As mentioned in our previous post we are going to take a step away from the sterile lab setting (though we will be back) and look at some more natural data. Data that has been occurring in my saves. As a result we will be able to use statistics around key performance indicators (KPI’s) to actually determine what our club DNA should be for for each of our units: defence and attack. Recruiting players FM19.
There are some great articles, blogs and videos out there about the use of a team DNA. The idea that there are certain properties or attributes you can reduce the team to. Essentially attempting to embody a particular football philosophy within your team by recruiting players with those attributes and traits. Whether that is the high press of Liverpool, the creative use of space of Man City, or the borderline aggressive psychopathy of the 1980’s Wimbledon Crazy Gang.
However, how do you go about determining what the club DNA should be? There are various approaches. For example you can simply distil the philosophy down into the key attributes that logically fit that approach. If you want a hard-working, never-say-die club that don’t crumble under pressure then your players would logically need to have high work rate, team work, and determination. Perhaps even bravery in addition to this. Following this decision you can then just recruit players with higher values in these DNA attributes. Hoping that the attributes selected translate to on the pitch performances.
Yet this approach misses an important factor, an important variable some of us nerds might say. You might know what philosophy you want the players to live and breath but are they actually playing that way? Your tactics will play a part in this. Obviously there’s no point having attacking DNA if your tactical set up doesn’t mirror this. Furthermore what you are asking your players to do tactically might throw up some surprises.
I’m going to work from the assumption that you should set up tactically, and get your team playing how you want them to play. You might use some common sense in how you value attributes for this but they are not your DNA yet. If you want the DNA that allows you to play beautiful no-holds barred hoofball that would mark you out as the secret love child of Pulis and Allardyce you firstly need to get your team to start playing like this. Or at least have the tactical framework present. You need to watch your matches and check your team is performing along the lines you want. Then, you can work out what KPI’s are that quantify that performance.
Finally you can then analyse your player performances to determine what attributes predict success in your chosen KPI’s. Once you know that you have your DNA. You know what attributes are core to the performance you want on the pitch.
To do this takes data, and it takes time. You need plenty of it from your own team for your DNA test. And remember, if you do this is only applies to your team and your selected set up. If you make a big change in style then the data collection starts again…
WHY THIS ORDER? AND WHY MY OWN DATA?
Simple. Other teams don’t play like you. They don’t have your philosophy and aren’t trying to do the same actions in the same scenarios as you. If you use their stats for KPI’s and compare to attributes generally you don’t know what influence their tactic is having (as well as other variables like competition level and the tactics they face)
Additionally you need plenty of data as we are going to be looking at some fairly complex interactions between variables. We’re not simply comparing groups anymore. We are going to be seeing what sets of attributes (our predictor variables) significantly influence our KPI’s (our outcome variables).
Finally whilst I can set up a tactic and get it playing roughly how my philosophy and eventual DNA dictates what I want is for that DNA to be stamped across the team. I want to make sure I know exactly what gets my team ticking, and then shamelessly exploit those attributes until every new player in my team bleeds it – from star player, to back up, all the way down to the youth team, maximising it’s impact.
MY DNA TEST
So how do I want my team to play? Well I normally play lower league in FM. In fact really lower league. I might be in the Scottish Lowland League, or Tier 7 of England, or even playing as minnow in Tahiti. As a result I’m not going to be playing much expansive samba football. If I get my team to gegenpress they’ll probably die of exhaustion. If I ask them to play vertical tiki-taka they’ll probably just boot the ball into the air and kill passing birds.
With this in mind I’m going to take inspiration from a relatively unsung and underappreciated source – The Crazy Gang. I want to get the ball forward quickly, I want to be direct, I want to tackle hard and without concern for the health of the opposing players. Essentially I want my team to play antagonistic angry hoofball. I have a feeling aggression is going to be a key attribute for this, but what else will be?
THE (DNA) SAMPLE
I’ve taken approximately 9 seasons worth of data from both my senior and youth team playing a tactic that I hope encourages hoofball. Watching endless passes to a target man from my centre and fullbacks, and the gradual decline into madness through boredom of my now bypassed midfielders, seems to confirm that I’m on the right track. Ultimately that gives me the data from essentially 235 individual players of my preferred playing style.
I’ve only got 9 or so seasons as the rest from that particular save are from when I took a more naive approach and didn’t fully embrace lower league long ball. I’m not trying to have develop a creative and expressive team so I don’t want any non-hoofball data messing this up. If I had more I could add more, and the more that is added the bigger, better and more sensitive the data set becomes.
The end of season data has been taken, in the form of attributes and KPI’s.
The KPI’s have been broken down by unit: Defence, and Attack. Whilst I want them all to have the same core DNA we have to acknowledge that the job we ask our players to do tactically will differ even though they have the same overall aim. This isn’t an exhaustive list but it is a fairly broad coverage for this example post to give you an idea of how this works. The midfield are also being missed out here as in my tactic midfielders are either defenders who are slightly forward, or attackers that arrive slightly late. Even the wingers cross from deep like the fullbacks.
For defence I will be focussing on: Interceptions per 90, Headers per 90, Tackle Success Rate, Key Headers and Key Tackles. This reflects that my players need to tackle hard, win the ball in the air and importantly, because I play against other rubbish teams, will have to face a lot of long balls and crosses.
For attack I will be focussing on: Goals, Goals per 90, Headers per 90, Key Headers and Offsides. This reflects the physical game being played and the standard scoring KPI’s you can’t get away. I’m expecting them to bully the opposition defence. I want the opposition to be more worried about protecting themselves than they are getting the ball . I want my strikers to get stuck into the tackle and I want them to terrify the defenders. Finally I also want my strikers to stay onside when the long ball comes through, otherwise it is just a wasted pass.
For all of my players I’m also checking on mistakes leading to a goal, average win rate, and average points won per game as a general check of effectiveness. I’m completely ignoring average rating as this is skewed toward favouring attacking metrics. A key tackle (that might save your team from a goal and therefore defeat) does not get the same weighting as an assist or goal when it comes to the match rating. I think using average rating alone for any statistics should be avoided, and when it is used it should be used with caution.
DNA TEST: DEFENCE
I mentioned previously in the FM Statistics Lab series that we would be using SPSS (specialist statistics software) and that we would be running what is known as a multi-linear (or multiple) regression. This allows us to see in groups of variables can predict another variable, and importantly which of these predictor variables are better at predicting the change in your outcome variables.
With the KPI’s set the next step is to enter our potential predictors. In this case I am entering every single visible player attribute bar goalkeeping data.
Running the regression we discover that first of all this combination of variables can significantly predict (p <.05) key tackles, interceptions, key headers but not general tackle rate (p >.05).
Importantly marking and positioning were the only significant predictors of the number of key tackles made. For interceptions positioning, pace, jumping reach and aggression were the only significant predictors (with positioning being much stronger than the rest). Finally for key headers positioning, jumping, heading, strength and aggression were the significant predictors but with heading actually being the weakest of them. Likewise with the general headers per 90. Jumping was the strongest predictor followed by positioning, heading, aggression and bravery.
So we have our first shock. Supposedly key defensive attributes like tackling have no significant impact on our play style (as measured by the KPI’s). Even more importantly positioning, jumping reach and aggression all seem to be key to our approach and generally more influential than heading. I used to regulary look for traditional attributes like tackling, marking and heading in a defender, but the analysis points to other considerations for my playstyle. Therefpre if I want to maximise my team DNA and philosophy then I need to change my searches.
DNA TEST: ATTACK
Running the regression we first discover that first of all this combination of variables significantly predicts goals scored, minutes per goal, goals per 90, headers per 90, shots on target per 90 and number of offsides.
For goals it is revealed that along with the expected finishing attribute, jumping, heading, dribbling and agression are all significant predictors, with little between them. With goals per 90 finishing, and jumping are again involved, as are strength and teamwork. Offsides can be significantly predicted by pace, and bizzarely heading and finishing though this is likely because strikers and simply more likely to be offside and generally speaking they are also more likely to have a higher attribute for finishing (at least) than defenders. Agility negatively predicts the amount of offsides, with more agile players being offside less. Finally for shots on target per 90 it is predicted by finishing, heading and aggression.
Headers per 90 has also been covered above.
Importantly for us aggression makes a frequent appearance, and whilst finishing and heading may seem obviously predictors composure is oddly missing from the picture (as is pace and acceleration in a positive sense). Again I used to search for strikers by using finishing, composure and the pace and acceleration. I know now not to bother with composure for this tactic and value aggression and jumping more.
DNA TEST: GENERAL
Testing against some general KPI’s like points won per game by a player we discovered that the attributes that most significantly contributed to an increase in points per game were vision, aggression, stamina, natural fitness, balance and off the ball.
Secondly if we look at mistakes, mistakes leading to a goal or win rate we are not able to predict them based on our data.
Again we see some general dark horses. Aggression seems to be key for our approach which is what we’d expect but players with better vision (for long ball passes perhaps), better positioning and off the ball work (and tactical awareness/reading of the game), fitness and stamina, and balance (keeping on their feet after aerial battles and long passes) seem to contribute more.
From the above I now know what I need to look for when recruiting if I want to enhance the performance of my hoofball tactic, and by extension embodied that Crazy Gang spirit. As a result of all of this I know that if I want my team to effectively embody a club philosophy of being hard in the tackle with aggressive direct football that does not hold back then the key attributes that make up that DNA are:
- Jumping Reach
With secondary attributes being:
- Marking (defenders)
- Finishing (attackers)
Some of that is common sense, but some of it is a surprise that would be missed. Importantly statistics back all of this up. It won’t hurt me to add other nice attributes like determination and work rate to my DNA. I just know that they don’t have the same on pitch impact for my philosophy
WHY DOES THIS MATTER?
Sometimes in FM there is a disconnect between what you think should work, and what actually works. There are so many variables and some much complexity and interaction between them that at times what we want to implement can be lost.
However looking at your own data can help you address this, and from this you can change your approach. I used to recruit defenders almost purely on defensive technical attributes like tackling, marking and heading. Whilst some of these are still obviously important I now know that actually other attributes have a bigger impact on my footballing performance and philosophy. For example I’m never going to turn down a technically good defender, but when I have limited resources and options I know how to maximise my recruitment and get not only a better performance but the performance I want. I know not to worry about my strikers composure as other attributes have a much better impact.
Hoofball is just the example I’ve used here but you can do this with any approach. If you have a way you want to play you can trace the steps back from your tactics and KPI and run your own FM DNA test.
Any suggestions, questions then let us know below in the comments. To find out more about some of the statistical terminology then check out Decisions, Decisions, Decisions: FM Statistics Lab.
Make sure you follow us on our social media platforms and let us know if this article has helped you:
Other articles you may enjoy:
Decisions, Decisions, Decisions: FM Statistics Lab
Dictate the Game Podcast 11 | The FM Editor & DTG Cup
Dynamo Project: Introducing the Club DNA
Guide to the Wide Target Man and the Flo Pass Tactic
15 CommentsLeave a Reply
Hi, this is thought provoking (and makes me want to play #hoofball) but it would turn my save upside down…
Any chance of you doing av similiar test run for other playing styles?
I could do this with other styles. In theory it’s pretty simple. Data goes in and we make sure we are comparing to the KPI’s that are important for that tactic. So if it’s more of a wing play situation then the KPI’s I would be focussed on would include crosses completed, attempted, and dribbles for example.
The hard bit is having enough data so I have about 9 seasons of data, all from my teams playing hoofball, using pretty much the same tactic. If I want to look at another style then I would need data from teams playing that style. I did start using a more creative passing tactic but got 3 seasons in and had to revert back to pure hoofball after getting promoted beyond my resources! So if anyone has any data to donate I could use that….
Alternatively I could set up a few tactics in my lab leagues (from Lab posts 1 and 2) and holiday to get some data that way. What sort of styles do you play? What would you be interested in seeing?
As of now I play a Mourinho-like approach with Esbjerg fB in Denmark. I would expect that a few of the attributes you have singled out will trigger/enhance the same KPI’s regardless of playing style, though. Your thoughts on this? As you point out, the ME calculations are complex, but I’d be massively surprised if say interceptions in a posession oriented system require another set of attributes for the defensive unit to perform.
I think you’re right about some of the key attributes. I think positioning for example, and jumping reach will always be pretty useful for interceptions. But what might change is how important other fringe attributes become – like off the ball and anticipation. They might be more important if you have to deal with more creative players. Likewise with pace and acceleration that might become more important as you face more players at the top of those attributes.
I think for my tactic, and the level of opposition, a good structure to the defense with good positional awareness can deal with most of the balls in without needing to be too fast or too technical (tackling for example) but against more creative teams, or teams that can dribble and pass around on the ground more things might be different.
What sort of tactics do you come up against in Denmark?
It’s a good mix really of 442s – FCK, Randers, 4231s – FCM, Brøndby and I often see teams changing to 3 striker formations when in need of goals late in the game.
I think the defensive DNA might be pretty similar then. I faced a lot of 4231’s over my 9 seasons, and 442’s, and 4141’s. I still think there might be a slight shift as there are more creative players in that league than in mine (all in Tahiti!) but positioning, jumping, marking and aggression are still likely to be key.
Going forwards though…that might be worth looking at. I think I might try and look at a few different approaches/styles for my next post.
I love this idea and together with the Moneyball approach this is the way I’ve started to play the game and I find ite more fun now.
What I try to get my head around is how the program calculated which attributes was “active” with which KPI; or have I just gotten that wrong?
But I love this lab series!
Thanks! I love a bit of moneyball as well. And right wage control if I can. Nerdy but rewarding when your team punches above its weight!
You are pretty close. So instead of active we would say which ones are significantly related or able to predict a change in the KPI. Beyond random chance. So all the ones in our analysis that we’ve picked out we are at least 95% sure are real relationships.
The analysis basically looks at all the possible relationships between your KPI of choice (say key tackles) and the attributes you have entered (all outfield and non-dead ball ones usually). It works out which are reliable relationship that have some predictive power and then tells you how much those attributes can be relied on, as one big model. So attributes A to ?, but not… predict improved scores for key tackling. And you can rank them based on which contribute more than others.
It’s a type of analysis that’s used often for predicting risk factors in health outcomes or forensic outcomes but can be used for any situation where you have lots of potential relationships you want to look at in one go