Definition
The expected value of a random variable X, written E(X) or mu, is the probability-weighted average of its possible values: each outcome multiplied by the probability of that outcome, then summed (or integrated, for a continuous variable).
For a discrete X, E(X) is the sum over all values x of x times P(X = x); for a continuous X with density f, E(X) is the integral of x times f(x). The number it returns is rarely a value the variable can actually take — a fair die has E(X) = 3.5, which never appears on a face — but the law of large numbers guarantees that the average of many independent repetitions converges to it. That convergence is what makes expectation more than an arithmetic trick: it is the bridge between a single uncertain trial and the predictable long-run behaviour of many.
Why it matters
How it works
The arithmetic — enumerate, weight, sum
To compute an expected value, list the possible values of the random variable, multiply each by its probability, and add. For a fair six-sided die, E(X) = (1 + 2 + 3 + 4 + 5 + 6) / 6 = 3.5. For a Bernoulli trial with success probability p, E(X) = p. For a binomial(n, p) — the count of successes in n independent trials — E(X) = np. For a Poisson(lambda), E(X) = lambda. Once the distribution is named, the expected value is a one-line formula.
Expectation also behaves well under linear operations: E(aX + b) = aE(X) + b, and crucially E(X + Y) = E(X) + E(Y) for any X and Y, independent or not. This linearity of expectation is the workhorse that lets you decompose a complicated random quantity (the total wickets in a match, the total claims in a portfolio, the total wait time across a queue) into simple pieces, take expectations piecewise, and add. No other summary statistic composes this cleanly.
As the centre of a distribution — Haigh's chance experiments
In Probability: A Very Short Introduction, Haigh introduces expected value as the natural answer to the question "where does the cloud of possible outcomes sit?" A chance experiment is reasoned about by replacing the bare outcome with a random variable and a distribution; the expected value is then the centre of mass of that distribution. The shift from asking "what will happen?" to "what is the shape of the cloud, and where is its centre?" is the move that separates folk reasoning from statistical reasoning.
Centre alone is not enough. Two distributions can share an expected value and still differ wildly — a fair coin paying plus or minus one pound and a fair coin paying plus or minus one thousand pounds both have mean zero, but the second is far riskier. Risk lives in variance, not expectation. Expected value tells you where outcomes cluster on average; variance (and its square root, the standard deviation) tells you how tightly. Any honest decision needs both numbers.
As the rule of rational action — Priest's decision theory
In Logic: A Very Short Introduction, Priest treats expectation as the cornerstone of practical reasoning. When an act has several possible outcomes, you can usually estimate two things about each — how likely it is, and how much you value it. Score values on an open-ended scale (positive good, negative bad, zero indifferent), multiply each by its probability, sum, and you have the expectation of the act. Priest is careful to flag that this technical "expectation" has almost nothing to do with the ordinary English word; it is a weighted average of how much you stand to gain or lose.
The rational-choice rule that follows is unembarrassed in its simplicity: pick the action with the greatest expectation; break ties at random. This single rule underlies economics, game theory, insurance, and risk management — anywhere choices are made under uncertainty. It also sharpens classical philosophical arguments. Pascal's Wager urges belief in God on the grounds that disbelief risks an infinitely bad outcome that swamps every other figure. Decision theory exposes the flaw not by rejecting expectation but by demanding a complete table: there are many possible gods, several jealous of the wrong worship, and once every relevant possibility is included, no specific theistic belief comes out on top.
As the price of a gamble — Haigh's games people play
Every commercial game of chance is engineered around one number: the house edge — the operator's expected gain per unit wagered, or equivalently the player's expected loss. On a European roulette wheel, a one-pound bet on red wins one pound with probability 18/37 and loses one pound with probability 19/37, giving E = 18/37 minus 19/37, or about minus 2.7 pence. That is the famous 2.7 percent edge. The American wheel adds a double-zero pocket and the same calculation gives roughly minus 5.3 pence — two pockets, one wheel, almost double the long-run skim. Lotteries are sharper still: a typical 6/49 game returns about half of takings as prizes, so a one-pound ticket has expected value near 50p, a minus-50 percent return per pound.
Two corollaries follow. First, expectation is additive across independent bets, so no sequence of negative-expectation wagers — Martingale doubling, progressive systems, hot-streak chasing — can be combined into a positive-expectation strategy. The arithmetic is closed. Second, roulette spins (and lottery draws) are independent: the wheel and the machine have no memory. If red has come up twelve times running, the chance of red on spin thirteen is still 18/37. The Gambler's Fallacy — "black is due" — is exactly the belief that expectation somehow remembers the past and corrects.
Where expected value stops being enough — utility
A lottery ticket has an expected value of about 50p on a one-pound stake, yet millions of people buy one every week. They are not all making a mistake. The resolution is utility: the satisfaction money provides is not linear in the amount. The pound spent is barely missed; the fourteen million pounds, if won, would be worth far more than fourteen million times the utility of any single pound. Multiply the tiny probability by the huge utility, add the large probability of near-zero loss, and the sum in utility terms can credibly be positive even when the sum in money terms is sharply negative. Add the small pleasure of dreaming between purchase and draw, and the trade settles into something close to "50p of expected return plus 50p of fantasy equals one pound."
The same person who buys lottery tickets also buys insurance with a negative expected money value — paying a premium above the expected payout to avoid a rare large loss. There is no contradiction: a utility curve that is concave over the bulk of the distribution (smoothing ordinary gains and losses, hence risk-averse) but convex in its tails (life-changing windfalls or ruinous losses valued out of proportion) reconciles both behaviours. Expected utility, not expected value, is what real agents maximise. The St. Petersburg paradox — a game with infinite expected money value that nobody will pay much to play — was the historical jolt that forced this distinction into the open.
A portable checklist for any risky decision
Insurance, options trading, R&D portfolios, lawsuit settlements, career bets, even clinical-trial economics are all instances of paying a known cost for a probabilistic payoff. The same two-question discipline that demystifies the casino carries over. First: what is the expected value per unit stake? Multiply each outcome by its probability, sum, subtract the cost. If the answer is negative, you are paying for something other than the money. Second: what is the variance, and what shape are you buying? A small negative expectation with high variance (lottery, venture round, long-shot lawsuit) buys a very different experience from a small negative expectation with low variance (slot machine, index fund, settled claim). Decide whether your utility function values the shape enough to cover the price — and only then commit.