#Project4704 – methodology.
Generic football image. Original - lawinsport.com. Effects - kusocartoon.com

#Project4704 – methodology.

To recap, then:

Imagine that I have no knowledge of what works, and what doesn’t, in Football Manager.

I have no knowledge of what works, but I do have a mind full of preconceptions: I am convinced that asymmetrical formations are the work of the devil, and that a system with inverted wing-backs is likely to overwhelm the central areas of the opponent’s midfield. What my preconceptions do, though, is to cost me time, and suffering, when I put them into practise in an actual saved game, only to be picked apart again by the AI. I must do whatever I can do to set aside my preconceptions.

I am someone who loves the slow build-up of the classical Balkan national teams in real-life football. (I love it, but critics would call it keeping possession for the sake of it.) In my mind’s eye, I see a central defender strolling forward with the ball, only for him to think better of playing a creative pass, and rolling it out to another central defensive colleague, stationed in one of the channels. As far as I am concerned – it’s the combined weight of memory, and history – this is the best way to play football. Why should players exert themselves physically, when it’s better to make the ball do the work?

My problems are obvious. Firstly, there is no single best way to play football – if there were, everyone would be doing it, but only until someone came up with the antidote – and, anyway, Football Manager is a game, and there’s no guarantee that laconic passing, the revival of Bulgaria 1994, or Croatia 1998, is going to get me anywhere in FM21.

What, then, to do?

Well, you can play the computer against itself a few thousand times, dump the results into a spreadsheet, and look at what the numbers are telling you, thus scattering your preconceptions to the wind.

This is my methodology:

I edited a saved game, in the Slovakian second division, so that I had two teams – Team Opposition, and Team Test – set up to play a pre-season friendly against each other.

Team Opposition and Team Test are identical. They both have 40 players, all of which are copies of each other, and there are enough players in the squad that if, say, a LWB is lost to injury, there’ll be another copy of that LWB on the bench to replace him.

Every player attribute has been fixed at 10.

Aside from club doctors, the only staff member is a clone of Jose Mourinho, with all his attributes – pressing intensity, defensive line, the type of passing he likes to play, everything – all set to 10. (I imagine he would be horrified that his star has fallen so far, but he is now called either Test Manager, or Opposition Manager, though he was born in Setúbal, and his darkness will somehow live on.) So, everything is as equal as it can be. So, whatever the results are, it can only be to do with the formations, because the players and manager are the same every single time.

Team Opposition are at home to Team Test in the friendly.

Football Manager 21 gives AI managers one of 28 default formations to be going on with – you can change it in the Staff Details when you are using the in-game editor. So what I have done, is to play all 28 of those formations against each other, with six repetitions per formation – so when Team Test is playing 442, and Team Opposition is playing 4141, I’ll repeat that configuration six times, in order that the data I am generating has some sort of depth to it. And, of course, 28 * 28 * 6 = 4704.

This presents us with 4704 repetitions of what I would call top-line stats: goals scored, shots on, shots off, possession, corners, fouls, passes completed, passes attempted, and, perhaps most interesting of all, xG. It’s possible to extend these top-line, team-based stats, so that I am looking at the chalkboard stats for individual players, and that’s definitely something I am thinking about (but it does take a long time to do, about eight hours per 28 * 6 game cycle. It’s like having a baby, getting up in the middle of the night to check that it’s running okay, and setting another 168-match run going.)

Over the next few days, I’ll talk about my findings in a number of posts. The dataset is only relatively small, but it should give some idea about what might work when the AI throws a five-man defence at you, and what might counter a midfield diamond. There are, of course, various ways of cutting the numbers. Paradoxically, I am of the opinion that xG is at least as important as the final scoreline, but I’ll be making use of both.

The dream situation would be to find a single formation which is universal, more or less, because the less tactical shuffling about you have to do, the more your squad retains its tactical familiarity. I’d like to think it’s possible, but I am very far away from it. But the flexibility you are given in a tactical sense means that your double pivot can be very defensive, or extremely aggressive – it depends how you alter the roles and duties. Let’s see what conclusions can be drawn from the top-line statistics, when I am sufficiently awake to examine them, and write about them.

Leave a Reply