Partition dependence
How the options are grouped sways the probabilities we assign them.
What it means
The tendency for probability judgments to depend on how the space of possible outcomes is partitioned into categories, because people lean toward spreading belief evenly across whatever options are presented — an 'ignorance prior' of equal odds per branch. Split one outcome into two sub-outcomes and its combined probability rises; lump several into one and their summed probability falls, even though the underlying uncertainty is identical. The effect is a structural cousin of the unpacking effect and a manifestation of support theory, but it highlights the role of the menu's architecture rather than the vividness of any single branch. It matters acutely for forecasting and survey design: the choice of how to slice the possibilities — which is often arbitrary — can steer the elicited probabilities, so the same expert can give different forecasts depending only on the partition handed to them.
Examples
Asked to forecast tomorrow's weather, people give 'rain' a higher chance when the alternatives are listed as sun, clouds, and fog (three options) than when the alternative is simply 'no rain' (one).
Investors handed a menu of one bond fund and four stock funds put more into stocks than those shown one stock fund and four bond funds; money spreads evenly across whatever boxes appear.
Asked who will win the category, a manager gives the market leader half the chance against 'all rivals'; list the four rivals separately and the leader's odds slide toward a fifth.
First described in Fox & Rottenstreich (2003).