🧪 This platform is in early beta. Features may change and you might encounter bugs. We appreciate your patience!
subadditivity_effect
The tendency for the sum of probability judgments of individual events to exceed the probability of the overall event, or for the judged probability of the whole to be less than the sum of its parts. When people break an event into components, they assign higher probabilities to each component than is warranted, and the parts add up to more than 100%.
When asked the probability of dying from any cause in the next year, a person might say 5%. But when asked separately about heart disease, cancer, accidents, and other causes, the individual estimates sum to 15% or more.
An insurance customer, when asked how likely it is that his apartment will be damaged in the next year, estimates 4%. But when asked separately about fire, water damage, theft, and vandalism, his individual estimates add up to over 20% — each specific peril feels plausible in isolation, inflating the total.
A project manager asks her team to estimate the overall probability that a product launch will be delayed. The team says 25%. When she later asks them to estimate delays from supplier issues, software bugs, regulatory approval, and staffing problems separately, the combined probabilities sum to over 70%.
Binary (yes/no) questions an LLM must answer to identify this aspect:
Do the probabilities assigned to individual components sum to more than the whole?
Type: binaryIs an event judged as more likely when described in detailed sub-parts?
Type: binaryWould the probability estimate decrease if the event were described holistically?
Type: binaryThe tendency for the sum of probability judgments of individual events to exceed the probability of the overall event, or for the judged probability of the whole to be less than the sum of its parts. When people break an event into components, they assign higher probabilities to each component than is warranted, and the parts add up to more than 100%.
Thinking about specific scenarios makes them more vivid and available, inflating their perceived probability. People also use the representativeness and availability heuristics differently when evaluating wholes versus parts.
When estimating probabilities of component events, check that they sum to no more than the total. Use top-down estimation (start with the whole) to anchor component estimates.
This effect is exploited in insurance marketing (itemizing risks makes total risk seem larger) and affects risk assessment, strategic planning, and legal reasoning when separate claims are evaluated independently.
Use these tools to detect, analyze, or train this aspect.