Basic Probabilities and True Odds in Double-Zero Roulette

An American roulette wheel contains 38 pockets: the numbers 1–36 plus 0 and 00. For any single-number (straight-up) bet the probability of winning is 1/38, and the probability of losing is 37/38. For other bets the probabilities follow directly from counts: a red bet wins if the outcome is one of the 18 red pockets, so P(win) = 18/38 = 9/19; an even-money bet (red/black, odd/even, high/low) wins with probability 18/38 and loses with probability 20/38 because 0 and 00 cause losses on these bets.

The distinction between "payout odds" and "true odds" is key. A straight-up bet pays 35:1, meaning a $1 winning bet gives $35 profit plus the returned $1 stake. True odds against winning are 37:1 (37 losses to 1 win), so the casino’s payout understates the true cost of winning by 2 units per win. For even-money bets the casino pays 1:1 but the true odds against winning are 10:9 (20 losses vs 18 wins), again producing an edge. Probabilities for compound bets (splits, corners, dozens, columns) come from counting how many pockets are covered and dividing by 38. This counting model underlies every further calculation: expected values, waiting times, and variance all derive from those basic probabilities.

Expected Value, House Edge, and Long-Term Behavior

Expected value (EV) quantifies average outcome per bet in the long run. Take the canonical $1 straight-up bet: it wins $35 with probability 1/38 and loses $1 with probability 37/38. EV = (1/38)*35 + (37/38)*(-1) = (35 - 37)/38 = -2/38 = -1/19 ≈ -0.05263. This is about −5.263% of the stake; that fraction is the house edge. The same house edge applies to virtually all bets on an American wheel (exceptions are special side bets or promotions), because payouts are scaled so that the casino keeps 2 units of expectation per 38-unit fairness cycle.

For an even-money $1 bet: EV = (18/38)*(+1) + (20/38)*(-1) = (18 - 20)/38 = -2/38 = -1/19. Thus the percentage loss per dollar wagered is independent of the bet type; smaller or larger bets scale the absolute EV linearly. Law of large numbers: over many independent spins the average return per spin converges to the EV (−5.263% per dollar wagered) with high probability. But convergence can be slow and sample paths show large fluctuations: the house edge only guarantees loss on average, not for any single session.

A useful framing: the casino’s edge is a transfer rate from players to casino per unit wagered. If you bet $1000 repeatedly, expected loss per spin is $1000 × 0.05263 = $52.63. Over n independent identical bets the expected cumulative loss is n times that. This is why no strategy that only reallocates bets (without changing the underlying odds) can overcome the house edge in expectation.

Bet Types, Payouts, and Their Mathematical Comparisons

Bet types differ by payoff magnitudes and coverage, but their math is derived from the counts on the wheel. Examples:

- Straight-up: covers 1 pocket, payout 35:1, P(win)=1/38.

- Split: covers 2 pockets, payout 17:1, P(win)=2/38.

- Street (3 numbers): payout 11:1, P(win)=3/38.

- Corner (4 numbers): payout 8:1, P(win)=4/38.

- Dozen/column (12 numbers): payout 2:1, P(win)=12/38.

BobaBet Premium

- Even-money (18 numbers): payout 1:1, P(win)=18/38.

Each payout is less than "true" fair payout computed as (1/P(win)) - 1, and the shortfall produces the same fractional EV of −2/38. This uniformity means that while payoff and variance differ across bets, the expectation per dollar is identical. Variance is higher for bets that pay larger multiples (e.g., straight-up has high variance: rare big wins), while even-money bets have lower variance per dollar. For a straight $1 bet the variance can be computed: let X be net gain. X = +35 with p = 1/38, X = −1 with p = 37/38. E[X] = −1/19. E[X^2] = (1/38)*35^2 + (37/38)*1^2 = 1262/38 ≈ 33.21. Var(X) = E[X^2] − E[X]^2 ≈ 33.21 − (0.05263)^2 ≈ 33.21. Standard deviation ≈ 5.76 per $1 bet. For even-money $1 bets the variance is much smaller: X = +1 w.p. 18/38, X = −1 w.p. 20/38, producing Var(X) = 1 − (E[X])^2 ≈ 0.9997, giving SD ≈ 1.0. So choice of bet controls volatility but not expected loss rate.

This explains why high-payout bets can produce dramatic short-term swings (big wins possible but rare), whereas even-money bets produce steadier but still negative drift. Any strategy claiming to exploit bet selection must change the underlying probabilities or payouts; in a fair casino those are fixed, so EV remains negative.

Models for Multiple Spins: Variance, Risk, and Strategy Limits

When analyzing sequences of spins you use standard discrete models. Hitting a specific single number follows a geometric distribution: the number of spins until the first success has mean 1/p = 38 and memoryless property. The number of successes in n independent spins is binomial with parameters n and p (p depends on the bet). So expected count = n p and variance = n p (1 − p). Probability of at least one hit in n spins on a single-number bet is 1 − (1 − 1/38)^n.

Risk metrics of interest to players include probability of ruin, drawdown sizes, and standard deviation of cumulative winnings. For a sequence of identical independent bets of size b, the cumulative EV after n spins is n b (−1/19), and the cumulative variance is n b^2 Var_single. Thus standard deviation grows like sqrt(n), while expected loss grows linearly with n, so eventually expectation dominates, but fluctuations can be large for moderate n (this is why short sessions can appear profitable by chance).

Common strategies: Martingale (doubling after each loss) attempts to recover previous losses plus a unit profit when a win occurs. Mathematically, Martingale does not change EV: because it scales bets to chase a fixed return but the underlying expectation per spin is negative and the strategy is limited by finite bankroll and table limits. The probability of catastrophic loss (long losing streak) increases exponentially with required streak length and leads to ruin risk that makes long-run EV negative and large losses likely eventually. One can compute ruin probability given bankroll and table cap: if you need k consecutive losses to hit the cap, probability of that streak is (q)^k where q is the loss probability per spin for the chosen bet (e.g., q = 20/38 for even-money). Expected return conditioned on not hitting cap is still negative.

Kelly staking prescribes fraction of bankroll for a positive-expectation bet to maximize long-run growth; since roulette EV is negative, Kelly suggests betting zero. That formalizes why no fractional staking rule recovers positive expectation from a negative-EV game.

Practical consequences: the mathematical model explains why casinos profit steadily — small negative EV per bet compounded over huge numbers of bets yields reliable profit — and why players can have short-term wins but not a guaranteed long-term edge. Understanding p, EV, variance, and distributions enables precise calculation of odds for specific questions (e.g., “what’s the chance of hitting my number at least once in 100 spins?” = 1 − (37/38)^100), and clarifies the limitations of betting systems that do not alter the underlying probabilities.