The Affect Heuristic

“The affect heuristic produces an inverse relationship between perceived risk and perceived benefit that does not exist in the world.”Paul Slovic, Risk Analysis 1994

In 1994 Ali Alhakami and Paul Slovic asked subjects to rate twenty-three hazards on two separate scales: how risky each was, and how beneficial each was. In the real world the two correlate weakly positively — technologies that deliver high benefit have usually been pushed harder, exposed more people, and accumulated more risk. In subjects' heads the correlation came out the other way. The mean within-subject correlation between risk and benefit ratings was r = -0.40. Things they liked, they rated low-risk and high-benefit. Things they disliked, they rated the opposite. The two ratings, supposed to be independent, were being read off the same underlying feeling.

Six years later Finucane Alhakami Slovic and Johnson (2000) ran the same paradigm under five-second time pressure. The negative correlation strengthened to r = -0.55. They also ran an information manipulation: tell subjects an item has high benefit, their risk rating for it drops; tell them it has high risk, their benefit rating for it drops. The two ratings are coupled at the substrate. The paper named the underlying mechanism the “affect heuristic”: when judging risk and benefit, the mind asks itself how it feels about the item, and reads both numbers off that feeling.

Kahneman (2011) Thinking Fast and Slow chapter 12 calls the move “substitution”: when System 1 is asked the hard question “what is the risk of X?” it answers a different easier question, “how do I feel about X?” The two answers come out coupled because they were the same answer.

You are about to see ten items: nuclear power, childhood vaccines, GM food, social media, AI, alcohol, pesticides, electric vehicles, smartphones, microwave ovens. For each you move two sliders: harm (0 to 100) and benefit (0 to 100). After each lock, WIZ shows the documented public rating and the documented expert rating on the same scale. At the end, WIZ computes the Pearson correlation between your harm and benefit ratings across all ten items and places you in the literature bands. A calibrated rater produces r near zero. The 1994 founding-study mean was r = -0.40.

Harm and benefit are supposed to be two questions. The affect heuristic makes them one.