Bounded Rationality

Topics for this week

If people can’t (or won’t) optimize (maximize expected utility), what alternative decision rules might they use? And how successful are they? What biases might they introduce?

• What are “heuristics”?
• What is “bounded rationality”?
• In what ways might heuristics create problems in judgment and decision making?
• In what ways might heuristics help people to make good decisions with limited information?
• How have ideas about heuristics and bounded rationality been applied to political decision making?

Heuristics

A heuristic is a decision strategy that ignores part of the information - two broad traditions of research:

• Heuristics and biases (Kahneman & Tversky)

  • Kahneman: “Our research attempted to obtain a map of bounded rationality, by exploring the systematic biases that separate the beliefs that people have and the choices they make from the optimal beliefs and choices assumed in rational-agent models.”

• Bounded rationality / “fast and frugal” decision making (Simon, Gigerenzer)

  • Gigerenzer & Gaissmaier: “A heuristic is a decision strategy that ignores part of the information, with the goal of making decisions more quickly, frugally, and/or accurately than more complex methods.”

Heuristics and biases: Availability

The availability heuristic uses the ease with which something comes to mind as a substitute for information about prevalence, importance, probability, goodness, etc.

• It is not unreasonable to use such a strategy
• Higher frequency events have higher baseline accessibility in long-term memory
• Familiarity is often a sign of safety - we often have more experience with good and non-threatening things
• But ease of recall is also shaped by irrelevant factors, e.g., vividness of presentation, media biases

Examples of applications to politics

• Frequency of repetition and desirability

  • Candidate name recognition
  • Repeat it until it sounds true

• Importance or weight of different considerations

  • Agenda-setting, focusing illusion
  • Seen and unseen, e.g., unintended consequences, opportunity costs

• Taking risks:

  • Overweight easily imagined risks
  • Optimism bias: easy to imagine path to success, hard to imagine all the ways you can fail

Heuristics and biases: Representativeness

The representativeness heuristic assesses the probability that an instance \(A\) belongs to a class \(X\) by the degree to which \(A\) is “representative” of \(X\)

What does it mean for something to be “representative”?

• resemblance to prototype or stored representation of \(X\) in memory (e.g., a stereotype of a group)

• degree to which \(A\) possesses features diagnostic of \(X\)

  • features are “diagnostic” when they have high relative frequency compared to a comparison class
  • Not the same as high absolute frequency!

Examples: Randomness

Which of the following sequences of tosses of a fair coin is most likely?

• H-T-H-T-T-H
• H-H-H-T-T-T
• H-H-H-T-H-H

Examples: Linda problem

Examples: baserate neglect

https://thedecisionlab.com/biases/base-rate-fallacy

Examples: Stereotypes and biases

Stereotypes often focus on large relative differences between groups - traits that are diagnostic rather than common for a group

• The most common traits in a group are rarely core aspects of a stereotype (e.g., 2 arms and 2 legs)
• Diagnositicity = \(\Pr(trait | group A) \space / \space \Pr(trait | group B)\)

Diagnosticity and prevalence

Pernicious stereotypes

Many negative stereotypes may be fostered by this kind of cognitive process

• A small absolute difference between groups in a negatively valued attribute (e.g., crime rates) generate large differences in representativeness via diagnosticity
• People vastly overestimate the probability that an out-group member is a member of the broader class defined by the attribute

Example (using Bordalo et al. (2016), “Stereotypes”)

# Group A has 1% crime rate, Group B has 2% crime rate
A <- c(0.01, 0.99)
B <- c(0.02, 0.98)

# representativeness of criminality for each group
R_A <- A / B
R_B <- B / A

# stereotype: what is probability that randomly drawn member of group is criminal?
S_A <- A * R_A / sum(A*R_A)
S_B <- B * R_B / sum(B*R_B)

# true difference in probability between groups
true_diff <- B[1] - A[1]

# distorted (stereotyped) diff in probability between groups
stereo_diff <- S_B[1] - S_A[1]

# compare
comp <- round(cbind(true = c(A[1],B[1],true_diff), 
                    stereotype = c(S_A[1],S_B[1], 
                                   stereo_diff)), 3)
row.names(comp) <- c("A","B","B-A")
comp

    true stereotype
A   0.01      0.005
B   0.02      0.040
B-A 0.01      0.035

False polarization in politics

Ahler (2014)

Bounded rationality (Herbert Simon)

Catching a baseball

https://www.wired.com/story/how-do-people-actually-catch-baseballs/

Catching a baseball

https://www.wired.com/story/how-do-people-actually-catch-baseballs/

General heuristic decision rules

Decision theorists have studied a number of general heuristic decision rules: strategies that can be applied across many decision contexts

• Defined by differences in what information (and how much) is ignored

A running example

A running example

Who would the voter choose if they use all information (give +1 times weight for matches and -1 times weight for mismatches)?

Equal weights

Who would voter choose with equal weights?

Satisficing

Satisficing

Let’s say the voter’s aspiration level is utility greater than 0

• Who will the voter choose (using satisficing) if they get +1 times weight for match and -1 times weight for mismatch?

Lexicographic

Lexicographic rules consider alternatives on one attribute at a time, in order of importance,

• Choose the alternative that is better than all others on the most important attribute
• If there is a tie, go to the next attribute
• Continue until a decision is made

Lexicographic

Who will the voter choose using lexicographic?

Elimination by aspects

Elimination by aspects is a rule created by Amos Tversky

• Set aspiration levels for each attribute
• Consider attributes one at a time, eliminating alternatives that do not meet the threshold
• Continue until only one is left

Elimination by aspects

Who will the voter choose using EBA (aspiration level is agreement)?

Recognition / Fluency heuristic

Which of these two cities in Wales has a larger population?

• Wrexham
• Bangor

Accuracy / Effort tradeoffs

How can we evaluate how successful these heuristic strategies are?

• Define “accuracy” as how well the heuristic rule does compared to the full-information, expected value rule (i.e., the optimizing rule)

• Define “efficiency” as the number of “cognitive operations” required to implement the rule, e.g.,

  • find a piece of information
  • add two numbers together
  • compare two alternatives

Strategy success is conditional

Payne et al. (1993)

Interest groups

Lupia (1994)

Interest groups

Lupia (1994)

Interest groups

Arceneaux and Kolodny (2009)

AI?

Issue publics

Converse (1964)

…large portions of an electorate do not have meaningful beliefs, even on issues that have formed the basis for intense political controversy among elites for substantial periods of time…where any single dimension is concerned, very substantial portions of the public simply do not belong on the dimension at all. They should be set aside as not forming any part of that particular issue public…One man takes an interest in [racial issues] and is relatively indifferent to or ignorant about controversies in other areas. His neighbor may have few crystallized opinions on the race issue, but he may find the subject of foreign aid very important. Such sharp divisions of interest are part of what the term “issue public” is intended to convey.

One good reason

Much work suggests partisanship may be a “standing decision” - but when do people “defect”?

• If people care mostly about a small number of issues important to them, they will defect when:

  • the important issue becomes salient in the campaign
  • their position on the issue conflicts with their party’s
  • the candidate running is typical of their party on the issue

Hillygus and Shields (2008), The Persuadable Voter

Hillygus and Shields (2008), The Persuadable Voter

Studying the process of decision making

Process-tracing is the study of decision making by observing the intermediate steps decision makers take to reach a decision

• Eye-tracking
• Static or dynamic information boards
• Virtual internet or social media environments

Static information boards

Jenke et al. (2024)

Eye-tracking

Virtual social environments