FM20 Statistics in Focus
I’ve already covered with the By the Numbers post and series that I’m a bit of a nerd. I love statistics and use them in my real job fairly frequently. I’ve generally extended that to Football Manager as well. It’s not simply because I love a good spreadsheet though. Whether we realise it or not our lives are pretty much governed by statistics. Medical treatments are approved because they are (statistically) significantly better than alternatives. Big business makes decisions because they’ll make significantly more money from it. And now football clubs make transfer and business decisions based on the numbers.
A lot of the world is moving away from common sense and gut instinct because of the biases inherent in them. We make lots of cognitive errors as the brain likes a good shortcut. We get fixated and ignore alternatives, we think memorable events are common events (they are not the same), and make bad decisions because we are loss averse. But, if we rely on statistics, the numbers, we can overcome this.
Or in FM20 terms I can use statistics to lead my newgens to glory on a shoestring budget. There’s plenty of data in the game we’ve just got to get it out and use it correctly.
Statistics not attributes
That right. I’m not talking about the technical, mental or physical attributes when I’m talking about statistics or stats. At least not directly. I’m talking about the in-game measures of performance, or the metrics we have to add like xG. Or I’m talking about using really nerdy tests like regression and t-tests to work out what attributes do, what impact they have, and what the team DNA should be.
Statistical Significance not One-Off Experiments
When I talk about differences or relationships, predictors and outputs, I’m talking about statistically significant results. Ones based on big data sets rather than one off-seasons. And differences or relationships that aren’t based on chance, but are very likely (95%) to be representative of real effects and relationships.
This means I break out the stats software, and that small samples and averages generally don’t work for the stuff I’m looking at.
What FM currently has
FM20 does a decent job of giving you a range of player and match statistics to look at. There’s a lot already going on in the game that you can use to try and improve your performance or guide your key decisions like team selection and transfers.
You can see above that on a player level the base profile already collates information like dribbles per game, shots on target and pass percentage. Below you can see that which some tweaking of your squad views you can expand this even further and include even more information like Points Per Game, win percentage, mistakes made leading to a goal and so on.
Within the match analysis, you can break it down by actions such as shooting and passing. This is great for looking at combinations or problem areas of the pitch. Try overlaying crosses and aerial battle stats for example.
Or if you are feeling a little nerdier you can look at heatmaps, passing combinations and movement. This can be useful for seeing your build-up play, and importantly if your players are moving as you intend. What you tell them to do and what they actually do don’t always align.
You can take into account the break down of assists and goal types, and through in some locations.
So there’s plenty within the game, so much so you could probably have a good at playing without attributes. There are skins out there that remove the attributes so you can play the purest game. I’m not likely to go that far as attributes are such are core part of FM but I will and do lean heavily on the actual output and metrics. We’ve all had players who have great attributes but that don’t deliver – this is why we need statistics.
Statistics FM is missing
Whilst there are lots of in-game statistics for you to play around with there are plenty that haven’t made the cut into FM20. For whatever reason Football Manager doesn’t offer the full suite of metrics the likes of Opta, StatsBomb and other football analytic agency have at their disposal.
Maybe there think it’s too nerdy, maybe they think there isn’t the demand. But from books like Football Hackers and Outside the Box it is clear to see the huge impact some of these missing metrics can have. They’re all about gaining an edge and squeezing out a little more from your team. As a frequent manager of smaller teams on shoestring budgets that’s what I’m all about. That and hoofball inspired violent play
xG
Expected Goals is a key metric for understanding how many goals a team should score. Or to be more precise how many goals on average you would expect them to score considering the chances they had. It considers the location and situation shots were made in, considers how likely each individual shot from that area and situation is likely to go in (as a decimal value) and then gives an overall value for all the shots in a match.
Sites like Understat keep a good record of them in the top flight, allowing you to see whether a team over or underpeformed. The same can be applied to individual players to assess their performance and contribution as well.
It’s better than simply shots on target or conversion rate as it takes into account the quality of the chance rather than simply the number of chances. Shots close and central to the goal are better quality chances, they are more likely to go in (33% of the time for immediately in front of goal for example), compared to long-distance shots (outside the area might be between 5 and 11%).
As mentioned here, and for Dictate the Game, I’ve demonstrated how to look at this in Football Manager, and how to extract it using my xG spreadsheet. It’s easy enough to assess what impact your tactical changes have had, if you’re in a bad patch or whether a player needs replacing.
xA
Often overlooked expected assists follow in the footsteps of xG. Every pass has the chance of leading to a goal, of being an assist. This is linked to where the pass is made from and to, the type of ball and a myriad of other factors. But it gives a value for each pass just like xG.
I looked at this recently for Dictate the Game and used it to assess my strikers. Again it a metric that takes into account quality and quantity. Lots of aimless passes have less impact in terms of chances created and goals score than a few penetrating key passes.
xG against
This flips the approach of xG around and looks at keepers instead of strikers. Keepers can be very hard to assess. If they let goals in they could still have had a good match but have been overwhelmed with shots because of a poor defence (or exceptional attack) in front of them. Likewise, a keeper might only concede one goal but that might have been a sloppy and unnecessary goal statistically. A weak keeper can hide behind a strong defence.
Assessing the Keeper
To assess this we can look at all the shots on target that your keeper faces. All the shots that require a save of some sort. Blocked shots and off-target shots don’t matter as they’re not a test for the keeper. Once we have the on-target shots we can work out the xG, we can work out how many goals on average we would expect to concede from these shots. We can then compare this to how many goals were actually conceded. If the keeper concede less goals than the xG then we have a ‘saving’. The keeper is efficient and is doing better than expected. If the keeper concedes more then we have an inefficient keeper who is essentially underperforming.
Do this across enough matches and you can work out how many extra goals compared to the average your keeper is saving or costing you. Keepers costing you goals need to go. And keepers that are ‘saving’ you can carry on.
I show you how to do it in this Dictate the Game article. This process eliminates the issues with the influence of the defence and team quality. You can assess your keeper for your newly promoted side even if they are picking the ball out of the net a lot. It’s not skewed by the amount conceded, it takes into account how many you would expect to concede (which could be high or low).
Pitch Impact or Received Passes
Something I started taking into account around the same time as xA was the position players, or my strikers, received passes in. Not just where they got their shots off from but were they receiving the ball in dangerous locations or not? This could be a good indication of whether a player is good at getting into goal scoring positions. A comparison of their Received Passes or Pitch Impact to their xG might also reveal which players are wasteful and which add value by improving their shot locations.
There are metrics relating to pitch location and the control players exert from this position and their passing. It’s not quite as simple as where are they when they get the ball, but FM20 is limited in the information you can get out. So I just used the same measure as I did based on location for xG and xA and applied it to where players received passes.
As you can see in this article it is an extra metric you can use to break down and analyse player peformance.
Assessing Players
All the above statistics give you an extra edge when assessing your players. Want to know whether your striker is actually cutting it then you can check not only what the xG is for them but how that compares to what you would expect. Worried your Keeper is rubbish? Then out comes the xG against which lets you cut through the confusion caused by how good (or not) the defensive line in front of them are.
It doesn’t have to be things like xG or xA that you rely on either. You can go back to the in-game metrics and use the statistics you think are relevant for assessing performance by role. If you play a tactic with lots of wing play, where you rely on beating players and then getting a cross in for your striker then you’ve got three key performance indicators (KPI’s) that you want to look at. Dribbles completed crosses completed and aerial battles won. Combined these vanilla statistics will let you know how the key battles on the pitch are going for you.
And importantly these can all let you know who is stealing a wage, and who is good value for your side.
Assessing Tactics
You can take a much more holistic approach though and use these statistics to breakdown your tactic. One thing I always worry about with tactical changes is whether they are actually making a real tangible difference? Have all my tweaks and changes actually changed the things I hoped they would change?
Going even further to assess DNA
When looking at tactics you can make a clear link to recruitment. If you want to play a certain way, you need to recruit certain types of players. You need principles and guidelines for your recruitment so you have a cohesive side that is going to meet the objectives of your tactic. To ensure your players produce the output, and the key performance indicators (KPI’s). The hallmarks of success for your playing style.
Sometimes this is referred to the team’s DNA. You’ll have seen it in great blog posts by the likes of FM Grasshopper all the way back in FM17, and referred to as fibra. This is recruitment that focusses on four select attributes that define the type of players you want, and the playstyle you want.
My DNA
I had a look at this for my own teams. But I started from a different point. Whereas most consider the style and philosophy they want and work from that to the attributes I decided to look at the outputs, the KPI’s. I know that I want to play aggressive hoofball. I love direct football and crunching tackles. If I could get each team to play as the reincarnated version of the 1988 Wimbledon side then I’d be a very happy man. I know it’s not perfect football, but it is entertaining.
I had a 4-4-2 that tackled hard and lumped it forward to a big man/small man combination. Peak crazy gang. If this tactic was working well I would be making a lot of key tackles, interceptions, winning aerial battles and headers etc. So I decided to look at lots of data I had over seasons and season of playing this 4-4-2. Then using a form of statistical analysis (a regression) work out which attributes predict success with these KPI’s. These would then become the attributes that formed the DNA I would look out for when recruiting. If that all sounds a bit confusing you can read about it here. But in short, rather than going for the gut feeling or common-sense view of what attribute should work for my DNA and playstyle I actually tested it. And I got a few surprises.
Moneyball
I alluded to this earlier. Metrics, or the output of your players, are key. It’s not enough to look like they’ll play well, they have to actually do the job you need on the pitch. Billy Beane was able to overlook the aesthetics and gut instincts scouts and coaches had about baseball. He instead relied on players who could meet key outputs. Outputs that correlated with winning games.
Arguably real Moneyball is about finding players who hit these targets for you that are for some reason undervalued. Or, by finding metrics that others undervalue, and players who hit them, and recruiting them. As is mentioned in Football Manager circles it’s not about buying low and selling high, it’s about getting value.
Moneyball in FM20
This links back to team DNA. If we know what metrics or KPI’s predict success for our squad, or our tactic, we can find players who are good at these aspects. We can find the ones that are undervalued for some reason and bring them in. Maybe interceptions per 90 is a KPI for our tactic. We want to either find players who are good at this but have been disregarded and undervalued (age, personality, footedness, undervalued league etc.) or find players with the attributes we know are significant predictors of these KPI’s that are likewise undervalued.
Whilst I’ll not likely have any all-out Moneyball saves, there will be an element of it for any save driven by statistics. If you want more about Moneyball in Football Manager though take a look at FM Vars’ articles on Dictate the Game.
Experiments?
I do love the odd experiment as well. But they need to be real experiments. There are a lot of ‘experiments’ out there in FM that are really just one season simulations. Often really interesting ones where a variable or factor is changed to see what happens. But this isn’t an experiment in the truest sense though they can give really useful indications for what is going on. For an experiment, variables need to be controlled for, and then repeated measures taken. One season or set of data often isn’t really enough to know whether the altered variable is the reason for the change. Or whether any differences you find are just down to chance. That’s why you repeat or have a larger data set.
It’s not enough to know that two different teams, or players etc. have scored a different amount of goals for example. What you need to know is if that difference is a statistically significant difference. One that reflects a real impact of the variable you are interested in rather than being a difference due to pure chance.
It’s not something I dive into often but I have in the past for Dictate the Game. I created an experimental league with 6 identical teams under a transfer embargo to keep everything controlled. Down to the staff and facilities so that the only differences in end results and metrics would be because of whatever variables I had altered.
Professionalism Experiment
In one experiment I played around with the player professionalism to see what impact it had on performances. This was over the course of several seasons (10 seasons, giving over several thousand matches worth of data). The personalities were changed so that some teams had good ones, and some had the bad. This was the only factor that differed between the teams.
In the end, by seeing what happened over thousands of matches, we discovered a statistically significant effect of personality. In this case professional teams were gaining around 15 extra points per season compared to teams with less professional players.
Decision Making Experiment
Using the same league we made a change to the decision making attributes. I took a couple of experiments due to some potential related mistakes. But we eventually concluded that there was no significant difference in performance between those with High, Medium and Low decision-making ability. In isolation decision making is not a game-changer, it looks like it is reliant on other factors. But we did discover it is a big ability sink.
Striker Traits Experiment
This time we altered the ability of our strikers and then gave them a range of player preferred moves or traits. Some liked to lob the keeper, some liked to round them, other placed shots and some shot with power.
Initially, we found no difference between the traits in terms of goal scored, and no interaction in ability and traits. Despite other theories about shooting with power and placing shots being useful for different ability players.
We did follow it up with more of a natural experiment, using several seasons worth of data from the vanilla database, across the top two divisions in England, Spain and Italy. We were able to narrow down which attributes were predictors of success for players with these different traits. This was very similar to the approach we took for the team DNA analysis.
FMTahiti and Statistics
So after this long ramble (sorry), I think it’s fair to say I love statistics. I’ll keep adding to this article and to this site. Whilst not all of my saves will have a stats angle its something I’ll always be visiting. Even if it is just in my By the Numbers save.