A Preview of the 2014 Draft Class

[DCEBB2001]

Quote: Originally Posted by DNMHIII

I found this interesting and it appears that maybe your efforts to control OVR is in part achieved by lowering measurables, is that right? I've found the OVR really means nothing to my gameplay experience and the only way to separate elite players from the the others is to just focus on position specific attributes. Personally I like having skill players that stand out because of their positional attributes combined with their physical attributes. Just my thoughts

Your first statement is not correct. The OVR is not simply controlled by the attributes. Sure the attributes are set statically by some criteria, but the OVR is a separate value that also has to be satisfied. Call this the "x-factor" that can raise a player's attribute quality to better satisfy an OVR "target".

The FBG system utilizes the scouting data to show how different players really are from one another. They feel more individualistic and less cookie-cutter. Not every WR, for example is rated as 85+ in all important position-specific attributes. The scouting data paints a very different picture, in fact. It shows that the margin of differentiation is much wider than EA proposes. A simple peak at the source material's distribution proves this.

The only other thing I can do is to recommend that you try them out for yourself. There are several roster-related topics out there with people who are creating these rosters. Check them out and let me know what you think.

[DNMHIII]

Quote: Originally Posted by DCEBB2001

I completely understand. I just cannot release any of the details. Acquiring the scouting data, for example, required 2 separate NDAs just to obtain it and publish derived results. I wouldn't argue that the measurables have been "neutered" as you put it. Instead, all players are rated on a data-driven, realistic scale that utilizes the entire 0-99 range. It isn't like I placed the top of the scale at 90. Players can still be rated a 99 in any category. These instances are just more rare, which allows for more differentiation between the elite and average players. I have always contested that the game plays better with these ratings, which is why I have devoted so much time to it. You don't need my word for it though. There are quite a few forum members who have used them and are more than willing to express their views on it. Doing a forum search for "FBG Ratings" will bring up a myriad of topics with some reviews thrown in there.

I gotcha, If 90 is elite and only a few players will ever get into that level in any attribute you're essentially using a formula lowering attributes across the board and as a result also controlling the the OVR to ensure it stays bellow 90.

Are you also using any subjective form of determining measurable ratings or is it just based on scouting numbers?

[charter04]

These ratings used for rosters are awesome. It sucks that we can't edit draft classes. That's the only reason they don't work as well in CFM. It's amazing how much better the game plays with realistic ratings. I can vouch for it for sure

[DCEBB2001]

Quote: Originally Posted by DNMHIII

I gotcha, If 90 is elite and only a few players will ever get into that level in any attribute you're essentially using a formula lowering attributes across the board and as a result also controlling the the OVR to ensure it stays bellow 90. Are you also using any subjective form of determining measurable ratings or is it just based on scouting numbers?

That isn't necessarily correct either. I am not trying to control anything, really. I just follow the data. The entire thing is data-driven, not Dan-driven.

Let me explain. If the data dictated that say 33% of the top 3000 players (or 1000 players) should be rated above 80, then we would ensure that players were rated that way. However, the scouting data dictates, by examining the data-distribution, that elite players are extremely rare.

Last season, we used an equal-interval method to rating players. That means that the distance between each rating point is the same. The difference between a 70 AGI and a 71 AGI is the same as the difference between a 69 AGI and a 70 AGI; the intervals are equal.

What that did, when we used it across the board for all attributes and ratings, was give us about 7 players rated 90+ in the entire data set of over 70,000 players.

One thing we did for this season was change the method back to the "traditional" way of rating players at FBG. This is the true-mean method where each attribute is set to a standardized mean. For all raw, physical, attributes this is set to 70. The advantages of this methodology is that it is more closely mirrored to EA's system, but isn't as broken (rating inflation).

Look at ACC for example. The average player EA rated last year had an ACC of 81 while other attributes were closer to 70 or 75 (depending on the attribute). That tells me that their method is skewed toward acceleration. However, the scouting data tells us a different story. It tells us that the average rate of acceleration, if set to a true-mean, should be 70. That is an 11-point swing. Now, what we are able to do is compare not just player-to-player for each attribute, but also compare attribute-to-attribute and really see who is better at what.

Utilizing the entire population, we find that NFL players actually are better at obtaining a top-end velocity because the differentiation is smaller than that of their rate of acceleration. EA rated their players exactly the opposite of what the real data tells us! They are saying that NFL players are better at accelerating than they are at maintaining a top-end velocity!

Somebody here is incorrect, as they both cannot be right. With all of the raw physical attributes set to 70, you can now see who is REALLY better at what. Some players, like Chris Johnson are actually better at accelerating than they are at achieving a top-end velocity. Others, like Adrian Peterson, have a better top-end velocity than acceleration. The really cool thing, IMO, is that FBG ratings allows you to realistically compare each player based on the same criteria from the data instead of relying on EA's ability to watch youtube videos.

My point in short - the data dictates the ratings and we allow you to REALLY see how players should be rated.


The scouting data is both objective (combine/pro day data) and subjective (the scout's opinion). All attributes have a relative scale and every player has a grade on that scale. Believe it or not, the scouts that compile this data actually have a grade for some of the mundane things like the catching ability of a Kicker. Scouts watch EVERYTHING from the time you are warming up to workout, to the time you are goofing off throwing the ball around after an official workout. It was quite amazing to find that out. I never imagined that I would have some grades for players that you would think would never utilize these skills in a game. Go back to the FAQ section of our webpage. It explains where we get the data if you need more info.

[DCEBB2001]

Quote: Originally Posted by charter04

These ratings used for rosters are awesome. It sucks that we can't edit draft classes. That's the only reason they don't work as well in CFM. It's amazing how much better the game plays with realistic ratings. I can vouch for it for sure

Thanks man. You are not alone. Maybe I need a testimonials section on the website or something because a lot of people enter skeptics but leave believers.

I am HOPING that EA will allow us to edit draft classes with NCAA Football gone for the foreseeable future. That would REALLY allow us to do some awesome things.

[DNMHIII]

Quote: Originally Posted by DCEBB2001

One thing we did for this season was change the method back to the "traditional" way of rating players at FBG. This is the true-mean method where each attribute is set to a standardized mean. For all raw, physical, attributes this is set to 70. The advantages of this methodology is that it is more closely mirrored to EA's system, but isn't as broken (rating inflation). Look at ACC for example. The average player EA rated last year had an ACC of 81 while other attributes were closer to 70 or 75 (depending on the attribute). That tells me that their method is skewed toward acceleration. However, the scouting data tells us a different story. It tells us that the average rate of acceleration, if set to a true-mean, should be 70. That is an 11-point swing. Now, what we are able to do is compare not just player-to-player for each attribute, but also compare attribute-to-attribute and really see who is better at what. Utilizing the entire population, we find that NFL players actually are better at obtaining a top-end velocity because the differentiation is smaller than that of their rate of acceleration. EA rated their players exactly the opposite of what the real data tells us! They are saying that NFL players are better at accelerating than they are at maintaining a top-end velocity!

I think its important to separate measurables from skill attributes to get a baseline average and it appears that you're trying to do that as well, but I think its a flawed system to begin with if you use the entire population to determine an average as each position should have its own average IMO. Why would you want to have an average for for measurables across the board when each position is vastly different in what physical abilities they obtain? Why should the ACC of offensive lineman have anything to do with determining the average for a wide receiver? If you used hard data to calculate the average ACC of a WR and an offensive lineman or a QB for example you'd get completely different averages. I could go on but Im sure you get the point.

Measuarbles should be completely separate from skill attributes when determining averages and they should also be established within their own position if you want an accurate rating IMO. The next step would be to combine the subjective date to the hard data of measuarbles to establish and OVR that IMO means absolutely nothing relative to gameplay.

Honestly though why is there a need for averages for measurables if you use hard data? Now I'm really confused...lol

[DCEBB2001]

Quote: Originally Posted by DNMHIII

I think its important to separate measurables from skill attributes to get a baseline average and it appears that you're trying to do that as well, but I think its a flawed system to begin with if you use the entire population to determine an average as each position should have its own average IMO. Why would you want to have an average for for measurables across the board when each position is vastly different in what physical abilities they obtain? Why should the ACC of offensive lineman have anything to do with determining the average for a wide receiver? If you used hard data to calculate the average ACC of a WR and an offensive lineman or a QB for example you'd get completely different averages. I could go on but Im sure you get the point. Measuarbles should be completely separate from skill attributes when determining averages and they should also be established within their own position if you want an accurate rating IMO. The next step would be to combine the subjective date to the hard data of measuarbles to establish and OVR that IMO means absolutely nothing relative to gameplay. Honestly though why is there a need for averages for measurables if you use hard data? Now I'm really confused...lol

Measurables for raw physical data are the only ones set at 70 for the true mean for the entire population. This is because the measurables are the same for every player no matter position, size, etc. Each position has their own average as a result of the population data. Not sure how that is flawed if you include the entire population. Now, as you can expect, certain positions will have a higher mean for certain attributes than others. The average ACC for the entire population is 70 with the max of 99 and the min of 1. DBs, however, have a range that is currently from 47 to 99 with 79 being the average for a DB. So as you can see, each position has a different average as a result of using the same scale as everyone else. Another example for QBs would be THP. That range, as the data shows, is 64-96 with the mean of 84. RBs, however, have a range of 8-72 with a mean of 38 whilst utilizing the same data that contrives the THP for QBs.

The point is that so long as you rate everyone on the same scale with the same parameters, you will get differentiation for each sample of your data. Should you sample by position, you will find that some samples have higher means than others and wider ranges. Others will have the opposite. The key thing is that everyone is rated the same and the data dictates the results, not vice versus.

Now, to your example, the ACC of a lineman contributes to the ACC of the entire population the same way that of a WR does. The data shows that, on the average, WRs accelerate better than OL, which seems to make sense if you think about it. They utilize the same scale.

The position-specific attributes (like THP for a QB) is measured based on that position's data set as you say above. Not sure how you thought I did it otherwise. The only thing is that all positions utilize the same population scale of 1-99 instead of being "capped" like EA does when rating players. What sense is it to set a cap if the data doesn't illustrate that caps on skills are necessary?

The OVR is a product of the attribute calculations and the "It-Factor" that can raise or lower your OVR. Consider this to be similar to a "production" rating that enhances or decreases a player's overall value. It is miniscule, however, in comparison to the actual attribute calculations.

Setting an average is useful because it helps us determine what to set the data values to. The data set is in a 0.0-5.0 scale with 0.1 increments. We could do it linear and equal-interval like last year where you set 5 = 99 and 0.0 = 0, but you will get players rated in the 50s as starters. Instead, by setting the average score equal to a number that makes sense for the position, we get something that allows us to work "better" within Madden's in-game "box". If I developed a game after the data, I would have 5.0 being the highest a player could score on any attribute and 0.0 being the lowest and using an equal-interval. But, since there is not another game out there and Madden uses a 0-99 scale, we have to make our data best fit that limiting scale.

I know is sounds complicated and confusing, but I can assure you that I sought professional consultation on this project. Dr. Craig Larner, formerly of Texas A&M University, helped set up the modeling for the data using his professional standards in statistical analysis and data interpolation. The bottom line is you can try it out and see what you think. If you like it, that's cool...welcome! If not, well that's cool, too. The only reason I continue to do this whole project is because I want to see how Madden players really should be rated. The fact that it offers better game-play is simply a byproduct, IMHO. The current game is crap, IMO, and if this helps polish the turd that is Madden, then I guess it is a subtle step in the right direction.

[DNMHIII]

My point is why is there any need for an average for measurables? If you rated players based on hard data theres no need to create an average baseline correct?