Just an interim post as a thought for the league, looking for responses from both the experts amongst us and also any pro stat wranglers, as I suspect we have both amongst the 12 of us.
In the EFFL we have a big mix of real time nfl fan experience amongst the coaches. From the die hards to the more casual observers like myself. However, I suspect that it’s likley that even our most ‘expert’ coaches come pundits take at least a glance at predicted scores or rankings.
Unfortunately the predicted scores are a complete nightmare, has anyone ever hit their prediction? Individual and team scores are either way higher or lower, never on the money, why is that?
Well first and most obvious is that the games themselves are not controlled from the big Xbox in the sky by some dispassionate observer, they are real people making real decisions. So although we can have a stab at thinking how an offensive coach and QB might think against a particular team, in the heat of the moment, things are going to change; lucky breaks, penalties, injuries and ‘personality differences’ in an offensive line are going to affect the final score for a player.
In fact, the very thing that makes the real game, is the thing that screws up the predictions; touchdowns. That’s not to say that touchdowns shouldn’t be included in predictions, but just that they are so much harder to predict than the yardage for a particular player. For WRs in particular the a touchdown is often 7-10 pts rather than just the 5, as often as not coming from a big yardage play as a short yardage red zone formation. Even for RBs though, a big breakthrough play of say 30 yds will not affect the the predicted vs actual too much, but if that results in a touchdown its going to be a half to a third of most predicted scores in one play.
This suggests to me that predictions might be better expressed as 2 separate values, a yardage value and a touchdown value. The touchdown value might even be better expressed for the whole team, rather than individual players.
In any case, what you are looking at in a predicted score is not some magic number plucked out of thin air, it’s the mid point in a likely spread of values, probably best expressed with a bell shaped probability curve. Now maybe you’ve all noticed this already, but if you look at it that way it becomes more obvious that the higher a predicted score, the wider that spread is going to be, so a player with a predicted score of 5 might have an 80% chance of scoring between 3 and 7, but a player with a predicted score of 20 might have an 80 % chance of scoring between 12 and 28. If you’ve got that first player in your team and he has a bad day, it’s not going to have too much impact on your team, but the second player could really make the difference between a win and a loss. Of course a td for a low predicted scorer can easily mess up the predictions too, another reason to keep them separate.
Now I’m guessing that the pro sites have reasonably complicated algorithms for calculating predictions, but they’ve got to be including some of these variables:
Average number of carries per game
Average yds per carry
Relative strength of opposition against that position
Maybe less obvious
Likelihood of team being in front in second half (positive for rbs, negative for WRs?)
Likelihood of special formation red zone carries
There will be others of course, but what are they, maybe between us we could have a stab at the EFFL predcition equation?