Expected Value: A Practical Guide for Everyday Decisions

Expected value is one of the most useful ideas for making decisions under uncertainty, and its definition fits on one line. The expected value (EV) of a choice is the sum, across every possible outcome, of that outcome’s value multiplied by its probability. Written out, EV = Σ (probability × value), summed over every outcome. Because the probabilities of all outcomes add to 1 (or 100%), the result is a weighted average: the value you would land on, on average, if you could face the same choice again and again under identical conditions.

That last clause — “again and again under identical conditions” — is doing more work than it looks like, and most of what follows in this guide is about taking it seriously rather than skipping past it.

A worked example: the extended warranty

At the checkout you are offered a two-year extended warranty on a $500 phone for $80. Should you take it? Break the decision into its parts. Suppose there is roughly a 10% chance the phone suffers a fault the warranty would cover, and that such a repair would cost about $250 out of pocket without it. The expected value of the warranty’s payout is then 0.10 × $250 + 0.90 × $0 = $25. You are being asked to pay $80 for protection worth $25 on average, so the EV of buying it is $25 − $80 = −$55. On expectation, the warranty is a losing proposition — which is more or less by design, since the retailer prices the premium above the expected payout in order to turn a profit across everyone who buys one.

That does not automatically mean skipping it is the smart move for you personally — see the risk-of-ruin and utility sections below — but it does mean you should know the arithmetic is against you before you decide to buy anyway for other reasons, such as wanting the certainty of a fixed repair cost.

A second worked example: raising your insurance excess

The same arithmetic works just as well in the other direction — not deciding whether to buy protection, but deciding how much of it to keep. Suppose your annual policy lets you raise your excess (the amount you pay out of pocket before the insurer covers the rest) from $100 to $500, in exchange for $120 off this year’s premium. If you do make a claim, the higher excess costs you an extra $400 compared with today’s terms. Suppose your realistic chance of making any claim this year is about 8%.

The expected extra cost from the higher excess is 0.08 × $400 = $32. Against that, you are guaranteed to save $120 on the premium regardless of whether you claim. The expected value of switching is $120 − $32 = $88 in your favor. Unlike the warranty, this is a case where the arithmetic favors taking on more risk yourself, precisely because the insurer’s premium for the extra protection is priced well above what that protection is worth on average — which is the ordinary state of affairs for insurance, since insurers have to price in their own costs and margin on top of the raw expected payout.

When expected value is the right tool

EV shines when a decision repeats, or is small enough relative to your resources that a bad single result would not meaningfully hurt you. An insurer writing thousands of policies, a shop pricing thousands of items, a business making the same small purchasing call every month — each of them actually lives near the average, because the law of large numbers grinds the luck out over enough repetitions. If you face a small, low-stakes choice many times over — which warranty to buy on a $40 kitchen gadget, which of two short routes to take on a daily errand — take the option with the higher expected value each time and let the arithmetic compound quietly in your favor. Over a hundred such small calls, the ones with better EV will, on average, leave you ahead, even though any single one of them could go either way.

When expected value is the wrong tool

The average is a promise about the long run, and at least three situations break that promise.

• One-shot decisions. If you will make a choice exactly once, you do not get the average — you get a single draw from the distribution. A choice can carry a genuinely positive EV and still leave you with the bad outcome on the only occasion that counts, because “on average” describes a population of trials, not the one trial you actually get. For a decision you will not repeat, the spread of possible outcomes matters as much as their mean, and a narrower spread around a slightly lower EV can be the more sensible pick.
• Risk of ruin. A positive-EV choice that can wipe out your ability to keep going is still a trap, because ruin ends the game before the long-run average has any chance to rescue you. Picture a single irreversible commitment with a 50% chance of a strong result and a 50% chance of an outcome so damaging you could not attempt anything similar again — committing an entire year’s operating budget to one unproven, unbacked supplier with no fallback if it fails, say. The expected value across many repetitions of that kind of commitment might look excellent on paper. But you are not making it many times; you are making it once, and half of the possible futures end with nothing left to try again with. Never accept a proposition whose downside removes your capacity to keep participating in the kind of decisions that reward playing the average.
• Nonlinear value. A dollar, or an hour, or a unit of any resource is not worth the same amount to you in every state you might end up in. The first chunk that covers your fixed costs matters far more than an equivalent chunk stacked on top of a comfortable surplus. When outcomes are large relative to your resources, their raw numeric value stops tracking their real importance to you, and averaging the raw numbers quietly misrepresents the choice.

Expected value versus expected utility

That last point leads to a deeper idea: expected utility. Instead of averaging the raw amounts, you average their usefulness to you — how much each outcome actually improves your position. Consider a guaranteed $1 million against a 50/50 chance of $2.5 million or nothing. The gamble’s expected value is 0.5 × $2.5M = $1.25 million, which beats the sure million by a comfortable margin on paper. Yet most people rationally take the guaranteed million, and they are not being timid or bad at arithmetic. For someone of ordinary means, the first million changes what is possible in a way the second million on top of it does not; the practical benefit of each additional dollar shrinks as the pile grows. Once you weight outcomes by their utility rather than their raw size, declining the higher-EV gamble in favor of the sure thing is the more defensible choice, not a mistake the EV number is correcting you out of.

This is also, in miniature, why the risk-of-ruin case above is not really an exception to expected value theory so much as a reminder that "value" was never supposed to mean "raw dollar amount" in the first place — it was always meant to mean something closer to utility, and dollars are only a convenient stand-in for utility in the ordinary range where they track each other reasonably well.

A limitation built into the tool, not just the idea

The expected value calculator on this site takes a list of outcomes, their probabilities, and their payoffs, and returns the weighted average along with the single best and single worst outcome in your list. That is the whole of what an EV number can honestly tell you, and it is worth being explicit about what it cannot: it cannot tell you how likely you are to land near the worst outcome versus the best one beyond the two numbers you already gave it, it says nothing about whether the worst outcome is merely disappointing or actually ruinous for you specifically, and it has no opinion on whether this is a decision you get to make once or a thousand times. Those three judgments — the real shape of the distribution, what “ruinous” means for your situation, and whether you are looking at a repeatable or one-shot choice — are exactly the ones a calculator cannot make for you, and skipping them is the single most common way an EV calculation gets misused. Before you lean on a positive number, look again at the best and worst rows the tool reports and ask honestly whether the worst one is survivable and whether you are really going to get to run this more than once.

A short checklist before trusting an EV calculation

• Is this decision repeated, or close to it? If not, treat the EV as one useful data point rather than the verdict.
• Is the worst-case outcome survivable? If a bad result would meaningfully damage your ability to keep going, a positive average does not offset that.
• Are your probabilities more than a guess? EV is only as trustworthy as the numbers you feed it; a confident-looking average built on a rough guess at probability is still a rough guess.
• Does the value scale roughly linearly for you here? For small, everyday amounts it usually does. For anything that would represent a large fraction of your resources, weight it by how much it would actually matter to you, not just its face value.

Run through those four questions and expected value stops being a trap that dresses up bad advice in tidy arithmetic, and becomes what it was always meant to be: a clear way to compare the small, repeatable, survivable choices that make up most of ordinary life.