Sports
Research NFL player props across every book
An NFL prop is a claim about opportunity that gets settled by efficiency. Get the opportunity projection wrong by ten percent and no amount of statistical refinement downstream will recover it, because the error is upstream of everything else.
Analytics only. quant+ never places bets or holds funds.
- cover line. The main number the market trades on this player and market.
- alt lines. Every other rung the book posts, each with its own price.
- best price. The highest number available anywhere for the side you want.
- consensus line. Where the market broadly settles across books.
- fair price. The market's own probability with the margin removed.
- projection. An independent estimate from expected opportunity and matchup.
- movement. How the number and the price have changed, with timestamps.
- book count. How many venues post this exact market.
Field names and definitions only. No odds, projections or hit rates are hardcoded anywhere on this page.
Opportunity, then conversion, never the reverse
A count projection is a product of how many chances a player gets and how well they convert them. Most prop research inverts the order, starting from a recent output and working backward to a conclusion.
Snap share
The ceiling on everything else. A player who is on the field for a third of the snaps cannot have a full-time workload regardless of how efficient they are. Snap share moves on injuries, personnel groupings and coaching decisions, and it moves faster than any season-long average.
Route participation and carry share
For pass catchers the relevant count is routes run, not snaps, because a blocking snap produces no target opportunity. For running backs it is the split of carries and the share of passing-down work, and those two are frequently different jobs held by different players.
Target or carry conversion
How often the opportunity becomes a statistic. Target share tells you how much of the passing game flows through a player; catch rate converts those targets into receptions; yards per target and per carry convert them into yardage. Efficiency is applied to opportunity, not the other way round.
Opponent and game script
The defence's tendency by position, and the game state the player is likely to be in. A team expected to trail will throw more, which lifts receiving volume and suppresses rushing volume. The implied team total is a more informative input than the opponent's defensive ranking, because it tells you the shape of the game.
The reason the order matters is that opportunity and conversion have different stabilities. A player's role over the last three games is a reasonably informative guide to their role next week, because roles persist. Their conversion rate over the same three games is close to meaningless, because catch rate and yards per carry fluctuate around a personal baseline and three games is not enough to distinguish a change in the baseline from ordinary variation.
So a projection built from a rolling output average is mostly a projection of recent luck. A projection built from an opportunity estimate that has been updated for the current week's expected game script, then multiplied by a shrunk conversion rate, is doing something different and more defensible.
Anytime touchdown is not a count market
Passing yards, rushing yards, receptions and receiving yards are counts. They take a range of values, they occur often enough within a game to be modelled as rates, and the distribution of their totals across games is reasonably well behaved.
A touchdown is not that. It is a rare event, usually zero or one per player per game, and its occurrence depends on a team reaching a specific field position and choosing a specific play. Treating it with the same machinery as a yardage prop imports assumptions that are wrong in a direction that matters, particularly in the tail.
The correlation structure compounds the problem. Every skill-position player on a team competes for the same small pool of scoring opportunities, and that pool is itself a function of the team's implied total. Pricing two touchdown props from the same team as though they were independent double-counts the same information, once in each direction.
Why a naive rate model fails here
- A touchdown is a rare event, so a per-game average reflects a few tallies spread across a season rather than a stable rate.
- Touchdowns are heavily concentrated: within a team, a small number of players account for most of them, and who those players are changes with formation and personnel.
- A rushing touchdown and a receiving touchdown by the same player are different events with different drivers, and a market that combines them is pricing a sum of two things that behave differently.
- Touchdown markets correlate with the team total, so two players on the same team are not independent propositions and combining them compounds the error.
The workable approach prices the team's scoring distribution first, then allocates it across the players who can plausibly score, using role and formation data rather than season-long touchdown totals.
Why the shape of the distribution decides the answer
Two players can share the same projected mean and have different probabilities of clearing the same line. Any method that reports only a mean is discarding the part of the answer that changes the bet.
Count statistics are non-negative and discrete. Their variance is frequently larger than their mean, which is the definition of overdispersion. The Poisson distribution, whose single parameter forces variance to equal the mean, therefore produces tails that are too thin: it understates how often a player has a very large game and understates how often they have a very small one.
A negative binomial adds a dispersion parameter and lets variance exceed the mean, which matches how these counts actually behave. For a line close to the projection the difference between the two may be modest. For a line in the tail, which is exactly where alternate rungs live, the difference is the entire answer.
There is a second source of spread that a simple count model ignores: the uncertainty in the projection itself. Expected targets are not known, they are estimated, and an honest model propagates that uncertainty into the final probability rather than treating its own point estimate as fact. This is why a fair price from a good model is usually a range, and why a model that offers suspicious precision on a volatile market should not be trusted further than its arithmetic can be checked.
For a longer treatment of the general problem, and why rolling hit rates are a weak substitute, see the expected value guide.
- Mean versus spread. Two distributions with identical means but different dispersions give different probabilities of exceeding any given line. The width is not a refinement; it is a primary input.
- Alternates amplify the error. Rungs far from the projection are the ones where a thin-tailed assumption does the most damage, and they are also the rungs carrying the largest quoted prices. The two effects point the same way.
- Every rung is a separate bet. A ladder of alternate lines is not one market with variations. Each rung has its own probability, its own price and its own margin, and the best price on the board is frequently on a rung nobody is looking at.
Five things that move an NFL prop, and what each implies
Reading movement is a way of asking what the market learned. The implication column is the part that changes a decision.
| Driver | Typical effect | What it implies for a position already held |
|---|---|---|
| Injury to a teammate | Volume shifts toward the remaining players at the same position | Your original projection is stale. Re-derive opportunity before deciding whether the position still has value. |
| Practice participation reports | Small moves in the projected opportunity of the player and their replacements | Reports are deliberately imprecise. A limited participant is not the same as a non-participant, and the market usually prices the distinction. |
| Weather forecast revision | Lower passing volume and totals in high wind | Affects both teams' passing props in the same direction, so it does not create a relative edge on its own. |
| Team total movement | Repricing of every scoring-dependent prop on that team | A team total move is the cleanest single signal that the market has changed its view of the game script. |
| Sharp-side pressure | A number moving against the direction of public money | Worth noting and not worth copying blindly. Movement is information; it is not an instruction. |
A weekly prop routine
Football's schedule is a gift here: the same sequence works every week, and the information arrives in a predictable order.
Project opportunity first
Build the expected snap, route and carry counts for the week before looking at any price. The market's number should not be an input to your opportunity estimate.
Shrink the conversion rate
Pull efficiency toward a positional baseline rather than using the recent sample directly. Small samples of a noisy rate carry very little information.
Produce a distribution, not a point
Convert the projection into a full distribution with dispersion calibrated to the market, and read probabilities off it for every tradable rung.
Compare rung by rung
Price each alternate separately. A ladder is a set of bets, and one rung is often priced out of line with the rest.
Log the movement
Record the line and price when you act and again at the close. The gap between them is the clearest available measure of whether your process is adding value.
Re-check before kickoff
Inactives publish shortly before the game. A position decided on Wednesday has not been checked against the information available on Sunday.
What prop research cannot do
It cannot remove variance. A model with a genuine edge on a receiving yards prop is right perhaps a little more often than the price implies, and that difference is invisible in a single Sunday. It shows up, if at all, across a season of bets, which is a sample small enough that a good process and a lucky one look identical for long stretches.
It cannot fix bad opportunity estimates. If the market knows something about a player's role that you do not, no distributional refinement compensates. This is why the discipline of re-checking before kickoff matters more than any modelling choice on this page.
And it cannot tell you which side of a coin flip to take. Some of the best-executed prop research in a given week produces no positions at all, and a workflow that always finds something is a workflow that has loosened its threshold.
quant+ provides the research surface: lines, alternates, best price, consensus, fair price, projection and movement, with the inputs that produced them. It does not place bets. See also the NFL game model for the team-level side of the same slate.
Scan for prop edges · Compare prices across books
Project the opportunity, then price it
Build the usage estimate first, turn it into a distribution, and read every rung of the ladder against fair value instead of against a streak.
Questions people ask
Why is hit rate over the last ten games a weak NFL prop signal?
Three reasons. Ten games is a small sample of a count variable with real game-to-game variation. The window blends the player's ability with the game situations they happened to be in, and those situations are not stable. And a hit rate measured against a past line is describing a bet that may no longer exist, because the market has moved the number. It is a useful sanity check on a projection and a poor substitute for one.
How should anytime touchdown props be modelled?
Not as a rate multiplied by games. Price the team's scoring distribution first, then allocate the scoring across the players who can plausibly produce it using role and formation data. Touchdowns are rare, concentrated among a few players per team, and correlated through the team's implied total, so a count model applied directly to them will misprice both the tail and any combination of two players from the same team.
What data does quant+ show for an NFL prop?
The cover line and its alternates, the best price available across books, the market consensus line, a fair price with the margin removed, an independent model projection, recent line movement with timestamps, and the number of venues posting that market. The pattern across those fields is usually more informative than any one of them alone.
Do alternate lines matter in NFL props?
They are where a mispricing is most likely to survive, for two reasons. Each rung carries its own price and its own margin, so a long ladder is not guaranteed to be internally consistent. And rungs far from the main line sit in the tail of the distribution, where a thin-tailed modelling assumption does the most damage to a naive probability estimate.
Can quant+ place my NFL prop bets?
No. Regulated United States sportsbooks prohibit automated wagering and do not offer public bet-placement interfaces, and quant+ does not attempt to work around either. The platform produces the research and you place the wager yourself.
Sources
- Research player propositions with Outlier's Prop Finder, help.outlier.bet
