By Jason Hreha

Glossary

Probability weighting

Reviewed by Jason Hreha

Published 2 min read

What is probability weighting?

Probability weighting describes how probabilities contribute to the value assigned to uncertain options in a decision model. A decision weight is not necessarily a belief about how likely an event is.

Probability weighting in prospect theory

Tversky and Kahneman (1992) developed cumulative prospect theory with weights based on cumulative probabilities. Their evidence describes a fourfold pattern: risk seeking for low-probability gains, risk aversion for low-probability losses, risk aversion for high-probability gains, and risk seeking for high-probability losses. These are patterns in the modeled choices, not rules followed by every person.

Probability weighting examples

A person can correctly state a lottery’s small winning probability while still giving the prize substantial weight in a purchase decision. Establishing that account requires evidence about both the options and the choices.

Four hypothetical choices

Each pair below has the same expected monetary outcome. The uncertain option otherwise pays or loses nothing. The examples show how to recognize the four risk attitudes, not choices that every person will make.

  • Low-probability gain: a 1% chance to gain $100 versus a sure $1 gain. Choosing the chance illustrates risk seeking.
  • Low-probability loss: a 1% chance to lose $100 versus a sure $1 loss. Choosing the sure loss illustrates risk aversion.
  • High-probability gain: a 90% chance to gain $100 versus a sure $90 gain. Choosing the sure gain illustrates risk aversion.
  • High-probability loss: a 90% chance to lose $100 versus a sure $90 loss. Choosing the chance illustrates risk seeking.

For example, the first gamble’s expected value is 0.01 × $100 = $1. Someone choosing it may know the probability exactly. Probability weighting concerns how that probability enters the choice model, not necessarily an incorrect estimate of the odds. The model’s treatment of gains and losses also matters, so these choices alone do not isolate a weighting function.