Image by Author
The previous four posts in this series built a statistical toolkit for evaluating decisions at every stage of a hand. You can now calculate the probability of improving a hand using outs and the Rule of 2 and 4, determine whether calling is justified using pot odds and expected value, assess starting hands by their frequencies and positional value, and estimate equity against a range of opponent holdings using combinatorics. Taken together, these tools tell you how to make correct decisions.
What they can’t do is guarantee short-term results. That gap between correct decisions and actual outcomes is explained by variance. Understanding it, and structuring your approach to account for it, is the final statistical piece of a complete picture of Texas Hold ‘Em.
Key Definitions
Variance is a statistical measure of how spread out a set of results is around the expected value. High variance means individual outcomes can deviate significantly from the average, even over a large number of hands. Low variance means results cluster tightly around what’s expected.
Standard deviation is the square root of variance and provides a more intuitive unit: the typical size of a single result’s deviation from the mean. In poker, a player’s standard deviation per 100 hands quantifies how much their results can swing in either direction over that sample.
The law of large numbers states that as the number of independent trials increases, the observed average converges toward the true expected value. In poker, this means correct decisions eventually produce results that reflect their positive expected value, but “eventually” can require tens of thousands of hands.
Bankroll refers to the total amount set aside specifically for poker, separate from living expenses and other financial commitments. Bankroll management is the practice of sizing individual game stakes relative to the total bankroll so that variance can’t eliminate it before the law of large numbers has enough hands to operate.
Risk of ruin is the probability of losing an entire bankroll before recovering to a profit, given a specific win rate and variance level. It decreases as bankroll size increases relative to the stakes being played.
Variance in Practice
Even a player making consistently positive EV decisions will experience significant losing stretches. This isn’t a failure of the decision-making framework introduced in this series. It’s the expected behavior of any statistical process with variance.
Consider a player who holds a 55% chance of winning any given significant decision, slightly better than a coin flip. Over 10 decisions, winning 5.5 times is impossible, and the actual result might be 3 wins or 8 wins. Over 1,000 decisions, the result will be far closer to 550. Over 10,000, closer still.
Poker works the same way. A session of 200 hands is a small sample. A month of play might represent 5,000 to 10,000 hands. Even at that volume, running 2 standard deviations below expectation is a statistically normal event, roughly a 1-in-20 occurrence, not proof that a strategy is wrong.
The implication is important: short-term outcomes are poor signals of decision quality. The correct evaluation framework is the one built across this series, not the results of any single session.
The Law of Large Numbers and Convergence
The law of large numbers provides both reassurance and a caution.
The reassurance: if you consistently make positive EV decisions, results will converge toward that positive expectation over a large enough sample. The frameworks in this series are worth applying because they produce winning results across a sufficient number of hands.
The caution: the sample required for convergence is larger than most players expect. A winning player running below expectation for 10,000 hands is not in unusual statistical territory. Research on professional poker results suggests that meaningful conclusions about a player’s true win rate require samples of 50,000 to 100,000 hands at minimum.
That doesn’t mean you can’t learn from shorter samples. It means being careful about conclusions drawn from them, particularly conclusions about whether a strategy or decision type is actually profitable.
Bankroll Management
Bankroll management is applied variance control. Its purpose is to ensure that normal downswings, which are statistically inevitable, don’t eliminate your ability to keep playing before the long run arrives.
The general principle: the more variance in the format you play, the larger your bankroll needs to be relative to the stakes. Cash games carry less variance per session than tournaments, because tournaments have binary outcomes (win a share of the prize pool or lose the buy-in entirely), but both formats require meaningful buffers.
Widely used guidelines suggest:
- Cash games: 20 to 30 buy-ins as a minimum for the stakes being played
- Tournaments: 50 to 100 buy-ins, reflecting the higher variance of the format
These figures represent the bankroll size at which the risk of ruin drops to an acceptable level given typical win rates and standard deviations. Moving down in stakes when a bankroll is depleted, rather than continuing at the current level, is the behavioral expression of bankroll management. It preserves the ability to keep playing through downswings until variance has had the sample size it needs to even out.
Practical Examples
Example 1: A Normal Losing Stretch
A player wins an average of 5 big blinds per 100 hands, a healthy cash game win rate, with a standard deviation of 100 big blinds per 100 hands. Over 1,000 hands, they expect to win 50 big blinds. But one standard deviation represents a realistic swing of roughly minus 50 to plus 150 big blinds over that stretch. A losing 1,000-hand session is well within normal statistical range for this player.
Example 2: Risk of Ruin with Different Bankroll Sizes
At $1/$2 cash games with a $200 buy-in, a player with 10 buy-ins ($2,000 total) faces a substantially higher risk of ruin than a player with 30 buy-ins ($6,000). Both have the same win rate. The difference is purely the buffer against variance. The player with 10 buy-ins can be eliminated by a single normal downswing. The player with 30 cannot.
Example 3: Evaluating Results Over a Short Sample
After 5,000 hands running significantly below expectation, a player questions whether their strategy is flawed. Statistically, 5,000 hands is a modest sample. A 2-standard-deviation downswing over that period is roughly a 1-in-20 occurrence. It’s worth reviewing specific decisions for EV accuracy. It’s not evidence that a statistically sound strategy is broken.
Common Misconceptions
Variance is not bad luck. It’s a mathematical property of any probabilistic process. Labeling losing stretches “bad luck” and winning stretches “skill” is a cognitive bias that misreads normal statistical variation as something meaningful.
The long run is further than most players think. Many players overestimate how quickly results converge toward expectation. A month of play is not the long run. A year might not be either. The law of large numbers requires a genuinely large number.
Bankroll management isn’t conservative play. Moving down in stakes when a bankroll shrinks is statistically rational, not timid. A player who continues at stakes their bankroll can’t support is taking on a risk of ruin that no win rate can fully offset.
Connecting to the Broader Game
This final post completes the series by placing individual hand decisions within their long-run statistical context. The probability framework from the first post, the EV analysis from the second, the pre-flop frequencies from the third, and the range-based equity from the fourth all contribute to a win rate. Variance explains why that win rate produces uneven short-term results. Bankroll management ensures you’re still at the table when the law of large numbers has had enough hands to do its work.
Conclusion
Variance, the law of large numbers, and bankroll management form the statistical backbone of long-term poker thinking. They explain why correct decisions don’t always produce winning sessions, and why the size of your bankroll matters as much as the quality of your strategy. Together with the frameworks from the first four posts in this series, they complete a statistical picture of Texas Hold ‘Em that extends well beyond individual hands and into the mathematics of the game over time.
