Guides
Why two sites give the same player different projections
One of the forecasters this site reads has George Pickens down for 292.7 points this season. Another has him on 226.2. Same player, same PPR scoring, same season. Sixty-six and a half points between them, and for years my answer to a gap that size was to believe whichever tab I had open.
Neither of them is broken. They disagree because they are guessing at different things, and the size of the disagreement tells you more than the average of it does. Everything below came off this site's consensus feed on 15 August, and all of it will have moved by the time you draft.
Two receivers, six points apart
Pickens and DeVonta Smith land within six and a bit points of each other once the four sources are averaged.
| Player | Lowest | Consensus | Highest | Spread |
|---|---|---|---|---|
| George Pickens | 226.2 | 241.1 | 292.7 | 66.5 |
| DeVonta Smith | 229.2 | 234.9 | 241.3 | 12.1 |
Read the middle column on its own and those are interchangeable receivers. Read the outside columns and they are not the same kind of asset at all. Every forecaster who published a number on Smith landed inside twelve points of every other one. On Pickens they are five and a half times further apart than that.
So which of the two would you rather spend the pick on? I take Smith, and I hold that loosely, because "they all agree" and "they are all right" are not the same sentence.
What they are actually arguing about
The four numbers on Pickens are not four opinions about how good he is. They are four different stat lines that were never going to add up the same.
The highest has him catching 94 balls for 1,388 yards and 10 touchdowns. The lowest has 74 for 1,102 and 7, plus a yard lost on a carry. Twenty catches, 286 yards, three touchdowns and that stray yard. Run the difference through PPR scoring and it comes to 66.5 points, which is the entire gap, to the tenth.
Touchdowns are where forecasters part company most, and touchdowns are the part nobody predicts well. The career stat line guide put a number on that from this site's own career data: touchdowns per target repeat from one year to the next at 0.16, against 0.70 for targets per game.
Lamar Jackson, agreed on and not agreed on
Lamar Jackson is the tightest consensus among the sixty players going earliest in this month's drafts. The four sources have him on 317.4 points and they fit inside 5.5 of them.
| Source | Pass yds | Pass TD | Rush yds | PPR |
|---|---|---|---|---|
| Highest total | 3,886 | 26 | 671 | 320.4 |
| Second | 3,704 | 26 | 646 | 318.8 |
| Third | 3,380 | 27 | 668 | 316.0 |
| Lowest total | 3,368 | 33 | 602 | 314.9 |
Now look left. The source with the most passing yards has him throwing 26 touchdowns. The source with 518 fewer yards has him throwing 33. Seven touchdowns of daylight, and the two totals finish within six points of each other.
Agreement on the total is not agreement about the player. It is four different seasons that happen to add up the same.
Is the range a probability?
No, and the site says so out loud rather than in a footnote. Ask this site's API for any one player and the answer carries a field called range_meaning. Its value is "source disagreement, not a calibrated prediction interval".
That is not modesty for its own sake. The range is a minimum and a maximum over at most four numbers, so it belongs to whichever source sits furthest from the other three. Drake London has the widest range of those same sixty players, 248.2 to 328.4. One source has him scoring 14 receiving touchdowns. The other three say 7, 7 and 7.3. Take that one row out and most of the 80 points goes with it.
I have not measured how often a real season lands inside a player's range. Until somebody has, a wide range means the forecasters disagree and nothing else. It is not a 90% band, and the methodology page says the same thing about every range on the site.
Four is the best case
Everything above is a four-source player. Most of the board is not. Of the 300 highest projected players this site publishes, 181 have four independent forecasters behind them, 68 have three, and 51 have one or two.
Jaydon Blue is being drafted at 149.3 with two numbers behind him: 44.9 and 46.5. A 1.6 point spread is not agreement. It is two people. The consensus board labels those rows rather than averaging them into something that sounds settled, and I would not reach for anybody on the strength of two opinions.
What I do with it
I read the width before I read the number. A tight range means the argument is over and I am buying a known quantity at a known price. When it is wide, somebody who does this for a living is seeing something the other three are not, and I have to work out whether I want to be on that side of it.
Where I have genuinely not made up my mind: whether a wide range should pull me towards a player or away from him. Upside I cannot get anywhere else, or variance I do not need. I have argued myself into both positions inside the same week, and I am not going to pretend to a rule I do not have.
Four people who forecast football for a living cannot agree on George Pickens to within sixty-six points. I find that reassuring rather than irritating. It means nobody has this solved, including the ones who sound like they do.