Projections by category
Consensus
Tap a category to add or remove it. The columns follow, and the value at the end of the row is the z-sum over what is checked. Field-goal and free-throw percentage are attempt-weighted; turnovers are flipped so that up is always good. Click a column header to sort by it.
A greyed figure is our model's line on a player the outside consensus does not cover. Blank: neither line carries the stat.
Field guide
How to read this
A category manager does not draft a points total; he drafts a line. Twenty-six points, seven rebounds, six assists, two threes, a steal and a block on 48 percent shooting is a line you can weigh against your roster, column by column, and this page shows that line for every projected player, per game or as a season total. Sources counts the independent outside forecasters behind a row, and for basketball that number is one today: a single forecaster publishes a full season line for every rostered player, so the outside consensus on this page is that forecaster's line, not an average of many. We say so on every row rather than dress one opinion up as a crowd.
The second line is our model. It never votes in the consensus, because a model of ours should not lend itself the authority of a crowd, but it is shown beside the outside line so you can see where the two disagree, and it stands alone, greyed, for the players the outside forecaster does not cover at all.
Read the rest
Read the rest
Two lines, and why neither is folded into the other
Every outside forecaster we read gets one vote, and the consensus is the median, stat by stat, across the forecasters who published that stat. A median rather than a mean, so one forecaster who is far out on a player cannot drag the figure toward him. With one forecaster the median is that forecaster's line, which is fine as long as the page says so, and it does: the source chip on every row reads one, not three, and the warning above the table counts the rows that are an opinion rather than a consensus. Our own model is kept out of that vote on purpose. It is the least-evidenced input on the page, it has no beat reporters and no depth-chart intel, and blending it in would drag the consensus toward its errors while making the row look better supported than it is. It is still shown, on the Line chip, and it is the fallback: a player no outside forecaster has published takes the house line whole, greyed, with a source count of zero, so the deep end of the pool is not silently missing from a page that claims to list every projected player.
Per game or season total
Basketball managers think per game, because lineups are daily and a matchup is a handful of games per player, so the page opens per game: the season total divided by the projected games played, which is itself a projection and the one that moves most when a player has a history of missed time. The Figures chip flips every column to the season total, which is the number a roto league is actually scored on and the one to read when comparing a 65-game player with an 80-game one. The two percentages are the same in both views, because a rate is a rate; they are re-derived from the consensus makes and attempts rather than voted on directly, so a median of two percentages never appears, and the makes and attempts are what the value column weighs them by.
What the value column is
Under a category format the value is the player's summed z-score over the categories you have checked in the picker: in each one, how many standard deviations his consensus line sits above or below the pool, added up. Field-goal percentage is weighted by his projected field-goal attempts and free-throw percentage by his free-throw attempts, so a low-volume shooter with a pretty percentage cannot swing the column the way a twenty-attempt starter can; turnovers are flipped so that a bigger number is always better. It is the same arithmetic the category board ranks by, minus the punts and weights, and it is always computed on the consensus line even when the Line chip shows our model's figures, so the order of the table does not change when you flip lines and the disagreement stays readable. Under the Points format the value is the line scored with the table you picked, as a season total, and the columns show the stats that table pays for.
Where basketball forecasters disagree, and why
The gaps between the two lines cluster in four places. Minutes, first: a projection is a rate times a role, and a forecaster who has a player at 30 minutes and one who has him at 24 will disagree by a fifth in every counting stat while agreeing on his percentages. Games played, second, which is where a history of load management or a surgery in the summer shows up as a season-total gap with no per-game gap at all. Rookies and players who changed clubs, third, because there is no last season on the new roster to anchor to, and the forecaster with a reporter in the building is guessing better than the model without one. And usage, fourth, when a club has lost or added a star and the shots have to go somewhere. Read a disagreement as a question about one of those four, and it usually answers itself from the player's own page or the news.
The draft-position view
The second view is average draft position, the same players in the order real drafts are taking them, with the sample behind each figure. A projection says what a player should produce; a draft position says what a room is willing to pay, and the two are kept apart because they answer different questions. A player with a top-thirty value on the category board and a draft position in the sixties is a bargain the room has not noticed, usually because his line is built from blocks and percentages rather than points; the reverse is a scorer the room likes more than the categories do. Draft position is a single market today, from a points-league default, so a category manager should read it as a price rather than as a ranking in his format.
What to do with a disagreement
Take it to the category board, which prices the same line against your league's categories, punts and roster rule, or to the printable sheet for draft night. Then decide which line you believe and draft accordingly. The arithmetic, and what we refuse to publish, is in the methodology.