Using Data Visualisation and Coding in R as a scouting strategy.

[StatSav]

Hello,

As FM20 heads towards the end of it's life cycle and I find myself living life as an unemployed graduate (thanks covid!) I've decided to start applying my novice coding knowledge to Football Manager. Whilst "plug and play" systems, 4-2-3-1 gegenpressing and so on can be fun, I feel that using the plethora of data that FM provides us can potentially provide an efficient and effective strategic advantage over the AI and other players. Whilst I will be using this thread to post some of my own progression and ideas, I'd like to strongly encourage other likeminded community members to contribute also. I truly am a novice to coding, data visualisation and general data science so all feedback and learning from others is massively appreciated.

So. A case study.

I've decided to target a somewhat niche area that I believe is overlooked during squad building, training and tactical brainstorming - the full backs. It is no secret that over the past couple of decades in football there has been a massive emphasis on using full backs as the "extra man", overloading wide areas of the pitch and generally providing genuine creativity to a side. Nobody needs this explaining to them, but this is the basis of the "investigation". For my first piece of "work" on these forums, I've decided to simulate in full match detail every single match in the top 5 European leagues for the 2019/2020 season using a relatively up-to-date transfer database (post-Bale arriving at Spurs anyway!) as a means of having a thorough data set to manipulate.

The Process.

Once the season was simulated, these are the filters that I selected to find the appropriate database of full backs. It's a simple set of filters really; the player must have played at least 2000 minutes of football in all competitions across 2019/2020, be currently contracted to a side in the top 5 European leagues and be either a natural right back or left back. I opted not to include wing back as I personally only ever play a back 4 myself and I felt as if including wing backs may have 1) included more advanced wide players also natural at wing back which would have skewed the data set and/or 2) meant that the data set was somewhat biased against out-and-out full backs in a creative sense, which is what I intended on looking at.

Filtering done, it was time to pick a couple of variables that would be appropriate for assessing the overall build-up play inclusion and creativity of the 218 full backs discovered within the data set. In the end, I opted for "passes completed per 90 minutes" and "key passes per 90 minutes". I believe that these two variables speak for themselves - I wanted to see which full backs completed a lot of passes and also which full backs were creative with their passes. With filtering and variables selected, it was time to create a graph that was 1) visually appealing and 2) actually painted a picture of what was going on.

Here's what I came up with.

Apologies if this isn't the prettiest - I did warn you I'm a novice!

I believe that the graph is quite self-explanatory, but I'll delve a little deeper. The idea of splitting the scatterplot in to quadrants rather than using a line of best fit was to create a simple way of identifying which players were above/below average in certain categories. Towards the bottom left of the graph we have players that completed a low number of both passes and key passes, whilst towards the top right we can see players that completed both a high number of passes and key passes. The other 2 quadrants are labelled directly for ease of understanding. With that out of the way, we can start to highlight some players of high importance within this data set.

Gaetano Letizia

I believe this player to be outright the most impressive full back SPECIFICALLY focusing upon these two variables within the data set. Note that I am not suggesting he is the best overall full back whatsoever, but when considering just these two variables - his output is outstanding. He is head and shoulders above the competition in regards to creative passing (well - he is followed closely by Kimmich but I haven't yet checked how many games he played in midfield for Bayern!). Let's take a deeper look.

Player Profile and Report

Whilst I do not need to state that he is nowhere near the levels of players such as Trent Alexander-Arnold, I propose the question - does he HAVE to be?
Let's bring this post back a little towards training and tactics. What Letizia IS good at is being creative. A vision attribute of 15, crossing of 16, acceleration and pace of 16 and 15 respectively partnered alongside traits often undervalued by the FM player base such as "Gets Forward Whenever Possible", "Tries Long Range Passes" and "Likes To Beat Man Repeatedly" have caused him to produce truly elite creative output (including 7 assists) for a shoestring-budget Benevento side predicted to place 18th in Serie A before ultimately placing 15th. FIFTEENTH and he produced elite creative output "nearly" matched by only Joshua Kimmich of 2nd-placed FC Bayern.

I feel like this can teach many lessons about what it's important to look out for in a player. Whilst star ratings and wonderkids can be tempting, you can still get a 29-year-old full back with just 12 passing to be a creative force for your side. Letizia is a player that also enjoys big matches, is fairly consistent and does not really show any signs of injury proneness, mixed with a fairly ambitious personality. Let's take this just one step further before wrapping things up by excluding any full back valued above £4m in the data set.

Full backs valued at under £4m

A league of his own

I'm going to wrap this up here and let the data vis do the talking - but taking the time to create things like this can, in my opinion, provide the player valuable insight in to potential transfer targets. Whilst there is no guarantee the player will succeed for the player - is there EVER a guarantee a player will succeed (other than Rhian Brewster on loan, nothing tops that lets be real!)? The next steps for me are to begin creating data vis that present more of a "full picture" about a player, including a mixture of defensive, attacking and other output. Alongside that, taking the time to investigate the player further such as thoroughly checking their coaching/scouting reports, traits, individual attributes and so on will aid in hopefully getting closer to guaranteeing the success of a player.

I'm open to any ideas - but I think I want to focus specifically on creating more "complete" data vis to help analyse players like Letizia that stand out next time, as well as starting to review the player on a match-by-match basis and see what is causing them in-game to post such output. Thanks to anybody that read this far, as mentioned previous I'm a genuine novice at doing this and would appreciate any feedback on the methodology, visualisation, presentation and so on.

[sporadicsmiles]

I just love me some good proper data mining. This is pretty cool and interesting. I'd actually be interested to see it used in a save game, where you target specific roles for players, identify what you think is the best player, sign them and see how it works out. 

I wonder if there would be any interest in a PCA of some of these data too. I mean, I have no idea why or what it would mean. But it would be fun. 

[StatSav]

2 minutes ago, sporadicsmiles said:

I just love me some good proper data mining. This is pretty cool and interesting. I'd actually be interested to see it used in a save game, where you target specific roles for players, identify what you think is the best player, sign them and see how it works out. 

I wonder if there would be any interest in a PCA of some of these data too. I mean, I have no idea why or what it would mean. But it would be fun. 

I'm considering starting a save that has somewhat "masked" attributes. I believe there are skins that instead of showing direct attribute numbers, they simply show a colour for that attribute range. So instead of 11, 12, 13, 14, 15 being shown outright, you'd just get a singular colour to go off of for attributes within the 11-15 range. This would allow you to see where someone is roughly at attributes wise, and then make decisions tactically based upon that without making it too tempting to just avoid the data side of things and sign players with high attributes.

[Big Yellow]

Out of curiousity, how does one extract all that data? I have never looked to be honest....

[mouli]

I believe this is how Liverpool signed Robertson from a relegated Hull City for around €7 million. A starter for a relatively low tier team who had determination to succeed. 

[ferrarinseb]

Its really interesting and great work.

I would like to see player vs player comparisions say Letizia and Kimmich in terms of their performances. To see what exactly Kimmich did or what Letizia. 

[StatSav]

I'll reply to all comments later - thank you for some good ideas, I was not expecting any interest if I am honest!

In the meanwhile, I've been playing around with different colours/layouts for the scatterplot. Excluding goalkeepers, here is every player that played 2000+ minutes in all competitions in Europe's top 5 leagues in my simulation to demonstrate a new layout idea. Better or worse than those in the OP?

 

[Big Yellow]

lighter background is the way to go!

[Pasonen]

2 hours ago, StatSav said:

I'm open to any ideas

Completed passes and keypasses both include freekicks and throw-ins too? At least completed passes do. This makes those numbers unfortunately a bit unreliable. Anyway great work! And Letizia wow!

Edited October 15, 2020 by Pasonen

[StatSav]

56 minutes ago, Pasonen said:

Completed passes and keypasses both include freekicks and throw-ins too? At least completed passes do. This makes those numbers unfortunately a bit unreliable. Anyway great work! And Letizia wow!

They do. I can't really think of any practical way around this, however when applying context it isn't the biggest issue. I can manually enter matches to see which passes were considered "key" and find out pretty soon if a large amount of them are from set pieces. I think in the case of Letizia it actually makes his stats more impressive as he was not likely to be taking free kicks for Benevento due to his attribute being 9 whilst they have players like Oliver Kragl and Gianluca Caprari with 16 and 13 respectively.

[Pasonen]

1 minute ago, StatSav said:

not likely to be taking free kicks for Benevento due to his attribute being 9 whilst they have players like Oliver Kragl and Gianluca Caprari with 16 and 13 respectively.

Yeah true. Crossing 16, vision 15 natural in left and right side. Preferred foot right so he is not bad with left either.

[StatSav]

25 minutes ago, Pasonen said:

Yeah true. Crossing 16, vision 15 natural in left and right side. Preferred foot right so he is not bad with left either.

I'm going to post a deeper look at him later on today or in the early hours of tomorrow morning - I've had a look at a few of his games and it's interesting how the AI chose to use him!

[Bot Makel]

Very interesting work!!

It allows you to understand which players perform well in the game and to find the player you need for your tactics.

It would be interesting to combine different stats to create particular player styles. For example, who in midfield recovers more balls but at the same time is a creative solution.

[icidamien]

This is very interesting.

You are proving that we don't need to have the player with the highest CA to have the best performance. 

Having the 4-5 appropriate attributes for the right role on the pitch is also more important than having the player with the highest CA.

Thanks for your work.

[nugatti]

I'd like to pitch the idea of doing the same data extract more than once, to see if good performers actually do well over multiple tests/time. Then one could perhaps distill which attributes needed for performance combined with roles in a team? What do you think?

Letizi's performance be a one-off in your game, right?

Also I tend to use the various stats 'per 90' as these are able to use effectively as filters in the player search screen. 

Btw, do you know this guy? https://afmoldtimer.home.blog/category/ac-milan/

He's also using stats heavily for squad building and recruitment

Edited October 15, 2020 by nugatti
Typo

[davidbarros2]

I have nothing of value to add to the discussion here, just stopped here to congratulate the OP for such a wonderfull idea and a very well laid out post.

Looking forward for more!

[Sebas]

Honestly fantastic work. Reminds me of Betis scouting Lo Celso after selling Fabián Ruiz and trying to find a player around europe with similar performing stats for that role.

Would love to be able to process in game data and being able to make graphs like that one. As someone who has never touched R (or never really did much coding before) how hard would you say it would be to learn enough to do what you did?

[StatSav]

The deeper look on Letizia will have to wait - I decided to start learning to code radar plots and it is certainly difficult for an R novice. 

Here's a REALLY early, ugly, but working radar. I hope that in a day or two I can get these looking as good as I want to, right now it's genuinely horrible but I thought they were interesting enough to post anyway. The data isn't 100% accurate, I boosted some of Salah's numbers just for more contrast. Full post at some point explaining the radars once I can make good looking ones efficiently.


All variables are calculated as per 90 minutes (other than PassingAccuracy and ShotsonTarget) and presented as percentiles. For example, I set Salah to be in the top 99% of shots on target, so he gets a score of 99 in "ShotsonTarget" on the radar. I'll be making this look much prettier and giving proper titles, stats names etc.

Edited October 16, 2020 by StatSav

[Ö-zil to the Arsenal!]

Wonderful, wonderful post. Thank you for sharing. Really enjoyed reading! 

I'll be following with great interest. Particularly to see if you take this as far as to make signings and build a side.

I would go as far as to say that in-game stats is perhaps the most under-utilised aspect - functional aspect, anyway I am not talking about my social media presence  - of the game for me personally.

I've always found it extremely interesting when players consistently perform beyond their 'ability' level. My judgement on this has been due to observations in the match engine, rather than stats, but I would be interested to see if the stats add up.

During my Benfica save, I used their academy to build up a world class squad but some of my favourite players were actually rated far lower ability and attributes than the others, but fit the system perfectly having grown up with it since 16 years old. Performances of players like Pedro Rodrigues, Luis Pinheiro and Goncalo Oliveira are particularly fond memories.

Playing devils advocate here - how much do you trust FMs reading of a 'key pass'?

Similar to a 'Clear Cut Chance' I have always found the FM definitions of the more subjective statistics a bit spurious. I have never found a way I have felt confident in measuring my playmakers. Particularly deeper playmakers. Passes per 90 is good to show involvement in the game and the percentage completed also. But I have never found anything to measure the quality of those passes. Key passes tends to omit my deeper playmakers, which is typically my preference. Thinking out loud here, I don't actually have a better suggestion 

[StatSav]

On 15/10/2020 at 05:51, Big Yellow said:

Out of curiousity, how does one extract all that data? I have never looked to be honest....

If you click the "FM" button at the top right of the screen, you can "print screen" the page. I printed out the pages as html and then converted them to csv. I typically use it on the player search screen where I can filter players, add the stats I want to look at and so on.

On 15/10/2020 at 05:58, mouli said:

I believe this is how Liverpool signed Robertson from a relegated Hull City for around €7 million. A starter for a relatively low tier team who had determination to succeed. 

I think the vast majority of top sides now have strong analytics departments. Pretty much a career goal of mine to be involved in something like that

On 15/10/2020 at 06:06, ferrarinseb said:

Its really interesting and great work.

I would like to see player vs player comparisions say Letizia and Kimmich in terms of their performances. To see what exactly Kimmich did or what Letizia. 

I'm looking to do more long form content like this to really apply context to the numbers. I think alongside creating radar plots, watching highlights of the players in their matches and so on we could get a good idea of what they're doing well.

On 15/10/2020 at 06:42, Big Yellow said:

lighter background is the way to go!

I'm honestly not so sure... I quite like the idea of the darker background to match the FM aesthetic!

On 15/10/2020 at 09:06, Bot Makel said:

Very interesting work!!

It allows you to understand which players perform well in the game and to find the player you need for your tactics.

It would be interesting to combine different stats to create particular player styles. For example, who in midfield recovers more balls but at the same time is a creative solution.

Creating radar plots for exactly this reason. It would be very interesting to see who the "complete" players are that potentially go missing in the scatterplot, but perform reasonably well across multiple areas instead.

On 15/10/2020 at 10:46, icidamien said:

This is very interesting.

You are proving that we don't need to have the player with the highest CA to have the best performance. 

Having the 4-5 appropriate attributes for the right role on the pitch is also more important than having the player with the highest CA.

Thanks for your work.

I think it'll take a lot more work to get a full understanding of exactly why certain players seem to perform well, but I agree. A good combination of attributes both visible and hidden, personality, consistency, etc will go a long way.

22 hours ago, nugatti said:

I'd like to pitch the idea of doing the same data extract more than once, to see if good performers actually do well over multiple tests/time. Then one could perhaps distill which attributes needed for performance combined with roles in a team? What do you think?

Letizi's performance be a one-off in your game, right?

Also I tend to use the various stats 'per 90' as these are able to use effectively as filters in the player search screen. 

Btw, do you know this guy? https://afmoldtimer.home.blog/category/ac-milan/

He's also using stats heavily for squad building and recruitment

I've already got some ideas for running multiple simulations of a season. I think the data could be incredibly useful for a few reasons that hopefully I can get in to in some later posts! I haven't heard of that person in particular, but I will definitely give it a read having skimmed through!

21 hours ago, davidbarros2 said:

I have nothing of value to add to the discussion here, just stopped here to congratulate the OP for such a wonderfull idea and a very well laid out post.

Looking forward for more!

Thank you - honestly much appreciated. It's my first time really trying this type of thing so I hope that the passion gets me through messing code up time after time lol.

18 hours ago, Sebas said:

Honestly fantastic work. Reminds me of Betis scouting Lo Celso after selling Fabián Ruiz and trying to find a player around europe with similar performing stats for that role.

Would love to be able to process in game data and being able to make graphs like that one. As someone who has never touched R (or never really did much coding before) how hard would you say it would be to learn enough to do what you did?

I'm genuinely a complete novice in R still myself. There's a lot of material to help get you started, as well as guides that there are no shame in "copying" as long as you're changing things, figuring out what they do, etc. Give it a go!

3 hours ago, Ö-zil to the Arsenal! said:

Wonderful, wonderful post. Thank you for sharing. Really enjoyed reading! 

I'll be following with great interest. Particularly to see if you take this as far as to make signings and build a side.

I would go as far as to say that in-game stats is perhaps the most under-utilised aspect - functional aspect, anyway I am not talking about my social media presence  - of the game for me personally.

I've always found it extremely interesting when players consistently perform beyond their 'ability' level. My judgement on this has been due to observations in the match engine, rather than stats, but I would be interested to see if the stats add up.

During my Benfica save, I used their academy to build up a world class squad but some of my favourite players were actually rated far lower ability and attributes than the others, but fit the system perfectly having grown up with it since 16 years old. Performances of players like Pedro Rodrigues, Luis Pinheiro and Goncalo Oliveira are particularly fond memories.

Playing devils advocate here - how much do you trust FMs reading of a 'key pass'?

Similar to a 'Clear Cut Chance' I have always found the FM definitions of the more subjective statistics a bit spurious. I have never found a way I have felt confident in measuring my playmakers. Particularly deeper playmakers. Passes per 90 is good to show involvement in the game and the percentage completed also. But I have never found anything to measure the quality of those passes. Key passes tends to omit my deeper playmakers, which is typically my preference. Thinking out loud here, I don't actually have a better suggestion 

Thank you, I've been a keen reader of your Benfica save(well, saves at this point!) as a guest on the forums myself

In terms of trusting FM for it's statistical outputs... I think it's fair to say there are some issues. As pointed out in a previous post, key passes in specific can include things such as set pieces, throw-ins and what not which unfortunately are not going to be filterable without looking at EVERY key pass for EVERY player. I would however assume that due to the nature of thousands of "key passes" being completed across the data set per simulation, if say 10 or even 100 simulations worth of data was collected then we would get a pretty accurate picture as to what is going on.

Another thing worth noting is that as much as statistics such as a key pass may be unreliable in a sense, it ultimately shouldn't really matter. The purpose of creating scatterplots, radars, etc is to simply find players of note easily. Once that's done (if this were a save) I'd take a deeper look at the player, watch their key passes, chances created, etc and make a decision myself. I think as much as statistics are important in football, they aren't the ultimate consideration overall. Letizia is a great example and I'll demonstrate it in a further post about him at some point, but pretty much none of his creative output in terms of key passes comes from set plays/throw-ins as mentioned prior. I've watched each of his ~60 key passes across the season and the majority are a great demonstration of the value of a high vision attribute, he repeatedly plays balls in to channels and/or over the opposition defence from deep positions. So in the case of starting a save, I'd like to feel that having watched the TYPE of key passes he plays, the system/position/role he plays in for his current side Benevento would essentially help paint the picture as to what to expect from him and perhaps how to deploy him tactically myself. If I'd already created a tactic however that didn't plan on utilising that type of player, I'd probably have a look through some of the other top performers on the scatterplot and see if they fit my ideas better.

Apologies if this was a bit... waffle-y? My sleep pattern is a disaster but I'm excited even thinking about the game this way haha.

[StatSav]

Thought I'd leave a WIP radar plot using some accurate data from the 2019/2020 simulation and explain it a little. Who better to use as an example than Messi and Ronaldo? I chose these 2 as quite frankly they're 1) the most easily recognisable, 2) seem to perform quite averagely on FM compared to real life and 3) have completely different playstyles too. Will the data show this?

 

Well to an extent, yes. First off, I'll explain a few things about how the plot works (or is supposed to work). The radar operates on a linear scale from 0-100 with 1 being the bottom 1% of the data set for a statistic and 99 being the top 1% of the data set for a statistic. I've tried to make that sound as simple as possible, but essentially the closer to the edge, the better a player performs in that category and vice versa. In order to get these scores from 1-100, I first extracted each player within the data set that was "natural" at striker and had played 2000+ minutes in all competitions for a club in the top 5 European leagues. With that done, 169 natural strikers were left in the data set for the 11 categories - note this does not mean the player played all 2000+ minutes at striker, figuring this out would require me to manually check every single player in the data set, including non-natural strikers which just isn't going to happen. Still - we can assume the vast majority of natural strikers played in attacking positions, likely on attacking duties or pushed out wide for top sides with good attacking output nonetheless. With that... ramble out of the way (which I still think was important to mention when understanding the results) I then calculated which percentile all 169 players had reached for each category. I won't go in to much detail here, but essentially a score of 99.7 in goals per 90 minutes would indicate the player is in the top 0.3% of the data set for goals per 90 minutes.

 

All statistics are per 90 minutes other than "PassAccuracy" and "ShotAccuracy" which are an overall % calculated on accurate passes % and shots on target %. Doing so allows us to compare player contribution per 90 minutes rather than the player with the most minutes played over the course of a season having an unfair advantage by nature of them having more opportunity to well... play football and rack up stats. Explaining done - let's have a look to see if there are any conclusions to draw.

"1) They're the most easily recognisable". I don't really have anything to add to this - this was not something I wanted to find out, but more of an accessibility choice.

"2) Seem to perform quite averagely on FM compared to real life". Whilst I still believe it's true that the ridiculous output of Lionel Messi and Cristiano Ronaldo over the years is rare to ever see matched on FM other than in freak cases, both players within THIS data set have performed incredibly well in certain areas. Ronaldo can be seen in the top 1-2% of players for goals and shots per 90 and Messi can be seen in the top 1.8% for both assists and key passes per 90. Both players perform quite admirably across other metrics also, regularly finding themselves in the top 10-20% for multiple statistics.

"3) have completely different playstyles". Whilst I haven't reviewed the matches in any real detail, I think this has to absolutely be considered as a genuine possibility (if not truth) just by looking at the data alone without context. In Ronaldo we can see a player that can be wayward with his shooting and passing, but attempts a significant amount of aerial battles (top 11.8% of all strikers in the data set) and is an outright shooting, goalscoring machine. Messi on the other hand produced elite assists and key passes numbers, whilst being somewhat more accurate in his passing and shooting than Ronaldo. I'd like people to come up with some of their own observations so I won't examine EVERY detail - but I thought that both players placing in the top 25-30% for tackles and interceptions was incredibly interesting. Both are aging, nowhere near as physically prominent as the majority of other strikers in the data set however... they seem to be very astute pressers? Both players also have VERY poor work rate, marking and tackling attributes - so what gives?

If I was to predict - I'd suggest that both players play for the league winner within their division and likely have an AI manager fielding them with a high mentality which causes them to be aggressive defensively off the ball (think counterpressing vs regroup). Anybody that is versed in possession adjusted statistics (PAdj) would also find it interesting that Barcelona and Juventus averaged just 49% and 52% of possession respectively in their league campaigns - more on PAdj another time once I have included it within the model!

Thanks again for reading. I really want to make the radar plots FAR more aesthetically pleasing and functional over time, but I'm now open to any player requests. If there's a single player you'd like to see the output of within the data set (preferably strikers for now), or a comparison matrix like Messi vs Ronaldo then let me know and I'll get it posted asap.

[stevenvdijk]

How do you extract the data from FM?

[StatSav]

4 minutes ago, stevenvdijk said:

How do you extract the data from FM?

If you click the "FM" button at the top right of the screen, you can "print screen" the page. I printed out the pages as html and then converted them to csv.

[stevenvdijk]

3 minutes ago, StatSav said:

If you click the "FM" button at the top right of the screen, you can "print screen" the page. I printed out the pages as html and then converted them to csv.

Cheers!

[StatSav]

How strange that Lyon chose to only start Toko Ekambi in 10 league matches... he still managed to rack up over 2000 minutes played, so it makes you wonder how much of an impact the sample size could have even had on his numbers here!

[StatSav]

27 years old, shoots, scores, presses very well, makes a tonne of aerial challenges, dribbles well, creates chances and plays key passes... starts 10 games for a struggling Lyon side? Very interesting

A very low assist count probably points towards his key passes and chance creation flying under the radar. Seems as though he must lose possession quite a lot with his pass accuracy amongst the very worst in terms of top 5 European league strikers - but maybe that's a by product of him being used as a substitute so often? Used to chase games too often that he HAS to play very risky passes? An interesting situation regardless!

[EnigMattic1]

I think, the beauty of a radar plot is that you can compare your own player to another player in the same position when you are looking to upgrade. Could be handy after achieving promotion or something. 

[StatSav]

2 minutes ago, EnigMattic1 said:

I think, the beauty of a radar plot is that you can compare your own player to another player in the same position when you are looking to upgrade. Could be handy after achieving promotion or something. 

Definitely. I think they're also useful to compare potential targets at a glance quickly that you've picked out as being really good in one area i.e goals per 90 and see what other aspects they have to their game.

[Fatkidscantjump]

Very interesting post, thanks for sharing.

Now one can only hope that in the future, SI would integrate this kind of data based analysis into the game, with the accuracy of it depending on the stats data analysts you hired have.

That sounds very appealing for the tech illiterate that I am

Edited October 16, 2020 by Fatkidscantjump

[Zemahh]

Fantastic stuff! 

@StatSav I assume the only way to make stats relevant (to a degree, since different tactical styles can have a big impact as well) is to have the leagues loaded and simulated in Full Detail, right? I've made a Player Search view filled with all kinds of stats rather than attributes before and trying to find players you need based just on those can be a really fun and much more realistic experience. Another favorite thing of mine to do is set up Screen Flow and track stats from youth leagues from all over the world, so I can keep tabs on best performers. However, I was never really fully confident stats from more obscure leagues were actually relevant, without having them simulated in Full Detail, which can slow down the experience quite a bit a few years into the save.

I'm not fully familiar with it, so I guess my question is, how does FM generate stats for players from non-active leagues? Are they at least partially based on players' attributes or are they completely randomized?

[zyfon5]

14 hours ago, Fatkidscantjump said:

Very interesting post, thanks for sharing.

Now one can only hope that in the future, SI would integrate this kind of data based analysis into the game, with the accuracy of it depending on the stats data analysts you hired have.

That sounds very appealing for the tech illiterate that I am

Agreed on the data analyst part especially with how data analysis is more important than ever in sports now. And you look at in game figuring out what benefits you get with a better data analyst. But I think the better stats data analyst should give you more stats that you can choose from as most data points that you can get rarely deviates too much from one another in real life. So for example, a better data analyst instead of only able to display xG will be able to track individual xG and give you some generic advice from there.

[Deego619]

I have a limited experience with R from my uni days and inspired by this thread, I decided to use it to explore (badly) some insights on my team's performance in the first half of the season.

Background

My Kaiserslautern team are back-to-back Bundesliga champions, being only the 6th highest spenders in the league. This season however, we are being challenged heavily by a resurgent Bayern Munich being led by a superhuman American regen hitting his prime. Current league standing just before the winter break.

Inspired by the infamous Arrigo Sacchi thread, my players are expected to contribute to all phases of play.

Using ggplot, I have created a very simple plot chart to assess my players performance in some key areas, using this formula:

Quote

 

Goalkeepers - Saves/90 v Goals Conceded/90

Best Tacklers - Tackles/90 v Tackle Won %

Best Headers - Headers Won/90 v Header Win %

Ball Winners - Tackles/90 v Interceptions/90

Wing Play - Dribbles Per Game v Assists/90

Playmakers - Key Passes/90 v Chances Created/90

Goalscorers - Goals/90 v Shot On Target %

 

 

Whilst I am still working my way through creating presentable analysis, my current messy work has still provided me interesting insights, especially for Wing Play.

 

It seems that R isn't cooperating with me on accents in player's names but it's still insightful nevertheless.

Pedersen and Rosso are not first teamers so I assume their stats to be a bit skewed. What really stood out to me here was the contrast in creativity between Ihatteren and Vargas, my two starting XI wingers who unsurprisingly lead the way in dribbles per game.

Ruben Vargas

Ruben Vargas is an unassuming winger signed for £5m from relegated Augsburg as cover for both flanks. He also tended to play quite well against me! 

Oliver Batista Meier, a player in a similar mold (as you can see from the chart), was injured early on and Vargas came into the team. Vargas does not take corners or free kicks, and his job is to simply play a traditional winger on the right hand side and use his work rate, determination and aggression to contribute to the team overall.

Ihattaren is his counterpart on the left hand side. He is not a traditional winger, with the idea being that number 10s are quite hard to pull off in a gegenpress system (the number 10 in this system is essentially a deeper pressing forward for DMs that are prevalent in Bundesliga) so his creativity and playmaking skills are better used on the flanks. He also takes corners and the odd free kick when my regular taker Cardozo is off the pitch.

How did they perform in reality?

If you gave Ihattaren Vargas' stats and vice versa, I would have believed you 100%.

Vargas completes 4.97 dribbles per game contributing in 0.26 assists per game. 

Ihattaren completes 4.77 dribbles per games contributing in 0.14 assists per game. 

Vargas' market value is almost a full £70m less than Ihattaren's, yet is performing much better in general wing play, why is this happening?

Vargas' general attributes for playing as a winger are good, 14 for crossing, 15 dribbling, 15 flair, 17 agility as well as some traits which make him unpredictable, combining runs with ball down the right in addition to cuts inside from both flanks.

Ihattaren only has tries killer balls often, but still has good dribbling-based attributes (dribbling, flair, agility, acceleration)

Seeing as my idea originally for playing Ihattaren on the left flank was for creativity purposes, how have they fared in this regard?

Vargas actually creates more chances!

Whilst Ihattaren's metrics are not bad per say, it's again an example like Gaetano Letizia from the OP, of an unremarkable player on the surface being extremely good value being played in the right system and role, whilst a luxury player being played outside of his natural role being less effective.

So does Ihatteren provide any value to my side at all?

Strangely, his output is considerably better as a goalscorer than Vargas. This may be skewed slightly by the odd free kick, but I do know from watching games is that Ihattaren finds himself in the channel taking shots much more than Vargas does, for example this is where he scored his goal against Manchester City in the Champions League:

Considering this, should I play Ihattaren at number 10 for a spell and see how he does there? You can see my current number 10 Juan Jose Macias from the graphs above does not produce much in the way of output, but I'm thinking perhaps a more natural player there may help the cause.

Anyways, thanks for this thread @StatSav , a great way to get even more enamored with this game!

Edited October 28, 2020 by Deego619

[cmason84]

Hi @StatSav - great post. Thanks too to @nugatti for pasting the link to my blog site. That save is now completed, so feel free to dive into the metric analysis I've done on there using Excel and PowerPoint. I'm really looking forward to diving into the xG and presumably xA and other metrics that look set to be in FM21.

@Deego619 - I'd probably keep Ihattaren where he is if he's producing for you. I've found that I can always have one wider player producing the goods and the other one doesn't do quite so well. For me, in the 4-1-2-2-1/4-3-3 DM wide, it was typically the AMR who lit up the metrics with the AML struggling comparatively, which is almost certainly down to the tactic structure.

Edited October 28, 2020 by cmason84

[GianniM]

Interesting, but you're playing both players in a different position. So therefore you can't really compare the two in this way. 

[Toin]

Absolutely loving this! Inspires me for FM21, and I hope the xG measures are done properly for more analysis!

[Sloak]

On 28/10/2020 at 11:38, cmason84 said:

Hi @StatSav - great post. Thanks too to @nugatti for pasting the link to my blog site. That save is now completed, so feel free to dive into the metric analysis I've done on there using Excel and PowerPoint. I'm really looking forward to diving into the xG and presumably xA and other metrics that look set to be in FM21.

@Deego619 - I'd probably keep Ihattaren where he is if he's producing for you. I've found that I can always have one wider player producing the goods and the other one doesn't do quite so well. For me, in the 4-1-2-2-1/4-3-3 DM wide, it was typically the AMR who lit up the metrics with the AML struggling comparatively, which is almost certainly down to the tactic structure.

I’m having this exact same problem, but i chalked it up to footedness. Have you found a solution or is one side always going to struggle

[cmason84]

On 04/11/2020 at 04:06, Sloak said:

I’m having this exact same problem, but i chalked it up to footedness. Have you found a solution or is one side always going to struggle

Hi @Sloak - as I say, I presume it's down to the tactical set up, but I don't pretend to be a tactical nous.

[cmason84]

On 03/11/2020 at 09:08, Toin said:

Absolutely loving this! Inspires me for FM21, and I hope the xG measures are done properly for more analysis!

Looking at the stream, I'm a little undecided, but Miles did say that the xG was going to be harsher in FM than in real life. Not sure on the logic behind the reasoning for that, but that's what he said.

[Kolarov's missing tooth]

Interesting topic. I use R every day for work and am a stats geek so am interested in looking at this myself. Problem; I have no idea where to extract the kind of stats you're using. Is there anywhere they can be accessed? Apologies if someone has already answered, a scan through the thread didn't throw up anything obvious.

[CowShedBarmyArmy]

This is excellent. I've been doing similar in Python

Just started a new job so haven't had much time recently.

[cmason84]

On 07/11/2020 at 11:39, Kolarov's missing tooth said:

Interesting topic. I use R every day for work and am a stats geek so am interested in looking at this myself. Problem; I have no idea where to extract the kind of stats you're using. Is there anywhere they can be accessed? Apologies if someone has already answered, a scan through the thread didn't throw up anything obvious.

@Kolarov's missing tooth if you want the squad data then you need to go to the Chalkboard option on the player squad screen (and add a few more columns in that for whatever reason also aren't included) then press Ctrl + P on a Windows machine. Save it as a webpage and then you can open it in Excel.

If you want the same data for the Player Search Screen, create a search screen view with the same chalkboard settings (and to save yourself time for comparison put them in the same order as your squad screen) and then do the same Ctrl + P thing to extract your data. I would recommend you set a minimum number of minutes played to avoid 'noise' and select positions in the filter to make for easier comparison.

[StatSav]

Bit of a bump on an old thread from this time last year...

As much as this thread never stayed active - the response was great and motivated me to continue getting better at these things. I'm now due to study a masters degree in sports analytics within the next couple of weeks and expect to get even better from that experience! Here's something I created tonight after seeing the "data hub" feature for FM22:

All rankings are per 90 (apart from average rating of course) including data from 1026 outfield players with 2000+ minutes competing in the Sky Bet Championship, Ligue 2, 2. Bundesliga, Serie B or LaLiga 2.



I'd love to see what others have come up with in the past year since posting this. All data is 100% accurate with full match detail using a fully up to date transfer database, so if you would like to see a certain player (if they are within the sample) let me know!

Thanks for the great response last time around guys. As much as I was (and still am) a complete novice, it no doubt helped shape what I have decided to do with further education and so for that I am forever grateful!

[Rashidi]

On 15/10/2020 at 14:22, Pasonen said:

Completed passes and keypasses both include freekicks and throw-ins too? At least completed passes do. This makes those numbers unfortunately a bit unreliable. Anyway great work! And Letizia wow!

Yeah you got to do a bit of internal number crunching with key passes and knock off free kicks, maybe even corners too, cos I remember that being an issue at one point too.

Goal scorers needs drilling down too. SoT per 90 vs xG per 90

[Rashidi]

Got to say I love these kind of analyses.

[StatSav]

Same data set as the previous post - this time for a centre back.


 

This is a great example highlighting some of the nuances associated with scouting via data in Football Manager. Looking solely at the above image, Pulido comes across as a player above average in the air that is below average in terms of the amount of tackles that he wins and the fouls he commits. This however in comparison to other centre backs could not be much further from the truth.



Filtering the data set in to only players that are "natural" at DC, we see an entirely different picture. Before diving any deeper, the important thing to note here is that this data set includes the 266 players that CAN play at DC naturally - this does not mean that they all DID play at DC. With this considered, it is not PERFECT but it is the best that we have from a "macro" standpoint that I can find in Football Manager currently as the majority of this data set will have played their matches at DC. When applying this in-game, you can of course initially refer to the data but then perform more "micro" analysis by checking the player's profile and matches played.

What this does show us is that, actually, Pulido is not a centre back that was aerially dominant throughout the campaign, but did however produce an above-average number of tackles per match, including a very high amount of "key" tackles (as well as fouls!).

So, why bother with doing all of this in the first place?

 

Currently, I have Football Manager set to not display attribute values perfectly (skin available from @sebastian_starttrbts thread!). This for me is a way of playing the game that I find incredibly interesting as it presents new challenges and thus ways to play the game. A few things that we can look at with Jorge Pulido which may explain the statistics that he produced are his bravery, physical attributes and player traits. Pulido is a brave player with a strong positional sense that is not physically dominant - we would expect a player like this to read the game well (high interceptions) and throw his body on the line when required most (key tackles). In terms of traits, Pulido possesses "does not dive into tackles" which the game presents as a decrease in frequency of tackles, which would perhaps explain why despite his aggression and bravery, he ranks in only the top 26.1% for tackles won per 90, despite his ability to perform key tackles.

With all of the above said, there are also many other ways to spin the values that we see and it is important to get a "full" picture when profiling a potential signing. One such way is combining Pulido's traits of "plays short simple passes" and "tries to play way out of trouble" alongside Huesca's division-high average possession of 57% across the campaign as an explanation of how Pulido can rank in the top 1.6% of passes attempted and 0.8% of passes completed across all natural DC's despite his visibly low passing, first touch and vision attributes. Just some food for thought - but this is what I enjoy currently within Football Manager and believe it adds an entirely new dynamic to tactics and transfer strategy.
 

6 hours ago, Rashidi said:

Got to say I love these kind of analyses.

Thanks! I'm a long-term YouTube viewer of yours and associate a lot of my deeper understanding of traits, tactics, etc from your work. Maybe we will see you play the game in a way I have outlined in this post soon?