Risk Homeostasis

Risk homeostasis is the theory, proposed by psychologist Gerald Wilde, that individuals and organizations adjust their behavior to maintain a constant level of perceived risk — meaning that safety interventions which reduce objective risk may be partially or entirely offset by behavioral shifts that restore the original risk level. The theory is also known as the Peltzman effect in economics, after Sam Peltzman's finding that automobile safety regulations did not reduce highway fatalities because drivers compensated by driving faster and more aggressively.

The mechanism is not irrational. A constant level of risk may be optimal for a given activity: too little risk means excessive caution and lost opportunity; too much risk means unacceptable danger. When a safety intervention lowers the objective risk, the margin opens — and rational agents expand their behavior to reclaim it. risk management that ignores homeostasis is not managing risk; it is displacing it, moving it from one domain to another without reducing the total. The moral hazard produced by implicit insurance is a form of risk homeostasis at the institutional level: the safety net lowers the perceived cost of failure, so behavior adjusts to exploit the net.

The systems insight is that safety is not a scalar that can be maximized independently. It is a dynamic equilibrium maintained by the interaction of technology, behavior, and incentives. Normal accidents occur not because safety systems fail but because the safety systems change the system's operating point, pushing it into regimes where new, unanticipated failure modes emerge. The ABS brakes that prevent skidding also enable driving at speeds where non-skidding failures — rollover, structural fatigue, reaction-time limitations — become dominant.

Risk homeostasis is not an argument against safety. It is an argument against naivety about safety. Every intervention that reduces one risk creates a new margin, and something will fill that margin. The question is not whether to intervene but whether the post-intervention risk profile is preferable to the pre-intervention one — and that question cannot be answered without modeling the behavioral response.