Saturday 1 June 2013

Drunkard's Walk - reviewed

The Drunkard’s Walk: how randomness rules our lives by Leonard Mlodinow.  Vintage Books 2009.  Amazon link:
It ain't what you don't know that gets you into trouble. It's what you know for sure that just ain't so.”  Mark Twain

The day after I finished The Drunkard’s Walk, we were driving along in the car and I asked the family if they had any idea of how many people died in the events of 9/11.  The Beloved was the only one prepared to play and her estimation of 2000-and-something was close enough to merit a silver star.  But I then took a leaf (p.159 to be exact) out of Mlodinow’s book to suggest that the Wikipedia answer of  2,974+19+1 (the last as the result of lung damage from dust inhalation) was too low by about 1000.  Because fear of flying was being given extra attention in the 2001 weeks following 11 September, LOTS more Americans decided to drive long distances to go about their business rather than take the plane.  Accordingly, there was an excess of about 1000 road-deaths in the following months compared to the same period in 2000.   The Beloved’s response to this piece of information was something to this effect: “You wonk, only someone detached from reality would think in those terms AND go on to make the calculations”.  And my rhetorical riposte was that there were 1000 Americans who’d be alive today if they were willing to do the math rather than drift through life buffeted by emotive responses to a troubling and potentially dangerous world.  As it happens the girls were booked to fly to London on 14/09/2001 to celebrate Dau.II’s 6th birthday in Legoland.  We did briefly consider wimping out and cancelling the trip but they went and had Legoland almost entirely to themselves – no queues, great fun, not to be missed.

Here’s one example of this book's embrace of the counter-intuitive: in 1989, Mlodinow applied for life insurance and was asked to take a blood test (for HIV).  A week later he was refused cover and later told by his GP that it was 999/1000 certain he’d be dead within the decade.  The doctor had reasoned from this statistic: the HIV test shows positive when the blood is in reality not infected in only 1 case out of 1000. But this confounds two probabilities: a) the chance of testing positive if you’re not HIV-positive and b) the chance of not being HIV-positive if you test positive.  They sound like they’re mirrors of each other but they’re not.  Mlodinow is after all American, white, heterosexual and doesn’t share needles.  In that cohort the rate of HIV is about 1 in 10,000.  The rate of false positive is, as above, 1 in 1000 or 10 in 10,000.  So if you test 10,000 such blokes you’ll reveal one bona fide HIV carrier, 10 healthy men who will have difficulty getting insured and 9,989 who are genuinely not carrying the virus.  All things being equal, therefore, if you test positive you’re ten times more likely to be HIV-clear than HIV-carrier.  So don’t rush off to get measured for your coffin.

Mlodinow uses similar reasoning to demolish the case against Sally Clark who in 1999 was banged up in chokey convicted of murdering her two children after they had succumbed to SIDS.  The prosecution in that case had called as an expert witness a consultant, but mathematically incompetent, paediatrician who calculated that the chance of two babies dying of SIDS was 73 million to 1 and the jury had been led to conclude that therefore the most reasonable explanation was murder.  Subsequent statistical analysis indicated that as infanticide is even rarer in our society than SIDS, it was 9 times more likely that the unfortunate children had been victims of cot-death and eventually their mother was released.  And if you want it in pictures there is an excellent TED talk by Peter Donnelly about the case:

Here’s another weird finding.  The high-point of the baseball calendar is an event called the World Series. Americans are endearing that way, as if baseball is played in Germany, Botswana and China.  This championship requires that two very good teams compete against each other until one or other has won 4 games.  The mathematics indicates that if, in some cosmic sense, one team is better than the other in the ratio 55:45, nevertheless the weaker team will win the cup/pennon/crown 4 times out of ten.  It also suggests that you’d need a series of 269 games (and they say cricket is slow) to achieve any statistical certainty about which team really was better. 

Most importantly for RecEiression Inc., Mlodinow reviews some of the compendious literature on the psychological baggage that we’ve been carrying around since the Paleolithic.  A major drive is that we are predisposed to make sense of the world by maintaining, almost at any cost, the illusion that we are in control, so we are adept in finding patterns, particularly positive results, in random data.  Mlodinow shows that, although financial fund managers believe they are entitled to a hefty fee, they are really no better at spotting winning stocks than a fly landing on the listings in the Financial Times. This has an all too relevant bearing on the zany salaries that are (still!) paid to banking executives and other nabobs of the Irish state.  If you could pick from a pool of executives who would accept a salary of, say, 50% of what the chairman of Any Other Bank considered his due, you’d probably do no worse by your share-holders. 

No comments:

Post a Comment