Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
RSS Statistical fallacies meeting 1
The Improbability Principle Returns ‐
Luck, Lotteries, and Laura
David J. Hand Imperial College, London
and Winton Capital Management
May 2015
RSS Statistical fallacies meeting 2
Émile Borel: Borel measure
Borel‐Cantelli lemma
Heine‐Borel theorem
. . . . .
RSS Statistical fallacies meeting 3
Borel’s law: sufficiently improbable events are impossible
Nonsense?
RSS Statistical fallacies meeting 4
Borel’s law: sufficiently improbable events are impossible
Nonsense? What he meant was, we should behave as if such events are impossible:
If an event is so improbable that you’d expect it to occur just once in the history of the universe, then it would be irrational to behave as if it would happen
But nonetheless people do see extraordinarily improbable events happen
RSS Statistical fallacies meeting 5
‐ John Ironmonger: The Coincidence Authority
‐ about a London‐based professor who studies coincidences ‐ whose birth day is 30th June, my birthday ‐ and the female protagonist works at UCL, where my wife works
‐ King James Bible was published in the year that Shakespeare turned 46. The 46th word from the start of Psalm 46 is "shake" and the 46th word from the end is "spear"
‐ Shakespeare and Cervantes both died on 23 April 1616 ….?
RSS Statistical fallacies meeting 6
The RSS and ASA 104th presidents each held office in the same year RSS Presidents (normally) serve 2 years ASA Presidents (normally) serve 1 year The RSS started in 1834 The ASA started in 1839
RSS Statistical fallacies meeting 7
Edward Jarvis served 30 years as ASA president: 1852‐1882 Other early presidents served more than 1 year as ASA pres.
RSS Statistical fallacies meeting 8
Seeking to understand these apparently impossible events, people have come up with all sorts of bizarre explanations:
‐ new physical forces (synchronicity)
‐ clairvoyance
‐ intuition
‐ supernatural beings
‐ morphic resonance
‐ etc
RSS Statistical fallacies meeting 9
But none of these things are necessary The highly improbable events can be explained in perfectly sound mathematical terms In fact, the improbability principle says
extremely improbable events are commonplace and why
RSS Statistical fallacies meeting 10
The improbability principle An intertwining of five mathematical laws and the human mind
RSS Statistical fallacies meeting 11
1. The law of inevitability One of the complete set of all possible outcomes must occur How to guarantee a lottery win More subtle variants: the pigeonhole principle: if n items are put into m containers, with n > m, then at least one container must contain more than one item
RSS Statistical fallacies meeting 12
2. The law of truly large numbers
If a highly improbable event has enough opportunities to happen, it becomes almost certain
The Silverton family: 4 generations, 16 male offspring, no female
216 = 65,536 7 bn people: track forward for 16 births for each: about 100,000 such families
Alex and Donna Voutsinas, looking at old family photos a week before getting married, come across a picture of 5 year old Donna at DisneyWorld In the background is Alex, in a stroller
RSS Statistical fallacies meeting 13
3. The law of selection
Changing what you choose from (the reference class) can dramatically alter probabilities
The map at a railway station saying “You are here”
Penis enlargement emails
Stephen Jay Gould
More subtle variants: regression to the mean:
RSS Statistical fallacies meeting 14
4. The law of the probability lever Changing the circumstances a little can have a huge impact on probabilities The Sally Clark case and the independence assumption Gilbert and Sullivan opera The Yeomen of the Guard, Act 2: Fairfax: ‘And didst thou see all this?’ Point: ‘Aye, with both eyes ‐ this and that. The testimony of one eye is naught ‐ he may lie. But when it is corroborated by the other, it is good evidence that none may gainsay’
RSS Statistical fallacies meeting 15
5. The law of near enough
Relax your criteria for a match and you increase the chance of a match
What’s a lottery “win”? “A man is celebrating winning the lottery twice in four months by matching the same five numbers and the bonus ball. Mike McDermott's initial win of £194,501 on the Lotto was followed by another £121,157 last weekend.” Telegraph 12 Oct, 2002
“we take wins at face value but come up with excuses to think of almost wins as actual wins or should‐have‐been wins” Matthew Hutson
RSS Statistical fallacies meeting 16
Combinations of the laws
The classic case of Laura Buxton Laura Buxton, a ten year old girl from Stoke‐on‐Trent, released a balloon with a label attached, saying ‘ please return to Laura Buxton”, and gave her address on the other side It was blown 140 miles south west, landing in Milton Lilborne, where it was found by a ten year old girl named Laura Buxton
The two Laura Buxtons: ‐ are both fair haired ‐ are both in year 5 at school, ‐ both have three 3 year old female black labradors ‐ both have a rabbit ‐ and a guinea pig
RSS Statistical fallacies meeting 17
Incredible coincidence ! To give some perspective, there are just 28 voters called Laura Buxton in the UK
RSS Statistical fallacies meeting 18
But
The law of truly large numbers applies ‐ how many people release balloons each year?
What’s the chance that someone releases a balloon which is found by someone with the same name?
‐ 42 million people on the electoral register ‐ 12 million different names in the UK
And the law of near enough also applies:
‐ actually it was a neighbour of Laura 2 who found it. Was about to bin it when he saw the label, so delivered the balloon to Laura 2
‐ Laura 1 was only nearly 10
RSS Statistical fallacies meeting 19
People notice and remember the matching characteristics Why would you notice and remember that the Lauras ‐ had different named parents ‐ had different age guinea pigs ‐ families had different types of car ‐ took holidays in different places ‐ fathers had different jobs ‐ had differently named friends ‐ . . . . . . ‐ and so on endlessly
RSS Statistical fallacies meeting 20
And on top of this there is the human mind A collection of heuristics created by aeons of evolution To make quick rough judgements which are often correct But which also take consequences into account And so are vulnerable to false positives etc
RSS Statistical fallacies meeting 21
Hindsight bias Self‐serving bias Overconfidence bias Base rate fallacy Confirmation bias Conjunction fallacy Gamblers’ fallacy The need for narrative – everything has a cause . . . . . .
RSS Statistical fallacies meeting 22
Hindsight bias
A manifestation of the law of selection
“It is all too common for caterpillars to become butterflies and then to maintain that in their youth they had been little butterflies” George Valliant
“Life is lived forwards, but understood backwards” Kierkegaard
“After the event, of course, a signal is always crystal clear ... but before the event it is obscure and pregnant with conflicting meanings.” Leonard Mlodinow
RSS Statistical fallacies meeting 23
“History is written by the victors”
“Hindsight is 20/20”
“Should have known”
RSS Statistical fallacies meeting 24
Why? ‘Your act was unwise,’ I exclaimed, ‘as you see by the outcome.’ He solemnly eyed me. ‘When choosing the course of my action,’ said he, ‘I had not the outcome to guide me.’
Ambrose Bierce
Looking back, one can always find a sequence leading to the known now
‐ Recognise this in statistical modelling ?
RSS Statistical fallacies meeting 25
World Trade Centre:
‐ 10 July 2001, CIA received intelligence that Al‐Qaeda might be planning an attack on the US
‐ numerous publications describing deliberately crashing planes into buildings
‐ unusually large numbers of shares in insurance companies and airlines sold before the attack
RSS Statistical fallacies meeting 26
World Trade Centre:
‐ 10 July 2001, CIA received intelligence that Al‐Qaeda might be planning an attack on the US
‐ numerous publications describing deliberately crashing planes into buildings
‐ unusually large numbers of shares in insurance companies and airlines sold before the attack
But how many other strands of intelligence were there?
RSS Statistical fallacies meeting 27
Pearl Harbour:
‐ Autumn 1941: US intercepted message asking a Japanese spy to divide Pearl Harbour into five areas, and summarise the ships in those areas;
‐ weeks later US monitors lost track of radio messages from Japanese fleets;
‐ December, Japanese change call signs for second time in a month (instead of every 6 months);
‐ two days later, messages to Japanese diplomats instructing them to destroy codes and secret documents;
‐ etc
RSS Statistical fallacies meeting 28
Same issue as overfitting in statistics
RSS Statistical fallacies meeting 29
Same issue as overfitting in statistics One can always fit a conspiracy theory to the data
RSS Statistical fallacies meeting 30
Also
People have an inbuilt inclination to consistency: when asked about their earlier state of mind, they tend to reconstruct their current beliefs
After an event occurs, people give a higher probability to their earlier belief that the event would occur
One consequence is that people underestimate how much they are surprised by events
Self fulfilling prophecies
RSS Statistical fallacies meeting 31
Also We choose which events to remember:
The German “Karicartoon” calendar, published by a small Berlin‐based company, gives a witty cartoon for each day of the year.
On Sunday 10 February 2011, it showed a drawing of the Pope sitting in front of the TV while the lottery numbers are announced.
On finding that he has all seven numbers correct, the caption shows him exclaiming: “Holy s**t! I’ll resign tomorrow!”
RSS Statistical fallacies meeting 32
Also We choose which events to remember:
The German “Karicartoon” calendar, published by a small Berlin‐based company, gives a witty cartoon for each day of the year.
On Sunday 10 February 2011, it showed a drawing of the Pope sitting in front of the TV while the lottery numbers are announced.
On finding that he has all seven numbers correct, the caption shows him exclaiming: “Holy s**t! I’ll resign tomorrow!”
11 Feb 2013, Vatican announced that the Pope would resign
RSS Statistical fallacies meeting 33
And which to emphasise: In The Old Curiosity Shop, Charles Dickens describes the conversation between the mothers of two characters, Tim and Barbara, who are meeting for the first time:
RSS Statistical fallacies meeting 34
And which to emphasise: In The Old Curiosity Shop, Charles Dickens describes the conversation between the mothers of two characters, Tim and Barbara, who are meeting for the first time:
“And we are both widows too!” said Barbara’s mother. ‘We must have been made to know each other’....tracing things back from effects to causes, they naturally reverted to their deceased husbands, respecting whose lives, deaths, and burials, they compared notes, and discovered sundry circumstances that tallied with wonderful exactness; such as Barbara’s father having been exactly four years and ten months older than Kit’s father, and one of them having died on a Wednesday and the other on a Thursday, and both of them having been of a very fine make and remarkably good‐looking, with other extraordinary coincidences.”
RSS Statistical fallacies meeting 35
Let’s take the famous Lincoln/Kennedy coincidence:
Both were assassinated in office ‐ Lincoln's name has 7 letters ‐ Kennedy's name has 7 letters ‐ In Lincoln's and Kennedy's names the vowels and consonants fall in exactly the same place; in the order c, v, c, c, v, c, c ‐ Lincoln was elected to Congress in 1846 ‐ Kennedy was elected to Congress in 1946 ‐ Lincoln was elected president in 1860 ‐ Kennedy was elected president in 1960
RSS Statistical fallacies meeting 36
‐ Kennedy had a secretary named Lincoln ‐ War was thrust upon Lincoln almost immediately after inauguration ‐ War was thrust upon Kennedy almost immediately after inauguration ‐ Lincoln ordered the Treasury to print its own money ‐ Kennedy ordered the Treasury to print its own money ‐ Lincoln gave black Americans freedom and legalized equality ‐ Kennedy enforced equality for black Americans ‐ Lincoln delivered the Gettysburg Address on November 19, 1863 ‐ Kennedy was assassinated on November 22, 1963
RSS Statistical fallacies meeting 37
‐ Lincoln was loved by the common people and hated by the establishment ‐ Kennedy was loved by the common people and hated by the establishment ‐ Lincoln was succeeded, after assassination, by vice‐president Johnson ‐ Kennedy was succeeded, after assassination, by vice‐president Johnson ‐ Andrew Johnson was born in 1808 ‐ Lyndon Johnson was born in 1908 ‐ Andrew Johnson's name has 13 letters ‐ Lyndon Johnson's name has 13 letters ‐ Andrew Johnson had a pug nose and slicked‐back hair ‐ Lyndon Johnson had a pug nose and slicked‐back hair
RSS Statistical fallacies meeting 38
‐ Lincoln was sitting beside his wife when he was shot ‐ Kennedy was sitting beside his wife when he was shot ‐ Rathbone, who was with Lincoln when he was shot, was injured (by being stabbed) ‐ Connally, who was with Kennedy when he was shot, was injured (by being shot) ‐ Rathbone's name has 8 letters ‐ Connally's name has 8 letters ‐ Lincoln's wife held his head in her lap after he was shot ‐ Kennedy's wife held his head in her lap after he was shot ‐ Lincoln was shot on a Friday ‐ Kennedy was shot on a Friday
RSS Statistical fallacies meeting 39
‐ Lincoln was shot in a theater named Ford ‐ Kennedy was shot in a car named Ford ‐ Kennedy was shot in a car also named Lincoln ‐ Lincoln's bodyguard was away from his post at the door of the President's box at the theater ‐ Kennedy's bodyguards were away from their posts on the running‐boards of the President's car ‐ Lincoln was shot in a theater and his assassin ran to a warehouse ‐ JFK was shot from a warehouse and his assassin ran to a theater ‐ Lincoln's assassin had a three‐word name, John Wilkes Booth ‐ Kennedy's assassin had a three‐word name, Lee Harvey Oswald
RSS Statistical fallacies meeting 40
‐ John Wilkes Booth has 15 letters ‐ Lee Harvey Oswald has 15 letters ‐ John Wilkes Booth was born in 1839 (s/b 1838) ‐ Lee Harvey Oswald was born in 1939 ‐ Lincoln didn't die immediately after being shot ‐ Kennedy didn't die immediately after being shot ‐ Lincoln and Kennedy died in places beginning with the initials P and H ‐ Lincoln died in Petersen's house ‐ Kennedy died in Parkland Hospital ‐ Booth was shot and killed in police custody before going to trial ‐ Oswald was shot and killed in police custody before going to trial
RSS Statistical fallacies meeting 41
‐ Andrew Johnson was a heavy drinker with crude behavior ‐ Lyndon Johnson was a heavy drinker with crude behavior ‐ There were conspiracy theories that Johnson was knowledgeable about Lincoln's assassination ‐ There were conspiracy theories that Johnson was knowledgeable about Kennedy's assassination ‐ Days before it happened Lincoln told his wife and friends about a dream he'd had of being shot by an assassin ‐ Hours before it happened Kennedy told his wife and friends it would be easy for an assassin to shoot him from a crowd ‐ Shortly after Lincoln was shot the telegraph system went down ‐ Shortly after Kennedy was shot the telephone system went down
RSS Statistical fallacies meeting 42
‐ Kennedy's father had been the Ambassador to England at the Court of St James ‐ Lincoln's son became the Ambassador to England at the Court of St James ‐ Lincoln and Kennedy were 2 of the greatest presidents of the nation ‐ Lincoln's wife tastefully and expensively re‐decorated the White House ‐ Kennedy's wife tastefully and expensively re‐decorated the White House ‐ Lincoln loved great literature and could recite poetry by heart ‐ Kennedy loved great literature and could recite poetry by heart ‐ Lincoln had young children while living at the White House ‐ Kennedy had young children while living at the White House ‐ Lincoln's sons had ponies they rode on the White House grounds ‐ Kennedy's daughter had a pony she rode on the White House grounds
RSS Statistical fallacies meeting 43
‐ Lincoln lost a child (12 year old son) to death while President ‐ Kennedy lost a child (newly born son) to death while President ‐ Lincoln had 2 sons named Robert and Edward. Edward died young and Robert lived on.
‐ Kennedy had 2 brothers named Robert and Edward. Robert died young and Edward lived on
‐ Lincoln let his children run and play in his office ‐ Kennedy let his children run and play in his office ‐ After Lincoln's assassination the nation experienced an emotional convulsion ‐ After Kennedy's assassination the nation experienced an emotional convulsion
RSS Statistical fallacies meeting 44
‐ Lincoln's funeral train traveled from Washington DC to New York ‐ Kennedy's brother's funeral train traveled from New York to Washington DC ‐ Lincoln Assassination conspiracy theories are believed all these years later ‐ Kennedy Assassination conspiracy theories are believed all these years later ‐ Abraham was the first name of the man who filmed Kennedy's murder in the Lincoln ‐ The man running alongside Kennedy's car snapping pictures with his 35mm camera was a salesman of Lincoln cars ‐ Jefferson Davis was the name of the President of the Confederate States while Lincoln was president of the United States ‐ Jefferson Davis Tippit was the name of the police officer killed by Kennedy's assassin
RSS Statistical fallacies meeting 45
‐ Lincoln was famous for his wit and for telling hilarious stories and anecdotes ‐ Kennedy was famous for his wit and for telling hilarious stories and anecdotes ‐ Lincoln was sitting in a rocking chair at Ford's Theater when he was shot ‐ Kennedy had a special rocking chair he sat in at the White House ‐ Kennedy's seat in the Lincoln he was sitting in when he was shot is in Ford's museum ‐ Lincoln's seat in the Ford he was sitting in when he was shot is in Ford's museum ‐ John Kennedy is the name of a character in a 1951 movie about a detective traveling by train to thwart the assassination of President Lincoln ‐ John Kennedy is the name of the real‐life detective who traveled in the train with President Lincoln in 1860 to thwart his assassination
RSS Statistical fallacies meeting 46
You will note, however, that the names Abraham Lincoln and John F. Kennedy have 12 and 14 letters respectively.
RSS Statistical fallacies meeting 47
You will note, however, that the names Abraham Lincoln and John F. Kennedy have 12 and 14 letters respectively. However, as one of my correspondents helpfully points out Abraham is an elongated form of Abram, so that Abram Lincoln has 12 letters. My correspondent added “this is what’s known as a ‘clue’ ”
RSS Statistical fallacies meeting 48
Hindsight bias is not merely a failure of the naive: The eminent historian E.H.Carr said “When I studied ancient history in this university many years ago, I had as a special subject ‘Greece in the period of the Persian Wars’. I collected fifteen or twenty volumes on my shelves and took it for granted that there, recorded in these volumes, I had all the facts relating to my subject. Let us assume ‐‐ it was very nearly true ‐‐ that those volumes contained all the facts about it that were then known, or could be known. It never occurred to me to inquire by what accident or process of attrition that minute selection of facts, out of all the myriad facts that must once have been known to somebody, had survived to become the facts of history.”
RSS Statistical fallacies meeting 49
Hindsight bias as a form of selection bias Survivorship bias Length bias Dropout bias Publication bias . . . . . .
RSS Statistical fallacies meeting 50
The tendency to regard escaping from a disaster as good luck
Many lucky events are associated with narrowly avoiding bad outcomes.
RSS Statistical fallacies meeting 51
The tendency to regard escaping from a disaster as good luck
Many lucky events are associated with narrowly avoiding bad outcomes.
Divide a roulette wheel into (i) three wedges of equal size, red yellow blue (ii) 18 sections of equal size, 6 of each colour
85% of subjects felt it was luckier to land on the red in case (ii) than on case (i).
[Because in case (ii) landing on red is always a close call ?]
RSS Statistical fallacies meeting 52
Subtleties ! ‐ the Monty Hall problem ‐ the exchange paradox ‐ the sibling gender problem ‐ . . . . . . . . . . . . . . . . . . . . . . H T H H T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . H H H H H. . . . . . . . . . . . . . . . . . . . . .
RSS Statistical fallacies meeting 53
So: Statistical fallacies are only statistical in subject matter They are not failures of statistics Any more than “probability paradoxes” represent contradictions in mathematics It’s a question of improved understanding, education, and intuition