The availa­bil­ity heuris­tic is judg­ing the fre­quency or prob­a­bil­ity of an event by the ease with which ex­am­ples of the event come to mind.

A fa­mous 1978 study by Licht­en­stein, Slovic, Fischhoff, Lay­man, and Combs, “Judged Fre­quency of Lethal Events,” stud­ied er­rors in quan­tify­ing the sever­ity of risks, or judg­ing which of two dan­gers oc­curred more fre­quently. Sub­jects thought that ac­ci­dents caused about as many deaths as dis­ease; thought that homi­cide was a more fre­quent cause of death than suicide. Ac­tu­ally, dis­eases cause about six­teen times as many deaths as ac­ci­dents, and suicide is twice as fre­quent as homi­cide.

An ob­vi­ous hy­poth­e­sis to ac­count for these skewed be­liefs is that mur­ders are more likely to be talked about than suicides—thus, some­one is more likely to re­call hear­ing about a mur­der than hear­ing about a suicide. Ac­ci­dents are more dra­matic than dis­eases—per­haps this makes peo­ple more likely to re­mem­ber, or more likely to re­call, an ac­ci­dent. In 1979, a fol­lowup study by Combs and Slovic showed that the skewed prob­a­bil­ity judg­ments cor­re­lated strongly (0.85 and 0.89) with skewed re­port­ing fre­quen­cies in two news­pa­pers. This doesn’t dis­en­tan­gle whether mur­ders are more available to mem­ory be­cause they are more re­ported-on, or whether news­pa­pers re­port more on mur­ders be­cause mur­ders are more vivid (hence also more re­mem­bered). But ei­ther way, an availa­bil­ity bias is at work.

Selec­tive re­port­ing is one ma­jor source of availa­bil­ity bi­ases. In the an­ces­tral en­vi­ron­ment, much of what you knew, you ex­pe­rienced your­self; or you heard it di­rectly from a fel­low tribe-mem­ber who had seen it. There was usu­ally at most one layer of se­lec­tive re­port­ing be­tween you, and the event it­self. With to­day’s In­ter­net, you may see re­ports that have passed through the hands of six blog­gers on the way to you—six suc­ces­sive filters. Com­pared to our an­ces­tors, we live in a larger world, in which far more hap­pens, and far less of it reaches us—a much stronger se­lec­tion effect, which can cre­ate much larger availa­bil­ity bi­ases.

In real life, you’re un­likely to ever meet Bill Gates. But thanks to se­lec­tive re­port­ing by the me­dia, you may be tempted to com­pare your life suc­cess to his—and suffer he­do­nic penalties ac­cord­ingly. The ob­jec­tive fre­quency of Bill Gates is 0.00000000015, but you hear about him much more of­ten. Con­versely, 19% of the planet lives on less than $1/​day, and I doubt that one fifth of the blog posts you read are writ­ten by them.

Us­ing availa­bil­ity seems to give rise to an ab­sur­dity bias; events that have never hap­pened are not re­called, and hence deemed to have prob­a­bil­ity zero. When no flood­ing has re­cently oc­curred (and yet the prob­a­bil­ities are still fairly calcu­la­ble), peo­ple re­fuse to buy flood in­surance even when it is heav­ily sub­si­dized and priced far be­low an ac­tu­ar­i­ally fair value. Kun­reuther et al. sug­gest un­der­re­ac­tion to threats of flood­ing may arise from “the in­abil­ity of in­di­vi­d­u­als to con­cep­tu­al­ize floods that have never oc­curred . . . Men on flood plains ap­pear to be very much pris­on­ers of their ex­pe­rience . . . Re­cently ex­pe­rienced floods ap­pear to set an up­ward bound to the size of loss with which man­agers be­lieve they ought to be con­cerned.”1

Bur­ton et al. re­port that when dams and lev­ees are built, they re­duce the fre­quency of floods, and thus ap­par­ently cre­ate a false sense of se­cu­rity, lead­ing to re­duced pre­cau­tions.2 While build­ing dams de­creases the fre­quency of floods, dam­age per flood is af­ter­ward so much greater that av­er­age yearly dam­age in­creases.

The wise would ex­trap­o­late from a mem­ory of small haz­ards to the pos­si­bil­ity of large haz­ards. In­stead, past ex­pe­rience of small haz­ards seems to set a per­ceived up­per bound on risk. A so­ciety well-pro­tected against minor haz­ards takes no ac­tion against ma­jor risks, build­ing on flood plains once the reg­u­lar minor floods are elimi­nated. A so­ciety sub­ject to reg­u­lar minor haz­ards treats those minor haz­ards as an up­per bound on the size of the risks, guard­ing against reg­u­lar minor floods but not oc­ca­sional ma­jor floods.

Me­mory is not always a good guide to prob­a­bil­ities in the past, let alone in the fu­ture.

1 Howard Kun­reuther, Robin Hog­a­rth, and Jac­queline Meszaros, “In­surer Am­bi­guity and Mar­ket Failure,” Jour­nal of Risk and Uncer­tainty 7 (1 1993): 71–87.

2 Ian Bur­ton, Robert W. Kates, and Gilbert F. White, The En­vi­ron­ment as Hazard, 1st ed. (New York: Oxford Univer­sity Press, 1978).