Numeracy · 0 connections

Expected value

Multiply each possible outcome by its probability and sum the results: that number is what the choice is worth on average — not what it will deliver this time.

Explanation

A fair die has six faces, each equally likely. The expected value is (1+2+3+4+5+6) ÷ 6 = 3.5 — a number no single roll ever lands on, but the one the running average moves toward as you keep rolling. That same move works for anything you can attach numbers and likelihoods to: an investment, a medical treatment, a pricing decision. Weight each outcome by its probability, sum the results, and you have the expected value. The result also adds cleanly — the expected value of two payoffs combined equals the sum of each one's expected value on its own, whether or not those payoffs are linked. That makes large, tangled decisions workable: break them into parts, compute each, then add. One condition: the probabilities must be real estimates, not invented ones. Where likelihoods are truly unknown — not just hard to pin down, but simply not there — the formula still runs. It just produces a confident-looking number with nothing behind it.

Watch

No video has yet earned its place here.

When it applies

**Repeated decisions with knowable odds.** When the same type of choice comes up again and again, and you can genuinely estimate the probabilities, expected value is the right frame. The actual average of your outcomes will move toward the expected value as those decisions accumulate — not quickly, not smoothly, but reliably. **Risks pooled across many independent cases.** When separate risks can be averaged together — an insurer pricing policies, a lender setting rates — the average realised loss across the whole pool tends toward what the expected value of a single case predicted. Pooling converts individual uncertainty into a stable number. **Any single decision where you can estimate both probability and magnitude.** Even a one-off choice — a job move, an investment, a medical decision — becomes clearer when you ask explicitly: what might happen, how likely is each outcome, and how much does each cost or gain? The discipline works even with rough estimates, as long as you know they are rough.

Where it stops

**When a single extreme outcome dominates the math.** Some choices carry a tiny chance of an enormous payoff. The expected value is real, but it can be driven almost entirely by that far-edge outcome — one that may never be realised in a lifetime of decisions. When the expected value is mostly tail, it is not a useful guide to the choice in front of you. **When probabilities are genuinely unknown.** Expected value requires probability estimates. When you face real uncertainty — not "difficult to estimate" but "the shape of what might happen is unknown" — the formula has no inputs. Invented probabilities produce invented answers. **When you are living one path, not averaging across many.** Expected value is the average across many independent runs of the same situation. If a single bad outcome could end your ability to keep playing — financially, professionally — the average across all those parallel runs was never a description of your situation. It was a description of a different, collective calculation. **When the question is what people will do, not what they should.** Expected value describes how a rational agent should choose under certain conditions. It does not describe how people actually choose. People reliably prefer a certain outcome over a higher-expected-value gamble, weight losses more heavily than equivalent gains, and treat known odds differently from unknown ones. These are stable features of how people decide, not mistakes that disappear with education. When the task is to predict or explain behaviour, expected value is not the tool.

The misuse

A bettor finds a game where the odds genuinely favour them — a real, verifiable edge. Each round carries a positive expected value, and a larger stake means a larger expected gain. If the bet is good and more money on a good bet is better, stake as much as the bankroll allows. Staking everything on a favourable bet, round after round, drives the chance of eventual ruin toward certainty. The expected dollar figure keeps climbing — because it is averaged across many parallel versions of the game — while the bettor walks only one of those versions. A single bad stretch wipes out the bankroll, and with it every future round the average quietly assumed would continue. The error is not in the calculation. Expected value ranks the direction of a choice — it says *this way*. It does not size the commitment — it says nothing about *how much*. A user who has run the numbers correctly, confirmed a real edge, and still staked too much has made a mistake the expected value calculation could not catch, because it was never asked the right question.

A worked example

A mathematician at the blackjack table worked out that a deck's composition shifts as cards leave it. Count what remains, and certain combinations carry a positive expected value for the player rather than the house. That was the edge. He did not then push every available dollar onto every favourable hand. He confirmed the edge was real, decided what he was willing to risk on a single round, and sized each bet so that a bad run could not end the game. Compounding the edge required surviving the bad runs first. He applied the same two-step discipline in financial markets: find a mispriced position where the expected value sits in your favour, confirm it, then size the stake so that no single rough stretch could end the run. One method. Two settings.

Push

Take the positive-expected-value option. Then size your commitment at the largest stake you could survive losing in one go.

Veto

Never accept a stake so large that a single bad outcome could take you out of the game, no matter how high the expected value.

Connects to

Connections appear as the deck grows.

Go deeper

Mastery question

**Question:** You face a one-time decision with a clearly positive expected value. You take it, and it goes badly. A friend who also ran the numbers tells you: "There — you were wrong to do it." Were you? And, separately: was there a version of that same positive-expected-value bet that you would have been wrong to take? **The answer:** Two things are true at once. A single bad outcome does not make the decision wrong — you judge the decision by the probability-weighted picture you faced at the time, not by the one draw you got. Your friend is reading the decision backward from its result. But yes, there was a version you'd have been wrong to take: if the bet was sized so that a bad draw could take you out of the game, it was a poor decision despite the positive expected value. Expected value ranks the direction of a choice, not the size of the stake. And because this was a one-shot decision, the long-run average may never be realised at all — so the expected value number was never simply "the value" of this one path. **The answer that misses it:** "No — the expected value was positive, so it was the right call. Outcomes don't change that." **Why the difference matters:** The rule "judge the decision, not the outcome" is the easiest part of expected value to memorise — and the wrong answer wields it correctly. What it misses is everything the rule doesn't say: that expected value sizes no bet, that it presupposes you survive to repeat the trial, and that a one-shot path is not the ensemble the average describes.