Decision weights
Also known as: Weighting function, Pi function
The transformed probabilities people actually act on, not the stated odds.
What it means
Decision weights are the nonlinear transformation of stated probabilities that prospect theory substitutes for raw probabilities when valuing a prospect. The weighting function is inverse-S shaped: it overweights small probabilities and underweights moderate-to-large ones, and it is discontinuous near certainty and impossibility. Crucially, weights are not probabilities — they need not sum to one and are not interpreted as beliefs, only as the impact a chance has on choice. In cumulative prospect theory the transformation is applied to cumulative (rank-ordered) probabilities rather than to each outcome separately, which repairs the original version's violations of stochastic dominance. Decision weights are what give the theory its grip on the certainty effect, the possibility effect, and the fourfold pattern.
How an outcome's weight is computed
In cumulative prospect theory a single outcome does not draw its weight from its own probability alone. Outcomes are ranked from worst to best, the weighting function is applied to the cumulative probability of doing at least as well as each one, and that outcome's decision weight is the gap between two adjacent cumulative transforms. The result is rank dependence: the same ten-percent chance carries a different weight depending on whether it sits near the top of the ranking or the bottom. This machinery is what lets the model inflate the extreme tails of a distribution while compressing the middle, and it is the specific repair that stops the theory from ever preferring a dominated gamble.
Two knobs: curvature and elevation
Gonzalez and Wu showed the curve is better described by two parameters than one. Curvature sets how sharply it bends, meaning how sensitive a person is to changes in probability away from certainty and impossibility, while elevation sets how high the whole curve sits, an index of how attractive the gamble looks. The two move independently: someone can be broadly optimistic yet still barely tell a forty-percent chance from a sixty-percent one. Prelec separately derived a compact two-parameter form from an axiom he called compound invariance, giving the shape a footing in choice principles rather than curve-fitting alone. Most estimates place the crossover, where weight equals probability, near a third.
What the evidence shows
The inverse-S is among the more reproducible findings in decision research, recurring across the original gambles of Tversky and Kahneman, the individual-level estimates of Gonzalez and Wu, and many parametric forms since. The 1992 data put the gain-side curvature parameter near 0.6, well under the value of one that linear weighting would demand. But the aggregate curve hides wide variation: a sizable minority weight probabilities almost linearly, or even in an S-shape that overweights the middle, so median parameters describe a population rather than a person. Fitted values also shift with the elicitation method and the parametric family assumed, which is why point estimates differ from study to study.
Where the curve moves
Decision weights are not a fixed dial. Fox and Tversky found they depend on the source of the uncertainty: a chance tied to one's own expertise is weighted differently from a matched numerical probability, so identical odds attract different weight in a familiar versus an unfamiliar domain. The shape can even invert. When probabilities are learned by sampling rather than stated in words, the description-experience gap documented by Hertwig and Erev, people behave as if they underweight rare events, the mirror image of the overweighting seen with described risk. That reversal matters, because much of the standard evidence rests on explicitly described gambles, and real decisions often are not.
Reading a weight correctly
A decision weight measures impact on choice, not stated belief, and conflating the two misleads practitioners. Someone who insures against a one-percent hazard is not necessarily misjudging the odds as larger; the thin chance simply carries outsized pull. So you cannot read a person's probability estimate off their choices, and correcting their beliefs may not move behavior when the weighting, not the estimate, is doing the work. The more reliable lever is usually how the probability is framed, whether described, experienced, or bundled with certainty, rather than the number itself.
Examples
A jump from a 0% to 1% chance of winning feels far larger than a jump from 50% to 51%, though both add one percentage point.
A treatment that removes the last sliver of risk feels worth far more than one cutting risk from 30% to 20%, though the second helps ten times as many patients.
Shoppers pay extra to insure a phone against a 5% chance of theft, then shrug at the 40% chance they will crack the screen themselves — small chances loom, big ones flatten.
Facing a two-percent chance of a ruinous jury verdict, a defendant settles for far more than the expected payout, because that thin possibility of catastrophe dominates the calculation.
A one-in-a-million vaccine side effect draws more decision weight than the far larger disease it prevents, so a rare harm can stall an otherwise sound immunization program.
First described in Kahneman & Tversky (1979); cumulative version, Tversky & Kahneman (1992).
Key references
- Hertwig, R., & Erev, I. (2009). The description-experience gap in risky choice. Trends in Cognitive Sciences, 13(12), 517-523. doi.org/10.1016/j.tics.2009.09.004
- Gonzalez, R., & Wu, G. (1999). On the shape of the probability weighting function. Cognitive Psychology, 38(1), 129-166. doi.org/10.1006/cogp.1998.0710
- Prelec, D. (1998). The probability weighting function. Econometrica, 66(3), 497-527. doi.org/10.2307/2998573
- Fox, C. R., & Tversky, A. (1998). A belief-based account of decision under uncertainty. Management Science, 44(7), 879-895. doi.org/10.1287/mnsc.44.7.879
- Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5(4), 297-323. doi.org/10.1007/BF00122574