Here is a scene you have lived through. A decision is on the table. Someone — maybe you — says the reasonable-sounding thing: “Let’s get more data first.” Another study. Another opinion. One more test. It always feels responsible. And a startling amount of the time, it is pure waste dressed up as diligence.
The value-of-information model is the antidote. It comes from decision analysis (Ron Howard coined “value of information” in 1966) and it hands you one blunt question to ask before you spend a cent or a day gathering more: could the answer actually change what I do? If the honest answer is no — if you’d make the same move whatever the test says — then the information is worth exactly zero. Not “a little.” Zero. However accurate, however cheap, however interesting.
The one-sentence version
Information has value only if it could change your decision. Its worth is the size of the mistake it lets you avoid — no more, no less.
Before you read — take a guess
Before we start — take a guess. A surgeon will operate no matter what a 2,000-dollar scan shows. What is that scan worth to the operating decision?
Value comes from averted mistakes — nothing else
Sit with why that surgeon’s scan is worthless, because the whole model is hiding in it. Information can only help you by changing a choice. A changed choice only helps if the new choice is better — that is, if it steers you away from a mistake you were about to make. So the value of any test is capped by one thing: the cost of being wrong that it could save you from.
That gives the model its shape. The value of information is:
the expected payoff of your best decision with the information minus the expected payoff of your best decision without it.
Two consequences fall straight out, and we’ll lean on both all course:
- It is never negative. Worst case, you learn something useless and ignore it — you were free to make your original choice anyway. (More information never hurts your decision; only its price can.)
- It is capped by the stakes. You can’t save more than the mistake was going to cost you. A test that could avert a $100 loss is worth at most $100, no matter how clever it is.
Why 'never negative' isn't a licence to test everything
Information itself never worsens a decision — but gathering it costs money, time, and delay. That’s why the interesting quantity is always the net: what the answer is worth minus what it costs to get. Half this course is learning to compute the worth; the other half is remembering to subtract the cost.
What you’ll walk away with
By the end you’ll be able to look at any “should we gather more data?” question and answer it with a number, not a vibe. Here’s the map:
- Would it change your action? — the core test, and why a doctor who’ll operate regardless gains nothing from the scan. Learn to kill worthless tests on sight.
- The perfect-info ceiling (EVPI) — what a flawless answer, a literal crystal ball, would be worth. It’s the most any test could ever be worth, worked out on a real payoff table.
- The value of a real, noisy test (EVSI) — no test is perfect. Fold in its error rate with Bayes, and find what an imperfect test is actually worth. It’s always below the ceiling — and the net is that minus its price.
- When is it worth it? — value peaks when you’re genuinely uncertain, the decision is close, and the stakes are high. Kill any of the three and the value falls to zero. Plus the tie to optionality.
- Where the model lies — the traps: the cost of delay, analysis paralysis, the very accurate test of the wrong question, and the answer you’ll rationalise away.
One habit to unlearn as you go
“More information is always good” feels obviously true. It isn’t. More information is only good when it can change something — and past that point, gathering it is procrastination with a spreadsheet. This course retrains the reflex.
A quick taste: the umbrella that tells you nothing
You’re leaving the house. You’ll take an umbrella if rain is at all likely and leave it if not. Now someone offers to sell you a hyper-accurate forecast for tomorrow — but you’re deciding about right now, and the sky is already black with clouds. You’re taking the umbrella regardless. The forecast, however good, is worth nothing to this decision, because your action is already locked.
Change one thing — make it a genuinely uncertain, patchy sky where you truly might go either way — and suddenly the forecast is worth something, because now it can flip your choice. Same forecast, same accuracy. The value lives entirely in whether the answer can still move you.
In the umbrella story, why is the accurate forecast worth nothing under the black, cloudy sky but worth something under a patchy one?
How to use this course
Every lesson opens with a quick guess (don’t skip it — guessing before you know is one of the most reliable ways to remember), explains the idea with a concrete worked example and real numbers, and checks that it stuck. There’s arithmetic here, but nothing worse than multiplying and adding payoffs — and an interactive lab does the heavier counting for you, so you can see a test’s value climb and collapse as you drag the dials.
When you’ve finished the five teaching lessons, a graded final exam pulls it all together. It’s one-way — once you submit an answer it’s locked — so treat every practice question along the way as exactly that: practice.
Ready? Lesson one states the core test carefully and teaches you to spot a worthless test in seconds — starting with a surgeon who really will operate no matter what.