wazzaflow10

I can't continue this conversation in good conscience. You have no idea what you're talking about. There is no amount of telling you that you're incorrect that you'll accept.

Because you're missing key components of how attributes work together on top of having an incredibly small sample size. First the gap between 20 and 18 is relatively small in the game world. Secondly passing again isn't just passing, technique, and vision. That determines the players ability to spot and hit the pass on a technical level in a vacuum. If a player has low composure they wont do well if they are being closed down or under pressure. If their decisions are low they might pick the wrong time or option for a high risk pass. If they are one footed and forced onto their weaker foot it won't be as successful. If they have low flair they might not try to curl a ball around a defender. If they have low anticipation they might not know a passing opportunity is occurring in a few ticks of game time. Then we get to what their teammates are capable of and what the defenders are capable of. The game isn't as simple as I made the "best" technical passer in the world why isn't he creating 100 assists per game?

You've changed the inputs into the game to such a degree that you're bound to get weird results. I don't know how else to explain this more simply. Its not about 20's or 1's or 10's. Its that you've gone outside the realm of what the game has been trained on. The game has likely never been tested with the inputs you've given it. Its not to say that nothing can be changed but you've gone from 0 to 100 without documenting anything in between. You have you just don't realize you did. Passing isn't just the passing attribute. Its a combination of attributes including footedness, flair, vision, composure, anticipation, decisions, technique. Keeping things "fixed" doesn't make it a control group. Your control group would be where nothing is changed from the base game. You'd have to run many, many trials to figure out the expected range of results. There's a lot of RNG components that go into a single match you can't control for anyway as an end user. To overcome that it requires a significant amount of trials to find the limits of the game as they occur naturally. Its a lot of the reason these tests are somewhat pointless as an end user. You're trying to reverse engineer the game but have no mechanism to keep things in a truly stable state. Anyway, once a natural limit is discovered, then you can go about experimenting the effects of changing a single attribute and measure their effects on results compared to the control groups. Eventually once you've documented incremental changes, then you can go about doing groups of changes to see how they interact and affect the engine. There's numerous ways to do it depending on how complex you wish to test the effects. A weird result would be if a small change completely knocks the game off balance rather than just slightly nudges it in an expected way. If it takes getting to changing lots of inputs to something that wouldn't occur naturally then all that you've shown is the match engine hasn't been trained to handle those inputs and more or less broke the game. Its fine that the engine has been broken because a certain set of inputs caused unexpected results but its not a valid result to say the entire engine is broken for all inputs.

A lot has been addressed already namely by santy and ninecloudnine. To reiterate some of their stance: Your "control" variables aren't realistic. These inputs don't occur within the game naturally so it isn't something that would happen except in maybe very, very extreme cases. There certainly wouldn't be a team filled with this sort of distribution of attributes, let alone two. In short it's not an internally valid model. What I mean by that is by running an experiment that falls outside of something that has been trained you're inevitably going to find something "weird" because the model has never seen anything like this before. As a concrete example suppose I'm growing tulips. I want to measure how tall they grow based on watering schedule, sunlight exposure, soil conditions etc. After a few trials I have an idea of what affects height. If I all of a sudden start planting petunias instead, none of my data for tulips is relevant. There may be general principles like all plants need water, sunlight, soil but you can't really claim to know anything about how a petunia would grow. FM is the same way. The match engine has been trained and tuned on a particular set of data that exists. Just because you can change an input doesn't mean you should. That's not a control variable. FM experiments need to exist more or less naturally inside the game world. Putting in extreme values is going to yield an extreme result. You'd have to make a pretty compelling case as to why this initial state of inputs should exist. Additionally, I have no idea what your tables are telling me in conjunction with the text. Is there some sort of regression? Is it just descriptive statistics - if so is there any association at the raw match level? Which group is which? Are the value statistically significantly different? There's probably more I could address but I really don't have an idea what you're trying to show to direct you further here. Its just very unclear other than you've shown me a table and claim something is weird.

I know what they're asking. I would have said asking if I didn't understand the question. Its a poorly conducted experiment hence why I said I don't know what they're trying to prove.

I've read this post 5 times and still can't figure out what they're trying to prove.

All great points from both of you, @perpetua and @NineCloudNine. I'll certainly take your word for this as I'm obviously not privy to what goes on behind the scenes on how a player is constructed. I wonder if it could be something done programmatically that can functionally separate the top from the middle from the bottom that doesn't require a change but clearly I'm out of my depth here. I suppose in some regards research could be easier because there's not the need to languish if an average player should be 11 or 12 since the gap between those would be relatively small comparatively. More effort would be around discussions about the upper end and lower end of the scale where it should, theoretically be easier to say yes world class or barely professional. I find this point below to be constructive in thinking about this challenge conceptually if we can just slightly open the can of worms. This suggests to me (perhaps incorrectly) that world class players are better all around rather than truly excellent in the same aspects as their real life counterpart. So hypothetically lets say Luka Modric is the best passer in FM (which we know includes attributes of passing, technique, flair, vision, composure etc) but because those skills are capped at 20 to get the target CA he needs points elsewhere rather than make him really stand out in his best attributes. So the effect I think is that Luka Modric is more well rounded in FM than he is in real life (not that he isn't good elsewhere but just in comparison). A thought I'm having as I type this out (so apologies of a half baked idea) is that players are created/modeled based on archetypes. What I mean by this is the archetype dictates the cost of CA for attributes either in addition or in place of the position familiarity. A deep lying playmaker would have an easier time adding to attributes that govern passing but finds it more difficult/expensive to add to certain physical attributes. Likewise an anchorman will find it easier to add CA to defensive attributes than more attacking attributes. As a side effect - this could help the AI train players more effectively since it would be able to view the archetype and want to focus training to a players strengths. I don't mean for everyone to be able to reach 20. The high CA cost for reaching world class should still exist. I do want to prevent a lot of world class players from becoming jacks of all trades to compensate for the lack of room at the top to differentiate their greatness. Its not to say that all around players can't or shouldn't exist either.

yes sorry i meant team. You're right.

This has been in existence forever. Stars are relative to your league.

I don't know that's why I was asking the head researcher if that's how the game views attributes or not under the hood. Well there's the range of top level professionals (say attributes 12-20) but there's not enough separation between levels of say a League One team or a Serie C team. Outside of physicals - which should not be bounded by the league - there is likely not enough of a difference in technical or mental attributes to really show the difference that truly exists. If we could view say Regionalliga data and below I'm curious what the bottom end of pace would look like even getting down into semi-pro leagues.

The separation of the top few to the squishy middle to the bottom few pretty much exists in all facets of life. That's why its easy to name a few of the best in anything but hard to rank once you get past the first 5 to 10 or so. the game is meant to be a simulation of real life. Assigning values according to real life wouldn't be arbitrary. SI doesn't have to give the exact formula for it to be transparent. The general concept that attributes follow a similar pattern to the graph when being interpreted by the match engine is sufficient . Functionally too, players aren't defined by one attribute, so its not like you'll break Mbappe by giving him a 19 instead of 20 in pace while keeping all else equal. Its that there should be a clear separation between world class abilities compared to players that exists even just below that tier.

There seems to be at the user level not enough separation between attributes to really show that the world class players are truly head and shoulders above your average top division players. To the user the difference between a 19 and 20 is only one "point" (or 10 using the 1-200 scale) but is that how the match engine interprets that? In other words is each step in attribute interpreted as a linear increase - such that for each tenth of a point in pace is an increase in running speed of .01 km/h? Using the research about pace from above, the difference between the fastest player and the 10th fastest player in the Bundesliga is .45 km/h. On average that's .05 km/h per rank. When we get to the bottom end of the scale the difference between the the 250th and 300th fastest player is also .45. However there's 50 players between those two ranks which means its on average a .009 increase per rank. Does the game reflect that the difference in pace per rank .05 km/h between the top player (Alphonso Davis) and the 10th ranked player (Jeremie Frimpong) is larger than the difference per rank .45 km/h between Frimpong and Kilian Fischer? If Davis is assigned a pace of 20 and Frimpong is assigned a pace of 19 is the gap between 20 and 19 reflective of this difference? Perhaps it is better to show this phenomenon graphically: Also note the decreasing at an increasing rate at the bottom end of the scale - slow players should be getting slower relative to the next jump in attribute. In other words someone assigned a pace of 1 there should be a huge difference between them and someone with a pace of 2. In my opinion all attributes should be reflective of this sort model.

Excellent points all around. Well said. SI face a real challenge threading the needle between the CPU "figures" out your tactics the moment you set them up and the current state. It'd be nice to see lower level managers struggle to make necessary changes or take much longer while the Pep's/Klopp's/Ancelotti's of the world are fairly adaptable without breaking their core principles. It'd make moving up the pyramid and managing a top team a real challenge.

Has nothing to do with beginning of the save. There is less of a problem if you are a lower level team because the players that are good enough are plentiful. If you're looking for a top player at a big club we've been told they must be interested in joining before they even get scouted. That leads to zero results in recruitment focuses. Worse than that is when you look in your scouted players pool no one shows up either. The game weirdly ignores your CA/PA filters when setting up a focus and decides who is good enough for you before being scouted. It would be a much better system if the game actually 1) used the filters you applied and 2) used the full scale of letter grades for recommendations. If a player isn't interested put it in the near matches with a lower grade. Player interest should only affect the recommendation, not that actual scouting process.

The OP reminds me a certain poster who I believe got banned around the time this account was opened. Would hate to see the mods look into it further...