Decision Weights

What it is

The attribution is Kahneman and Tversky jointly, and Tversky is routinely dropped from secondary coverage of this material. He should not be.

The idea explains why the same person insures a house and buys a lottery ticket. That is not inconsistency. It is one weighting function read at two points.

The table that circulates runs probability from zero to a hundred against the weight the mind assigns. Two readings do most of the work. The bottom end is inflated: a two per cent chance carries a weight of about eight, an overweighting by a factor of four, which is the machinery under lotteries, insurance premiums and a good deal of what gets called irrational fear. The middle is flattened: the stretch from five per cent to ninety-five per cent maps onto a span of roughly two thirds the width a rational agent would use, so across almost the whole range a change in the odds moves the decision less than it should.

Zero and one hundred are the only points where weight and probability agree, which is exactly why moving from ninety-five per cent to certainty feels like far more than the five points it is.

In effect

The qualification matters more than the table, and it is the part that gets lost. Kahneman qualifies his own figures immediately. Probabilities below one per cent and above ninety-nine per cent are a special case, no unique decision weight can be assigned to very rare events, and rare events are sometimes not overweighted at all but ignored entirely and given a weight of zero.

So the function is not a smooth curve running cleanly to the axis. The very low end is unstable, and the same event can be inflated fourfold or dropped to nothing depending on whether it has been brought to mind. Anyone using the table to argue that people always overweight small risks is contradicted on the following page. Kahneman returns to the rare-events account twice more in the same book, saying that the original theory did not specify the conditions and that his current view has been strongly influenced by later research. That is an author telling a reader the parameter is open.

Where this bites is risk communication. A one in a million risk presented in isolation and a one in a million risk presented after a vivid story are the same number and not the same weight, and the difference is not captured by any fixed curve. That is a health and money question, so the house standard for such material applies.

What it does not say

It does not say that people misjudge probabilities. Weighting is about how a probability enters a choice, not about whether the person can state it correctly. Someone can know the odds precisely and still weight them the way the function describes.

It does not say that small risks are always overweighted. Its own author says the opposite half the time, and the zero-weight case is the more dangerous one, since a risk given no weight at all produces no precaution.

It does not come with a verifiable measurement in its popular form. The table is presented as measurement and the reader is given no paper, no sample size and no year to check it against. Treat the numbers as illustrating a shape rather than as data.

And it does not have the neuroscience behind it that popular accounts imply. The claim that brain responses resembling these weights have been found appears with no researcher, no laboratory, no imaging method and no design named. That claim converts a behavioural model into a physical fact while carrying no evidence a reader can follow, and this publication does not repeat it.

---