Thinking Fast and Slow Notes (Part 2)

Jan. 6, 2020

Thinking: Fast and Slow

Daniel Kahneman, 2011 Thinking Fast And Slow The following are sections and ideas in condensed form that I found particularly salient.

Part 2 - Heuristics and Biases

The Law of Small Numbers - Scientists use their judgement instead of calculations to determine sample size, yielding a 50% risk of failure to confirm hypotheses. Strong recommendation to “regard their statistical intuition with proper suspicion and replace impression formation by computation whenever possible.”

Kidney cancer highest and lowest incidence in rural areas. System 1 rationalizes explanations. The argument for low: Simple living, no pollution, fresh diet
The argument for high: Poverty, high-fat diet, limited medical care access, alcohol, and tobacco.

Extreme outcomes are more likely in small samples than in large ones.

Insensitivity to sample size: “In a telephone poll of 300 seniors, 60% support the president.” What if it were 150? 3000? We pay more attention to the content of the message rather

than the reliability of the information.

Seeing patterns where none exists: WWII bombing of London map, basketball’s “hot hand”, air squadron effectiveness.

Many facts of the world are due to chance.

Anchoring

Consider two sets of questions posed at the San Francisco Exploratorium, displayed at different times:

Is the height of the tallest redwood tree greater than or less than 1200 ft? What is your best guess of the tallest redwood tree?

and

Is the height of the tallest redwood tree greater than or less than 180 ft? What is your best guess of the tallest redwood tree?

How does the anchor, 1200 ft and 180 ft, affect a person’s guess?

Anchor of 1200 ft yields an average guess of 844 ft.
Anchor of 180 ft yields an average guess of 282 ft.
(Real answer? 379.3 ft. just shy of 380 by woodpecker damage)

The ratio of the differences between anchors and best guesses is the anchoring index (844-282) / (1200-180) = (562) / (1020) = (.55)
(.55) x 100% = 55%
An anchoring index of 100% corresponds to a best guess equal to the anchor, where 0% corresponds to a best guess with no influence from the anchor. An anchoring index of about 50% is found in many real world situations.

When shown low and high asking prices for a home, professional real estate agents evaluated prices which showed an anchoring index of 41%, despite claims that the asking price has no effect on their evaluation of value. The same test given to a group of business school students with little real estate experience showed an anchoring index of 48%.

Anchoring is very powerful in dealings with money. It is a priming effect.

Anchoring was also proven to affect how judges evaluate a jail sentence. Judges were given a case about a woman caught shoplifting. They then rolled two loaded die so that every outcome was a 3 or a 9. They were then asked a pair of questions,

“Would you sentence this woman to a jail term longer or shorter than the number on the die?” “How many months will you sentence this woman to jail?”

Anchor of 9 yields an average sentence of 8 months. Anchor of 3 yields and average sentence of 5 months.

(8-5) / (9-3) = 3/6 = 50% anchoring index
Anchoring can sway a judge’s or jury’s decision.

Anchoring is a powerful tool in negotiating, which often favors the party or individual with first-mover advantage. If you are presented with an outrageous proposal, the best advice is to make a scene and storm away, refusing to negotiate. It is not suggested to make an equally outrageous counter offer which makes a large gap difficult to bridge in further negotiations.

Much more so than we want to believe, our thinking and behavior is heavily influenced by our environment. The easiest and best way to change our habits, is to change our environment.

Availability

We have no problem remembering salient, dramatic events that attract attention. Therefore, it becomes more available to our mind. Some available events include: Political sex scandal, celebrity DUI, plane crash, nuclear disaster.

We make decisions and judgements based on what information is easily available to us, not based on statistic. Now we are more apt to distrust politicians, take the train vice the plane, denounce nuclear power.

Availability Bias - famous study on marriage asks, “How large was your personal contribution to keeping the place tidy, in percentages?” Independent, self-estimated contributions of husband and wife always added up to more than 100%. The same answers were true for similar questions about initiating social interaction and taking out the trash.

Even without ability to control the feeling, It is useful to know that each member of a team is likely to have the personal feeling of carrying the weight in that team.

Norbert Schwarz experiments on self-rating.

“First, list six instances in which you behaved assertively. Now, evaluate how assertive you are.”

compare with

“First, list twelve instances in which you behaved assertively. Now, evaluate how assertive you are.”

Self-rating in assertiveness was dominated by the cognitive ease (availability) in which instances came to mind.

While being able to list twelve instances is a more impressive feat, the cognitive difficulty experienced in recalling twelve instances rather than six made people feel that they were less assertive.

Schwarz’s paradox also found in people who:
1) believe they use their bicycles less often after recalling many rather than few instances.
2) are less confident in a choice after being asked for more arguments to support it.
3) are less confident that an event was avoidable after listing more ways that it could have been avoided.
4) are less impressed by a car after listing more of its advantages.

UCLA professor boosted his approval rating by asking students to list more ways to improve his class. The cognitive difficulty in coming up with more ways to improve the class made students rate the class better, even though they physically listed more ways to improve it.

Schwarz’s paradox does not always hold true. Basing a judgement from cognitive ease is a system 1 trait. Sometimes system 2 gets involved, which focuses more on content to make the judgement. People are more apt to behave in a system 1 way if they:
1) are concurrently engaged in a mentally challenging task.

2) score low on a depression test.
3) in a good mood after thinking about something happy. 4) are knowledgeable novices on a task (not experts). 5) score high on a faith in intuition test.
6) are or are made to feel powerful.

Availability, Emotion, Risk

Public richer conception of risk than do the expert. (Paul Slovic)

Experts measure risk by # of lives lost
Public distinguishes between "good death" (random accidental) and "bad death" (voluntary activity)

Evaluation of risk depends on choice of measure.

Ways of defining mortality of toxic material release "death per million people"
"death per life year" (i.e. save the young)
"death per million dollars of product produced"

=> defining risk is an exercise in power
counter-argument by Cass Sunstein (the man with the attribute of intellectual fearlessness)

guide risk management by life years saved and dollar cost to economy policy makers are overly responsive to irrationalities of citizens.

Availability Cascade

Love Canal Affair as a pseudo-event versus environmentalists exaggerations
Alar Scare "safe to pour apple juice down the drain? Or take to toxic waste dump?"

=>Net effect was to hurt public health due to fewer apples consumed!

Numerator vs. Denominator

Amount of concern not adequately sensitive to probability of harm. Policy makers must endeavor to protect public from fear, not only real danger
"Democracy is inevitably messy"
Tom W's Speciatly

Choose Tom W's major from a list of nine majors - you know to take relative size of enrollment into account.

But if you add a description...
"intelligent but lacks creativity, neat and tidy but dull, with little sympathy for others but a deep moral sense."

BOOM, now he's a computer scientist.

Somebody reading NYT on the subway, what's a better bet? 1. She has a PhD
2. She does not have a college degree

Representativeness vs. statistics - Information about the individual case weighed much more heavily than base rate information

How to discipline your intuition.

1. Anchor your judgement of the probability of an outcome on a plausible base rate. 2. Question the diagnosticity of your evidence. (often we over exaggerate this)

Linda: Less is More

more representativeness vs. statistics

people say Linda more likely to be "feminist bank teller" than just a "bank teller".
Violation of the logic of probability (89% of undergraduates did this! 85% of Stanford Business School Students)

"The most coherent stories are not necessarily the most probable, but they are plausible, and the notions of coherence, plausibility, and probability are easily confused by the unwary."

Addind detail to scenarios makes them more persuasive, but less likely to come true. more likely? Mark has hair or Mark has blonde hair

roll a six sided die with 4 green faces and 2 red bet on...
1. RGRRR
2. GRGRRR

3. GRRRRR

sequence 2 has more green faces so people choose it over sequence 1. But it was created by appending a green roll to the front of sequence 1, which automatically makes it less probable.

Probability does not mean Plausibility.

Causes Trump Statistics

Bayesian Interference - The Taxi Cab Problem

85% of cabs in city are green
15% of cabs in city are blue
A hit and run occurred, witness identifies the cab as blue.
Witness reliability in distinguishing colors is correct 80% of time, wrong 20% of time

What is the probability that the cab in the accident was blue?

Both of the following situations involve the hit and run cab as a blue cab:
Situation 1: Witness correctly identifies hit and run cab as blue = 0.80 x 0.15 = 0.12 (12%) Situation 2: Witness incorrectly identifies green cab as being blue = 0.20 x 0.85 = 0.17 (17%)

12% + 17% = 27% chance that the witness will identify the cab as blue.
Probability that the hit and run cab identified as blue is actually blue = 0.12/0.27 = 0.41 (41%)

Stereotyping

Written in law that stereotyping in hiring and profiling is illegal. This is good to avoid drawing possibly erroneous conclusions about somebody due to group statistics.
However, completely denying stereotypes is not scientifically defensible.
Neglecting valid stereotypes results in suboptimal judgements.

Individual Microphone booths, 2 minutes to speak, one insider who pretends to have a stroke during his turn and pleas for help.
only 4/15 people respond, the others imagine somebody else will help.
Individuals feel relieved of responsibility when they know others are available and hear the request for help.

Statistically, YOU are more likely to NOT respond (27% likelihood)

Changing one's mind about human nature is hard work, and changing one's mind for the worse about oneself is even harder.

Students who watched a video of the experiment and were told the results still did not change their behavior.

We "quietly exempt ourselves" from the conclusions of experiments that surprise us.

However, a story about individual cases that is surprising, two nice people who did not help, helped students realize the generalization, that helping is more difficult than they thought.

"Subject's unwillingness to deduce the particular from the general was matched only by their willingness to infer the general from the particular."

Regression to the Mean

One of Kahneman's eureka moments when teaching Israeli Air Force instructors about psychology of effective training.

Research supports that rewards for improved performance are more effective than punishments for mistakes.

One instructor noted that when he praises cadets for a good maneuver, they do worse the next time. When he scolds them for poor performance, they do better the next time.
What he observed was Regression to the Mean = Random Fluctuations in quality of performance. However, he attached a causal interpretation to these outcomes. It is much more likely for someone to improve after a bad performance, or deteriorate after a good performance, without any praise or punishment.

"Because we tend to be nice when people please us, and nasty when they do not, we are statistically punished for being nice and rewarded for being nasty."

-------------------------------------

Success = Talent + Luck
Great Success = a little more Talent + lots of Luck

Day 1 Golf Scores do not usually reveal the winner
The more extreme the score, the more regression to the mean on Day 2

The Sports Illustrated Jinx - Players on magazine covers are notorious for not meeting expectations. Their time above the mean has already passed, then they regress back to it.

Correlation Coefficient (0-1)

Highly intelligent women tend to marry men who are less intelligent than they are. The correlation between intelligence scores of spouses is less than perfect.

Depressed children treated with an energy drink improve significantly over a three month period "It is also the case that depressed children who spend some time standing on their head or hug a cat for twenty minutes a day will also show improvement."
You need a control group. who is expected to improve by regression alone. Therefore the experiment determines whether treated patients improve more than regression can explain.

Howard Wainer's list of eminent researchers who confuse causation with correlation.

Judgement in Managerial Decision Making by Max Bazerman- Predicting next-year sales for 4 locations of a department store chain knowing that overall sales will increase by 10%.
The mistake of adding 10% to each of last year's figures.

Francis Galton's discovery that the concept of Regression to the Mean really is not so obvious. (Correlation and regression are not two different concepts - different perspectives on same concept).

Taming Intuitive Predictions

Sources of Power by Gary Klein - Chessmasters' skilled intuitions come to mind as a result of familiar cues.

Correct your Intuitive predictions

Julie is a senior in a state university. She read fluently when she was four years old. What is her GPA?

System 1 reaction = 3.7 or 3.8.

System 2 reasoning
1. Estimate an average GPA (3.0 from statistical data)
2. Determine a GPA that matches your impression of the evidence (the 3.7 or 3.8 System 1 reaction)
3. Estimate correlation between reading precocity and GPA (assume 30%)
4. Move the percentage distance from the average GPA to the one formed by impression. (3.0 x 0.30 = 0.9)

So the result would give 3.09 round up to 3.1, for a more realistic result.