Expected value is a powerful decision tool, but the number is meaningful only when you know what wager, event, rules, stake basis, strategy, and time unit it describes. A statement such as “the EV is −$0.20” can refer to one bet, one resolved decision, one hand, one roll, one hour, or a model built on assumptions that do not match the game in front of you.
The arithmetic can be correct while the conclusion is wrong because the context was omitted.
What expected value actually measures
For outcomes with probabilities p₁, p₂, …, pₙ and net results x₁, x₂, …, xₙ, expected value is:
EV = p₁x₁ + p₂x₂ + … + pₙxₙ
The probabilities must add to 1, and the x values must use the same accounting basis. If wins are recorded as net profit, losses must also be net loss. Mixing total return with net profit produces a false result.
Suppose a $10 even-money wager wins $10 with probability 18/37 and loses $10 with probability 19/37:
EV = (18/37 × $10) + (19/37 × −$10)
EV = −$0.27027… per $10 wager
That is an average loss of about 27 cents each time the specified wager resolves. It does not mean the next result will lose 27 cents. The actual next result is either a $10 win or a $10 loss.
The expected-value glossary entry covers the core definition. This page is about the conditions that determine whether an EV number is useful.
Context begins with the unit
Every EV statement needs a unit. Common units include:
- per dollar wagered;
- per $10 bet;
- per resolved decision;
- per hand or spin;
- per roll;
- per hour;
- per promotion cycle;
- per trip.
These are not interchangeable.
A craps wager may stay active across several rolls. Reporting expected loss “per roll” answers a different question from reporting it “per resolved bet.” A slot’s RTP is usually expressed per unit wagered over a large volume, not per hour. A table-game rating may estimate theoretical loss from average wager, pace, and time.
Before comparing two numbers, make sure they use the same unit.
House edge is EV expressed as a percentage of action
For a fixed wager, house edge can be written as:
House edge = −EV per wager ÷ Amount wagered
Using the $10 example above:
House edge = $0.27027 ÷ $10 = 0.027027 = 2.7027%
The percentage helps compare the price of one dollar of action. It does not show how many dollars will be wagered.
To estimate hourly cost, context must include pace:
Expected loss per hour = Average wager × Decisions per hour × House edge
A low-edge game played quickly and heavily can have a higher hourly cost than a higher-edge game played slowly. Why Game Speed Can Matter More Than House Edge develops that comparison.
The rules must match the number
Casino game names are not complete mathematical descriptions. Blackjack edge changes with payout, decks, doubling rules, surrender, dealer action, and player strategy. Video-poker return changes with the paytable and hold decisions. A slot title may be offered in different return configurations. Carnival-game side bets can have separate paytables at neighboring tables.
An EV calculated for one version should not be transferred to another merely because the name looks familiar.
A useful EV statement identifies:
- the exact wager;
- the exact rules or paytable;
- the assumed strategy;
- the stake basis;
- whether side bets are included;
- the source and date of any changing inputs.
Without those details, precision after the decimal point can create false confidence.
Strategy is part of the probability distribution
In games with decisions, expected value depends on how those decisions are made.
A blackjack figure based on correct basic strategy does not describe a player who regularly stands, hits, splits, or doubles incorrectly. A video-poker return based on optimal play does not describe holds chosen by intuition. Poker EV depends on opponents, position, rake, and future decisions.
This does not mean every decision must be perfect before EV is useful. It means the model should state what behavior it assumes.
A strategy table can improve expectation without guaranteeing the hand. Good decisions and bad results can occur together.
EV is not a session forecast
Expected value is the center of a probability distribution, not the path a bankroll must follow.
Suppose a game has an expected loss of $20 for a one-hour session. Several outcomes may still be plausible:
- win $500;
- win $40;
- lose $20;
- lose $300;
- lose the entire session bankroll.
The exact range depends on variance and game structure. Two games with the same EV can produce very different short-term experiences.
One may pay small amounts frequently. Another may return most value through rare large prizes. The average alone does not describe that difference.
How Variance Tricks You explains why a negative-EV game can produce a winning session and why a positive-EV opportunity can lose repeatedly.
A positive EV can still be financially dangerous
Positive expected value means the probability-weighted average is favorable under the stated assumptions. It does not mean the bankroll is large enough to survive the distribution.
Consider a promotion with two outcomes:
- 99% chance to lose $100;
- 1% chance to win $10,100 net.
EV = (0.99 × −$100) + (0.01 × $10,100)
EV = −$99 + $101 = +$2
The opportunity has positive EV of $2 per attempt. It also loses $100 on 99% of attempts. A player with one $100 stake is overwhelmingly likely to finish with nothing.
The decision needs at least three pieces of context:
- expected value;
- variance and loss distribution;
- bankroll and number of available attempts.
Positive EV without adequate bankroll can be mathematically favorable and practically unusable.
Zero EV does not mean zero risk
A fair coin wager paying even money has EV of zero before costs. The player can still lose the entire stake.
Likewise, a hedge may reduce or neutralize expected profit while changing the range of outcomes. An insurance bet can have negative standalone EV but reduce a specific exposure. Context determines whether you are evaluating the wager alone or the combined position.
The phrase “no expected loss” should never be translated into “cannot lose.”
Costs outside the paytable belong in the decision
A gambling opportunity can have favorable game EV and unfavorable total economics after costs.
Relevant costs can include:
- travel and accommodation;
- tournament or entry fees;
- taxes;
- tips;
- financing costs;
- data, software, or subscription expenses;
- time and alternative income;
- mistakes and execution failures;
- limits on cashing, redeeming, or repeating the offer.
Suppose a promotion has an expected gain of $80, but the trip costs $120:
Net expected value = $80 − $120 = −$40
The game component is positive. The decision as a whole is negative.
Comps and rebates can change EV, but not by face value alone
Cashback, free play, match play, rooms, meals, and loyalty benefits can reduce expected cost. Their value depends on redemption rules and personal use.
A $100 free-play award is not automatically $100 cash. If the original stake is not returned and the credit must be wagered once, the expected cash value depends on the game used and the treatment of winnings.
Similarly, a room priced by the casino at $300 may be worth only $120 to a player who would otherwise book a $120 room. Context uses personal replacement value, not marketing value.
The restored article What Casino Comps Are Really Worth provides full examples.
Conditional EV changes with the current state
Some games and promotions contain persistent information. The expected value after observing the state can differ from the unconditional value before observation.
Examples include:
- a progressive jackpot meter;
- a collected-symbol or must-hit-by feature;
- remaining cards in a known deck composition;
- a tournament stack and payout position;
- a promotion with limited qualifying inventory;
- a poker decision after seeing opponents’ actions.
The correct calculation is then conditional:
EV given current information = Σ P(outcome | current state) × Net result
This does not justify treating ordinary independent outcomes as “due.” The condition must actually change the probabilities, payouts, or available decisions.
Estimated EV inherits uncertainty from its inputs
Many casino EV figures are exact only under a mathematical model. Real-world estimates may use uncertain inputs:
- approximate hands per hour;
- sampled average bet;
- unknown slot configuration;
- estimated promotional participation;
- incomplete opponent data;
- uncertain error rate;
- changing jackpot contribution or competition.
If the input is a range, the output should often be a range.
Suppose pace could be 50 to 80 decisions per hour, with a $20 wager and 2% edge:
Low estimate:
$20 × 50 × 0.02 = $20 expected loss per hour
High estimate:
$20 × 80 × 0.02 = $32 expected loss per hour
Reporting exactly $26.00 without explaining the pace assumption would overstate certainty.
Research and educational context
Expected value is a probability-weighted average, as shown in the OpenStax explanation of expected value. The mathematical definition is stable. The difficult part in casino decisions is identifying the correct outcomes, probabilities, costs, and unit of analysis.
That distinction matters because a clean formula does not repair bad assumptions.
What EV can and cannot tell you
Expected value can help you:
- compare wagers on the same basis;
- estimate long-run cost;
- identify a genuine mathematical advantage;
- price promotions and rebates;
- separate a good decision from a lucky result;
- evaluate the effect of wager size, speed, and time.
Expected value cannot by itself tell you:
- what happens next;
- how large a bankroll you need;
- the chance of reaching a loss limit;
- how severe short-term swings may be;
- whether the assumptions match the real game;
- whether the entertainment is personally worth the cost;
- whether continued gambling is safe for a particular person.
Those are separate questions.
A context checklist
Before using an EV number, ask:
- What exact wager or decision does it describe?
- Is the result net profit, total return, or expected loss?
- Is the unit per bet, per decision, per hour, or per session?
- Which rules, paytable, and strategy are assumed?
- Are side bets, comps, fees, and other costs included?
- How volatile is the distribution?
- Is the bankroll sufficient for the downside?
- Are any inputs estimated or change-sensitive?
- Is the state independent, or does current information genuinely alter the probabilities?
- Am I using the number to compare decisions or to predict a session?
Expected value becomes useful only when its context is defined
Expected value does not need to be abandoned because it is incomplete. It needs to be used correctly.
EV tells you the average value of a defined probability distribution. Context defines the distribution, the unit, the assumptions, and the practical consequence. Without that context, a precise number can be less useful than an honest range. With it, expected value becomes one of the clearest tools for seeing what a casino decision really costs.