FINITE MATHEMATICS • PROBABILITY AND STATISTICS

Expected Value in Decisions — Compute expected value in decision contexts

Quantify uncertain outcomes to make rational choices by weighting payoffs with their probabilities.

Historical Context & Motivation

Long before modern probability theory existed, merchants, gamblers, and rulers grappled with decisions under uncertainty. Should a ship owner insure a cargo against piracy? Should a gambler accept a particular wager? These questions demanded a framework for comparing outcomes that were not guaranteed — a way to assign a single numerical value to a gamble whose result depended on chance. The concept that eventually answered these questions is expected value, one of the most consequential ideas in the history of mathematics and rational decision-making.

1654
The Pascal–Fermat Correspondence
Blaise Pascal and Pierre de Fermat exchanged a series of letters on the 'Problem of Points' — how to fairly divide the stakes of an interrupted game of chance. Their solutions laid the groundwork for the notion of weighting outcomes by probability.
1657
Huygens' De Ratiociniis in Ludo Aleae
Christiaan Huygens published the first formal treatise on probability, introducing the concept of 'expectation' (Latin: expectatio) as a fair price for participating in a game of chance.
1738
Bernoulli's St. Petersburg Paradox
Daniel Bernoulli posed a game with infinite expected monetary value, yet no rational person would pay an unlimited sum to play. His resolution — using expected utility rather than expected monetary value — revealed the limits and extensions of the expected-value concept.
1944
Von Neumann–Morgenstern Utility Theory
John von Neumann and Oskar Morgenstern axiomatized expected utility in their landmark work Theory of Games and Economic Behavior, anchoring expected value as the backbone of modern decision theory and game theory.
Present
Ubiquitous Decision Tool
Expected value now permeates finance, insurance, operations research, machine learning, and public policy. It remains the default criterion for comparing alternatives whose outcomes involve randomness.

The central question that drives this lesson is deceptively simple: given several possible actions, each leading to different outcomes with different probabilities, how do we compress all of that uncertainty into a single number that lets us compare alternatives? Expected value provides the answer, and this lesson develops the concept from definition through computation to application.

Core Principles & Definitions

Before computing expected values in decision contexts, it is essential to internalize the foundational ideas. Expected value is, at its core, a probability-weighted average of all possible outcomes associated with a random variable or decision. Unlike a simple arithmetic mean, expected value accounts for the fact that some outcomes are more likely than others. The following principles underpin every expected-value calculation you will encounter.

1

Random Variable & Outcomes

A random variable X assigns a numerical value to each outcome of a random process. In decision contexts, each action yields a set of possible payoffs (gains or losses), each with an associated probability.
2

Probability Distribution

The probabilities of all possible outcomes must sum to 1 (or equivalently 100%). Each outcome xᵢ has probability P(xᵢ) ≥ 0. This complete listing of outcomes and probabilities forms the probability distribution of the decision.
3

Weighted Summation

Expected value E(X) is computed by multiplying each outcome by its probability and summing the products. This weighting ensures that high-probability outcomes contribute more to the average than low-probability ones.
4

Long-Run Interpretation

E(X) represents the average outcome you would observe if you repeated the decision (or experiment) infinitely many times. It is a theoretical mean, not a guaranteed single-trial result.
5

Decision Criterion

When comparing two or more alternatives, the rational (risk-neutral) choice is the one with the highest expected value. If E(X) > 0 for a gamble, it is 'favorable'; if E(X) < 0, it is 'unfavorable.'
KEY TAKEAWAY
Think of expected value like planning a road trip with multiple possible routes. Each route has a different travel time (outcome) and a different likelihood of being open (probability). Rather than optimistically assuming the fastest route will be clear, you compute the probability-weighted average of all travel times. The 'best' route is the one whose weighted average — its expected travel time — is the shortest. Expected value works identically: it gives you the single number that best summarizes an uncertain situation.

Visual Explanation — Decision Tree

A decision tree is one of the most intuitive tools for visualizing expected-value calculations. Decision nodes (squares) represent choices the decision-maker controls; chance nodes (circles) represent outcomes governed by probability. The diagram below models a classic decision scenario: an entrepreneur deciding whether to launch a new product or keep the current one, where the new product's success depends on uncertain market conditions.

The decision tree shows a square decision node (D) branching into two choices: Launch and Keep Current. The Launch branch leads to a circular chance node (C) that fans out into three possible demand levels. Each terminal node displays its payoff. The expected value boxes on the right aggregate the probability-weighted payoffs, revealing that launching ($69,000) dominates keeping the current product ($40,000).

Notice how the decision tree structures the problem. At the decision node (square labeled D), the entrepreneur exercises free will — she chooses Launch or Keep Current. At the chance node (circle labeled C), nature determines the outcome according to the given probabilities. By computing the expected value at each chance node and then comparing across decision branches, we convert a complex, multi-outcome problem into a straightforward comparison of two numbers: $69,000 vs. $40,000. This procedure — sometimes called 'folding back' the tree — is the standard technique for expected-value analysis in finite decision problems.

Mathematical Framework

We now formalize the computation. Let X be a discrete random variable with n possible values x₁, x₂, …, xₙ and corresponding probabilities P(X = x₁), P(X = x₂), …, P(X = xₙ). The expected value — also called the mathematical expectation or mean — of X is defined as follows.

EXPECTED VALUE (DISCRETE)
E(X) = Σᵢ₌₁ⁿ xᵢ · P(X = xᵢ) = x₁P(x₁) + x₂P(x₂) + … + xₙP(xₙ)
xᵢ = value of the i-th outcome; P(X = xᵢ) = probability of that outcome; n = total number of distinct outcomes. The constraint Σ P(xᵢ) = 1 must hold.

When comparing two or more alternatives (decision branches), we compute E(X) for each alternative and select the one that maximizes (or, in cost contexts, minimizes) the expected value. Two additional properties are routinely exploited.

LINEARITY OF EXPECTATION
E(aX + b) = a · E(X) + b
a, b are constants. This property is invaluable when payoffs are scaled or shifted (e.g., a ticket price subtracted from potential winnings).
EXPECTED VALUE OF A SUM
E(X + Y) = E(X) + E(Y)
This holds regardless of whether X and Y are independent. It allows decomposition of complex payoffs into simpler components whose expectations are individually computable.
⚠️ Important Distinction
Expected value does not tell you what will happen on any single trial. A lottery ticket with E(X) = −$0.50 may win you $10 million once — but if you bought millions of tickets, your average payoff per ticket would converge to −$0.50. The expected value is a long-run average, justified rigorously by the Law of Large Numbers.

Detailed Breakdown — Payoff Tables & Multi-Alternative Comparison

While decision trees offer a visual narrative, payoff tables (also called payoff matrices) provide a compact, tabular alternative that is especially useful when comparing multiple strategies against multiple states of nature. Each row represents a decision alternative, each column represents a state of nature (with its probability), and the cell entries are payoffs. The expected value of each alternative is computed by taking the dot product of its payoff row with the probability column.

Payoff table for three investment alternatives under three economic states. E(X) is computed as the probability-weighted sum of each row.
AlternativeBoom (P = 0.25)Stable (P = 0.50)Recession (P = 0.25)E(X)
Stocks+$12,000+$4,000−$8,000$3,000
Bonds+$5,000+$3,000+$1,000$3,000
Savings+$2,000+$2,000+$2,000$2,000

In this example, Stocks and Bonds share the same expected value of $3,000, both exceeding the $2,000 expected return of Savings. Based on expected value alone, a risk-neutral investor would be indifferent between Stocks and Bonds. However, note that Stocks range from −$8,000 to +$12,000 while Bonds range from +$1,000 to +$5,000 — the variance of Stocks is much larger. Expected value intentionally ignores spread, which is both its power (simplicity) and its limitation (insensitivity to risk). We will revisit this in Section 7.

Each vertical line segment represents the range of possible payoffs for an investment. Green dots are boom outcomes, yellow are stable, and red are recession outcomes. The purple bar marks E(X). Stocks and Bonds share E(X) = $3,000, but Stocks exhibit a much wider range, illustrating that expected value alone does not capture risk.

Worked Example — Insurance Decision

Consider the following scenario: A homeowner is deciding whether to purchase a flood insurance policy that costs $1,200 per year. Historical data indicate a 2% chance of a major flood causing $50,000 in damage, a 5% chance of a minor flood causing $10,000 in damage, and a 93% chance of no flood. The insurance covers all damage in full. Should the homeowner buy the policy based on expected value?

Expected-Value Analysis of Flood Insurance
1
Step 1 — Identify the decision alternativesThere are two alternatives: (A) Buy insurance, paying $1,200 regardless of what happens, but any flood damage is covered; (B) Do not buy insurance, paying $0 upfront but bearing the full cost of any flood damage.
2
Step 2 — List outcomes and probabilities for Alternative B (no insurance)The outcomes for the uninsured homeowner are: (1) Major flood: loss = $50,000 with P = 0.02; (2) Minor flood: loss = $10,000 with P = 0.05; (3) No flood: loss = $0 with P = 0.93. Note that Σ P = 0.02 + 0.05 + 0.93 = 1.00 ✓
3
Step 3 — Compute E(cost) for Alternative BE(cost without insurance) = (0.02)($50,000) + (0.05)($10,000) + (0.93)($0) = $1,000 + $500 + $0 = $1,500.
E(cost | no insurance) = $1,500
4
Step 4 — Compute E(cost) for Alternative AIf the homeowner buys insurance, the cost is $1,200 with certainty (probability 1), since the policy covers all damage. Therefore, E(cost with insurance) = $1,200.
E(cost | insurance) = $1,200
5
Step 5 — Compare and decideSince $1,200 < $1,500, the expected cost with insurance is lower than the expected cost without insurance. The expected savings from purchasing the policy are $1,500 − $1,200 = $300 per year.
Decision: Buy the insurance. Expected annual savings = $300.
6
Step 6 — Interpret the resultOn any given year, the homeowner may end up paying $1,200 for nothing (93% chance), which can feel like a waste. But over many years, the average annual flood-related cost without insurance would converge to $1,500, exceeding the fixed $1,200 premium. The expected-value criterion tells us that insurance is favorable not because floods are likely, but because the probability-weighted damage exceeds the premium.

Strengths, Limitations, and When to Look Beyond E(X)

Expected value is a powerful first-pass decision tool, but it is not infallible. Understanding when it works well and when it falls short will make you a more sophisticated analyst. The following table summarizes the key trade-offs.

StrengthsLimitations
Reduces complex multi-outcome scenarios to a single, comparable number.Ignores the spread (variance) of outcomes — two alternatives with the same E(X) may carry vastly different risk.
Mathematically rigorous, with well-established properties (linearity, additivity) that simplify calculations.Assumes a risk-neutral decision-maker; most real humans are risk-averse (they would prefer lower-variance alternatives even at the cost of some expected return).
Justified by the Law of Large Numbers for repeated decisions — the sample mean converges to E(X).For one-shot decisions (e.g., a single surgery, a unique business opportunity), the long-run interpretation may be irrelevant.
Universally applicable across domains: finance, engineering, medicine, public policy.Can be infinite or undefined for certain distributions (e.g., the St. Petersburg Paradox), challenging its use as a decision criterion.
KEY TAKEAWAY
Expected value is to decision-making what a compass is to navigation: it reliably points you toward the best direction on average, but it cannot warn you about cliffs along the path. For high-stakes or one-time decisions, supplement expected value with measures of risk (such as variance or the worst-case loss) and consider whether your personal risk tolerance aligns with the risk-neutral assumption built into E(X).

Connection to Expected Utility & Advanced Decision Theory

The expected-value framework studied in this lesson is the foundation upon which more sophisticated decision criteria are built. The most important extension is expected utility theory, which replaces raw monetary payoffs with a utility function u(x) that captures the decision-maker's attitudes toward risk. Instead of computing E(X) = Σ xᵢ P(xᵢ), we compute E[u(X)] = Σ u(xᵢ) P(xᵢ). For a risk-neutral agent, u(x) = x and expected utility reduces to expected value; for a risk-averse agent, u(x) is concave (e.g., u(x) = √x), causing the agent to favor lower-variance alternatives.

FeatureExpected Value (This Lesson)Expected Utility (Advanced)
Payoff measureRaw monetary (or numerical) outcome xᵢTransformed utility u(xᵢ)
Risk sensitivityRisk-neutral (ignores variance)Can model risk aversion or risk seeking via the shape of u
FormulaE(X) = Σ xᵢ P(xᵢ)E[u(X)] = Σ u(xᵢ) P(xᵢ)
When identicalAlways (by definition)When u(x) = x (risk-neutral case)
Resolves St. Petersburg Paradox?No — predicts infinite willingness to payYes — diminishing marginal utility bounds the sum

Beyond expected utility, modern decision science offers additional frameworks such as prospect theory (Kahneman and Tversky, 1979), which accounts for the empirical observation that people weight losses more heavily than equivalent gains, and minimax regret, which minimizes the worst-case regret across all states of nature. Mastering expected value as presented in this lesson provides the essential quantitative baseline for engaging with all of these advanced theories.

Practice Problems

PROBLEM 1CONCEPTUAL
A game has two possible outcomes: you win $100 with probability 0.4 or lose $60 with probability 0.6. Without computing, explain in one or two sentences why the expected value of this game might be positive, negative, or zero. Then compute E(X) to confirm your reasoning.
PROBLEM 2BASIC CALCULATION
A raffle sells 500 tickets at $5 each. There is one grand prize of $1,000, two second prizes of $200 each, and five third prizes of $50 each. If you buy one ticket, what is your expected net gain (payoff minus ticket cost)?
PROBLEM 3INTERMEDIATE
A tech company must choose between two product development strategies. Strategy A costs $300,000 upfront; it succeeds with probability 0.6 (generating $800,000 in revenue) or fails with probability 0.4 (generating $100,000 in revenue). Strategy B costs $150,000 upfront; it succeeds with probability 0.8 (generating $400,000 in revenue) or fails with probability 0.2 (generating $50,000 in revenue). Which strategy maximizes expected profit?
PROBLEM 4APPLIED
A city is evaluating whether to invest $2 million in a flood barrier. Engineers estimate the following annual flood probabilities: no flood (P = 0.85, damage = $0), minor flood (P = 0.10, damage = $3 million), major flood (P = 0.05, damage = $20 million). The barrier has a 20-year lifespan and eliminates all flood damage. Using expected value, determine whether the investment is justified over 20 years (ignore the time value of money).
PROBLEM 5CRITICAL THINKING
Consider two investment opportunities: Investment X yields +$10,000 with probability 0.5 and −$10,000 with probability 0.5. Investment Y yields +$1 with certainty. Both have E(X) = $0 and E(Y) = $1, respectively. A purely expected-value analysis recommends Y. However, construct a scenario (by modifying the probabilities or payoffs of X slightly) in which expected value recommends X, yet most rational people might still prefer Y. Discuss what this reveals about the limitations of expected value as a decision criterion.

Summary

The expected value of a decision is computed by the formula E(X) = Σ xᵢ · P(xᵢ), where each possible outcome xᵢ is multiplied by its probability and the products are summed. Originating in the 17th-century correspondence of Pascal and Fermat, the concept has become the default criterion for rational, risk-neutral decision-making. It is operationalized through tools such as decision trees and payoff tables, and it benefits from powerful algebraic properties such as linearity and additivity.

While expected value is an indispensable first step, it assumes risk neutrality and ignores the variance of outcomes, making it insufficient for high-stakes, one-shot decisions where risk aversion matters. Extensions like expected utility theory and prospect theory build on this foundation to incorporate risk preferences and behavioral biases, but expected value remains the essential quantitative baseline for all probabilistic decision analysis.

Varsity Tutors • Finite Mathematics • Expected Value in Decisions