Per Game College Dominator Is A Sham
Like everyone else, I enjoy pretending I always know the answer. But the most important thing I know about fantasy football is how useful anything can be and how much we can the answer at all.
It’s been a thorny subject before but before now Per Game College Dominators have mostly only come up when someone asked why my calculations differ from a site they use. But with the 2019 draft class comes a lot of players, who stand out on tape, that simply didn’t play or produce well in college. This has brought Per Game College Dominators to the forefront.
Per Game Dominator, the theory seems to be, “prove” or attempt to prove, that these players were in fact very productive when on the field.
This has honestly never made complete sense to me.
I’m going to lay out what I think the flaws are and how I went about testing them compared to whole season Dominator calculation. In the end, I want to offer an opinion on how Per Game Dominators are and are not useful for evaluating players.
Why Not Per Game?
So here are some of the questions I have about Per Game Dominators.
- If few players are good in the NFL and underproductive in college, why would this add signal?
- If under productive WR’s hit in the NFL does boosting their Dominator ratings also improve everyone else on an even scale?
- Are we comparing two different metrics? One describing and one forecasting imaginary snaps?
- Wouldn’t the team numbers have changed?
Per Game dominators also seem to assume that the value of production (i.e what they can do) is the measure of what a player did. I’ve always assumed production was a proxy for other things like development and experience as much as raw talent.
It’s all well and good making an argument that sounds “right”. But both the joy and the chore of statistics is the ability to actually go and look to see if it’s true. You don’t have to trust me, which is always a plus.
I compiled a sample of wide receivers drafted between 2003 and 2016 (so a minimum of 3 years spent in the NFL). This is from my own database available to everyone. I then averaged their points per game (PPR scoring) over the first three years in the NFL. This will be my measure of how “good” players have been.
Any player I didn’t have at least 2 years of college data on was removed (because of some quirks about how College Dominator ratings are calculated). This left a group of 471 players at the position to test.
I’m assuming that there is a connection between production in college and success in the NFL. Also that the first three years of a wide receiver’s career is a measure of if they “Hit” or “Miss” for fantasy.
My theory is that Per Game College Dominator Ratings are a sham, and they will be less able to explain the variance in PPG over the first three years of a player’s career.
If they produce an equal or greater r^2 value when compared to AVG PPG then I will consider my assumptions disproven.
We’re all statistics folk here, but I’m not a mathematician or an actual statistician. So let me try to explain what r^2 is so we’re all on the same page.
R^2 is a mathematical value that expresses the amount of variance between two different value sets….not so helpful? I agree. In essence, the closer to 1 (100%) the R^2 number is, the more the one number “explains” the other number. So Draft Round “explains” about 28% of wide receivers average points per game over their first three seasons. If College Dominator scores an r^2 of 100% then it would tell us exactly who was going to score more points per game (it doesn’t, it’s about 11%).
I’m using what I think is the most common definition: a players’ combined Market Share (% of the team) touchdowns and yards in their best and final season in college.
Players Stat / Team Total Stat = Markets Share (MS)
If you’re interested the formula looks like this:
CD = ((Best MS TD + Final MS TD)/2) + (Best MS Yds + Final MS Yds)/2)/2
I think there may be some misunderstanding about Per Game College Dominator ratings. They are not calculated using “just the games the player played in”. Instead, they use the same calculation as whole Season College DOminator and apply it to stats on a per game basis.
In order to convert whole Season College Dominator to a Per Game metric (and therefore adjusting for games missed), we need to know what the player and the team did per game.
Per game = Stat / Number of Games Played
I actually don’t have Team Games Played easily available in my database. So I assumed a team played 12 games (no bowl games) if the player had not played more than 12 in any season.
Because this introduces potential variance I also ran the calculations assuming the teams had played 14 games if the player had played less than 12 games.
I think this gave a fair idea of the maximum and minimum range of outcomes’ for Per Game College Dominator ratings predictive ability.
No matter how I worked the sample size or adjusted the team’ games played, Per Game College Dominators were always significantly less predictive then whole season College Dominator Ratings.
Assuming 12 games played
Assuming 14 games played
Per Game College Dominators were unable to increase how predictive College Dominator is. Not only that but they lessened the signal already present in College Dominator. It takes the signal College Dominator has and makes it worse.
Is it failing to increase good players who missed games, or decreasing good players who did well somehow?
We can also see a lot of the problems with Per Game College Dominators in the data very easily. Here are the top 12 biggest adjustments made using my per game calculations.
Sure, it makes Dez Bryant look better. But it also makes all these other players who have not been good/great for fantasy football look better.
At the same time, players like Demaryius Thomas and Calvin Johnson who scored well in whole season College Dominator get no adjustment at all. In essence, it pulls together a bunch of bad players and makes them look more like the good players. This blurs the formers relevance and obscures the latter’s potential.
College Dominator, by itself, isn’t very predictive. So it may seem like a small thing. But prospect evaluation is about that very small difference. Almost nothing has a good signal and Per Game Dominator calculations have the potential to create a negative knock-on effect.
Or to put it another way, a college production measurement that pushes Calvin Johnson and Demaryius Thomas out of the top 12, and pulls Devante Parker and Michaell Campanro up, is flawed.
I had planned to collect team games played data so I could calculate exact Per Game College Dominators. But given the range of outcomes from very bad to very, very bad, I don’t think I need to. Why would I when the one I have is better?
As far as I can see, there is no way for Per Game College Dominator Ratings to be more predictive (or even as predictive) as whole season dominators in any sample size big enough to matter.
By making less productive players look more productive, all you do is hide and obscure those who were actually productive in college.
Outliers look different
If outliers all had something in common, then they would not be outliers. They would share something in common that we could search for to try and identify them.
A variety of players have below ideal college dominators and did well in the NFL. But, they all did it differently for different reasons. Worse, even if using Per Game Adjustments were able to highlight other good players, it would matter less. Per Game College Dominator essentially takes players on a long road trip to end up back where they started, surrounded by a lot of players who didn’t hit in the NFL.
On a side note, I’ve found the average for yearly Dominator calculations more predictive then college dominator. This reinforces the trend we’re finding here that the predictive power of College Dominator is how productive players actually were in college overall, not how they good they were in a smaller sample at their best. Per Game calculations also lessened the signal for yearly Dominator calculations.
My main issue with the idea behind Per Game Dominator calculations is that they are unfair. They are unfair to players who were healthy and able to put up good production. But they are also unfair to the players who weren’t.
Players who missed games (but definitely display interesting athletic and football related traits on tape) also missed time. The benefit of playing a season in college isn’t the production. That’s’ just our measure of it and the result we can use to compare them. Players learn, grow, and gain experience through playing. Production carries with it some of the benefits of doing a thing for a longer time, at a higher level, and that is part of what the signal of doing it better carries.
Expecting a player who didn’t get that time to compete, one for one, with those who did, is unfair. They are more likely to need time, at least.
To Be Fair
Not all metrics and stats are designed to be predictive. Per Game College Dominator does a great job of describing the type of career a player could have had if they had continued producing at the same rate.
It has a use and a value. Just not for predicting players or comparing them to whole season College Dominator results.
It does not show how dominant they were in college. But it does describe a potential ceiling. Which brings us back to the mystery of if they would have been able to keep it going. Which we don’t, and can’t know through calculation.
I think it’s a good idea to point out that DK Metcalf’s production Per Game was high, or however high it was at least. However, I don’t think we can then compare it to a metric that attempts to describe what players did in a college. It does not prove he was as productive. He wasn’t, he didn’t play enough. That’s not his fault, obviously, nor does it say he isn’t good or can’t be productive in the NFL.
I’ll say that again: This does not mean players with low College Dominators can’t be good in the NFL.
All it means is their production was low in comparison to others. But so was Michel Thomas, Andre Johnson, Brandon Marshall, and so was Tyreek Hill. And they were all pretty good in the end.
The case for players who missed time is that they would have been better. Since it didn’t happen we don’t know. There is no magic calculation that can recreate the stats from an alternative universe where it did. Or at least not one I can run in excel.
This takes nothing away from a player who has impressed on tape. Nothing. They still did all of that. And since I want to watch more good wide receivers, I always hope they can transfer those traits to the next level.
Everyone loves an underdog, after all.
2019 Wide Receiver Prospects sorted by whole season College Dominator