FBGRatings Seeks to Recalibrate Madden Ratings, Change Game

[DCEBB2001]

Quote: Originally Posted by ggsimmonds

Interesting stuff, thank you. I imagine due to encompassing all levels that standard deviation is quite large. Could you perhaps provide a ballpark figure for average OVR for NFL players and the SD? I am wondering how wide the gap is between players. I would guess that it is lower than one would think. This way you don't have to provide anything regarding specific players. Also in your opinion what position allows for the greatest variance in terms of overall grade and could that be an argument for it being most important? I hope it is not QB but if it is what is second? Do you ever do some just for the hell of it models? You know, just play with the data?

I am constantly playing with the data (that sounds kinda dirty). The average score for an NFL player on a roster right now is 477. The standard deviation is 100. The widest range is 925 by the WR/TE position group. Keep in mind that injuries play a role in this, so if someone is beat up pretty bad, it will lower his score and increase the range. The K/P range is pretty tight with a range of about 300. That means that the Kicker and Punter competition is pretty close. QBs came in at about 630 for a range.

[ggsimmonds]

Quote: Originally Posted by DCEBB2001

I am constantly playing with the data (that sounds kinda dirty). The average score for an NFL player on a roster right now is 477. The standard deviation is 100. The widest range is 925 by the WR/TE position group. Keep in mind that injuries play a role in this, so if someone is beat up pretty bad, it will lower his score and increase the range. The K/P range is pretty tight with a range of about 300. That means that the Kicker and Punter competition is pretty close. QBs came in at about 630 for a range.

I bet. Time series data going back what, 15 soon to be 16 years, with thousands of observations? That could lead to many late night "I wonder..." sessions.

Regarding the range, it could be due to a few injuries or something that creates a few ornery outliers? Without those, would you suspect that the WR/TE group still has the largest variability?

[michiganfan8620]

Quote: Originally Posted by DCEBB2001

Every player and their situations are different. Mathis is still graded out at 2.8 in his category correlating to the FMV attribute. Jordan is still graded at 1.4 in his. The suspension, injury, or whatever case happens to any player simply modifies the final value by a multiplier. Once he is designated good to go for a return, the multiplier will be adjusted. Keep in mind that we also have to work within the framework of EA's OVR formulas and can't just rate players in an equal-interval method without an OVR value to meet. Last time I did that, people were pissed because guys like Wes Welker were coming in at 75 instead of 90 or whatever, WHICH I CAN SEE as being an issue for contracts, trade values, etc.

That still does not make sense. Both players are in the same situation. Facing the same 4 game suspension for failing a PED test. Mathis is the better player. How come Jordan is a 1.4 yet gets a 76 FMV yet Mathis gets a 2.8 and gets a 63? Where is the logic in that? And once again, you see you rate on ability, not production. A suspension effects a player's production, not his ability. Therefore, lowering the rating of a suspended player is lowering it due to his production, which you say you don't do.

[DCEBB2001]

Quote: Originally Posted by ggsimmonds

I bet. Time series data going back what, 15 soon to be 16 years, with thousands of observations? That could lead to many late night "I wonder..." sessions. Regarding the range, it could be due to a few injuries or something that creates a few ornery outliers? Without those, would you suspect that the WR/TE group still has the largest variability?

Well, I would define a statistical outlier as anything that lies beyond 3 SDs of the mean, so if you throw those out, it tightens it back up on the low bound for sure. If you do the entire population with an average of 477, that means that an outlier is 177 and below or 777 and above, of which there are only a few players (Manning, Rodgers, Brady, etc). It seems to me that the WR/TE group still has the widest variance. I would have to do each group individually, but that would be my guess considering that Calvin Johnson is way up there and there are some WRs on rosters who come in at 2.00 without injuries. But if you think of it, that is the largest group, too. Most teams usually carry more WR/TEs than any other position group, so there is an opportunity for a wider variance as well.

[DCEBB2001]

Quote: Originally Posted by michiganfan8620

That still does not make sense. Both players are in the same situation. Facing the same 4 game suspension for failing a PED test. Mathis is the better player. How come Jordan is a 1.4 yet gets a 76 FMV yet Mathis gets a 2.8 and gets a 63? Where is the logic in that? And once again, you see you rate on ability, not production. A suspension effects a player's production, not his ability. Therefore, lowering the rating of a suspended player is lowering it due to his production, which you say you don't do.

I am growing pretty tired of addressing you on this topic. What you fail to realize is that the OVR multiplier is lowering Mathis' FMV rating to match the desired OVR. When the OVR goes back up, so will his FMV rating, which will most likely be closer to his original FMV grade of 91. Until that happens, though, I cannot change the rating back. That is not how this works. We have to meet the individual attribute values and the OVR values. If his OVR drops, so too must his attributes.

If you have further questions about this particular topic, I will refer you to the FAQ page.

[ggsimmonds]

Quote: Originally Posted by DCEBB2001

Well, I would define a statistical outlier as anything that lies beyond 3 SDs of the mean, so if you throw those out, it tightens it back up on the low bound for sure. If you do the entire population with an average of 477, that means that an outlier is 177 and below or 777 and above, of which there are only a few players (Manning, Rodgers, Brady, etc). It seems to me that the WR/TE group still has the widest variance. I would have to do each group individually, but that would be my guess considering that Calvin Johnson is way up there and there are some WRs on rosters who come in at 2.00 without injuries. But if you think of it, that is the largest group, too. Most teams usually carry more WR/TEs than any other position group, so there is an opportunity for a wider variance as well.

Good points. Thank you for taking the time to answer my questions.
One last question though, going back to the earliest year you have data on, have you noticed any trends develop? Besides perhaps mobility in quarterbacks.

[DCEBB2001]

Quote: Originally Posted by ggsimmonds

Good points. Thank you for taking the time to answer my questions. One last question though, going back to the earliest year you have data on, have you noticed any trends develop? Besides perhaps mobility in quarterbacks.

Sure have! QBs are a TON faster. The average 40 time for a QB in 1997 was 4.96. In 2014, it was 4.87. And we have 5 times the QBs in the database in 2014 than we do for 1997. That is a HUGE jump.

The average 40 time for a draft-eligible player in 1997 was 4.84
In 2014, the average was also 4.84. So, on a whole, players aren't getting any faster, but some positions are.

In 2014, the average OL had a STR grade of 53.6. In 1997, that number was 56.4. In Madden, that is like going from a STR rating of 83 in 1997 down to 81 in 2014. Maybe it is the crackdown on PEDs?

[ggsimmonds]

Quote: Originally Posted by DCEBB2001

Sure have! QBs are a TON faster. The average 40 time for a QB in 1997 was 4.96. In 2014, it was 4.87. And we have 5 times the QBs in the database in 2014 than we do for 1997. That is a HUGE jump. The average 40 time for a draft-eligible player in 1997 was 4.84 In 2014, the average was also 4.84. So, on a whole, players aren't getting any faster, but some positions are. In 2014, the average OL had a STR grade of 53.6. In 1997, that number was 56.4. In Madden, that is like going from a STR rating of 83 in 1997 down to 81 in 2014. Maybe it is the crackdown on PEDs?

Maybe more of an emphasis on an up temp passing game? Could be the increase in popularity of spread based offenses. Probably a combination of a few factors (of which PEDs may be a factor). My initial thought would be an increasing preference for speed but you just pointed out that speed has held constant.