cloppbeast
Footballguy
You missed the part above where this metric is not being considered as much as it should. NFL teams are not myopic about fantasy points like us, so I guess it's understandable.
Fantasy Football - Footballguys Forums
Welcome to Our Forums. Once you've registered and logged in, you're primed to talk football, among other topics, with the sharpest and most experienced fantasy players on the internet.
Spaceshuttle (1 Viewer)
- Thread starter Interseptopus
- Start date
Lash
Footballguy
yeah I remember Juju had a very nice breakout age, then his YPC dropped off the cliff his Junior year
so folks JUST looking at last-year tape and/or stats may have passed him over (even though his Jr stats were still decent)
(Eye Opening Fun Fact: turned out the reason for his falloff was just a huge dropoff in QB play from Kessler to Darnold - I remember this coming up when looking at drafting one of the rookie QBs last year and it bothered me then as far as projecting Darnold as a franchise QB as the NFL had him pegged and I couldn't pull the trigger on him in any of my leagues)
so folks JUST looking at last-year tape and/or stats may have passed him over (even though his Jr stats were still decent)
(Eye Opening Fun Fact: turned out the reason for his falloff was just a huge dropoff in QB play from Kessler to Darnold - I remember this coming up when looking at drafting one of the rookie QBs last year and it bothered me then as far as projecting Darnold as a franchise QB as the NFL had him pegged and I couldn't pull the trigger on him in any of my leagues)
Biabreakable
Footballguy
Where was that shown?
I do not think any proponent of breakout age has demonstrated that it is more efficient than draft position. Including the guys at rotoviz who did a much more rigorous study of this back in 2014.
The NFL teams are not drafting for fantasy points like we are thats true. It is our job to figure out what the teams are trying to do and to project fantasy points from that. Good news is the fantasy community is able to interpret the information and identify which players will score more fantasy points
barackdhouse
Footballguy
For the purposes of an analytically defensible predictive model, yes. For the purposes of gambling in FF leagues, you don't have to gain anything on the NFL model, you have to gain on that of your leaguemates. They are literally the only entities we are competing against. A tool such as what Dan has developed here could be a huge edge on an individual league to league basis. Even if marginal at best vs the NFL method(s).
Another consideration regarding the quality of the NFL draft as a predictor is that some years are inflated with OL talent, or DL talent, or DB talent (or others) such that there will be positional runs that render the magnitude of difference between say a late 1st and a mid 2nd WR seem bigger than it really is. In your example of whether we would take the 2.15 WR over the 1.15 based on their breakout age, I'd say probably not, but possibly. What if there was a big run of DBs and OL in between? And then you're talking about comparing the 1st/2nd vs 3rd or 4th receiver off the board, which is perhaps a smaller gap than looking at it as 1.15 vs 2.15. Maybe breakout age is a tiebreaker between 2nd and 3rd off the board for example.
And of course maybe it isn't. Situation is still huge for me. I'd rather have Deebo in SF than Brown in Baltimore. And as far as I know Brown has draft position AND breakout age in his favor.
kittenmittens
Footballguy
Draft capital can be fairly ineffecient because there are a lot of dumb teams making dumb picks. Players go off the board before they should all the time. It only takes one dumb team to like a player too much or draft for positional need and reach for him.
ConnSKINS26
Footballguy
It's literally the most efficient means we have to predict success.
It's not ideal but nothing beats draft pedigree, in a large sample size, that I'm aware of.
joey
Footballguy
I don’t think ConnSkins is saying draft position is an absolute measurement, just like you’re not saying your stats in this thread are the end all, be all.
kittenmittens
Footballguy
The two things are not mutually exclusive. The most predictive thing, draft capital, is still inefficient.
cloppbeast
Footballguy
Not more efficient, but not accounted for as much as it should be. Shown above, 54% of receivers meet criteria, 79% of wr1s do.
I'll repost it.
cloppbeast
Footballguy
Last edited by a moderator:
Biabreakable
Footballguy
*shrugs
54% is basically a coin flip. If the probability is this close that means theres really no trend present in the population of the sample.
The 79% of them being WR 1 being higher than the entire population doesn't really tell us much in my view. It could just be random.
cloppbeast
Footballguy
BigTex
Don't mess with Texas
The drop off in production had a lot to do with JuJu play most of the season with multiple injuries.
Tex
ConnSKINS26
Footballguy
Not what I was saying. Just bringing some perspective to someone's comment that using draft position is inefficient. On the whole that may be true but it's still literally the best thing we've got. It's more a commentary on how far we've got to go.
In reality, even though we tend to regard NFL teams as backwards dinosaurs resistant to change (and they are), their decision-making still is the best predictor of success. Why? Part of it is that guys with pedigree get more chances, kind of like a self-fulfilling prophecy. But part of it is also that with their vast resources they are probably working a bit more of these analytics into their decision-making than we realize on the outside.
It's possible that no matter how much progress is made in the analytics we have access to, that the NFL draft will still remain the most effective predictor of success, because they are also paying attention.
cloppbeast
Footballguy
I think you're right. With so many factors to consider, one metric encompassing all of 'em created by the NFL is the best we can expect.ConnSKINS26 said:
On the other hand, being that an NFL team's goals differ from ours, in theory there's a better system out there. Some sort of multiple regression. @ZWK has his formula for instance, but I cant speak to how its derived.
One could also take the approach of adjusting draft position based on metrics which lead to fantasy points more/less often. To me, it's not prudent to do so without examining how each of them correlate. It's a bit sketchy to downgrade Parris Campbell for his poor b.o.a. without knowing exactly the implication of a great 40 time, or whatever else.
Last edited by a moderator:
ZWK
Footballguy
My formulas are not based on linear regression. I basically take a weighted average of a bunch of variables, and I make up weights that seem plausible to me rather than pulling them out of a best fit to historical data. I also sometimes transform variables, e.g. turning RB height, weight, and BMI into a single number that represents how good or bad the RB's size is.
Football Outsiders does use linear regression to create their prediction formulas. I have written about some of the pros and cons of their method vs. mine here. The basic problem is that it takes lots of data points for regression to work well, especially when there are other things going on like many predictor variables that are correlated with each other. And what do you do when a stat like drop rate wasn't even recorded until recently?
I think that there are fancier statistical methods which are better than regression for the thing we're trying to do, e.g. Bayesian multilevel modeling. But I haven't learned the stats well enough to actually use them, and I don't know if they deal well with all the issues. I think that some of the intuitive/subjective aspects of my approach are basically mimicking things that are done more systematically by fancier statistical methods.
Biabreakable
Footballguy
54% of the players who were included in the sample met the criteria correct? 46% did not.cloppbeast said:
This is a random distribution where no clear trend is found. A coin flip.
The 79% of players meeting the criteria who were top 12 players does suggest that meeting the criteria is a prerequisite for achieving top 12 in fantasy. However the sample size for this is very small, only a handful of players who were top 12 over the time frame (there is overlap).
It is entirely possible that in the decade preceding this sample that players in the top 12 were not as strongly represented, it is also possible that over the next 10 years that players in the top 12 are not meeting the criteria of the sample.
If that is the case then the 79% number is actually random rather than showing a correlation. This is always a problem with small sample sizes.
To give an example lets say for the decade preceding the sample that only 44% of the players met the criteria, so very different than the 79% of the chosen time frame. Then 10 years from now we look at that time frame and find that 55% of the players met the criteria.
If this were the case, then we are right back to our 54% of the players meeting the criteria (more or less) and would find that the 79% for the time frame selected is actually just random variance in the sample.
Biabreakable
Footballguy
Maybe you should include those overlap seasons. Sorry I thought you already did that.
This would be another thing you could look at. How many of the players who had more than one top 12 season met the criteria?
I'm really not comfortable with the market share data and where those cut offs are to begin with and that is the main source of my skepticism.
cloppbeast
Footballguy
Couple things.
I didnt intend this was definitive, only it suggests a trend worth further inquiry. Theres a few things which need altered before we get to make any conclusions. Namely the threshold part and the double metric part, one at a time, please. And as Doc just mentioned, counting multiple occurances of wr1 seasons is one such improvement.
Also, theres no sample here. This is every wr drafted in the last 8 years, totaling more than 200. Well technically, it's a sample of all of them in the NFL ever, but not sure how anybody from the 80s or 70s applies today. Considering how much the game has changed, the most relevant information would come from recent data.
Concerning variance, using the chi squared test, these are not independent when using a level of significance of .001. Safe to say this is not a case of random chance.
Last edited by a moderator:
TakiToki
Footballguy
I'm always very conflicted on topics such as these. My first thought is that the NFL is a billion dollar organization, and a dude with some data analysis techniques isn't going to beat the system.
Then I see teams punting on 4th-and-1 on the opponent's side of the field and I re-think that notion.
Then I see teams punting on 4th-and-1 on the opponent's side of the field and I re-think that notion.
barackdhouse
Footballguy
Frankly I think that kind of thing cannot be overstated. The NFL is full of meatheads.
Biabreakable
Footballguy
You will have to unpack that for me. I think doing such a test would require more data.
As far as the data prior to the sample being used being relevant or not in today's game? It should not matter if the market share is actually the key to unlocking WR performance in the NFL
I kind of already think I know it isnt. Because there are tons of college players with impressive college market shares who dont do anything in the NFL.
If market share was actually a measure of WR talent then players with high market share in college would see that transfer to the next level.
Yet we already know that half of them in the top 24 did not meet the criteria.
cloppbeast
Footballguy
Over 100, or however the hell many colleges there are, narrow down to 32. Needless to say, a minority of college players make it in the NFL. Not everybody who does well in the NCAA will make it; but if you cant hack it there, more likely you wont at the next level either.
I cant argue with you about the vast majority of good college recievers failing in the NFL. But you're milking ducks. We are comparing those who do well in school vs those who dont. The former has a better chance in the pros. In other words, as bad a chance those who produce have of NFL success, those who dont produce have even worse chances. Pretty simple really.
And to correct the bolded, about a quarter of the top 24 dont meet the criteria.
Last edited by a moderator:
Biabreakable
Footballguy
What does milking ducks mean? Never heard that before.
The criteria is where the margins have been set. You said 54% of the top 24 met the criteria. Is that not correct?
Now you are saying the top 24 is something else.
Please make up your mind. Hard enough discussion without the chains constantly moving.
No I haven't checked anyone's math here.
If meeting the criteria is a random distribution at each level of the partitions (top 12 top 24 top 36 ect) then there isnt a consistent pattern here and it may just be coincidence that a high percentage of the players met the criteria in the top 12.
cloppbeast
Footballguy
You are misunderstanding me. 56% of wide recievers entering the NFL via the draft meet the criteria.
TheWinz
Footballguy
Picture yourself at a live FF draft. You remember those days at your local bar, with beer, wings, laptops, and cheatsheets. Every time it's your turn to pick, you enlist 100 friends, ranging from statisticians to scouts to diehard fantasy players to take turns crossing off a player, each using a different set of criteria, until only one player is left to draft. If you did this for the whole draft, I bet you would like your team. And it would be a team made up by eliminating players, not by including them. This is Dr Dan's theory, in a nutshell.
Ilov80s
Footballguy
Last edited by a moderator:
Catbird
Footballguy
Its like people are speaking different languages here. Several think the clear consistency of the data makes consideration of the cutoffs valuable and the other half is saying either that draft order has been shown to be the most consistent predictor of WR success and so there is no need to look at other factors, or that the sample size of the last 8 years is insufficient to rely on. It seems pretty simple from there. People should use what they are comfortable with.
Like everything, people are very unlikely to convince anyone with a strong opinion of anything that isn't consistent with it. People who are open to the predicability of the data, knowing the sample size and lack of absolute certainty, should be discussing whether that can be improved upon. People who aren't open to seeing the data as predictive or useful should go somewhere else and look at things they think are more useful to them. Using this thread to bicker between the two basic views just takes time away from both useful endeavors.
Since no one is going to prove the data non-predictive, and with it clear that many people think it non-predictive or that this can't be shown clearly enough to be useful, can we keep this thread to talking about how best to use this data and how to improve its usefulness, rather than yelling back and forth whether or not people should use it at all?
Like everything, people are very unlikely to convince anyone with a strong opinion of anything that isn't consistent with it. People who are open to the predicability of the data, knowing the sample size and lack of absolute certainty, should be discussing whether that can be improved upon. People who aren't open to seeing the data as predictive or useful should go somewhere else and look at things they think are more useful to them. Using this thread to bicker between the two basic views just takes time away from both useful endeavors.
Since no one is going to prove the data non-predictive, and with it clear that many people think it non-predictive or that this can't be shown clearly enough to be useful, can we keep this thread to talking about how best to use this data and how to improve its usefulness, rather than yelling back and forth whether or not people should use it at all?
@Dr. Dan For those of us that are blocked from being able to see the spreadsheet would it be possible to just give a current listing of the results for this year's rookie class? I tried skimming the thread but it jumps around quite a bit and gets lost in the debate whether or not this is a worthwhile tool. I believe it is and was hoping for a quick cliff notes version of the results. Thanks.
bicycle_seat_sniffer
Smells like chicken
Bigboy10182000
Footballguy
ex-ghost
Footballguy
Doc, thanks for taking the time both to do this and to publish it ahead of the draft. It's always great to see someone else's Secret Sauce.
You mentioned a couple times that you're not sure what a statistically significant difference is between the NFL hit rate and your model, could you please tell us how many WRs were used in each round? 10 years is a great sample size but I don't know how many WRs fall into each round on average.
You mentioned a couple times that you're not sure what a statistically significant difference is between the NFL hit rate and your model, could you please tell us how many WRs were used in each round? 10 years is a great sample size but I don't know how many WRs fall into each round on average.
barackdhouse
Footballguy
I forget, and I looked above and couldn't find it, but did you post a link to the sheet? Or if you are willing to PM me read-only access, I'd appreciate it. I think it's likely a very useful tool. I've been following this thread since you first started developing this and I definitely think there is something to it. Would love to see the sheet, though.
Snorkelson
Footballguy
barackdhouse
Footballguy
good stuff. shouldn't column B and C add up to 100% in every case? because they don't. aiyuk, duvernay, some others.......
bicycle_seat_sniffer
Smells like chicken
. 
az_prof
Footballguy
So, based on this avoid Ruggs, Aiyuk, PIttman, and Claypool.
One thing worth looking at with your model is a careful analysis of each "miss." So Tyreek Hill is a miss. Of course he was a miss for NFL GMs too. But his speed was something that might have suggested that maybe he would be an exception. Which is where Ruggs comes in. So he may end up bucking your model because his speed is other worldly. And he also has high draft capital so he will be given every chance to succeed. And as others have said, teams with many studs pose a challenge for Dominator measure.
One thing worth looking at with your model is a careful analysis of each "miss." So Tyreek Hill is a miss. Of course he was a miss for NFL GMs too. But his speed was something that might have suggested that maybe he would be an exception. Which is where Ruggs comes in. So he may end up bucking your model because his speed is other worldly. And he also has high draft capital so he will be given every chance to succeed. And as others have said, teams with many studs pose a challenge for Dominator measure.
ffmail4me
Footballguy
I really don't know what to make of this data. So last year, 5 predicted misses (McLaurin, Diontae Johnson, Hollywood Brown, Deebo Samuel, and Darius Slayton) ended up as some of the better WRs in this class. Not trying to knock you AT ALL Dr. Dan.. but how much confidence should we have in this system if so many "predicted misses" were great hits? Is there a factor I'm missing here? And I am very grateful of all the work you're doing here, swear I'm not trying to be snarky at all!
Last edited by a moderator:
ffmail4me
Footballguy
and no doubt AJB was grand slam homerun, drafted him everywhere I could, out of my 10 dynasty leagues think I got him in 5 of them, and never once passed on him.
