Every bettor with a winning stretch asks the same question eventually: is this actually skill, or am I just running hot? A p-value gives you a rigorous, honest answer instead of a gut feeling based on a streak that felt good.
Loading calculator...
The P-Value Calculator tests whether your observed results — a win rate or an average value like CLV — are statistically distinguishable from having no real edge at all.
It covers two common bettor scenarios: testing a win percentage against a hypothesized fair rate, and testing an average value like CLV or bb/100 against a hypothesized mean of zero.
📊 How to Use the P-Value Calculator
Choose Win % Test if you’re evaluating a hit rate, like a system’s ATS record or a prop-picking process, against some fair baseline percentage.
One-tailed tests answer “is my edge greater than zero,” which is what most bettors actually want to know; two-tailed tests answer the broader “is my result different from zero in either direction.”
Choose Average Value Test if you’re evaluating a numeric average like your Closing Line Value or bb/100 winrate against a hypothesized mean, almost always zero for a “no edge” baseline.
🔢 Calculator Fields Explained
Observed Wins – the actual number of wins in your sample, for the Win % Test.
Total Sample Size – the total number of bets, hands, or trials in your sample.
Hypothesized Fair Win % – the baseline win rate you’re testing against, often 50% or the implied no-vig probability of your bets.
Sample Mean – your average result per unit, such as average CLV in cents or average bb/100, for the Average Value Test.
Sample Standard Deviation – how much your individual results vary around that mean.
Sample Size – the number of individual data points in your average-value sample.
Hypothesized Mean – the baseline you’re testing against, almost always 0 for “no real edge.”
💰 Understanding the Results
| Result Field | What It Tells You |
|---|---|
| P-Value | The probability of seeing a result this extreme if you truly had no edge at all |
| Z-Statistic / T-Statistic | How many standard errors your observed result sits away from the hypothesized baseline |
| Significant at 95% Confidence | Whether your p-value clears the common 0.05 significance threshold |
| Significant at 99% Confidence | Whether your p-value clears the stricter 0.01 significance threshold |
The P-Value itself is the number to focus on: smaller values mean your result would be increasingly unlikely to occur by chance alone if you truly had no real edge.
A statistically significant p-value tells you an edge likely exists, but it says nothing about how large that edge actually is — a tiny, barely-profitable edge can still be statistically significant with enough sample size.
A high p-value does not prove you have no edge — it often just means your sample size is too small to distinguish a real (but modest) edge from pure variance.
📐 Calculation Formulas
| Test Type | Formula | Used For |
|---|---|---|
| One-Proportion Z-Test | (observed % − hypothesized %) ÷ standard error | Win rate vs. a fair baseline percentage |
| One-Sample T-Test | (sample mean − hypothesized mean) ÷ standard error | Average CLV or bb/100 vs. a hypothesized mean |
| Standard Error (Proportion) | √(p × (1−p) ÷ n) | Used inside the Z-test |
| Standard Error (Mean) | sample stdev ÷ √n | Used inside the T-test |
The T-Test is used instead of the Z-Test for averages because the sample standard deviation itself is an estimate, which the t-distribution accounts for more conservatively, especially at smaller sample sizes.
As sample size grows large, the t-distribution converges toward the normal distribution, so the practical difference between the two tests shrinks at big sample sizes.
📝 Practical Examples
Example 1 – Betting System Win Rate: 55 wins out of 100 bets, hypothesized fair rate 50%, one-tailed. Z-statistic ≈ 1.00, p-value ≈ 0.159 — not statistically significant at 100 bets.
Example 2 – Same Win Rate, Larger Sample: 550 wins out of 1,000 bets (still 55%), same 50% baseline, one-tailed. Z-statistic ≈ 3.16, p-value ≈ 0.0008 — now clearly significant.
Examples 1 and 2 show the same exact 55% win rate producing wildly different statistical conclusions purely based on sample size — this is the core lesson of p-value testing for bettors.
Example 3 – Average CLV Test: Sample mean CLV of +1.5%, standard deviation 8%, sample size 300, hypothesized mean 0, one-tailed. T-statistic ≈ 3.25, p-value < 0.001 — strong evidence of real positive CLV.
Example 4 – Small Sample, Big Average: Sample mean bb/100 of +8, standard deviation 90, sample size 40, one-tailed. Despite an impressive-looking +8 bb/100 winrate, the small sample size keeps the p-value well above 0.05, meaning this result alone isn’t yet statistically distinguishable from variance.
💡 Tips & Best Practices
Use a one-tailed test when your real question is “do I have a positive edge,” since that matches what most bettors actually want to know.
Don’t treat a single p-value calculation as the final word — recheck as your sample grows, since early results are the least reliable.
Choose your hypothesized baseline carefully; for spread bets, a genuinely fair coin-flip line is close to 50%, but for other markets the no-vig implied probability is a better baseline.
Running this calculation periodically as your sample grows is far more informative than checking it once and treating that single result as permanent proof either way.
Remember that statistical significance and practical significance are different things — a tiny but real edge might be statistically significant yet not worth the effort or bankroll risk to pursue.
Combine this with a variance calculator to understand not just whether your edge is likely real, but how wide your realistic range of outcomes still is going forward.
- Record your hypothesized baseline before running the test, not after seeing a favorable result
- Re-run the test at natural checkpoints (every 100, 500, 1000 bets) rather than whenever a streak feels notable
⚠️ Common Mistakes to Avoid
Testing After Cherry-Picking a Hot Streak
Running a p-value test only on your best recent stretch, rather than your full sample, badly inflates the apparent significance.
Selecting only your winning stretch before running a p-value test is a classic statistical error that manufactures false significance out of pure variance.
Always test your complete sample, not a cherry-picked subset chosen because it looks good.
Confusing Statistical Significance With Edge Size
A tiny edge can become statistically significant with a large enough sample, leading players to overstate how meaningful the actual edge is.
Treating a barely-significant p-value as proof of a large, exploitable edge conflates two genuinely different questions: whether an edge exists, and how big it is.
Always look at both the p-value and the actual size of the observed edge together.
Drawing Conclusions From Too Small a Sample
Players frequently run this test on 20-30 bets and treat a non-significant result as proof they have no edge.
A genuinely real, modest edge routinely fails to reach significance in small samples, which is a sample-size problem, not evidence the edge doesn’t exist.
Picking an Arbitrary Hypothesized Baseline
Using a generic 50% baseline for markets that aren’t actually fair coin-flips distorts the entire test’s conclusion.
Match your hypothesized percentage or mean to the actual fair or no-vig baseline for the specific market you’re testing.
🎯 When to Use This Calculator
Use this calculator whenever you want to check whether a winning (or losing) stretch is likely to reflect a real edge, rather than relying on gut feeling about a hot or cold run.
Data-driven bettors often treat statistical significance testing as a required checkpoint before scaling up stakes on any new system or strategy.
It’s also useful for evaluating a new betting system or handicapping process objectively, before committing serious bankroll to it.
🔗 Related Calculators
Standard Deviation Calculator, Confidence Interval Calculator, Poker Variance Calculator, CLV Calculator, Z-Score Calculator
📖 Glossary
P-Value – the probability of observing a result this extreme (or more) if the null hypothesis of no real edge were true.
Null Hypothesis – the baseline assumption being tested against, typically “no real edge exists.”
Statistical Significance – a result unlikely enough under the null hypothesis to be treated as real evidence of an effect.
One-Tailed Test – a test checking for an effect in one specific direction, such as a positive edge.
Two-Tailed Test – a test checking for an effect in either direction from the baseline.
Z-Statistic – a standardized score measuring how many standard errors a proportion result sits from the hypothesized value.
T-Statistic – a standardized score measuring how many standard errors a mean result sits from the hypothesized value, adjusted for sample-based variance estimation.
Standard Error – the estimated variability of a sample statistic around the true population value.
Sample Size – the number of observations used in the test; larger samples produce more reliable p-values.
CLV (Closing Line Value) – the difference between your bet’s odds at placement and the odds at market close, a common sharp-bettor skill metric.
❓ Frequently Asked Questions
What p-value counts as “statistically significant”?
The most common threshold is 0.05, meaning a result would occur by pure chance less than 5% of the time if there were truly no edge; 0.01 is a stricter, more conservative threshold.
Neither threshold is a magic number — some bettors prefer the stricter 0.01 standard specifically because sports betting samples are prone to hidden biases that inflate apparent significance.
Why does the same win rate give different p-values at different sample sizes?
Statistical confidence grows with sample size, so an identical percentage becomes increasingly unlikely to be pure chance as more data confirms it.
A 55% win rate over 100 bets and the same 55% over 1,000 bets tell very different stories, even though the raw percentage never changed.
This is why sample size, not just the raw percentage, is central to any meaningful edge evaluation.
Should I use a one-tailed or two-tailed test?
Use one-tailed when you specifically want to know if you have a positive edge, which is the most common bettor question; use two-tailed for a more general “is this different from baseline at all.”
One-tailed tests produce smaller p-values for the same data, so be consistent about which you’re using when comparing results over time.
Can a non-significant p-value mean I have no edge?
Not necessarily — it often simply means your current sample size isn’t yet large enough to distinguish a real, modest edge from ordinary variance.
A genuinely profitable long-term strategy can easily show a non-significant p-value for hundreds or even a few thousand trials, depending on the edge size.
What sample size do I need for a reliable p-value test?
It depends heavily on how large your true edge is — smaller edges require dramatically larger samples to reach statistical significance than larger, more obvious edges.
As a rough guide, many sports betting edge evaluations need at least several hundred to a few thousand bets before small edges become statistically distinguishable from variance.
⚖️ Legal Disclaimer
This calculator is provided for informational and educational purposes only. Statistical tests use standard approximations and do not guarantee the existence or size of any betting edge. Please gamble responsibly.
This is exactly the framework I’ve been using to evaluate my poker results over the last year and a half. The thing most grinders miss is that sample size requirements scale with the variance of what you’re measuring. When I’m tracking my bb/100 winrate, I need way more hands than when I’m tracking a simple binary outcome like ATS records. The t-test approach here accounts for that uncertainty in the sample standard deviation, which matters at smaller sample sizes where the t-distribution has fatter tails. I ran my 2023 cash game data through this same methodology: 47k hands, average win of 1.2 bb/100 with a standard deviation around 8.5 bb. The p-value came back at 0.038, just under the 0.05 threshold at 95% confidence. Technically significant, but barely. That modest edge is exactly what I’d expect from a solid regular playing 200nl-500nl mixed games against a fairly tough player pool. The real value here is tempering expectations. A lot of grinders see one good month and assume they’ve cracked the code when they’re actually just running hot. This calculator gives you the statistical rigor to separate signal from noise. Also worth noting: this same logic applies to evaluating which sites have the softest games or which times of day produce the best field composition. You need enough sample data before you can claim one venue outperforms another. Too many players jump rooms based on a 20-session downswing.
Regarding your 47k-hand sample and that 0.038 p-value result: you’ve highlighted something critical that many bettors and grinders overlook. The margin between statistical significance and practical insignificance is exactly where most players operate. Your 1.2 bb/100 edge with that standard deviation puts you in a zone where variance is still the dominant force over your next 10k hands, even though the historical data supports real edge existence. One nuance worth adding: the calculator uses a one-sample t-test assuming your individual hand results are independent, which is generally valid in poker, but your win/loss sequences do contain some autocorrelation depending on your table selection and position dynamics. That said, the independence assumption holds well enough for bankroll planning purposes. Your point about room evaluation is particularly sharp. Testing whether 200nl at Site A beats 200nl at Site B requires running the same statistical test on win rates across comparable sample sizes at each site, controlling for field composition variables. Most players conflate a good downswing recovery with actual venue superiority. The p-value approach forces you to account for the exact scenario you mentioned: that 20-session downswing is statistically predictable variance, not evidence of a broken game or bad coaching.