Sports betting variance is enormous. A break-even bettor (52.4% win rate at -110, the math floor) will see weeks where they're 8-2 and weeks where they're 2-8 — and neither result tells them anything about their actual skill. Most retail bettors confuse variance for signal in both directions: assuming hot streaks are skill and cold streaks are bad luck. Both can be true. Both are usually neither.
The math of variance
For a binary bet at -110 (52.4% break-even), the standard deviation per bet is approximately 1 unit. Across N bets, the total standard deviation scales with √N. So:
- 10 bets: stdev ~3.2 units. A 4-unit profit is within one standard deviation — variance, not signal.
- 100 bets: stdev ~10 units. A 12-unit profit is within one stdev.
- 1,000 bets: stdev ~32 units. A 50-unit profit is meaningful but not conclusive.
- 10,000 bets: stdev ~100 units. A 200-unit profit is a strong signal.
The key takeaway: short-sample results are mostly variance. You need 1,000+ bets before you can confidently say your edge (or lack of edge) is real.
What 'real edge' means
A bettor with a 'real' 2% edge over the long run will see this distribution at -110:
| Sample | Expected profit (units) | 1-stdev range |
|---|---|---|
| 100 bets | +2 | -8 to +12 |
| 500 bets | +10 | -12 to +32 |
| 1,000 bets | +20 | -12 to +52 |
| 5,000 bets | +100 | +30 to +170 |
That's a real-edge bettor. They can lose money over 100, 500, even 1,000 bet samples — and still have edge. Conversely: a break-even bettor can win 30 units over 1,000 bets purely from variance.
The sample-size question for any specific decision
If you're trying to decide whether your NBA player props edge is real:
- 20 bets: completely uninterpretable. Variance dominates.
- 50 bets: barely informative. ±5 units of variance per typical bet level.
- 100 bets: weakly informative. Look at CLV alongside W/L.
- 200 bets: meaningfully informative. CLV trend reliable.
- 500+ bets: high signal. W/L starting to converge to true edge.
The accelerated signal: closing line value. CLV converges much faster than W/L. A 200-bet sample with consistent +1.5% CLV is a stronger signal of edge than 200 bets with 56% win rate.
Standard deviation in practice
Three formulas every serious bettor should internalize:
- Single-bet stdev: approximately 1 unit at -110. (Computed: √(p×(1-p)×payoff_squared))
- N-bet stdev: single-bet stdev × √N.
- Z-score: (observed result - expected result) / stdev. A z-score of 2 means the result is at the 97.5th percentile or beyond — strong but not conclusive evidence.
For the math averse: any result within ±2 standard deviations of expected is consistent with variance, not edge. ±3 stdev is strong evidence of either edge or bad luck.
The illusion of momentum
A common mental error: 'I've been hitting lately, my reads are sharp, I should size up.' That feeling is almost always variance, not skill change. Skill changes slowly. Variance changes constantly. Sizing up after a hot streak is the textbook way to lose money — you're betting bigger right when reverse-variance is most likely.
Discipline: keep stake size constant unless your bankroll has changed materially. Hot streak ≠ size up. Cold streak ≠ size down (that's just chasing in disguise).
How to evaluate any 'edge' claim
When someone (you, a tout, a Twitter account) claims to have edge, ask:
- What's the sample size? (less than 200 bets = weak)
- What's the win rate AND the CLV? (W/L without CLV is incomplete)
- What's the variance profile? (10-bet streaks at break-even are normal)
- Is the edge concentrated in one market or distributed? (concentrated = higher chance of luck)
- How was it tracked? (post-hoc results curation = unreliable)
If the answers are weak on any dimension, the 'edge' is probably variance.
The bigger frame
Most retail bettors who think they have edge don't. Most who think they're unlucky are actually playing slightly -EV. Honest tracking and CLV analysis are the only way to know which category you're in. The good news: if you genuinely have edge, even a small one, the math works in your favor over thousands of bets. The bad news: variance can hide edge for hundreds of bets, both ways.
Discipline rules
- Don't size up on hot streaks.
- Don't size down on cold streaks.
- Track CLV alongside W/L. The fast signal.
- Compute z-scores on your samples periodically. See where you are statistically.
- Plan bankroll for 10-bet losing streaks. They will happen.