A point spread is a forecast compressed into a single number: the expected scoring margin between two teams. A favorite listed at −3.5 is expected to win by roughly 3.5 points on average — not to win by exactly that amount, and not even necessarily to win at all, since real outcomes scatter widely around that expectation from game to game. Understanding how that one number connects to a win probability, why certain point totals carry outsized weight, why the number changes before kickoff, and how a quoted price differs from a true probability is mostly an exercise in applied statistics, and none of it requires placing a bet to be useful.
What a Point Spread Actually Represents
A point spread is not a prediction of a final score. It’s a statement about the expected difference between two final scores — the favorite’s points minus the underdog’s points, averaged across the range of ways a given matchup could plausibly play out. A −3.5 favorite might win by 20, lose by 10, or win by exactly 4; none of those individual outcomes contradicts the spread, because the spread is describing a center of a distribution of outcomes, not a single deterministic result. That distinction matters because the interesting statistical question isn’t “did the favorite cover” — it’s how much real outcomes vary around that expected center, and what that variation implies about how confident anyone should be in the forecast.
From Expected Margin to Win Probability: the Normal Approximation
Converting a spread into a win probability requires a model of how final margins scatter around the spread, not just the spread number on its own. The standard approach treats the difference between the actual final margin and the spread as approximately normally distributed, centered on zero, with some standard deviation . Under that assumption, the probability that the favorite wins outright — that the actual margin ends up above zero — is:
where is the standard normal cumulative distribution function and “spread” is entered as a positive number representing how many points the favorite is expected to win by.
The value of is what turns this from an abstract formula into a usable estimate, and it has a well-documented history. Hal Stern’s 1991 study in The American Statistician, “On the Probability of Winning a Football Game,” measured the standard deviation of NFL final margins relative to the closing spread and found it close to 13.86 points — a figure that has been treated as the field’s reference value ever since. Measuring the same quantity directly on a larger, more recent sample — 7,017 NFL games from 2000 through 2025 with a recorded closing spread, drawn from nflverse’s published game data — gives a standard deviation of 13.2 points, with the average residual (actual margin minus spread) close to zero, exactly as the model assumes. The two figures, three and a half decades apart, are close enough to treat the underlying scatter of NFL outcomes as a genuinely stable statistical property of the sport rather than an artifact of any particular era.
Plugging real numbers into the formula makes the relationship concrete. A 3.5-point favorite has a win probability of . A 7-point favorite comes out to . A 14-point favorite comes out to . Checked against the actual 2000–2025 record, favorites in the 6.5-to-7.5-point range won about 74.3 percent of their games, and favorites in the 13.5-to-14.5-point range won about 85.3 percent — both a bit higher than the pure normal-curve estimate, particularly at the 7-point level. That gap is a real and interesting feature of the data rather than noise, and it traces back to the same phenomenon that makes 3 and 7 special numbers in the first place.
Why 3 and 7 Are Not Like Other Numbers
A field goal is worth exactly 3 points. A touchdown with the extra point is worth exactly 7. Those two facts alone are enough to break the smooth, symmetric shape a pure normal distribution would predict for final margins, because real NFL games end on exactly those numbers far more often than a bell curve would suggest. In the 2000–2025 nflverse data, a final margin of exactly 3 points occurs in roughly 15.1 percent of all games — the single most common margin by a wide gap — with 7 the next most common at roughly 9.0 percent, followed by 6 at about 6.0 percent, 10 at about 5.5 percent, and 4 and 14 tied at roughly 4.9 percent each. A smooth normal curve with a standard deviation of 13 points would spread that probability mass out far more evenly across nearby values; real football games instead pile up on the numbers that correspond to how the sport actually scores.
This is why a spread crossing from 2.5 to 3.5, or from 6.5 to 7.5, is treated as a structurally bigger move than a same-sized shift somewhere else on the number line, like 9.5 to 10.5. A team favored by 2.5 wins outright in every scenario a 3.5-point favorite would, plus every scenario where the actual margin lands at exactly 3 — a genuinely common outcome — which a 3.5-point line would classify differently. The normal approximation described above is a useful and reasonably accurate simplification, but it’s still a simplification precisely because it treats every point of margin as equally likely near the mean, when the sport itself does not.
Why Spreads Move
A point spread posted days before kickoff is rarely identical to the spread at kickoff, and the movement between those two points reflects two distinct forces that are easy to conflate. The first is new information: a star quarterback listed as questionable becomes a confirmed inactive, a forecast turns from clear skies to a driving crosswind, a starting cornerback is added to the injury report — all of these plausibly change the actual expected margin, and a spread that doesn’t move in response to them would be a worse forecast than one that does. The second force is structural rather than informational: a spread’s original commercial function was to divide betting action as evenly as possible across both sides of a two-outcome market, and a line can shift in response to lopsided money even absent any new information about the two teams. Looking at a spread’s movement alone doesn’t reveal which force was responsible for a given shift — that requires knowing what became public information at the same time the number moved.
The Spread as a Public Forecast, Not Just a Betting Line
Set aside its commercial origin, and a closing point spread is one of the most information-dense public forecasts available for any NFL game. It implicitly aggregates a huge amount of dispersed knowledge — team strength, injury situations, coaching tendencies, situational factors — into a single number, continuously updated as new information arrives. That’s precisely why the pregame spread shows up as an input feature in win-probability models built independently of any wagering context: a model that only knows two teams’ win-loss records is working with much less information than one that also knows the market’s aggregated expected margin for this specific matchup. The spread’s usefulness as a forecasting input has nothing to do with what anyone does with it commercially — it’s a property of how much real-world information gets compressed into that one number before kickoff.
A live, in-game win-probability model typically doesn’t use the pregame spread as a constant input, either — it lets the spread’s influence fade as the game clock runs down and the actual score becomes the dominant signal. Early in a game, the current score says very little (a 7–0 lead after one drive is mostly noise), so a model that ignores the pregame forecast entirely is throwing away real information; late in a game, the score has accumulated enough information on its own that the pregame forecast stops mattering much by comparison. Weighting the spread’s influence by time remaining, rather than treating it as either always-relevant or irrelevant the moment the ball is kicked off, is a direct, practical consequence of what the spread actually represents: a forecast made before any of the game’s real information existed, whose value naturally erodes as that real information accumulates.
Implied Probability and the Built-In Margin
A quoted price and a “true” probability are not the same thing, and the gap between them is arithmetic, not opinion. American odds of −110 mean a bettor would need to risk 110 to win 100; converted to an implied probability, that’s — the break-even win rate at which a price of −110 would be exactly fair, with no margin built in.
The catch is that both sides of a two-outcome market are typically priced at −110 simultaneously, and — a sum that’s structurally impossible for two mutually exclusive, collectively exhaustive outcomes, which by definition must sum to exactly 100 percent. That extra 4.76 percentage points is the built-in margin (sometimes called the vig or overround): a markup embedded in how the two prices are set relative to each other, not a claim that either outcome is actually more or less likely than the raw numbers suggest. Recognizing that a pair of quoted prices always sums to more than 100 percent — and being able to compute by how much — is the arithmetic step that separates a quoted price from an actual probability estimate.
The same idea extends past a simple 50/50-style line. Any pair of prices covering two mutually exclusive, collectively exhaustive outcomes can be checked the same way: convert each price to an implied probability using the appropriate side of the formula above, add them, and whatever the total exceeds 100 percent by is the margin built into that specific pair. A lopsided price — a heavy favorite at one number against a live underdog at another — still carries a margin; it’s just not split evenly across the two sides the way a pair of matched −110 prices is. The arithmetic doesn’t change with how lopsided the two numbers are, only the size of the resulting margin and how unevenly it happens to be distributed between the two outcomes.
Reading a Spread as What It Is: a Statistical Forecast
None of the ideas above requires treating a point spread as anything other than what it mathematically is: a compressed, continuously updated forecast of an expected scoring margin, convertible to a win probability through a reasonably well-validated statistical model, and distinct from the priced-in margin that shows up once that forecast is expressed as a pair of moneylines. The spread converter runs the normal-approximation math from this page interactively, in both directions — spread to probability and price, and a pair of moneylines back to their implied probabilities and built-in margin. The win probability calculator shows the pregame spread being used exactly as described above, as one input feature inside a model that updates as an actual game unfolds. And the spread-to-probability explainer walks through the normal curve and the shaded win-probability area step by step, for anyone who wants to see where the 60, 70 and 86 percent figures above actually come from on the curve itself.
Frequently asked questions
What is a point spread in football?
A point spread is a number attached to a matchup that represents the expected scoring margin between two teams — for example, a favorite at −3.5 is forecast to win by roughly 3.5 points on average. It's a single-number summary of a probabilistic forecast, not a guarantee, and actual final margins scatter widely around it from game to game.
How is a point spread different from a win probability?
A spread expresses a forecast in points (expected margin); a win probability expresses the same underlying forecast as a percentage chance of winning outright. The two are connected mathematically through the scatter of real outcomes around the spread — converting one into the other requires a model of how much margins vary, not just the spread number itself.
What is the standard deviation used to convert a spread to a win probability?
Hal Stern's widely cited 1991 study in The American Statistician found NFL margins-versus-spread scattered with a standard deviation close to 13.86 points. Measured directly on nflverse's game data from 2000 through 2025 — 7,017 games with a recorded closing spread — the actual standard deviation of (final margin − spread) comes out to 13.2 points, close to Stern's historical figure.
Why are 3 and 7 such important numbers in NFL spreads?
Because a field goal is worth 3 points and a touchdown-plus-extra-point is worth 7, final margins land on exactly 3 or exactly 7 far more often than a smooth statistical curve would predict — in the 2000–2025 nflverse data, a 3-point final margin occurs in about 15 percent of all games, more than any other single margin, with 7 the next most common at about 9 percent. That clustering is why sportsbooks treat crossing those two numbers as a bigger move than a same-sized move elsewhere on the scale.
Why do point spreads move between when they're posted and kickoff?
Two forces move a spread: new public information (an injury report, a weather forecast, a lineup change) that shifts the actual expected margin, and the balance of money placed on each side, since a spread's original commercial purpose was to divide action roughly evenly across a two-sided market. Distinguishing which force moved a given line requires knowing what information became public at the same time — the movement alone doesn't say which cause was responsible.
Do football analysts actually use the point spread in their models?
Yes — independent of any wagering use, the closing spread is one of the most information-dense public forecasts available for a game, because it aggregates whatever the market collectively knows about both teams' rosters, injuries and form. Win-probability models commonly use the pregame spread as a feature precisely because it carries information a simple season-record or power-rating input doesn't fully capture on its own.
What is implied probability?
Implied probability is the win probability embedded in a price when that price is stated as American (moneyline) odds. A price of −110 implies a probability of 110 ÷ (110 + 100), or about 52.4 percent — the break-even probability at which that specific price would be fair with no margin attached.
What is the built-in margin, and why is it 4.76 percent on a standard −110/−110 line?
Each side of a −110/−110 line implies a 52.4 percent probability on its own; the two sides together imply 104.76 percent, which is impossible for two mutually exclusive outcomes to add up to. That extra 4.76 percentage points above 100 percent is the built-in margin — a structural feature of how the two prices are set relative to each other, not a probability statement about either team.
Does the normal approximation for spread-to-probability conversion actually match real outcomes?
Reasonably closely, though not perfectly. In the 2000–2025 nflverse data, favorites of roughly 6.5 to 7.5 points won about 74.3 percent of the time and favorites of roughly 13.5 to 14.5 points won about 85.3 percent of the time — both a few points above what the pure normal-curve estimate using a 13.2-point standard deviation would suggest, a gap most plausibly explained by the same 3-and-7 clustering that makes the underlying margin distribution not perfectly smooth.
Is a point spread the same thing as a prediction of the final score?
No. A spread implies an expected margin, not a specific final score — a −3.5 favorite is forecast to win by about 3.5 points on average across many similar games, not by exactly that amount in any single game. The standard deviation of roughly 13 points around that expected margin is itself the more informative number for understanding how much a single game's outcome can vary from the forecast.