Lecture 15 Heuristics & Framing Effects

  • Heuristics
    • K&T's Heuristics & Biases program identified 3 heuristics, employed
      • to assess probabilities (of an uncertain event, value of an uncertain quantity, is a complex task)
      • to predict values
    • Reliance on a limited number of heuristic principles which reduce this complex task \(\to\) simpler judgmental operations
    • Heuristics may lead to severe and systematic errors!
  • Representative Heuristic
    • \(A\) is highly representative of \(B\) then... \(B \implies A\)
    • Probability that Steve is engaged in a particular occupation: farmer, salesman, airline pilot
      • \(P(\text{Steve = Librarian})\) is assessed by \(P(\text{Steve represents a Stereotype Librarian} )\)
    • Can be considered as similarity heuristic
    • To judge \(A \in B\), people ask themselves (an easier question), how similar is \(A\) to their image or stereotype of \(B\)?
    • 6-foot-8-inch Afro-American more likely to be a professional basketball player than a 5-foot-6-inch Jewish (not many of such Jewish Basket ball players these days)
    • This approach to probability leads to serious errors
    • Factors:1
      1. Insensitivity of prior probabilities
      2. Insensitivity to sample size
      3. Misconception of chance
      4. Insensitivity to predictability
      5. the illusion of validity
      6. Misconceptions of regression
  • Availability Heuristic
    • Assess frequency of class / probability of event by the ease with which instances or occurrences can be brought to might.
      • How easy it is to think of a situation? If it's difficult to get an example, it assigns a lower probability to that outcome.
      • Risk of heart attack among middle-aged people by recalling such occurrences among one's acquiantances
      • Evaluate probability of failure by imagining various difficulties that can be encountered
    • e.g. Self Driving cars
      • In 2022, 45,514 people died in motor vehicle traffic crashes:
        • 39% speeding-related crashes
        • 32% alcohol-impaired driving fatalities
      • Self Driving cars can reduce such deaths considerably
      • But there have been weird instances highlighted in the news about accidents related to self driving cars.2 But nobody talks about the positive side (the number of lives that could have been saved) since that is something that "could have happened but was avoided by the self-driving cars".
      • Google's Waymo & Tesla's Cybercab. Annual survey by AAA (from 2021 \(\to\) 2024)
        • Trust? 14% \(\to\) 9%
        • Afraid? 54% \(\to\) 66%
    • e.g. Personally experienced earthquake makes you feel that eq is more likely than reading about it in a weekly magazine
    • \(\implies\) Vivid and easily imagined causes of death get more probability. (e.g. tornadoes)
      • less-vivid (like asthma) receive low, even if they are more frequent
    • \(\implies\) Recent events have a greater impact on behavior, fears than earlier ones
    • The Automatic System is keenly aware of the risk without any statistical tables
    • AH is a useful clue for assessing frequency / probability. However it's affected by other factors (than freq/prob)
    • Leads to biases due to retrievability of instances, effectiveness of search set, of imaginability, illusory correlation
  • Adjustment & Anchoring
    • People make estimates by starting from an initial value and make adjustments to it to get to the final answer.3
    • Initial value
      • suggested by formulation for the problem
      • result of partial computation
    • Adjustments are insufficient \(\Leftarrow\) Different starting points yield different value.
    • E.g. Guess the population of Jaipur
      • First, you will think of a city that you know of. And check its population.
      • But, since people may know limited number of cities, the answer may vary about a large amount
    • A&A can also affect how we think about life - College students were asked two questions
      • How happy?
      • How often are you dating?
      • When the questions were asked in this order, the correlation between the two questions was 0.11
      • When reversed, the correlation was 0.62.
        • They used the "dating heuristic" to answer the question about how happy they are
    • Problem: Even obviously irrelevant anchors creep into the DM process.
      • e.g. Write the last two digits of Aadhar (social security number) and then value a certain set of commodities
      • Top 20% (80 to 99 last two digits) bid highest and the difference between them and lowest 20% (00 to 20) was 216% to 346%!
      • Can be explained in terms of a theory of arbitrary coherence. "I think this is related to the question I was asked..."
  • The Theory of Arbitrary Coherence
    • Valuations of goods and experiences have a large arbitrary component.
    • Subsequent valuations are coherent, they are scaled appropriately relatively to the first
    • e.g. '99' in context of hamburger price ($.99) but in context of a meal ($99)
  • K&T noted that anchoring leads to
    • Insufficient adjustment
    • Biased evaluation in conjunctive and disjunctive events
    • Assessment of subjective probability distribution
  • Framing effects
    • How you create the context and execute the process in asking questions determine people's values, attitudes & preferences
    • 90/100 individuals live after 5 years. 10/100 individuals die after 5 years. Automatic system: "A significant number of people are dead and I might be one of them!"
    • \(\implies\) choices depend on the way a problem is stated. (Matters in many domains)
    • Matters a lot in public policy
      • (a) Use energy conservation \(\implies\) save ₹10,000
      • (b) Don't \(\implies\) lose ₹10,000
      • (b) was found to be more effective
    • Why it works? People tent to be somewhat mindless, passive decision-makers
    • Reflective system should ideally check whether the response would change if the question is framed differently. But it doesn't.
      • Why doesn't it? If there is a contradiction, then what to do? Which choice to select?
    • e.g. Happiness & Dating question
      • If the first question influences answer to the second... called "procedural invariance", a violation of the standard model of EUT
    • FE \(\implies\) PR
    • Preference Reversal
      • \(A \succ B\) if posed in one way and \(B \succ A\) if posed differently
      • OR, when preferences are intransitive \(\implies\) PR
        • \(A \succ B\quad B \succ C\) but \(C \succ A\)
      • Due to FE or due to temporal differences (meaning, preferences change over a period of time)
    • Evidence: FE is widespread
      • e.g. Products are evaluated more favorably when surrounding environment contain related cues
      • Quantity consumed can be influenced by size of plates, packages / serving bowls used.
      • Large plate sizes increase consumption: 15% \(\to\) 45%
      • Principle operates even if cues are not intrinsically related to the product
        • When NASA landed the Pathfinder on Mars in 1997, the sales of Mars Bars increased even thought it was not named after the planet, but after the founder.

  1. Individual discussion is beyond the scope of the course ↩

  2. A lady came into the path of a self driving car, it got confused and dragged her for another 20 feet. People argued, that had there been a human, the life of the lady could have been saved. ↩

  3. A very good idea. If people don't have an opinion about you, you can actually create any impression on them, and then they believe it. However, if you create a bad impression initially, they will anchor to it and even after you do a lot of good work, they will adjust slightly but you will remain in the ballpark of a bad person. As they say, first impressions are always important ↩