finally, here are the number of possessions required to either score or assist per player:
Only 33 possessions of the ball between Scoring Contributions. I have nothing to compare this to, other than previous Celtic seasons. Last season, Griffiths equivalent was 27 possessions per Scoring Contribution. I doubt anyone in the league would come close over a season.
I should not that on very limited data, Edouard is contributing a SC every 22 minutes. Hopefully by the next International break, both he and Dembele will have registered enough minutes to be challenging Griffiths and Sinclair across these (top) CATegories.