The Truth of FM Attributes and Testing | FM24
Вставка
- Опубліковано 19 лют 2024
- Plenty of oppinions are going around about FM attributes, their usefulness and the use of tactical testing in FM. Here are my thoughts on the matter.
Join the discord!
/ discord
Streaming on Twitch and UA-cam:
/ stybb
How the league was Built:
• How To Edit the Perfec...
Check out the rest of the FM24 series: • Football Manager 24
CORNER TACTIC ON STEAM HERE:
steamcommunity.com/sharedfile... - Ігри
Great video man, I’m new to fm and it’s so damn hard haha, good to have someone explaining things moderately understandable 😂
Thanks! FM is a lot better nowadays at guiding people through all its pages but it is still a lot to take in. Lots of good UA-camrs with guides here though.
There is a reddit post from I think FM21 where a guy won the Premier league with 1 CA players by focusing a few specific attributes. Imagine masses of people found out about that one :)
I've seen somewhere (forgive my poor wording) that mentals will tell you where and when you need to be, physicals will get you there, and only then is there a question if you can technically do what you need to do or not. Physicals can also compensate a lot of lacking mentals. And another thing I've seen and am not 100% right now, physicals are not affected by consistency.
One more thing, I think people have very wrong view of what a value of 1 for an attribute means. 1 finishing doesn't mean he sends it to 65th row every time he shoots. The scale is tailored for footballers. I for example don't have 1 first touch, on this scale I'd probably be at like -15 or something.
Nah, set a stat to 1 and see how a player behaves. It's pure shit. Benny Hill show type of comedy. 1 in a stat is terrible.
I think Zealand mentions somewhere that his players with 1 in shooting can still hit the target etc. Interesting point you mention about consistency, do you have a link to anything on that? Put me down as a -20 for composure.
-50 work rate here :)
Another great info video. Keeping it real on da streets!
Thanks!
Great video, insightful and enjoyable 👍🏻
Glad you liked it!
I think that was very nice put, great job.
Thanks for the kind comment
Thanks for the sanity
Cheers!
Great! Nice explanation data based. Thanks
Cheers for the feedback :)
Everyone should give this a thumbs up to help more FM players see this video
Very good video! I've seen the reddit posts you mentioned (there are two main ones, one deleted and one still on the sub). I absolutely agree with your conclusions here, but a couple caveats. I understand that you would need tons of testing. The posts really started because of some FMArena posts that were done over 2400 matches I think. Wouldn't that classify as enough?
Secondly don't you think it is pretty bad that after running so many tests yourself, passing seemed to have a negligible effect at best? That's in line with the FMArena tests as well. Your Target Man tests serve as more proof that something is clearly wrong with the match engine.
Howdy good sir.
I'll take your own number of 2400 matches, although I don't know if it's the correct one either.
Depending on the impact of the variable "passing" over the results of the match, 2.4k sampe size could be considered negligible. I am running RIGHT NOW a regression (same process he's running to construct the tendency lines, in essence) where my own sample size is giving me all kinds of headache. And my sample is nearly 25k data points. The effect of the variable I'm interested just isn't big (or clear) enough to provide conclusions.
Secondly and by faaaaaaaar the most important thing is the concept of collinearity.
To put it way too oversimplified (to the point it's bordering innacuracy) collinearity is when two things explain the same effect but at the same time, one explains the other.
For example: We can agree "points per season" can be explained by passing, pace, composure .... all of them with their own particular "weight".
But here's the catch (and it's a big one, and I believe, the core of the question):
Assume passing 15. We could say that's "good enough", so the player succeeds in most passing attempts.
Now imagine this player just played a through ball to a forward putting him one-on-one with the GK.
If the Forward had low pace, the pass would not complete. Posession would be lost.
If the Forward has enough pace, the pass DOES complete PLUS he scores. Therefore points are gained.
Which of these two variables explains the "points gained in this match"? Is it PASS or PACE?
Was it PASS or Vision?
Was it PACE or Off The Ball ?
That's the jist of the concept of Collinearity. Where two variables explain the same phenomenon while also having an effect on each other.
There are ways to try to treat and isolate collinearity, which is a whole rabbit whole in itself. Long story short, it's the Holy Grail of Statistics, though.
Cheers!
It's a good point you make on the 2400 matches (if it is that, I know its big). Without seeing the distribution of those 2400 matches, it is impossible to know by eye if there is statistical significance. Ideally, if they tested multiple values of passing you could see whether there is an identifiable trend without doing the statistical analysis.
An interesting post, thanks. I feel the challenge with all of this is generating the match data. Do you do it manually or do you have a way of automating it? @@abhuyllongfellow4233
@@StybbGaming STATISTICS:
The current understanding on how to deal with collinearity is to make use of something called an Instrumental Variable.
An IV would be would something that closely represents the variable you're interested in, while not BEING that variable.
It's a whole field of study by itself. I'm an economist, so I can use an economics example to try to explain the concept.
Jacob Mincer wanted to explain the behavior of wages.
He found out that wages could be explained (amongst others) by Age and Experience.
However, as you age, you tend to gain more experience in your current field, thus bumping your salaries once more. So Age has a "double" effect. it explains wages AND explains "Experience", which in turn, also explains wages. (It also bypasses "how to even measure experience in the first place?")
The solution he found was to square the age of the individual as a proxy variable (an Instrumental Variable) to represent Experience in his equation. And it worked perfectly.
Nobel prizes are awarded to the "creation" of a working IV, so I wouldn't even begin to attemp to invent one for FM lol
HOW I'D RUN AN EXPERIMENT:
Were I to try to attempt to figure the importance of a single variable in FM, I would try to run several experiments, though.
My example above illustrates the correlation between PASS and PACE, for example. So I'd try to make an experiment for PASS by switching PASS values, while also keeping all relevant variables (on all players) as favorable as possible.
Thinking of evaluating PASS for an AMC:
Lets say PASS 10-20, while keeping vision, off the ball, pace, strength and all other relevant variables for everyone at 18, for example. In order to make it favorable. In this case, if the pass failed, it's definitly not the AF's fault, it's the AMC.
Then compare the results (based on XG, not points or goals) and judge an AMC based of XG averages over a couple seasons (we're not vying for the Nobel Prize, we're talking FM! 2 seasons should be more than enough to grasp the tendecy).
Posession lost could be a good indicator as well for midfielders. Passes completed. Offsides.
This would allow me to exacerbate the importance of PASS by providing it with the best possible environment for it to succeed. PROBABLY generating a clearer result. Which I could, then, average to try to squeeze some of the RNG out of it.
For Defenders, Headers won. Fouls comitted (good defending = less fouls). Tackles won. XG Against. Cards received.
For Forwards, non penalty XG, Conversion rate, Fouls received, Corners pro.
AUTOMATION:
As for automating the process.... not my expertise.
BUT, maybe create a database of 17 year-olds with Future Sales deals for 20 years in the future. Embargo the team they're on so they can't hire anyone.
Set their stats as you wish, their CA doesn't matter. Freeze them all. Injury Prone = 1. Have backups for each position just in case. Freeze the opponents at 10 while you're at it?
From there, shouldn't be too hard to code a Python to scrape the monitor, click the editor button, select "Attribute" and set it to X+1, starting at X=10 until X=20. And then go on vacation for 2 years. Pretty much a Macro with a variable X. Run overnight.
IDK, literally my first idea on the subject.
Thanks for coming to my TEDTalk, btw
Don't physical attributes contribute more to CA than mentals? So, the game basically confirms that physicals are more important than mentals. Doesn't mean that good mentals are useless.
They all contribute to varying degrees. In know Pace is quite important, things like Finishing less so. That gives a rough idea of how important Attributes are with respect to one another.
Quick question, possibly just a data vis issue, but why does it look like passing 9,10,11 & 12 are all absent from the graphs?
The pace demonstration is in line with the forum posts squirrel plays used as the basis for his python program.
Great video, if true we need to pressure SI for an explanation.
Thanks. I think he works out the coefficients for each stats importance based off of FM arena if I remember rightly? Even with the example of the target forward above, I think it's a long discussion about what stats are "best" for each position.
I would imagine hidden attributes play a big part(although injury proneness seems to be ignored). Plus right foot / left foot values will have an impact. Maybe reputation?
My big issue with virtually all these tests are that I imagine the way the game is coded all attributes work together. Very rare you’ll get someone with 1 jumping reach and 20 heading. Or 5 pace and 15 acceleration.
In addition no one I know plays FM by picking a team then holidaying until the end of the season. You would need to actually play matches. And all of them. Never holiday.
I have seen someone say the game pays more attention to Current Ability rather than where the points are allocated. Who knows?
Interesting topic. I like doing little editor experiments myself. I don’t think we will ever get an answer.
The coding I imagine is too complex with too many variables.
for the algo. great video!
Very kind
First of all, thank you for the video and your efforts.
I don't understand English very well, so I use a translation programme. I watched the video for about 5-10 minutes, but the graphics confused me a bit.
So, according to what you have analysed and concluded in the video, is the statement ‘META FEATURES CHANGE THE GAME’ shared on Reddit wrong for you? I'm thinking of deleting the game just for this reason, if the physical features don't affect the game to a high degree, I shouldn't delete the game. Right?
There is a lot more than just the physical, mental and technical attributes when combined with the hidden attributes. I've found players who are professional for example play better than those who are fickle for example even if their attributes aren't quite as good.
It also depends on the role. An enganche (hopefully spelt right) can get away with poor physicals as he largely sits there playing people in. His mental and technical attributes in this case are more important.
I see your point, and this is something I'd love to teest - though I might not have the time for every position and role. Just as a note, all the hidden attributes on the players are set to 13 in all tests so we can keep it controlled.
Glad to hear the controls in place @StybbGaming. In the vanilla game the random way players respond to scenarios due to the combination of attributes is what makes the game great. The physical attributes do give an unrealist boost to performance though I think.
Out of curiosity, I assume you have, but have you also controlled for hidden attributes?
Good question, should have mentioned that all hidden attributes are set at 13.
Great video, soo while I agree that passing probably does matter to some extend...that this attribute is increadibly weak... looking at tests at FM arena and even yours the difference is very very small... If you would change physical attribute it wouldnt instantly show an increase..
To the point where if u can choose between 2 pace and 8 passing you probably better of taking pace almost if not every role..
I've always assumed attributes combine to have certain effects. I.e. decisions plus vision, plus passing plus composure equals the ability to successfully pull off a pass.
Messing about with one attribute alone seems meaningless as there is also Random Number Generation plus other factors. I.e. is the game a big match putting pressure on players composure etc.
How about hidden attributes?
I totally agree, testing the effect of multiple attributes that combine together would be really interesting and insightful
They definitely combine. I think this is mentioned in FM itself. Way back, I signed a CM (for FM10) with 20 Long Shots and 15 Flair (13 Finishing ) as I wanted more goals from midfield in what was at the time a mid-table EPL side. Now, the guy went on to have a long, excellent career, helping to drive my side into Champions League contention on the way to becoming a club Legend; the new stadium we built was even named after him!
One thing he never did, however was score a lot of goals - probably averaged around 4 per season. He'd score only a couple from range, on average and some of those would be from direct FKs (he had an Attribute of about 15). Open-play bangers from 20-25 yards? Rare, though he did attempt them plenty enough. He invariably played either as an Attacking or Defensive Playmaker, depending on who else was playing alongside him. I'd always up the slider to encourage him to take either a moderate amount, or lots of long shots.
What's the reasoning for this, then? Hard to say, but his Technique wasn't amazing - around 13, so whilst in theory the quality of his long-range shooting might be very high, the chances of him getting it right technically, when in a very strong league, wasn't great. Throw in his Agility and Off the Ball were mediocre (about 12), maybe meaning he didn't get too many perfect opportunities to hit it from range.
All speculative, but my take away wasn't that Long Shooting was unimportant, but that it only factored in when a sequence of other events had been ticked off, in order.
Messing with one single attribute seems to be a little too simplistic for me.
I do think attributes aren't the be all and end all, however.
We all have had good players "on paper" that we can't get a tune out of.
And players that despite their attributes not being fantastic seem to show up, time and time again.
Hidden attributes definitely play a role I think
I always considered that tactics dont matter as you can setup Man City to something absolutely horrid and there is still a real chance they can get top2 in the league
great videos but why do you always mispronounce attributes its ATTRIButes
Haha, this is fun to read - because for me this just says "why do you misspronounce ATTRRIButes its ATTRIButes". Literally have no idea of any other pronunciation.
.made tons of tests on my own AND watched the games...attributes dont matter besides of 3 you need not to loose most if the games...I stopped playing FM...it really destroyed me.
EDIT : After testing much more I need to admit I was wrong. shame on me. ATTRIBUTES DO MATTER. But maybe the people do not understand what those attributes really are saying .. its all about % and statistics...1 ist not bad at all AND 20 is not soooo much more than you would think of...AND its all about attributes working together. Really shame on me...I WAS WRONG.
man, you just validated what the rumor is/was. pace and probably acceleration and other physicals matter more. Test is needed but those posting those reddits were also doing tests. FM arena also had over 2,000 tests. No one other than a caveman IQ guy is claiming something works with a season or so. Everyone acknowledges you need representative data points. So what's your -additional- comment here other than agreeing with them?
The way you pronounce "attributes" bothers me.
Another poster commented the same. I have no idea how else it's pronounced.
Haha . You say “a tribute “ when it’s more like “at ribute “ . Make sense ? Sorry to join in but comment caught my eye .
These tests are useless really. As with any computer program - Unrealistic input = unrealistic output. If you max out physical attributes and minimize all others, you'll never get a realistic result as the game isn't coded that way. Attributes on their own don't tell you that much anyway, it's how the attributes work together, that's how the game works. I've no idea why people waste so much time on this, just play the game and enjoy it.