Probability

    Subadditivity Effect

    🇳🇴Subadditivitetseffekt

    Definition

    The subadditivity effect is the tendency to judge the probability of a whole as lower than the sum of the probabilities of the parts it consists of. Breaking an event into explicit sub-events makes the total feel more likely.

    Real-world example

    Tversky and Koehler (1994) demonstrated this clearly. Participants were asked: 'What's the probability a randomly chosen person died of an unnatural cause?' The mean answer was about 32%. Another group got the same categories broken out: car accident, other accidents, homicide, suicide, other unnatural. Their sum was about 58% – nearly double – even though it's logically the same question.

    The effect is robust and shapes medical diagnosis (a doctor listing possible diseases gets sums over 100%), insurance judgments, and risk communication. The more explicitly you describe the possibilities, the more likely they seem in aggregate.

    Supplementary perspective

    The effect points at something deeper: probabilities are assigned to *descriptions*, not events. Two ways of describing the same underlying outcome produce different judgments. This violates a basic axiom of probability theory, but it's how people actually think.

    Practical advice

    Recognize

    • Notice if a detailed scenario feels more likely than a general description.
    • Be skeptical when the sum of 'independent' risks exceeds 100%.
    • Check whether lists of specific causes inflate perceived total risk.

    Counteract

    • Aggregate probabilities: check whether the sum of parts exceeds a reasonable total.
    • Compare estimates both aggregated and disaggregated to reveal inconsistency.
    • Use base rates from actual data rather than your own summation.

    Ethical use

    • In risk communication: don't break down to scare, or aggregate to soothe.
    • In sales: don't list many specific benefits to inflate perceived value.
    • In analysis: document both aggregated and disaggregated assessments.

    Related biases