On Cohesion.

[Ben E Lou]

I've gotten some PMs about my comments in another thread, and I thought I'd address the issue here. If you know what you're doing, it only takes about 5 minutes to replicate what I did, so have at it if you'd like. I just re-ran it on the latest version of FOF. Here are the steps:

1. I created an fdt file using a csv with the following characteristics:

Every player in the league has the same experience, year drafted, % developed (100), and ratings.
Players on ONE team (Atlanta) modified with their "on team" year as 1995.
Players on the 31 other teams modified so that their "on team" year is 2010.2. I started the league using the fdt created above with the following settings:

Alternate player file: YES
Random player file: NO
Full X-Factor: NO (wanted players to be as closely rated as is possible)
Personality/chemistry: NO (wanted to eliminate any "noise" from that)
Hiring/Firing: doesn't matter
Level Playing Field: YES (in case there's any noise there)3. Simmed the league with injuries off and me controlling a team in the AFC that doesn't play Atlanta, just to make sure no bias there. Everything on the human team was handled by AI.

OBSERVATION 1: Atlanta has 100 cohesion in every category. Every other team has right around 30 with a tiny bit of fluctuation here and there, I suspect due to a few AI signings of players.)

OBSERVATION 2: All rosters are virtually identical, as expected.

I simmed the season 10 times. Atlanta's worst record was 11-4-1, best was 15-1-0. The highest record any other team had was 13-3-0, and that only happened once. I wrote down the 12-win teams. No team other than Atlanta had more than one 12-win season, and in a few spot checks, I saw teams that won 12 games on run trial winning 5 on the next.

CONCLUSION: Cohesion absolutely affects team performance in FOF2K7, patch 6.4.

NOTE: Because the results were so overwhelming and consistent with what I was already virtually certain of, I didn't bother wasting my time calculating average wins for each team, or re-running the same test with one crap-cohesion team. If anyone still thinks I'm wrong here, then feel free to replicate the testing above.

How much should cohesion affect team performance? As with anything that can't be quantified in real life, we could argue that point until the cows come home and reach no definitive conclusion. All I wanted to do was see if it affects team performance, and it most certainly does.

[Alf]

Thanks for the experimentation.

[Dave Hansen]

Thanks for your work.

After reading some recent postings, I'm glad to see that cohesion does impact play and is not broken. It's a nice feature that adds depth to the overall experience and has a strategic impact on the best laid plans of mice and men.

[RedKingGold]

Thanks for your work.

After reading some recent postings, I'm glad to see that cohesion does impact play and is not broken. It's a nice feature that adds depth to the overall experience and has a strategic impact on the best laid plans of mice and men.

Exactly. This is why I reacted with a bit of vitriol towards Hammer in the other thread.

There are people newer to the game who could read Hammers claims as facts because he presents them as such. There could very well be some validity towards cohesion not being tuned perfectly, however Hammer's form testing and his conclusions posted were misguided at best and dubious at worst.

Hammer is entitled to his opinion, but when he goes out of his way to jump in in on threads which are not even really explaining our testing cohesion to dump his theory, it gets a bit annoying.

[Hammer]

I have to say that is great news. Very different results from the last time you did this then Ben (maxed cohesion team never hit 12 wins in 6 seasons). Your method is excellent for illustrating cohesion does has an effect. I thought your original work tied in nicely with my findings that there was a small cohesion effect.

I would suggest cohesion is a relatively minor factor to bear in mind, it shouldn’t stop you doing what you want with your team.

Its important to note that Bens method is not real world in that your never going to get all 15 year vets on one team and all rookies on the other teams. My study was more what you could expect to see in a real league. Hence, me finding a small effect and him finding a solid but consistent effect fitted well with my quote above in my opinion. Cohesion was there, but not scratching the paint work in a real MP environment.

Now, in 6.4, we see a much larger effect. Similar method as your last study. The high cohesion team never had 1 12-4 season last time. Now they are between 11-4-1 and 15-1. Very different results.

I wouldn't like to interpret whether this is the right amount of effect or not, but it sure the hell is better. I will duplicate this, aiming to see a 100 cohesion team versus 60 rated cohesion teams. A more real, game cohesion scenario, than 100 versus 30.

[Ben E Lou]

(maxed cohesion team never hit 12 wins in 6 seasons)No. Same thing, basically. It was so obvious I didn't bother to write down results, but I think they didn't win less than 12 games in any of my 5 or 6 trial seasons.

[MIJB#19]

With 15 years oif playing together, that team better should steamroll all over the entire league. Good to see proven that it works that way.

[Hammer]

Hmm, perhaps I misunderstand people more than what I realize. Reading again what Ethan said, his message read cohesion didn't change for the 6.3 patch. I read into it that it was broke and always had been. But a communication today details that this wasn't in fact his intended message.

Ethan wrote:
The use of cohesion was not changed in any way for 6.3. The same applies to 6.4 - cohesion has not been touched probably since the
first post-EA version. Its effect may not be as strong as some would like.
The effect may be stronger than others would like. But it's been working the
same way, exactly, for close to a decade now.

I would be surprised if there was anyone out there who actually thought the strength of cohesion was to high, but your mileage may vary.

Looking back on the work I have done, I also got significant results for win-loss totals on a number of occasions. However, I dismissed them as noise as stats never backed them up. When testing, I looked at categories separately (passing, running, defense against the run/pass). For example, I got little, actually almost no difference when looking at passing statistics (QB rating, avg per attempt for example). But there was significance between the win-loss column. Go figure. Maybe it wasn’t noise at all, but why the difference never came across in stats I have no idea.

[Koprnkc]

Is this figured with say all of your O-Linemen or just the starting 5?

Is there any difference in overall cohesion when you insert a new starter?

Or does the game take into account that your LT,LG, and C have been playing side by side for 8 seasons, and only the RT is new?

Would there be a cohesion hit if you signed a free agent G and T from the same team, and inserted them into your starting 5? Exp: I sign a new RG and RT from the Jets, they played the right side of the Jets line for 5 years next to each other, they would have a obvious repore with each other, and thus, work well together.

Do Veteran lineman inserted into the starting 5 have less of a drop then say rookie or 2nd year lineman? I would think a veteran C can assimulate into the O-Line faster then a Rookie would.

Or are these examples beyond a game engine that is gong on 7/8 years old.

[Ben E Lou]

Is this figured with say all of your O-Linemen or just the starting 5?Cohesion is calculated using the players on the field for the play in question, I believe.

Is there any difference in overall cohesion when you insert a new starter? See above.

Or does the game take into account that your LT,LG, and C have been playing side by side for 8 seasons, and only the RT is new? It takes every player into account, I'm fairly sure.

Would there be a cohesion hit if you signed a free agent G and T from the same team, and inserted them into your starting 5? Exp: I sign a new RG and RT from the Jets, they played the right side of the Jets line for 5 years next to each other, they would have a obvious repore with each other, and thus, work well together. Cohesion is calculated based on time with the current team only. That's why I isolated the data field that I did in testing.

Do Veteran lineman inserted into the starting 5 have less of a drop then say rookie or 2nd year lineman? I would think a veteran C can assimulate into the O-Line faster then a Rookie would. No difference in vets or rookies.

[AlexB]

I have to say that is great news. Very different results from the last time you did this then Ben (maxed cohesion team never hit 12 wins in 6 seasons). Your method is excellent for illustrating cohesion does has an effect. I thought your original work tied in nicely with my findings that there was a small cohesion effect.



Its important to note that Bens method is not real world in that your never going to get all 15 year vets on one team and all rookies on the other teams. My study was more what you could expect to see in a real league. Hence, me finding a small effect and him finding a solid but consistent effect fitted well with my quote above in my opinion. Cohesion was there, but not scratching the paint work in a real MP environment.

Now, in 6.4, we see a much larger effect. Similar method as your last study. The high cohesion team never had 1 12-4 season last time. Now they are between 11-4-1 and 15-1. Very different results.

I wouldn't like to interpret whether this is the right amount of effect or not, but it sure the hell is better. I will duplicate this, aiming to see a 100 cohesion team versus 60 rated cohesion teams. A more real, game cohesion scenario, than 100 versus 30.

Warning. Vehicle Reversing.
Warning. Vehicle Reversing....

[MIJB#19]

For a re-test, I think 5 years together versus completely fresh is where you're going to have an interesting figure. And by that I suppose a 5 years difference (like 5 years vs 10 years) would be an even more realistic figure.
Hence, 2005 vs 2010 and then 2000 vs 2005 would give us a signal on whether that's the same or not.

[Hammer]

Plus it might be worth increasing the talent level, see if that does anything. Would high talent make cohesion more likely to be lost in a bigger equation?

[MalcPow]

For a re-test, I think 5 years together versus completely fresh is where you're going to have an interesting figure. And by that I suppose a 5 years difference (like 5 years vs 10 years) would be an even more realistic figure.
Hence, 2005 vs 2010 and then 2000 vs 2005 would give us a signal on whether that's the same or not.

5 years is basically the same. The test team has 100 cohesion and the rest of the league is around 50. The resulting win-loss advantage is basically equivalent. The cohesive team consistently wins close to 12 games in what I've seen.

[Firefly]

FWIW, this discussion got me thinking and a couple of weeks ago I decided to do my own testing. I took a team from an SP career that wasn't very old (cause it's what I had). The team happened to have excellent cohesion. Then I proceeded to release and resign every player I possibly could and saved the game as a different career. Those players became my new starters in both teams.

The result was that I had two teams with the exact same roster and depth chart, but with different cohesion. I then simmed 8 seasons with each and kept some key stats: QB rating, avg per carry, pass rush% for offense and defense, and pass defense %.

The results were stunning. The team with the high cohesion did better in every area, and the biggest difference was in the area which had the highest difference in cohesion.

[Hammer]

I have replicated Ben's study, I did the 5 year cohesion difference, plus also played with talent levels. I have to admit my previous studies were a pile of bollocks and the poor testing methods obviously skewed the results. I got significant results in all cases. As Ethan said cohesion hasn't changed at all in years, I can't blame it on the patch either.

[MalcPow]

Cool, good stuff guys

[Hammer]

Does anyone have an opinion as to whether chemistry or cohesion has a bigger impact. Tough question I know, but which one would you be more concerned with?

[Ben E Lou]

Testing chemistry is a bit more work because the csv file doesn't allow for changing personality ratings. However, you could find the default team with the highest personality rating leaders and rig it up to have very good chemistry and see what happens.

[Ben E Lou]

DOLA:

Or you could set up the birthdays to make it easy to build a monster chemistry team than go with a pref draft so that the human-controlled team is first a great-chemistry team and then a chemistry disaster.

(And of course in both cases you'd want the entire league to have the exact same talent levels.)

[MalcPow]

I tried to mess with that a little. The cohesion effect jumps out and is obvious. Chemistry, at least as far as having a good number of affinities while the rest of the league has none, does not seem to have as much obvious impact.

As mentioned before, monster cohesion advantage seems capable of taking you from 8-8 to close to 12-4 or something. Clear chemistry advantage didn't offer results that were as easy to observe, and since it's kind of a pain to set up, that's enough of a conclusion for me. I did not attempt to put together a raging conflict team though. Maybe that's where the sparks fly.

[MalcPow]

For anyone looking to mess around, easiest way was to build a roster file with everyone having the same birthday and giving all the guys in the FA pool part of the file five more years of experience and a birthday with an affinity (or conflict if you go that way) match to the rest of the league's birthday. Then just fire up a season and go sign your new leaders from the FA pool.

[MIJB#19]

For anyone looking to mess around, easiest way was to build a roster file with everyone having the same birthday and giving all the guys in the FA pool part of the file five more years of experience and a birthday with an affinity (or conflict if you go that way) match to the rest of the league's birthday. Then just fire up a season and go sign your new leaders from the FA pool.Not to downgrade the idea, I think it's almost the best possible set up, but I can see one flaw in the eventual results: the AI could do the same thing and 'accidentally' also hire a couple of chemistry bumping players.

[MalcPow]

I definitely would have made the same assumption, but with no injuries the AI signs very few additional guys. And for better or worse, the handful of guys they do sign are the randomly generated additional rookie fillers that show up in any new start even if you're loading a roster file.

Part of the annoyance factor is the relatively low number of guys that show up with high leadership and high personality for each position group anyway. So the AI signing a new leader possibility ends up being an almost total anomaly and really doesn't affect things that as far as I could see.

[Wanderer]

DOLA:

Or you could set up the birthdays to make it easy to build a monster chemistry team than go with a pref draft so that the human-controlled team is first a great-chemistry team and then a chemistry disaster.

(And of course in both cases you'd want the entire league to have the exact same talent levels.)Oddly enough, I started doing this on Friday.

Results still pending, but there is something weird about the chemistry stuff - there's a cap. If I set my 3 QBs as Leos, and my position leaders all as one of the sympatico groups and all the other players as the 3rd, everyone should have affinities to the degree that their personality strength dictates.

They don't. I use a set player file, and the number of affinities varies dramatically from game start to game start. Part of it was an aborted attempt because who the position leader is in each game start shifted around a bit.

My solution was to have all my QBs Leos, then alternate between the two sympatico groups for all other players. Every leader should have affinities with every position leader. But they maxed out at 3. In every group, half (give or take 1) of the players should have affinity with their leader. This varied wildly, but I only saw the expected number once out of a ton of trials (just quick game starts).

Same is true for conflicts, but the caps are much lower. For conflicts I just alternated every player including QBs between the two conflicting signs.

Which leads me to the reason this is taking much longer than I anticipated. There is so much variation in both the number and strength of affinities and conflicts that I needed to note the variation, so I created new columns marking the number of mild, standard, strong, and extreme affinities/conflicts for each QB and position group for each trial. That's 6x4x30x2 manual entries - 1440 for those of you scoring at home - with some scrolling and counting for each one. Thinking it over now, I probably would only have to do the QB ones manually - I could just export to txt and use a count function/parser...but whatever.

I also did "nochem" seasons, with everyone's birthday the same. These worked perfectly - there was no chemistry in either direction in all trials.

As far as a trial, I didn't set everyone's ratings equal across the board, I just used (well, am using) sample sizes large enough to filter out noise: 30 full seasons for Affinity, Conflict, and NoChem. I probably should have upped it to 50 just because of all the enormous variations in so many factors, but 30 is considered the magic number, so eff it. It's just a game.

I'm not just tracking wins and losses, but all stats in the Team Summary (via direct import) as well as ALY and KRB%, because I thought the OL might reflect chemistry moreso than any other position grouping.

I'll post results and the raw data in the next couple of days.

If someone wants to send me 3 player files - conflict, affinity, and nochem - as I've described with all the ratings equalized, I'll even be really nice and do the whole thing a second time with those. :)



Also, hi everyone! I'm a professional data analyst and algorithm development specialist, a consultant hired by companies to do things like figure out relationships between consumer demographics, purchase patterns, pricing, and brand loyalty to target couponing in real time. This game is fun, and now that I'm playing multiplayer, it's challenging enough that I'm going to do some semi-serious research on it as time allows.

If I can figure out a decent way to get the data out of the game without individually print-to-txt-filing a given screen, I could do some real damage. As it is, I'm setting up some templates to do some...of the stuff I do. Blahbedee blah blah. *everyone falls asleep*

One thing I know I'm going to do is run some numbers to get a cluster of statistics that have the closest correlation with win loss totals, and see if that answer varies between playoff wins and regular season wins. That one will be very tricky.

All of this is very messy because there are only two ways to test this stuff: 1) Save a game at the start of a season and reload and reload and reload. In that case you don't get a lot of the variations that you're looking for.
2) Keep starting new games. But in this case you're introducing way too much variation - the 2 most important IMO being the X-factors, even if you don't use the full X-Factor, and one I haven't seen discussed at all - coaches. With each restart the coaches are randomly generated.

That's another trial to do, IMO: Set all players equal, set everyone nochem, same cohesion, etc. and keep doing restarts. With each restart, mark down the ranks (1-6) in each category for every OC/DC/HC for every team and simulate the season. Wash, rinse, repeat x30. Munch data to find effects. This will be no fun, because there is no way short of someone writing another stelmack type program to get the data out automatically. But I would be interested to know how much it actually matters.

Blah, blah. I talk too much.

[MalcPow]

Sounds great I think. I'm not sure if I understand the problems you were having with affinity "caps" as I've never seen any limitations there.

I would definitely recommend creating roster files with talent equalized (literally just set everything at 500 for everyone), and I'd also recommend setting the test league as an MP league so you can easily trade all the highest personality guys onto your test roster to maximize affinity/conflict. If you're up for doing the things you're talking about, these steps are well worth the time. Once you've set a particular test roster to your liking in an MP test environment like this, you can just swap the roster to another team in a reload to capture the effect of different coaches. It will absolutely still require a lot of clicking, but there'd be less searching for guys and you'd control for the specific players as much as possible.

Happy hunting. Interested to see what you find.

[Wanderer]

Sounds great I think. I'm not sure if I understand the problems you were having with affinity "caps" as I've never seen any limitations there.Alright, let me try to describe it a little better:

Set my QBs to 8/1
Set all other players alternating between 9/1 and 12/1 to account for affinity groups.

And for some reason I'm getting a db error when trying to attach screen shots, but, I just started a new game with the affinity players list:
- 1 of the QBs has affinities with 3 leaders (should be 5, but always maxed at 3)
- None of the other QBs have any affinities at all (they all should, all have same BDs as the 1st one.
- As one positional example, my OL leader is a 9/1 birthday, and the OL is all 12/1 or 9/1. Of the 5 OLmen with 12/1 birthdays, only 2 of them have affinity with the leader.

Another new game with the conflicts players list, where everyone is an 8/1 or a 1/1, alternating. Total of 3 base, 3 mild, and 2 strong conflicts.

They have to be either capped or determined on a probability basis once the birthdays match. Looks like about 60ish% chance for affinity once birthdays match, and maybe 30% of conflict? I haven't measured it out. It also might be that the first one is 100% chance, but it declines as more and more are added in. So if your roster is set up so that everyone would have affinity with one another, you might end up with maybe 30 actual affinities. I don't know. I just know that there is some sort of a limitation on it, because a lot of times there wasn't affinity when there should have been.

In my first few trials with the affinity roster I got 18, 20, 16, and 21 affinities total.

[Jughead Spock]

FOF only 'counts' (at least apparently) the offensive affinities with QBs, so that explains the max 3 - RB, REC, OL. It *says* so-and-so will be happy with a compatible def. leader, but it only ever indicates based on the offensive leaders.

The only explanation I can think of for the other affinities not showing would be personality strength - that determines level of affinity. If two players both have really low personality ratings, no affinity (reported, at least). Otherwise, makes no sense a'tall.

[cuervo72]

Personality strength changes would have been my guess too.

[gstelmack]

If I can figure out a decent way to get the data out of the game without individually print-to-txt-filing a given screen, I could do some real damage. As it is, I'm setting up some templates to do some...of the stuff I do. Blahbedee blah blah. *everyone falls asleep*

You called? What specific data do you need? Have you checked out my UtilitySuite? Extractor can grab all the player cards and dump to CSV (like the affinities) and I suspect is your most likely tool, while Interrogator or DBUpdater can grab all the stats stuff.

[aston217]

Cohesion is calculated based on time with the current team only. That's why I isolated the data field that I did in testing.


I think this should be clarified a bit - cohesion is calculated based on the "On Team" year. As Firefly suggested earlier, a player you've had on your team for five years, cut at some point then re-sign later in the season, seems to be treated for all intents and purposes the same way as some rookie you just picked up. (edit - oops, did not realize you said as much in another thread).

On affinities -

What determines how strong a player's affinity is with another? Random factors? I checked a roster of mine and it seems it's not strictly correlated to personality strength.

[Firefly]

What determines how strong a player's affinity is with another? Random factors? I checked a roster of mine and it seems it's not strictly correlated to personality strength.

I was going to say personality strength before I realized it was a trick question :)

[Ben E Lou]

Yes, it's the "On team" field. I should have been more specific.

On chemistry, I've never seen any indication that the strength is based on anything besides personality strength. What's making you say otherwise?

[MalcPow]

If it will help clarify anything, it's a combination of both the leader's personality strength and the player with an affinity match's personality strength. The higher they both are, the better. But I'm pretty sure those are the only two things that matter...

[aston217]

Ah - I was both wondering if it is more dependent on the Leader's strength than the guy who has the affinity with him, and if there was a measure of stochasticity involved.

Just something I see on my OSFL team now: DT, 62 personality, has 'strong affinity' with an 89 personality ILB defensive front leader. However, there's an average strength (44) DE who has exceptional affinity with the same guy. So is this randomness in the mechanism for determining affinity strength, or inaccurate numbers I'm seeing on the player cards for personality?

[QuikSand]

Weird, that would be the first I heard of that twist. Have a screenshot or something to share?

[Firefly]

Me too. Very odd.

[Rando]

Here are the players in question;

Leader

2261

Lower Personality Exceptional Affinity

2262

Higher Personality Strong Affinity

2263

Damndest thing, I've never seen that before either. Thought maybe it could have something to do with McDonald being Unhappy so I loaded up an earlier sim before he started getting all bitchy but no dice, still just a strong affinity.

[MalcPow]

Hmm, McDonald is smarter than Rushe, so he loves with less passion and more irony. His affections can only be so strong...

I got nothing, that's truly strange.

[Firefly]

...there's way too much information to decode the Matrix. You get used to it. I don't even see the code. All I see is blonde, brunette, red-head.

I can see the description of the screenshot, but not the screenshot.

[cuervo72]

I wonder if for signs A, B, & C a leader from A might have more affinity for B than C, in general (though to balance, leader from B might have more affinity to C than A, etc?).

Dunno. Not something I've noticed before either, but then I've taken the system someone for granted for a while.

[aston217]

Thanks for the assist rando! (you oughtta update the in-game graphics though :P)

I think cuervo's theory is interesting. This isn't something I've ever investigated to any serious degree, but could be something to keep an eye on in the future.