States of Play

How managerial influence changes how Celtic play given different game states, and what this may tell us about Martin O'Neill specifically going forward.

A couple of topics we touch on regularly are mashed up today in what the kids of my youth would call a portmanteau.

We muse often about the power of the manager and how much influence they really have. Famously, Ian Graham opined in his book “How To Win The Premier League” that it is likely surprisingly small in terms of measurable impact, and evaluating this is the “holy grail” of football analytics.

Secondly, we usually try and contextualise performance in relation to what is labelled “game state”. Simply, this is the score and time at a given match point. Teams chasing an equaliser, for example, can be expected to attack more frequently, and teams defending a lead might be expected to attack less frequently.

Running through those themes is the consideration of Martin O’Neill and his time at Celtic, and our attempts to attribute what exactly he brought to the club that can explain the gap between objectively spotty performances (high goals conceded, low xG differential) and outcomes (20 league wins in 23).

I do not want to overpromise here, and under deliver, in relation to the pursuit of Graham’s holy grail, which as everyone knows is a wooden drinking vessel buried somewhere on an Essex farm, but rather provide some nourishment for cognitive function.

In the latest pod, which I highly recommend, I presented several views that compared the team’s xG performance based on the three score-based game states – i.e. being level, being ahead and being behind. When I built those, I had some hare-brain notion of cars and accelerating and braking, basically stuff I have no clue or interest in, and ended up displaying xG per hour as some sort of proxy for F1 performance. To get back in my lane, I’ve reworked for the much safer and more familiar xG per 90 minutes. A concept we can all get behind!

I have contained this analysis to 2025-26 and league play only as one of the points of the exercise is manager impact, and having the same squad of players as the control group therefore makes sense. The trade off is some small sample sizes.