Variance and Sample Size in Sports Betting in 2026: How Many Bets Before a Result Means Anything

The quick answer

• Swing per bet: a $100 bet at -110 has a standard deviation of about $95.

• After 100 bets: a bettor who wins 54% of -110 bets expects to be up $309, with a standard deviation of $951. They finish behind 38.1% of the time.

• Proof of an edge: at that 3.09% edge, expected profit doesn't clear two standard deviations until about 3,790 bets.

• Luck with no edge: a break-even bettor at -110 wins 57 or more of 100 bets 20.5% of the time, which is an $882 month.

• Drawdowns: over 1,000 bets at 54%, the median worst peak-to-trough drop in our simulation was $2,455, about 25 stakes.

• Price: the same 54% read at an even-money price with no fee (Novig charges none on pre-game straights) is an 8% edge and clears two standard deviations in about 621 bets.

A few hundred bets won't tell you much about whether you have an edge. Each bet swings by roughly its own stake, and a small edge adds only a few dollars per bet, so the noise dominates until the sample gets large. How large depends on the ratio between the two, and the price you pay moves that ratio as much as your handicapping does.

What variance means for a bet

A straight bet either wins the payout or loses the stake. For a bet that wins W with probability p and loses L otherwise, the standard deviation of the result is (W + L) x √(p x (1 - p)).

At -110, a $100 bet wins $90.91 and loses $100, so W + L is $190.91. At a 54% win rate, √(0.54 x 0.46) is 0.4984, which puts the standard deviation at $95.15. The expected profit is 0.54 x $90.91 - 0.46 x $100 = $3.09.

So each bet carries $3.09 of edge and $95.15 of noise. Over n bets the expected profit grows with n and the standard deviation with the square root of n, the standard result for a binomial count, which is why 100 bets leave $309 of expected profit sitting inside $951 of swing.

How we checked

Every figure below rests on stated assumptions: $100 flat stakes, -110 on every bet, independent bets, and a true win probability of 54% against a break-even of 52.38%. That's a 3.09% return on stake. The 54% is an assumption for illustration and doesn't describe any particular bettor.

The binomial mean and standard deviation formulas come from the NIST/SEMATECH e-Handbook of Statistical Methods. We computed the chances of finishing ahead or behind exactly from the binomial distribution, so there's no normal approximation in them. Drawdowns come from 20,000 simulated sequences of 1,000 bets each; the eight-loss streak figure was also computed exactly and matched the simulation within 0.3 percentage points. Novig's fee schedule was opened on 7 October 2026.

How many bets before a result means anything?

The table follows a 54% bettor at -110 across four sample sizes.

After 100 bets, the 54% bettor's record lands between 49% and 59% about two-thirds of the time. A 59-41 stretch and a 49-51 stretch can come from the same person with the same skill. At 1,000 bets the band narrows to 52.4% to 55.6%, and only the bottom edge of it touches break-even.

The usual line for "probably not luck" is two standard deviations, and you can solve for it directly: n = (2 x SD / edge per bet)². With $95.15 of swing and $3.09 of edge per bet, that's (2 x 95.15 / 3.09)², or about 3,790 bets. A bettor placing 10 bets a day needs just over a year to get there.

Why a profitable month can happen with no edge

A bettor who wins exactly 52.38% of -110 bets has an expected profit of zero. Over 100 bets they finish ahead 49.1% of the time.

They also go 57-43 or better 20.5% of the time. At -110, 57 wins and 43 losses is 57 x $90.91 - 43 x $100, an $882 profit. One break-even bettor in five can post that month, and nothing in the record separates them from someone with a 3% edge.

How a winning bettor can lose a month

The 54% bettor finishes 100 bets behind 38.1% of the time. Over 500 bets it's still 22.3%, and over 1,000 bets it's 14.8%, about one run in seven.

Drawdowns and losing streaks

Drawdown is the drop from your highest balance to the lowest point after it. In 20,000 simulated runs of 1,000 bets at 54% and -110, the median worst drawdown was $2,455 and the 90th percentile was $4,082. At $100 a bet, one run in 10 gives back 41 stakes at some point while still holding a 3.09% edge.

Losing streaks run long even with an edge. Over 1,000 bets at 54%, the chance of at least one streak of eight straight losses is 66.2%, and of 10 or more is 20.4%. The median longest streak in the simulation was eight and the 90th percentile was 11.

A drawdown of that size fits a 3% edge and fits no edge at all, so the balance alone can't tell you which one you have. Raising stakes to win it back increases the swing on every later bet and leaves the probability estimate behind them unchanged.

How the price changes the sample you need

The edge in the calculation is your probability minus the probability the price implies. The vig sits in that gap, so cutting it raises the edge without any change in your handicapping.

Assume a coin-flip game whose fair price is 50%, and a bettor who rates one side at 54%. At -110 that's the 3.09% edge above. At an even-money price with no fee, a $100 bet wins $100 or loses $100, so the edge is 0.54 x $100 - 0.46 x $100 = $8.00 per bet, with a standard deviation of $99.68.

The swing per bet barely moves while the edge more than doubles, which cuts the sample needed to separate it from luck to about a sixth.

Novig's prices are set by users trading against each other, and its fee page says "There's no vig or juice built into our prices." Pre-game straight trades carry no fee for either side. Live straights carry a taker fee on a 0.03 coefficient, or 0.06 on live college football spreads and totals, while futures take 0.06 (golf and tennis futures carry no futures fee as of 10 September 2026) and parlays 0.10, with the parlay fee already inside the quoted price. Makers pay nothing on any trade type. The no-fee row above applies to pre-game straights only.

What to look for

• Edge against swing: divide your edge per bet by its standard deviation before judging any record.

• Sample size: treat a record under a few hundred bets as noise unless the edge per bet is large.

• Win-rate band: after 100 bets, your observed win rate can sit five points either side of your true one.

• Drawdown budget: plan for a peak-to-trough drop of 25 to 40 stakes over 1,000 bets at a 3% edge.

• Price: compare the break-even rate at each venue before comparing records, since a lower break-even turns the same read into a bigger edge. At an even-money price with no fee, the break-even is 50% instead of 52.38%.

Variance FAQ

How many bets do you need to know if you're a winning bettor?

You'll need about 3,790 bets at a 3.09% edge on -110, if the test is expected profit two standard deviations clear of zero. An 8% edge needs about 621.

What is the standard deviation of a -110 bet?

About $95 per $100 bet at any win rate near 50%: $95.35 at the 52.38% break-even and $95.15 at 54%. Over n bets, multiply by the square root of n.

Can a winning bettor lose over 100 bets?

Yes. A bettor who wins 54% of -110 bets finishes 100 bets behind 38.1% of the time, and 500 bets behind 22.3% of the time.

How long can a losing streak last for a winning bettor?

Over 1,000 bets at a 54% win rate, the chance of at least one run of eight straight losses is 66.2%. There's a 20.4% chance of a run of 10 or more.

Does lower vig reduce variance?

It barely changes the swing per bet, which goes from $95.15 at -110 to $99.68 at even money for a 54% bettor. It raises the edge from $3.09 to $8.00 per $100, which cuts the bets needed to separate skill from luck from about 3,790 to about 621.

Sources

Opened 7 October 2026.

NIST/SEMATECH e-Handbook of Statistical Methods, "Binomial Distribution" (mean np, standard deviation √(np(1 - p))): itl.nist.gov/div898/handbook/eda/section3/eda366i.htm

Novig fee schedule: support.novig.com/en/articles/16195057-fees-on-novig

All other figures are calculations from the assumptions stated in "How we checked": exact binomial probabilities, an exact streak calculation and a 20,000-run simulation.