Rare events resist prediction because their causes interact uniquely
Evidence
Rare events are the result of a unique constellation of factors interacting with each other such that, were one or a couple of the factors absent, the shooting would not occur. Unique constellations of causes provide few predictive commonalities. Moreover, factors contributing to a unique combination of influences lose some of their predictiveness when they are considered separately from each other, because some of their causative influence may derive from their mutual influence. For example, neither bizarre beliefs alone nor substance abuse alone may be conducive to aggression, but in combination, they may increase the likelihood of aggressive behavior. — Albert Bandura, Moral Disengagement, 1279
Categorization
Category: Causal Relationships
- XREF: This is the classic statistical argument against individual risk-factor prediction — echoes the logic in Nate Silver’s The Signal and the Noise and general probability-of-rare-events literature.
- QUESTION: The interaction claim — separate factors lose predictiveness because causation is synergistic — deserves deeper chase into interaction-term statistics and the limits of additive risk models.