The Lab · The math · updated 2026-08-02
Expected value in sports betting — why the biggest edge is always a longshot
Expected value is one line of arithmetic, and the price term inside it is why a longshot tops almost every +EV screen, including ours.
On August 2 the biggest edge on our board belonged to the Nationals at +250 — a team the model gave a 29.09% chance of winning.
Read that twice. The best number on the board that day sat on a team the model expected to lose 71% of the time. Both things are true at once, and the reason is a single line of arithmetic that almost nobody who uses the phrase "+EV" ever writes down.
Expected value is one identity
Expected value per unit staked is what you win times how often, minus what you lose times how often:
EV = p × b − (1 − p)
where p is your probability and b is the profit per unit if the bet wins (+250 pays 2.50; −110 pays 100/110 = 0.909).
That form is correct and tells you nothing. Rearrange it and the whole subject falls open:
EV = (p − break-even probability) × (1 + b)
The two are the same expression. Break-even is 1/(1 + b), so (p − 1/(1+b)) × (1+b) = p(1+b) − 1 = p × b − (1 − p). Same thing, different lesson.
Now there are two factors, and they do very different jobs:
- (p − break-even) — how far your number sits from the threshold the price already demands. This is the disagreement. This is the hard part, the part a model exists to produce. (If break-even prices are new, the vig and favorites vs. underdogs cover where that 52.4% comes from.)
- (1 + b) — a multiplier that comes from the price and nothing else. No opinion, no model, no information. Just the number on the screen.
Work the Nationals through it. At +250, b = 2.50, so break-even is 1/3.50 = 28.57%. The model said 29.09%. The disagreement is half of one percentage point — 0.52 points. Multiply by 3.50 and you get +1.82% EV.
Half a point of opinion. A 3.5× multiplier. That's the whole "edge."
Three edges from one board
Every game our model flagged with positive EV that day, in full:
| Game | Model p | Price | Break-even | Gap | × (1 + b) | EV |
|---|---|---|---|---|---|---|
| Nationals | 29.09% | +250 | 28.57% | +0.52 pt | 3.50 | +1.83% |
| Giants | 35.61% | +182 | 35.46% | +0.15 pt | 2.82 | +0.43% |
| Padres | 46.61% | +115 | 46.51% | +0.09 pt | 2.15 | +0.20% |
(The EV column is the board's, computed at full precision. Recomputing from the rounded columns shifts the last digit — 29.0943% is what produced +1.83%.)
Look at the Gap column. Nine hundredths of a point separates our number from the market's on the Padres. On the Nationals it's half a point. These are not disagreements. These are rounding errors with a press release.
The same disagreement, priced four ways
Hold the opinion completely fixed — 0.52 points, exactly the Nationals gap — and change only the price:
| Price | 1 + b | EV |
|---|---|---|
| +250 | 3.50 | +1.82% |
| +150 | 2.50 | +1.30% |
| −110 | 1.91 | +0.99% |
| −250 | 1.40 | +0.73% |
Identical conviction. Same model, same certainty, same half-point of disagreement. The reported edge swings by 2.5× purely on where the price sits.
Which produces a rule with no exceptions: any leaderboard sorted by EV is sorted by price length as much as by conviction. A longshot will sit on top of it almost every day, because a longshot's multiplier is two to four times a favorite's. That's arithmetic, not insight. It is also exactly why chasing the biggest number on any +EV screen — a tout's, a tool's, ours — walks you straight into longshots without ever telling you it's doing it.
What we got wrong
We called that column an edge. On August 2, it wasn't one.
Our MLB module had no live signal that day. With nothing adjusting the baseline, the model's probability is the de-vigged consensus of the same sportsbooks quoting the game. So a "+1.83% edge" on the Nationals was not our model disagreeing with the market. It was one book (hardrockbet) paying 0.52 points more than the other seven implied — measured against an average that included that book's own price.
That is a line-shopping observation. It is worth exactly what line shopping is worth, which is real but small, and it is not a prediction about a baseball game. Labeling it "edge" implies a disagreement that did not exist, and the word did more work than the math supported.
Why the floor exists
Our MLB policy requires EV ≥ 3% before anything publishes. The Nationals at 1.83% did not clear it. Neither did the other two. Nothing was published that day — which is the system working, not failing.
The floor is there because of the Gap column. De-vigging eight books that routinely disagree with each other by more than half a point produces a consensus with error bars wider than 0.52 points. Anything under 3% is inside the noise of the estimate. Publishing it would be reporting the measurement error as a finding.
What this doesn't tell you
The multiplier amplifies mistakes identically. This is the part the +EV framing hides. If the model's 29.09% is wrong by half a point in the other direction — 28.05% against a 28.57% break-even — that same 3.50 multiplier turns it into −1.82%. Long prices magnify error exactly as efficiently as they magnify edge, and you don't get to know in advance which one you're holding.
EV assumes p is right, and p is an estimate. Every number in the first column is a guess with error attached. When the error bar is wider than the gap, EV is measuring your uncertainty and calling it opportunity.
Positive EV still loses most of the time here. A genuinely +1.8% bet on the Nationals loses 71 times in 100. Nothing in this article makes a longshot likely.
Three games on one day is not a sample. The arithmetic says the top of an edge board should tend toward plus money, because the multiplier is larger there. It does not say how often, and we haven't counted it across our own boards. A tendency is not a frequency, and we'd rather leave that gap open than fill it with a number we haven't measured.
And the 3% floor is a judgment, not a proof. It's set where we believe the de-vigging noise ends; we have not demonstrated that 3% is the right cut rather than 2% or 4%. When we have enough graded picks to test it, that number will move or stay based on the data, in public.
The takeaway isn't a bet. It's a habit: whenever you see an edge percentage, ask what the gap was before the price multiplied it. Ours was half a point.
Model output is informational and entertainment content, not betting or financial advice. If you bet, bet what you can afford to lose.
Log what you actually bet — units and dollars — and see it graded beside a model that publishes every call it makes, wins and losses alike. Free, no card.
Start a free tracker