Tag Archives: statistics

Micromorts and Understanding the Probability of Death

Under­stand­ing prob­ab­il­it­ies is hard (viz.) – and it’s espe­cially so when we try to under­stand and take ration­al decisions based on very small prob­ab­il­it­ies, such as one-in‑a mil­lion chance events. How, then, to com­mu­nic­ate risks on a sim­il­ar level, too?

The answer is to use a more under­stand­able scale, such as micro­morts; “a unit of risk meas­ur­ing a one-in-a-mil­lion prob­ab­il­ity of death”. Some activ­it­ies that increase your risk of death by one micro­mort (accord­ing to, among oth­er sources, the Wiki­pe­dia entry):

  • smoking 1.4 cigar­ettes (can­cer, heart dis­ease)
  • drink­ing 0.5 liter of wine (cir­rhosis of the liv­er)
  • liv­ing 2 days in New York or Boston (air pol­lu­tion)
  • liv­ing 2 months in Den­ver (can­cer from cos­mic radi­ation)
  • liv­ing 2 months with a smoker (can­cer, heart dis­ease)
  • liv­ing 150 years with­in 20 miles of a nuc­le­ar power plant (can­cer from radi­ation)
  • drink­ing Miami water for 1 year (can­cer from chlo­ro­form)
  • eat­ing 100 char­coal-broiled steaks (can­cer from ben­zo­pyrene)
  • eat­ing 40 table­spoons of pea­nut but­ter (liv­er can­cer from Aflatox­in B)
  • eat­ing 1000 bana­nas, (can­cer from radio­act­ive 1 kBED of Potassi­um-40)
  • trav­el­ling 6 miles (10 km) by motor­bike (acci­dent)
  • trav­el­ling 16 miles (26 km) on foot (acci­dent)
  • trav­el­ling 20 miles (32 km) by bike (acci­dent)
  • trav­el­ling 230 miles (370 km) by car (acci­dent)
  • trav­el­ling 6000 miles (9656 km) by train (acci­dent)
  • fly­ing 1000 miles (1609 km) by jet (acci­dent)
  • fly­ing 6000 miles (9656 km) by jet (can­cer from cos­mic radi­ation)
  • tak­ing 1 ecstasy tab­let

Issue fifty-five of Plus magazine looked at micro­morts in more detail, thanks to Dav­id Spiegel­hal­ter (the Win­ton Pro­fess­or of the Pub­lic Under­stand­ing of Risk at the Uni­ver­sity of Cam­bridge) and Mike Pear­son, both of Under­stand­ing Uncer­tainty.

via Schnei­er on Secur­ity

Psychic Numbing and Communicating on Risk and Tragedies

I’ve been pre­oc­cu­pied lately with the devel­op­ing after­math of the Tōhoku earth­quake. Unlike oth­er dis­asters on a sim­il­ar or great­er scale, I’m find­ing it easi­er to grasp the real human cost of the dis­aster in Japan as my broth­er lives in Kanagawa Pre­fec­ture and there­fore there are less levels of abstrac­tion between me and those dir­ectly affected. You could say that this feel­ing is related to what Moth­er Teresa was refer­ring to when she she said “If I look at the mass I will nev­er act. If I look at the one, I will”.

If I had no dir­ect con­nec­tion with Japan I assume the dry stat­ist­ics of the size­able tragedy would leave me mostly unaf­fected – this is what Robert Jay Lifton ter­med “psych­ic numbing”. As Bri­an Zikmund-Fish­er, a risk com­mu­nic­a­tion expert at the Uni­ver­sity of Michigan, intro­duces the top­ic:

People are remark­ably insens­it­ive [to] vari­ations in stat­ist­ic­al mag­nitude. Single vic­tims or small groups who are unique and iden­ti­fi­able evoke strong reac­tions. (Think, for example, the Chilean miners or “baby Jes­sica” who was trapped in the well in Texas in 1987.) Stat­ist­ic­al vic­tims, even if much more numer­ous, do not evoke pro­por­tion­ately great­er con­cern. In fact, under some cir­cum­stances, they may evoke less con­cern than a single vic­tim does. […]

To over­come psych­ic numb­ing and really attach mean­ing to the stat­ist­ics we are hear­ing […] we have to be able to frame the situ­ation in human terms.

Zikmund-Fish­er links heav­ily to Paul Slov­ic’s essay on psych­ic numb­ing in terms of gen­o­cide and mass murder (pdf): an essen­tial read for those inter­ested in risk com­mu­nic­a­tion that looks at the psy­cho­logy behind why we are so often inact­ive in the face of mass deaths (part of the answer: our capa­city to exper­i­ence affect and exper­i­en­tial think­ing over ana­lyt­ic­al think­ing).

The Numbers in Our Words: Words of Estimative Probability

Toward the end of this month I will almost cer­tainly post the tra­di­tion­al Lone Gun­man Year in Review post. Exactly how likely am I to do this? Am I able to quanti­fy the prob­ab­il­ity that I’ll do this? By using the phrase “almost cer­tainly”, I already have.

To provide unam­bigu­ous, quant­it­at­ive odds of an event occur­ring based solely on word choice, the “fath­er of intel­li­gence ana­lys­is”, Sher­man Kent, developed and defined the Words of Estim­at­ive Prob­ab­il­ity (WEPs): words and phrases we use to sug­gest prob­ab­il­ity and the actu­al numer­ic­al prob­ab­il­ity range to accom­pany each.

Kent’s idea has had a mixed recep­tion in the intel­li­gence com­munity and the dis­reg­ard­ing of the prac­tice has been blamed, in part, for the intel­li­gence fail­ings that lead to 9/11.

The words by decreas­ing prob­ab­il­ity:

  • Cer­tain: 100%
  • Almost Cer­tain: 93% ± 6%
  • Prob­able: 75% ± 12%
  • Chances About Even: 50% ± 10%
  • Prob­ably Not: 30% ± 10%
  • Almost Cer­tainly Not: 7% ± 5%
  • Impossible: 0%

The prac­tice has also gained some advoc­ates in medi­cine, with the fol­low­ing list of defin­i­tions formed:

  • Likely: Expec­ted to hap­pen to more than 50% of sub­jects
  • Fre­quent: Will prob­ably hap­pen to 10–50% of sub­jects
  • Occa­sion­al: Will hap­pen to 1–10% of sub­jects
  • Rare: Will hap­pen to less than 1% of sub­jects

It would be nice if there were such defin­i­tions for the many oth­er ambigu­ous words we use daily.

The Statistics of Wikipedia’s Fundraising Campaign

Yes­ter­day, 15th Janu­ary 2011, Wiki­pe­dia cel­eb­rated its tenth birth­day. Just over two weeks before, Wiki­pe­dia was also cel­eb­rat­ing the close of its 2010 fun­drais­ing cam­paign where over six­teen mil­lion dol­lars was raised from over half a mil­lion donors in just fifty days.

The 2010 cam­paign was billed as being data-driv­en, with the Wiki­pe­dia volun­teers “test­ing mes­sages, ban­ners, and land­ing pages & doing it all with an eye on integ­rity in data ana­lys­is”.

Nat­ur­ally, all of the test data, ana­lyses and find­ings are avail­able, provid­ing a fas­cin­at­ing over­view of Wiki­pe­di­a’s large-scale and effect­ive cam­paign. Of par­tic­u­lar interest:

If you’re ever involved in any form of fun­drais­ing (online or off), this data­set is essen­tial reading–as will the planned “Fun­drais­ing Style Guide” that I hope will be released soon.

My favour­ite ban­ner, which got elim­in­ated toward the begin­ning of the cam­paign has to be:

One day people will look back and won­der what it was like not to know.

And if you’re inter­ested in what Jimmy Wales had to say about his face been fea­tured on almost every Wiki­pe­dia page for the dur­a­tion of the cam­paign, BBC’s recent pro­file on the Wiki­pe­dia founder will sat­is­fy your interest.

via @zambonini

Outliers, Regression and the Sports Illustrated Myth

By appear­ing on the cov­er of Sports Illus­trated, sports­men and women become jinxed and shortly there­after exper­i­ence bouts of bad luck, goes the Sports Illus­trated Cov­er Jinx myth.

‘Evidence’ of the myth comes in the form of many indi­vidu­als and teams who have died or, more com­monly, simply exper­i­enced bad luck in their chosen voca­tion shortly after appear­ing on the cov­er of the magazine.

The Wiki­pe­dia entry for the Sports Illus­trated Cov­er Jinx has a thor­ough list of some “not­able incid­ences” and also provides a con­cise, sci­entif­ic explan­a­tion of the phe­nomen­on:

The most com­mon explan­a­tion for the per­ceived effect is that ath­letes are gen­er­ally fea­tured on the cov­er after an out­lier per­form­ance; their future per­form­ance is likely to dis­play regres­sion toward the mean and be less impress­ive by com­par­is­on. This decline in per­form­ance would then be mis­per­ceived as being related to, or even pos­sibly caused by, the appear­ance on the magazine cov­er.

Related: The Mad­den NFL Curse.

via Ben Gol­dacre’s Bad Sci­ence