Correlation & Independence
Why parlay math breaks when outcomes are correlated — and how to exploit this in same-game parlays.
Every parlay you've placed assumes each leg is independent of the others. The payout is calculated by multiplying the implied probabilities together. But when two outcomes in the same game share a common cause—and they almost always do—that multiplication becomes wrong. Understanding when and why independence breaks is the key to seeing through same-game parlay pricing.
Independent Events: The Foundation
Two events A and B are statistically independent if knowing the outcome of one tells you absolutely nothing about the other. The defining equation:
This holds ONLY when A and B have zero correlation. Cross-game parlays (Lakers tonight + Packers tomorrow) are approximately independent. Same-game outcomes almost never are.
Approximately independent events in gambling:
- Two different sporting events on different days
- Consecutive roulette spins (memoryless process)
- Hands at two separate poker tables
Standard cross-game parlays are priced using this multiplication rule, and the math is mostly valid. The sportsbook's edge comes from the vig on each leg, compounding multiplicatively across legs—not from a correlation mispricing.
When Independence Breaks: The Same-Game Problem
Independence collapses when outcomes share a common cause or are logically linked. Within the same game, this happens constantly:
Positive Correlation (+ρ)
Team wins + Star scores 25+: When the star goes off, the team is more likely to win. Game over + QB 300+ passing yards: High-scoring shootouts produce big passing numbers. Team covers −7 + Team wins: Every team that covers −7 also won the game.
Negative Correlation (−ρ)
Team wins by 20+ + Opponent scores 100+: Blowouts mean the other team scored less. Running back 120+ yards + Game under: A dominant rushing attack burns clock, suppressing scoring. Starting pitcher 10+ strikeouts + Team loses: Aces often pitch in low-scoring games their team still loses.
The Math: P(A ∩ B) When Events Are Correlated
For two binary events with correlation coefficient ρ, the true joint probability diverges from the naive product. A useful first-order approximation using the Bernoulli variance of each leg:
When ρ > 0 (positive correlation), the true joint probability is HIGHER than naive multiplication. When ρ < 0 (negative), it is LOWER. The adjustment term can be substantial even for moderate ρ.
Worked Example: NBA Same-Game Parlay
You want to parlay "Lakers win" (−150, implied 60%) with "LeBron scores 25+" (+150, implied 40%). These are positively correlated—when LeBron dominates, the team wins more often. Assume ρ ≈ 0.30.
The true probability is 31.2%, not 24%. A parlay priced at +316 (based on 24%) actually hits 31.2% of the time. If you take it at those odds, you are paying a massive hidden premium.
Warning
Impact Across Different Correlation Strengths
Lakers Win + LeBron 25+ (P(A) = 60%, P(B) = 40%)
| ρ | True Joint Prob | Naive Product | Difference |
|---|---|---|---|
| 0.00 | 24.0% | 24.0% | — |
| 0.10 | 26.4% | 24.0% | +2.4 pts |
| 0.20 | 28.8% | 24.0% | +4.8 pts |
| 0.30 | 31.2% | 24.0% | +7.2 pts |
| 0.50 | 36.0% | 24.0% | +12.0 pts |
| −0.30 | 16.8% | 24.0% | −7.2 pts |
Strategy Insight
Negative Correlation: The Hidden Trap
Negative correlation is even more dangerous for bettors. If you parlay "Team A wins by 20+" with "Game goes over 220," these are negatively correlated—blowouts often feature garbage time and running clocks, suppressing the total.
The naive product overestimates the true probability. You're getting odds based on a probability that's higher than reality. This is the worst possible parlay construction: you think you're getting a fair price, but correlation is silently working against you.
Where Correlation Creates Value
The flip side: when a sportsbook underestimates correlation, the parlay is underpriced. This happens when:
- Prop-game correlation is stronger than the model assumes — Some player performances are more tightly linked to team outcomes than books model
- Weather/game-script effects are missed — Rain games suppress passing stats AND totals; this correlation may not be fully captured
- Newly correlated markets — When books introduce new SGP combinations, their correlation models may be less refined than established markets
Good to Know
Our SGP Optimizer helps evaluate same-game parlay pricing, and the Props Tool shows player prop pricing across sportsbooks. For deeper analysis of SGP vig, see our SGP Correlation & Pricing article. Build smarter parlays with the Parlay Calculator.
Sources & References
- Definition of statistical independence P(A ∩ B) = P(A) × P(B) and correlation coefficient ρ — standard probability theory, independently verifiable from any statistics textbook.
- Tetrachoric correlation and the binary joint probability approximation — Pearson, K. (1901), "Mathematical Contributions to the Theory of Evolution," Phil. Trans. Royal Society. The first-order formula P(A∩B) ≈ P(A)×P(B) + ρ√(Var(A)×Var(B)) applies to bivariate Bernoulli variables.
- Same-game parlay correlation effects in sports betting markets — sportsbooks use proprietary correlation models to adjust SGP pricing; the general principle that correlated legs alter true parlay odds is well-documented in advantage betting literature.
- Peta, J. & Kopriva, F. (2023). "Same-Game Parlay Pricing and the Correlation Problem." UNLV Gaming Research & Review Journal. Analysis of how correlation adjustments affect SGP expected value.
Mathematical claims are independently verifiable. BonusBell platform analysis reflects data from 220+ tracked platforms as of March 2026.
Key Takeaways
- 1Independent events satisfy P(A ∩ B) = P(A) × P(B) — cross-game parlays are approximately independent, same-game outcomes almost never are
- 2Positive correlation means the SGP hits MORE often than naive odds suggest — but sportsbooks adjust for this and add vig on top
- 3Negative correlation is worse: the true probability is LOWER than the naive product, so you overpay without realizing it
- 4Even moderate correlation (ρ = 0.30) swings true probability by 7+ percentage points per leg pair, compounding across multi-leg SGPs
- 5Value exists when sportsbooks underestimate correlation — look for prop-game linkages, weather effects, and newly introduced SGP markets