Expected Value
Expected value is the long-run average outcome of a random variable if the underlying random process were repeated indefinitely, calculated by multiplying each possible outcome by its probability and summing the results. It is not a prediction of what will happen on any single occurrence — a fair six-sided die has an expected value of 3.5, a value the die can never actually land on, since 3.5 isn't even one of its six possible faces — but rather a statement about the average outcome across a very large number of repetitions, which is precisely why expected value is sometimes described as "what you'd expect on average, not on any given trial."
Expected value's origins trace to the correspondence between Blaise Pascal and Pierre de Fermat in 1654, prompted by a gambler's question about how to fairly divide stakes in an interrupted game of chance — widely regarded as one of the founding moments of formal probability theory, and the concept of a game's expected payout was exactly the practical question that correspondence was trying to answer. The concept generalized from games of chance into the formal statistical definition used today: for a discrete random variable, expected value is the sum of each possible value multiplied by its probability; for a continuous random variable, the equivalent calculation becomes an integral over the variable's probability density function, but the underlying logic — a probability-weighted average across every possible outcome — stays identical in both cases.
Expected monetary value — multiplying each possible financial outcome of a decision by its probability and summing the result — is a standard decision-making framework in business, finance, and insurance, used to compare options when each carries a different probability-weighted mix of possible gains and losses, but it carries a well-known limitation worth stating explicitly: expected value treats every unit of probability-weighted payoff as equally desirable regardless of the variance or risk involved, so a guaranteed payment of $1,000 and a coin flip paying $2,000 for heads or $0 for tails share the identical expected value, despite most people, and most rational risk management frameworks, treating the two as meaningfully different in practice. This is why expected value is treated as a necessary but not sufficient input to most real financial and business decisions, rather than the sole deciding criterion — risk tolerance, variance, and the practical consequences of worse-than-expected outcomes all have to be weighed alongside it, which is exactly the gap utility theory and risk-adjusted decision frameworks in economics were developed to address.
In machine learning and modern applied statistics, expected value underlies the formal definition of an unbiased estimator — an estimator whose expected value, across repeated sampling, equals the true population parameter it's trying to estimate — which is the precise mathematical property that makes an estimation method defensible, since an estimator whose expected value systematically differs from the true parameter (a biased estimator) would, by definition, be wrong on average even before accounting for any individual sample's random variation around that already-incorrect average.
Linearity of expectation — the property that the expected value of a sum of random variables always equals the sum of their individual expected values, regardless of whether the variables are independent or correlated with each other — is one of the most useful and widely applied results built on top of the basic expected value definition, because it lets a complex combined outcome's expected value be computed by simply adding up the much simpler expected values of its individual components, without needing to account for how those components might interact.
Read expected value as the probability-weighted average outcome of an uncertain process, never a prediction of any specific single result: its roots in Pascal and Fermat's 1654 correspondence on fair gambling stakes mark one of the founding moments of formal probability theory, its well-documented blind spot — treating risk and variance as irrelevant, which the guaranteed-$1,000-versus-coin-flip example makes vivid — is exactly why it's used as a necessary input rather than the sole basis for real decisions, and linearity of expectation remains one of the field's most practically useful shortcuts precisely because it holds regardless of whether the underlying variables are independent or correlated.