What is "average"?

[Big FN Deal]

Quote: Originally Posted by DCEBB2001

For your 40 yard dash example, the range I have is 4.21 to 6.33 (a player actually ran that slow this year at a pro day). The average of ALL players going back to 1998 who ran was 4.81. My question is more in regards to what do you set that "average" number of 4.81 to. 70? 50? Kind of like how ATSteelers pointed out that 70 is a nice number for "average" ability for these measurables. Your thoughts? Now for the others what I am working with is a scale of 5 to 0 with .1 incriments for the scouting data. So for instance, a WR is evaluated on his "mundane use of the hands" or his CTH rating as I estimate. The highest I have seen of any player graded is 3.7. The lowest is 0.1. So I figured that I could set 3.7 or 3.8 at 99/100 for the upper bound and 0.1 at a lower bound. The average is 2.1 for all WRs. So if you use EAs high/low/avg you set 3.7 at 99, 0.1 at 55 and 2.1 at 79. Graph it to a nice parabolic curve and viola! You have a way to determine the CTH rating for any WR. Now do that for 20+ attributes for each position and you get an idea. The problem arises in where to set the high/low bounds and average numbers at. Also keep in mind we want to avoid ratings inflation and spread them out. THAT is the tough part! Do you just pick a random number that looks right and go from there? Thoughts?

In reference to the bold if 4.81 is the average 40 time then wouldn't the average speed rating come out to be around 69, using the 1-99 scale I posted earlier? That's what I meant by I am thinking converting the verifiable data into the 1-99 scale would yield the average versus trying to predetermine it.

I think I get that you are saying what would be average talent but I am saying that would the average sum be the best number to set as average talent, if that all makes sense, lol

With regard to the skill rating, what you use sounds solid considering whatever the heck EA/Tiburon's system is but I wonder why they only have it going to 55. Reason being is see that as unnecessarily limiting the range of those ratings, especially when every player has every rating. So the lowest route running rating even an OL can have is a 55? I just don't get the skill ratings system in Madden so it makes it difficult for me to articulate suggestions for improvement, as it is currently formatted.

[KBLover]

Quote: Originally Posted by DCEBB2001

*snipped for length* Your thoughts?

Ignore OVR?


OVR is just a BS mismash of the ratings anyway. Like you mentioned with the HB rating. I mean you listed the ones that actually impact OVR. Where's PBS/F? We know protection is an area where a HB can differentiate himself and earn a roster spot even if he doesn't touch the ball much - or it can turn a 2-down back into a 3-down back and free up a roster spot.

So already, I think OVR is not taking in all the aspects a HB needs.

Average rating per position should be whatever your verifiable data should be to start with and whatever that makes the OVR - so be it. After all, it's just relative and I assume you'll be making a full roster? So it will be relative to your roster.

But if you want to be concerned with OVR, perhaps chopping 10 points off everything would work. Make the max for most "normal" players 90. Only the truly greats/exceptional talents/once-in-a-generation type players can break 90.

OOTP ranks players like that. Scale for most players was 1-100, but the internal scale is 1 to 150 (at least). Most players, even the stars, fit in 1-100, but super elite freaks of nature type players can "break" the scale and get higher ratings.

Perhaps your ratings can work similarly? That would lower the inputs to OVR in general, maybe getting closer to the kind of scaling you want to see there.

I think 70 is a good point for average NFL grade. 50-60 would developmental and worth hanging around, but you don't usually want them to start, while 40 is likely an also-ran, 80 is good, 90 is elite/star material. 91+ once-in-a-generation ability.

[DCEBB2001]

Quote: Originally Posted by Big FN Deal

In reference to the bold if 4.81 is the average 40 time then wouldn't the average speed rating come out to be around 69, using the 1-99 scale I posted earlier? That's what I meant by I am thinking converting the verifiable data into the 1-99 scale would yield the average versus trying to predetermine it. I think I get that you are saying what would be average talent but I am saying that would the average sum be the best number to set as average talent, if that all makes sense, lol With regard to the skill rating, what you use sounds solid considering whatever the heck EA/Tiburon's system is but I wonder why they only have it going to 55. Reason being is see that as unnecessarily limiting the range of those ratings, especially when every player has every rating. So the lowest route running rating even an OL can have is a 55? I just don't get the skill ratings system in Madden so it makes it difficult for me to articulate suggestions for improvement, as it is currently formatted.

So basically you are saying use a linear regression for the measurables? So 6.27-9.15 for the 3 cone would place 50 at 7.71 and 75 would be at 6.99s. Do that for every attribute? How would we account for the skew of the data though? For the 3 cone, the range is 6.27-9.15 with the average being 7.30, obviously skewed toward the 6.27 with the tail of the data being toward the 9.15 end.

Allow me to clarify. The range according to EA for CTH FOR WR'S ONLY goes from 99 to 55 with the average being 79. The scouting data obviously does not evaluate an OT's CTH ability, but only the abilities needed to evaluate a player at a given position.

[DCEBB2001]

Quote: Originally Posted by KBLover

Ignore OVR? OVR is just a BS mismash of the ratings anyway. Like you mentioned with the HB rating. I mean you listed the ones that actually impact OVR. Where's PBS/F? We know protection is an area where a HB can differentiate himself and earn a roster spot even if he doesn't touch the ball much - or it can turn a 2-down back into a 3-down back and free up a roster spot. So already, I think OVR is not taking in all the aspects a HB needs. Average rating per position should be whatever your verifiable data should be to start with and whatever that makes the OVR - so be it. After all, it's just relative and I assume you'll be making a full roster? So it will be relative to your roster. But if you want to be concerned with OVR, perhaps chopping 10 points off everything would work. Make the max for most "normal" players 90. Only the truly greats/exceptional talents/once-in-a-generation type players can break 90. OOTP ranks players like that. Scale for most players was 1-100, but the internal scale is 1 to 150 (at least). Most players, even the stars, fit in 1-100, but super elite freaks of nature type players can "break" the scale and get higher ratings. Perhaps your ratings can work similarly? That would lower the inputs to OVR in general, maybe getting closer to the kind of scaling you want to see there. I think 70 is a good point for average NFL grade. 50-60 would developmental and worth hanging around, but you don't usually want them to start, while 40 is likely an also-ran, 80 is good, 90 is elite/star material. 91+ once-in-a-generation ability.

And I agree with you that 70 is a good "average point" for OVRs and the raw attributes. Due to all of the players I have data for and the database size, there are obviously going to be far more guys with 40 OVR (the low bound I pre-determined) because for every 2500+ players who get evaluated each spring for the NFL draft/training camps only a relative few will make it. OVR ratings for prospects are greatly skewed more toward 40 than they are toward 70+ as you can imagine.

With that established, the big question now is how do you scale the position-specific, non-measurable, ratings for each position? For WRs these are the CTH, CAR, PBK, RBK, RET, TRK, ELU, BCV, SFA, SPM, JKM, CIT, SPC, RTE, and REL ratings. I have loads of data for all of these attributes to be translated to Madden ratings. Now it is just a matter of determining the upper/lower bounds and the averages to account for the skew.

Like the example provided for before with CTH for WRs, say the scale for the scouting data goes from 5.0-0.0 with 0.1 incriments. The highest in the NFL so far is 3.7 and the lowest is 0.1. The average is 2.1. What do you set these at? 3.7=99? 5.0=99? 0.0=0? 0.0=50 because any WR below 50 for CTH shouldn't be a WR (IMO)? 2.1=70? 2.1=80?

These are the things that are sooo hard to determine! What do you set the bounds and averages at?

[Big FN Deal]

Quote: Originally Posted by DCEBB2001

So basically you are saying use a linear regression for the measurables? So 6.27-9.15 for the 3 cone would place 50 at 7.71 and 75 would be at 6.99s. Do that for every attribute? How would we account for the skew of the data though? For the 3 cone, the range is 6.27-9.15 with the average being 7.30, obviously skewed toward the 6.27 with the tail of the data being toward the 9.15 end. Allow me to clarify. The range according to EA for CTH FOR WR'S ONLY goes from 99 to 55 with the average being 79. The scouting data obviously does not evaluate an OT's CTH ability, but only the abilities needed to evaluate a player at a given position.

Using a 1-99 scale again for the 3 cone I would have 6.27 99- 9.15 1. There is a 2.88 difference between them so I would divide that into 10, for the ten groupings of ratings from 1-99 ( 90-99, 80-89, 70-79, etc). That would be .28 for each group using tenths that comes to =/- .028 or just round up to +/- .03 between each rating. Which would put 7.71 at 57/.028 or 61/.03.

I see it as letting the chips fall where they may because in the end, everything will be relative in-game. I am presuming the 3 cone drill would be determining the physical rating, agility, in this instance.

[KBLover]

Quote: Originally Posted by DCEBB2001

So basically you are saying use a linear regression for the measurables? So 6.27-9.15 for the 3 cone would place 50 at 7.71 and 75 would be at 6.99s. Do that for every attribute? How would we account for the skew of the data though? For the 3 cone, the range is 6.27-9.15 with the average being 7.30, obviously skewed toward the 6.27 with the tail of the data being toward the 9.15 end.

Well, you could set the average NFL ability at the mid point of the most common data range. So if most guys are between, say, 6.6 and 6.8, then you could put average at 6.7 and scale from there with 6.27 being 100 and 9.15 being 1. It would take more effort to be above average while easier to fall off. The rating would accelerate rapidly in response as the times fell below 6.7 while the down side of the curve would be more gradual, but easier to separate yourself ("I'm not great, but I'm way better than Mr. 8.5 over there")

I would do this by taking the difference between the top and my set average and the bottom and my set average.

So 6.7 - 6.27 = 0.43

If I say 70 is average, then that's 30 grades above average (71-100). 0.43 / 30 = .0143s per point above average. So a 6.5 time would get +13.9 points (from 70, since that's the "bottom" of this range), or a 83 (or 84 if you round instead of drop remainder) rating. A 6.3 would get +27.9 points or an 97 (or 98 if you round) rating.

There's 49 grades below average (1-49) so 9.15 - 6.7 = 2.45 / 49 = .05s per point under 50. So Mr. 8.5 would get +13 points (from 1 since that's the "bottom" of this range) for a 14. A 7.3 would get +37 points for a 38.

That would account for the "real" average (i.e. where most NFL times cluster while still scaling for any rating along the range.

Would that help with the scenario you present?

[DCEBB2001]

Quote: Originally Posted by Big FN Deal

Using a 1-99 scale again for the 3 cone I would have 6.27 99- 9.15 1. There is a 2.88 difference between them so I would divide that into 10, for the ten groupings of ratings from 1-99 ( 90-99, 80-89, 70-79, etc). That would be .28 for each group using tenths that comes to =/- .028 or just round up to +/- .03 between each rating. Which would put 7.71 at 57/.028 or 61/.03. I see it as letting the chips fall where they may because in the end, everything will be relative in-game. I am presuming the 3 cone drill would be determining the physical rating, agility, in this instance.

OK so 6.27 = 99. 9.15 = 1.

9.15-6.27 = 2.88

2.88/2 = 1.44

6.27+1.44 = 7.71 = 50 (half way between)

.288 = 1/10th. So a 90 would be 6.27+.288 = 6.56.

Got it. The only issue I see is skew of the distribution. If 7.71 is the MEDIAN (half way between 6.27 and 9.15) and the MEAN is 7.30, then the distribution is skewed toward the 6.27 end of the distribution.

If we were to set the MEAN to 70, then 7.30 = 70 AGI.

If we were to set the MEDIAN to 50, then the average NFL player (who on average recorded a 7.30 cone drill time) would have an AGI in the low 60s. Also, a player with an ELITE AGI rating (90 and above) would have to run a 6.56 or better (6.27+.29).



Meanwhile if the same method is applied to SPD, using the MEDIAN...

6.33-4.21 = 2.12

2.12/2 = 1.06

4.21 + 1.06 = 5.27 = 50

1/10th = .21

90 SPD = 4.42

The MEAN SPD is 4.81. Using the MEDIAN, the average 40 would result in a SPD in the low 70s. Pretty darn close.


For ACC, the 10yd split's range is 1.40 to 2.12.

2.12-1.40 = .72

.72/2 = .36

MEDIAN = 1.76

MEAN = 1.66

90 ACC = 1.47 10 yard time.

MEAN of 1.66 = ACC in the low 70s.


Pretty interesting results. The "let the chips fall as they may" approach is interesting. The only issue I have is the skew issue, but it doesn't yield results that are too far off aside from the AGI. Guess it shows you how RARE great agility is.

Thoughts?

[DCEBB2001]

Quote: Originally Posted by KBLover

Well, you could set the average NFL ability at the mid point of the most common data range. So if most guys are between, say, 6.6 and 6.8, then you could put average at 6.7 and scale from there with 6.27 being 100 and 9.15 being 1. It would take more effort to be above average while easier to fall off. The rating would accelerate rapidly in response as the times fell below 6.7 while the down side of the curve would be more gradual, but easier to separate yourself ("I'm not great, but I'm way better than Mr. 8.5 over there") I would do this by taking the difference between the top and my set average and the bottom and my set average. So 6.7 - 6.27 = 0.43 If I say 70 is average, then that's 30 grades above average (71-100). 0.43 / 30 = .0143s per point above average. So a 6.5 time would get +13.9 points (from 70, since that's the "bottom" of this range), or a 83 (or 84 if you round instead of drop remainder) rating. A 6.3 would get +27.9 points or an 97 (or 98 if you round) rating. There's 49 grades below average (1-49) so 9.15 - 6.7 = 2.45 / 49 = .05s per point under 50. So Mr. 8.5 would get +13 points (from 1 since that's the "bottom" of this range) for a 14. A 7.3 would get +37 points for a 38. That would account for the "real" average (i.e. where most NFL times cluster while still scaling for any rating along the range. Would that help with the scenario you present?

It may. The data dictates that most players lie along the average time (7.3) for the 3-cone, if you do a histogram of the data that is. For the 40, most players actually lie around 4.6, which is WAY below the average of 4.81. Keep in mind that there are far more CB/WR types tested each year because they make up higher percentages of a roster compared to other positions. This would most likely account for the skewing.