Understanding poker variance and why it shapes your results
When you play poker, you’re competing against both opponents and mathematical noise. Variance is the natural fluctuation in results caused by short-term luck: the hands you run good or bad, the unavoidable ups and downs of card distribution, and the stochastic nature of other players’ actions. If you treat every session like a verdict on your skill, you’ll be misled; instead, you should treat results as noisy samples of an underlying skill level.
Understanding variance helps you do three practical things: protect your bankroll, avoid tilt after inevitable downswings, and make reliable progress assessments. In plain terms, variance answers questions such as “How many hands do I need before I can say my winrate changed?” and “How big a losing stretch should I expect even if I’m a winning player?”
Core metrics you must track: winrate, standard deviation, and sample size
Winrate and expected value (EV)
Your winrate (usually measured in big blinds per 100 hands, bb/100) is the primary estimate of your long-term profitability. Treat it as the average outcome per 100 hands you expect once variance averages out. Record hands played, stakes, effective stack sizes, and any major format differences (e.g., cash vs. tournaments) because winrate is only meaningful when compared across like-for-like samples.
Standard deviation: the size of typical swings
Standard deviation (SD) quantifies how widely individual session or-hand results spread around your mean. In poker, SD is typically large relative to the mean; that’s why short samples are dominated by variance. Knowing an empirical SD for your games lets you predict the scale of expected swings. Commonly, cash games have an SD measured in big blinds per 100 hands or per session; you can estimate it from your tracked hand-by-hand results.
Sample size: when small-sample noise dominates
Sample size is the lever that converts noisy data into reliable estimates. With a tiny sample (few hundred or a few thousand hands), your observed winrate can be wildly different from your true winrate. As sample size grows, the sampling distribution tightens and your estimate becomes more precise. Two practical rules of thumb:
- Expect extremely noisy results under ~10k hands; use cautious interpretation.
- By ~50k–100k hands your winrate estimate starts to become much more stable for most cash-game formats, assuming consistent stakes and players.
To make this actionable you’ll want to log: total hands, bb won/lost, session lengths, and the empirical SD per 100 hands. Those numbers feed directly into statistical formulas for confidence intervals and sample-size calculators—tools that let you quantify how certain you can be that an observed winrate reflects your true skill rather than luck.
How confidence levels connect to realistic stretch goals and planning
Confidence levels (commonly 90% or 95%) tell you how sure you can be that your estimated winrate falls within a given range of the true winrate. When you pair confidence intervals with your target earnings, you can define stretch goals that are both ambitious and statistically grounded—so you don’t set yourself up for disappointment by chasing noise.
In the next section you’ll learn the step-by-step calculations for estimating required sample sizes, building confidence intervals from your tracked metrics, and turning those numbers into concrete stretch goals.
Calculating the sample size you actually need for a target precision
Turn the abstract “I need more hands” into a concrete number by using your empirical standard deviation and the margin of error you’ll tolerate. Work with SD measured over 100-hand blocks (σ100), because winrates are normally reported in bb/100. The logic: if you split your play into k = N/100 independent 100-hand blocks, the standard error (SE) of your mean winrate (in bb/100) is σ100 divided by sqrt(k). Algebra gives a compact sample-size formula:
N = (z σ100 10 / E)^2
Where:
- z is the z-score for your confidence level (≈1.96 for 95%, ≈1.645 for 90%),
- σ100 is the empirical standard deviation of 100-hand block results (bb/100),
- E is the margin of error you’re willing to accept (bb/100),
- N is the number of hands required.
Example: suppose σ100 ≈ 80 bb/100 (a reasonable ballpark for many cash games) and you want ±5 bb/100 at 95% confidence. Plugging in: N = (1.96 80 10 / 5)^2 ≈ 98,300 hands. That’s why the oft-cited 50k–100k rule of thumb exists: with realistic SDs you need tens of thousands of hands to narrow margins to single-digit bb/100. If you relax to ±10 bb/100 or drop to 90% confidence, the required hands fall dramatically—use the formula to set a level that matches your goals and timeframe.
Building confidence intervals from your tracked metrics
Once you have total hands (N), observed winrate r̂ (bb/100), and σ100, computing a confidence interval is trivial and informative. First compute the SE:
SE = σ100 * 10 / sqrt(N)
Then the CI at your chosen confidence is:
r̂ ± z * SE
Worked example: r̂ = 5 bb/100, N = 50,000 hands, σ100 = 80, z = 1.96 (95%). SE = 80 10 / √50,000 ≈ 3.58 bb/100. Margin = 1.96 3.58 ≈ 7.0, so your 95% CI ≈ [-2, 12] bb/100. Practically this means that after 50k hands you cannot statistically distinguish a modest 5 bb/100 winrate from a break-even player; variance still dominates.
Pairs of practical tips:
- Estimate σ100 directly from your hand-history by chunking into 100-hand blocks and computing the standard deviation of their bb totals.
- Recompute σ100 periodically—different stakes, time controls, or formats can change variance and thus required sample sizes.
Turning intervals into realistic stretch goals and bankroll actions
Confidence intervals let you set stretch goals that are ambitious but honest. If your CI lower bound is comfortably above break-even, you can reasonably plan to move up stakes or increase session volume. If the lower bound overlaps zero, your “stretch” should be to collect more hands or tighten strategy rather than jump stakes.
Concrete ways to use the math:
- Stake decision rule: require the lower bound of a 90–95% CI to exceed your target minimum winrate before increasing stakes.
- Stretch targets: choose an ambitious target (e.g., +10 bb/100) and compute the N needed to show that target within a chosen margin—then set that N as a learning milestone.
- Bankroll sizing: plan for downswings consistent with σ100 and your chosen confidence (e.g., what’s the worst expected 1-month drawdown at 95%?).
Finally, use these numbers to create behavioral stretch goals: commit to a fixed sample after which you’ll reevaluate strategy, or to a hands-per-week cadence that reaches your required N in a realistic timeframe. By grounding ambitions in variance-aware math, you avoid chasing noise and make progress that compounds over the long run.
Putting variance to work: a practical checklist
Use the following steps to turn the theory into a repeatable routine that improves decisions and reduces emotional noise.
- Track every hand and session: log bb won/lost, stake, effective stack, and session length so you can compute empirical σ100 and true sample sizes.
- Estimate σ100 from your hand-history by chunking into 100-hand blocks and calculating their standard deviation; update this periodically as stakes or formats change.
- Pick a confidence level and margin of error that match your goals, then use N = (z σ100 10 / E)^2 to compute the hands needed for reliable inference.
- Set stake-move criteria: require the lower bound of a 90–95% CI to exceed your minimum target winrate before moving up.
- Create concrete stretch goals tied to hands played (e.g., “reach 75k hands and re-evaluate strategy”), not to short-term results.
- Plan bankroll and session volume using expected drawdowns from σ100 and your chosen confidence; don’t base decisions on single-session swings.
- Refresh your understanding of the underlying math when needed (see variance (statistics)) and use calculators to avoid algebra mistakes.
Final notes on disciplined progression
Treat variance-aware analysis as a tool for better habits more than a crystal ball. The real benefit comes from committing to data-driven rules—consistent tracking, pre-defined stretch milestones, and stake decisions anchored to confidence intervals—that keep you playing optimally through the inevitable noise. Patience, frequent recomputation of your metrics, and honest adherence to the rules you set are what convert statistical insight into lasting improvement.
Frequently Asked Questions
How many hands do I need to know if I’m a winning player?
There’s no single answer because it depends on your empirical σ100 and the margin-of-error you accept. With a common σ100 around 70–90 bb/100, achieving ±5 bb/100 at 95% confidence typically requires on the order of 50k–100k hands. Use the provided sample-size formula with your own σ100 and chosen confidence to get a precise target.
How do I calculate σ100 from my hand-history?
Chunk your tracked hands into 100-hand blocks, compute the total big blinds won in each block, then calculate the standard deviation of those block totals. That number is σ100. Recompute after enough new hands or when you change stakes/formats because variance often shifts with game conditions.
Should I move up in stakes after a big winning run?
No—make stake decisions based on statistically reliable evidence, not short-term runs. Require that the lower bound of a 90–95% confidence interval for your winrate exceeds your minimum acceptable winrate before moving up. If your current sample is too small, prioritize gathering hands and improving play rather than reacting to short-term variance.