18.+Judgment+&+heuristics

April 24, 2015
 * 18. Judgment & heuristics**

Representativeness Extension neglect
 * Outline**
 * Applications
 * Sample size
 * Base rates
 * Duration
 * Scope


 * Sequences that alternate 50% are usually perceived as non-random[[image:xoxoxoxo.png width="213" height="25" align="right"]]
 * The "law of small numbers"
 * Not really a law, but laypeople think it is
 * We think...
 * Small samples should represent the larger pattern
 * 5 Os in a row shouldn't happen
 * Random sequences are perceived as containing non-random "streaks"
 * We are bad at judging randomness

Hot Hand
 * Gilovich, Vallone, Tversky (1985) - Do shooters actually get hot (and cold)? Or are shooting streaks actually just random sequences that we perceive as non-random?
 * Surveyed 100 basketball fans
 * 91% said shooting percentages are better after 2 or 3 makes than 2 or 3 misses
 * Estimated that a 50% shooter would shoot 61% after a make, 42% after a miss
 * 84% said pass to whoever has the hot hand
 * Looking at data from 2005/2006 NBA season
 * Found that people tend to shoot better after a miss (when they're "ice cold") than after they make a shot ("red hot")
 * [[image:hot hand.png width="310" height="214"]]
 * Are there streaks?
 * Everyone agrees: yes
 * XOXOXXXXOOXOOOOOXXXOX
 * Are they random or do shooters get hot?
 * Athletes and coaches: Shooters get hot!
 * Scientists, statisticians: They're random!


 * Representativeness Heuristics**
 * When making a judgment about an item or category, we rely on resemblance
 * Two types of representativeness
 * Using an example to draw inferences about a category
 * Using a category to draw inferences about an example
 * Bias against an individual based on race
 * Judge an example based on the population
 * "Man Who" arguments
 * Judge population based on example
 * Hot hand
 * We expect a small number of observations to look like the larger category
 * If any run (even a short one) doesn't have some makes and misses, we conclude it's non-random
 * Why it's called "representativeness"
 * Saying a sample "represents" a population means it has the same properties as the population. It resembles the population
 * We assume samples are representative of their parent populations
 * Heuristic - a mental shortcut that usually leads to the right answer
 * Example: "The subject is a 33-year old psychologist who loves to read. The person’s hobbies are cooking and watercolor painting. The person is especially interested in how children think, has taken many classes in this area, and likes to talk to children about their problems."
 * The subject described is rated as more likely to be a "female clinical psychologist" than a "clinical psychologist"
 * These ratings cannot be correct: "Female clinical psychologist" is a subset of "clinical psychologist"
 * Judgments of category membership based on resemblance
 * Conjunction fallacy: mistake of assuming a conjunction is more likely than a constituent event
 * Female clinical psychologist > clinical psychologist
 * Linda the feminist bank teller
 * Representative heuristic means two things
 * We make judgments based on resemblance
 * We exclude other relevant information

**Extension neglect**: we don't attend to the size of the problem
 * Four examples:
 * Sample size neglect
 * Base-rate neglect
 * Duration neglect
 * Scope neglect


 * Sample Size Neglect**
 * Sample: subset of a population that is being measured in a study
 * The law of large numbers: the larger the sample size, the closer the average value is to the average of the population
 * Sample size should influence answers in the hospital problem
 * 60 births is a larger sample than 15 births
 * Number of males should be closer to 50% in the hospital with more births
 * Truth: larger samples produce less extreme averages
 * Our bias: we naturally ignore sample size
 * You go to college for four years. Demonstrate your ability, etc.
 * Go in for a job interview
 * Based on a 30 minute "sample," employer makes a decision
 * Evidence suggests not interviewing can produce better hiring decisions


 * Base Rate Neglect** - we do not take into account information about the broad likelihood of a particular category or type of event (e.g., number of people doing different jobs
 * Example: "I have a friend who is about medium height. She loves to knit and enjoys independent and art house movies. Which of the following is her field?" (1. teaches piano OR 2. works in retail)
 * People tend to answer "teaches piano" even though there are many more people who work in retail
 * Mammogram question:
 * Mammogram question: Assume Penny has a 3% chance (base rate) of having breast cancer given her health, age, etc. Then she gets a positive mammogram. What's the chance that she has cancer?
 * [[image:mammogram.png]]
 * People tend to get this wrong because they ignore the base rate.
 * When we ignore base rates, we rely on the representativeness heuristic
 * All we care about is what does the description resemble
 * Lawyers vs. engineers (in your book)
 * Piano teacher versus retail
 * Are more graduate students first-born or second-born children?
 * Do more hotel fires start on the first 10 floors or the second 10 floors?


 * Duration Neglect**
 * We aren't sensitive to how long things last
 * What comes to mind is salient moments
 * We assume they are representative
 * Peak/End rule
 * retrospective ratings can be predicted based on
 * Peak experience (how bad/good did it get?)
 * End experience
 * Kahneman & Frederick (2002): Patients underwent colonoscopy. Reported their level of pain every 60 seconds. Afterwards, asked to rate overall experiences.
 * Condition A: Normal procedure
 * Condition B: left the probe in for an extra 25 minutes
 * Leaving the scope in for a while made it seem more enjoyable, in retrospect
 * Patients were more likely to choose colonoscopy months later (as opposed to alternative approach)
 * Also works with painful clamps, ice water, etc.
 * Chajut et al. (2014): Research assistant asked women giving birth to rate their pain every 20 minutes
 * Retrospective ratings of child birth depends on:
 * Peak
 * End
 * Overall average
 * But they don't depend on:
 * Duration


 * Scope Neglect**
 * A teenager was arrested for peeing in a 38,000/38,000,000 gallon reservoir. Should they drain the reservoir?[[image:gallon.png width="311" height="230" align="right"]]
 * 38,000: 12% said yes
 * 38,000,000: 11% said yes
 * An oil spill in the gulf is threatening migrating birds. How much would you pay to save birds?
 * 2,000 birds: $80
 * 20,000 birds: $78
 * 200,000 birds: $88
 * Would you support buying expensive new equipment to increase airport safety?
 * Condition 1: It will save 150 lives
 * Condition 2: It will save 98% of 150 lives (147 lives)
 * More likely to support condition 2. 150 seems arbitrary. But 98% is really good.
 * Small et al. (2007):
 * Story group: "Rokia, a 7-year-old girl from Mali, Africa, is desperately poor and faces threat of severe hunger or even starvation. etc..."
 * Statistics group: "More than 11 million people in Ethiopia need immediate food assistance (etc)"
 * Statistics and Story group
 * [[image:small.png]]
 * Statistics turn off emotional reactions
 * Problem size can be overwhelming
 * We neglect the scope of the problem