Cascarino's alternate timeline
A couple of weeks ago, Tony Cascarino made some comments that caused a litte bit of a stir in the Scottish press. The general gist being that Celtic have no real competition, and so their recent record-breaking achievements are less noteworthy. There was one comment in particular that caught my eye:
If Man City went unbeaten in League One do you think he would be having an open-top bus for that? That’s essentially the gulf between Celtic and the rest in Scotland.
This got me thinking, what if Man City were dropped into League One? How well would they do? And how would they compare to this season’s Celtic side?
To test this I estimated team strengths for the Scottish Premier League, and for the English leagues down to League One. The method is the same as I outlined in my previous blog post, and the code exists on a branch of the same blog’s repo.
With team strengths in-hand, I ran simulations of a modified League One and a modified Scottish Premiership 10,000 times. That is, simulating each game’s outcomes based on the relative team strengths. The modified League One consists of the best-rated 23 teams from League One (at the start of 2017/18) and Manchester City. Then, to make the leagues a bit more comparable, I simplified the league format in Scotland so that each team played every other team twice, meaning each team played a total of 44 games (close-ish to League One’s 46 games apiece).
So what happened?
Perhaps the easiest way is to compare the distributions on points-per-game tallied by Celtic and Manchester City in the sims. While it’s not a totally fair comparison (Celtic ‘played’ two fewer games, for example), I think this gives us a rough basis for comparison.
As you can see, Manchester City do significantly better in League One than Celtic do in our bastardisation of the Scottish Premier League. If you look closely, you can even see a non-zero number of simulations (27 out of 10,000 in this run, to be exact) in which Manchester City won every single game.
What about unbeaten seasons?
Well, Celtic went unbeaten in their 44 game ‘season’ in less that 2% of the simulations. Man City, meanwhile, went unbeaten in 15% of their simulations.
So it seems like Manchester City would be a much more dominant side in League One than Celtic would in a 44 game round-robin league (shocking, I know). This then begs the question: how much does this extra talent actually matter?
Looking at the simulations one last time, we can see how frequently each team won the league. While there wasn’t a single simulation in which City did not win their league, Celtic lost out a little over 2% of the time.
So yes, the gap between Man City and League One is much bigger than Celtic and the Premiership. I don’t think this is groundbreaking, and I am sure you’d find countless fans of Scottish Football telling you the same thing. Nonetheless, Celtic are good enough that a 44 game league ceases to be a competition anymore.
Of course, that isn’t the actual format of the league. And Man City don’t actually play in League One.
It is possible that Tony was referring to the difference in resources in the Scottish Premiership as well as just the quality of the teams. However, accounting for this would require more nuance and would be harder. I think my way is more fun anyway. ↩︎
I smoothed the histograms of simulated points by drawing kernel density estimates with ggplot2’s
geom_density. This is to remove some of the artefacts resulting from only doing 10,000 sims. It is a little sneaky but hopefully you can forgive me. ↩︎
I think this also highlights how impressive and unlikely Celtic’s 66-game (so far) unbeaten run is, regardless of you thoughts on the league. ↩︎
Rangers won in 227 of 10,000 simulations. Aberdeen won 30 (ties between Aberdeen and Rangers were decided at random). Hibs won once. ↩︎