Simple Heuristics

1
Simple Heuristics That Make Us Smart
Gerd Gigerenzer
Max Planck Institute for Human Development Berlin
2

• Which US city has more inhabitants,
• San Diego or San Antonio?

Americans 62 correct
Germans ? correct
Germans 100 correct
Goldstein Gigerenzer, 2002, Psychological
Review
3
Recognition Heuristic
If one of two objects is recognized and the other
is not, then infer that the recognized object
has the higher value.
Ecological Rationality
The heuristic is successful when ignorance is
systematic rather than random, that is, when lack
of recognition correlates with the criterion.
4
(No Transcript)
5
Wimbledon 2003
Correct Predictions
69
70
68
66
60
50
ATP Entry Ranking
ATP Champions Race
Seedings
Recognition Laypeople
Recognition Amateurs
Frings Serwe (2004)
6
Wimbledon 2003
Correct Predictions
72
69
70
68
66
66
60
50
ATP Entry Ranking
ATP Champions Race
Seedings
Recognition Laypeople
Recognition Amateurs
Frings Serwe (2004)
7
The Less-is-More Effect
The expected proportion of correct inferences c is
where
is the number of recognized objects is the total
number of objects is the recognition validity,
and is the knowledge validity
n
N
a
b
A less-is-more effect occurs when
a gt b
8
80
b.8
75
b .7
70
Percentage of Correct Inferences ()
65
b .6
60
55
b .5
50
0
50
100
Number of Objects Recognized (n)
9
Ignorance-based Decision Making Recognition
heuristic

• kin recognition in animals
• ? food choice failures of aversion learning in
  rats (Galef et al. 1990)
• overnight fame (Jacoby et al., 1989)
• less-is-more effect (Goldstein Gigerenzer,
  2002)
• advertisement without product information
  (Toscani, 1997)
• consumer choice based on brand name
• picking a portfolio of stocks (Borges et
  al.,1999 Boyd, 2001 Ortmann et al., in press)
• Institutions competing for the publics
  recognition memory

10
Gaze heuristic
11
Gaze heuristic
12
Gaze heuristic
13
Gaze heuristic
14
Gaze heuristicOne-reason Decision Making

• Predation and pursuit
• bats, birds, dragonflies, hoverflies, teleost
  fish, houseflies
• Avoiding collisions
• sailors, aircraft pilots
• Sports
• baseball outfielders, cricket, dogs catching
  Frisbees
• NOTE Gaze heuristic ignores all causal relevant
  variables
• Shaffer et al., 2004, Psychological Science
  McLeod et al., 2003, Nature

15
Three Visions of Bounded Rationality

• People act like econometricians
• As-if optimization
• under constraints

• People dont, deviations
• indicate reasoning fallacies
• Cognitive limitations

There is a world of rationality beyond
optimization Fast and frugal heuristics
Ecological rationality
Gigerenzer Selten (Eds.) (2001). Bounded
Rationality The Adaptive Toolbox. MIT Press.
16
When Is Optimization Not An Option?

• Well-defined problems
• NP-hard problems (e.g., chess, traveling
  salesman, Tetris, minesweeper)
• Criterion lacks sufficient precision (e.g.,
  Mills greatest happiness of all acoustics of
  concert hall)
• Multiple goals or criteria (e.g.,shortest,
  fastest, and most scenic route)
• Problem is unfamiliar and time is scarce (Selten,
  2001)
• In domains like mate choice and friendship,
  calculated optimization can be morally
  unacceptable.
• All ill-defined problems

17
Four Mistaken Beliefs

• People use heuristics because of their cognitive
  limits.
• Real-world problems can always be solved by
  optimization.
• Heuristics are always second-best solutions.
• More information is always better.

18
The Science of Heuristics

• Descriptive Goal The Adaptive Toolbox
• What are the heuristics, their building blocks,
  and the abilities they exploit?
• Normative Goal Ecological rationality
• What class of problems can a given heuristic
  solve?
• Engineering Goal Design
• Design heuristics that solve given problems.
• Design environments that fit given heuristics.
• Gigerenzer et al. (1999). Simple heuristics that
  make us smart.
• Oxford University Press.

19
The Adaptive Toolbox Heuristics, Building
Blocks, Evolved Abilities

• Gaze heuristic fixate ball, start running, keep
  angle constant
• Building block fixate ball
• Evolved ability object tracking
• Tit-For-Tat cooperate first, keep memory of
  size one, then imitate.
• Building block cooperate first
• Evolved ability reciprocal altruism

20
Sequential Search Heuristics

• Take The Best
• Search rule Look up the cue with highest
  validity vi
• Stopping rule If cue values discriminate (/-),
  stop search. Otherwise go back to search rule.
• Decision rule Predict that the alternative with
  the positive cue value has the higher criterion
  value.

• Tallying
• Search rule Look up a cue randomly.
• Stopping rule After m (1ltmM) cues stop search.
• Decision rule Predict that the alternative with
  the higher number of positive cue values has the
  higher criterion value.

Dont add Dont weight
21
The Adaptive Toolbox Heuristics

• Class
• Ignorance-based decisions
• One-reason decisions
• Tallying
• Elimination
• Satisficing
• Motion pattern
• Cooperation

• Heuristic
• Recognition heuristic, fluency heuristic
• Take The Best, Take The Last, QuickEst, fast
  frugal tree
• Tally-3
• Elimination-by-aspect, categorization-by-eliminati
  on
• Sequential mate search, fixed or adjustable
  aspiration levels
• Gaze heuristic, motion-to-intention heuristics
• Tit-for-tat, behavior-copying

22
The Adaptive Toolbox Building Blocks

• Heuristic
• Take The Best
• Take The Last
• Minimalist
• Fast and frugal tree
• Tally-3
• QuickEst
• Satisficing

• Building Blocks
• Search rules
• Ordered search
• Recency search
• Random search
• Stopping rules
• One-reason stopping
• Tally-n stopping (ngt1)
• Elimination
• Aspiration level
• Decision rules
• Tally-n
• One-reason decision making

23
The Adaptive Toolbox Building Blocks

• Heuristic
• Take The Best
• Take The Last
• Minimalist
• Fast and frugal tree
• Tally-3
• QuickEst
• Satisficing

• Building Blocks
• Search rules
• Ordered search
• Recency search
• Random search
• Stopping rules
• One-reason stopping
• Tally-n stopping (ngt1)
• Elimination
• Aspiration level
• Decision rules
• Tally-n
• One-reason decision making

24
The Adaptive Toolbox Building Blocks

• Heuristic
• Take The Best
• Take The Last
• Minimalist
• Fast and frugal tree
• Tally-3
• QuickEst
• Satisficing

• Building Blocks
• Search rules
• Ordered search
• Recency search
• Random search
• Stopping rules
• One-reason stopping
• Tally-n stopping (ngt1)
• Elimination
• Aspiration level
• Decision rules
• Tally-n
• One-reason decision making

25
The Adaptive Toolbox Building Blocks

• Heuristic
• Take The Best
• Take The Last
• Minimalist
• Fast and frugal tree
• Tally-3
• QuickEst
• Satisficing

• Building Blocks
• Search rules
• Ordered search
• Recency search
• Random search
• Stopping rules
• One-reason stopping
• Tally-n stopping (ngt1)
• Elimination
• Aspiration level
• Decision rules
• Tally-n
• One-reason decision making

26
The Adaptive Toolbox Building Blocks

• Heuristic
• Take The Best
• Take The Last
• Minimalist
• Fast and frugal tree
• Tally-3
• QuickEst
• Satisficing

• Building Blocks
• Search rules
• Ordered search
• Recency search
• Random search
• Stopping rules
• One-reason stopping
• Tally-n stopping (ngt1)
• Elimination
• Aspiration level
• Decision rules
• Tally-n
• One-reason decision making

27
The Adaptive Toolbox Building Blocks

• Heuristic
• Take The Best
• Take The Last
• Minimalist
• Fast and frugal tree
• Tally-3
• QuickEst
• Satisficing

• Building Blocks
• Search rules
• Ordered search
• Recency search
• Random search
• Stopping rules
• One-reason stopping
• Tally-n stopping (ngt1)
• Elimination
• Aspiration level
• Decision rules
• Tally-n
• One-reason decision making

28
The Adaptive Toolbox Building Blocks

• Heuristic
• Take The Best
• Take The Last
• Minimalist
• Fast and frugal tree
• Tally-3
• QuickEst
• Satisficing

• Building Blocks
• Search rules
• Ordered search
• Recency search
• Random search
• Stopping rules
• One-reason stopping
• Tally-n stopping (ngt1)
• Elimination
• Aspiration level
• Decision rules
• Tally-n
• One-reason decision making

29
Ecological Rationality
30
Sequential Search Heuristics

• Take The Best
• Search rule Look up the cue with highest
  validity vi
• Stopping rule If cue values discriminate (/-),
  stop search. Otherwise go back to search rule.
• Decision rule Predict that the alternative with
  the positive cue value has the higher criterion
  value.

• Tallying
• Search rule Look up a cue randomly.
• Stopping rule After m (1ltmM) cues stop search.
• Decision rule Predict that the alternative with
  the higher number of positive cue values has the
  higher criterion value.

Dont add Dont weight
31
Ecological Rationality
Take The Best Tallying
non-compensatory
compensatory
Weight
1
2
3
4
5
1
2
3
4
5
Cue
Cue
Martignon Hoffrage (1999), In Gigerenzer et
al., Simple heuristics that make us smart. Oxford
University Press
32
Reinforcement learning Which heuristics to use?
100
Non-compensatoryFeedback
90
80
70
60
Choices predicted by Take The Best ()
50
40
CompensatoryFeedback
30
20
10
0
0-24
25-48
49-72
73-96
97-120
121-144
145-168
Feedback Trials
Rieskamp Otto (2004)
33
Ecological Rationality

• Heuristic
• Recognition heuristic
• Take The Best
• Fast frugal tree
• Tallying
• QuickEst
• Imitation

• Environment
• alpha gt .5
• Noncompensatory information
• Scarce information Mltlog2N
• Compensatory information, abundant information
• J-shaped distribution of objects on the criterion
• Stable environments, reliable information

Boyd Richerson (2001) Goldstein et al. (2001)
Hogarth Karalaia (in press) Martignon
Hoffrage (1999, 2002).
34
Robustness
35
How accurate are fast and frugal heuristics?

• Homelessness rates (50 cities in the United
  States)
• Attractiveness judgments of famous men and women
• Average motor fuel consumption per person (all
  states in the United States)
• Rent per acre paid (58 counties in Minnesota)
• House prices (Erie, Pennsylvania)
• Professors' salaries (a midwestern college)
• Car accident rate per million vehicle miles
  (Minnesota highways)
• High school drop-out rates (all high schools in
  Chicago)

Czerlinski, Gigerenzer, Goldstein (1999), In
Gigerenzer et al., Simple heuristics that make us
smart. OUP
36
Robustness
75
70
Take The Best
Tallying
Accuracy ( correct)
Multiple Regression
65
Minimalist
60
55
Czerlinski, Gigerenzer, Goldstein (1999)
Fitting
Prediction
37
Czerlinski et al. (1999) Multiple regression
(MR) improves performance (76) but so does Take
The Best (76)Hogarth Karelaia (2004) Take
The Best is more accurate than MR if -
variability in cue validities is high - average
intercorrelation between cues .5 - ratio of
cues to observations is highAkaikes Theorem
Assume two models belong to a nested family
where one has fewer adjustable parameters than
the other. If both have, on average, the same
number of correct inferences on the training set,
then the simpler model (i.e., the one with fewer
adjustable parameters) will have greater (or at
least the same) predictive accuracy on the test
set.
Robustness What if predictors are quantitative?
38
Design
39
The heart disease predictive instrument (HDPI)
Chest Pain Chief Complaint EKG (ST, T wave
?'s) History STT Ø ST? T?? ST? ST?T??
ST??T?? No MI No NTG 19 35 42 54 62 78 MI
or NTG 27 46 53 64 73 85 MI and
NTG 37 58 65 75 80 90 Chest Pain, NOT Chief
Complaint EKG (ST, T wave ?'s) History STT Ø
ST? T?? ST? ST?T?? ST??T?? No MI No
NTG 10 21 26 36 45 64 MI or
NTG 16 29 36 48 56 74 MI and
NTG 22 40 47 59 67 82 No Chest Pain EKG
(ST, T wave ?'s) History STT Ø ST? T?? ST?
ST?T?? ST??T?? No MI No NTG 4
9 12 17 23 39 MI or NTG 6 14 17 25 32 5
1 MI and NTG 10 20 25 35 43 62
See reverse for definitions and instructions
40
Fast and frugal classification Heart disease
ST segment changes?
yes
no
Coronary Care Unit
chief complaint of chest pain?
yes
no
regular nursing bed
any one other factor? (NTG, MI,ST?,ST?,T)
yes
no
Coronary Care Unit
regular nursing bed
Green Mehr (1997)
41
Emergency Room Decisions Admit to the Coronary
Care Unit?
1
.9
.8
.7
.6
SensitivityProportion correctly assigned
Physicians
.5
Heart DiseasePredictive Instrument
.4
Fast and Frugal Tree
.3
.2
.1
.0
.0
.1
.2
.3
.4
.5
.6
.7
.8
.9
1
False positive rateProportion of patients
incorrectly assigned
42
Sequential Search HeuristicsOne-reason decision
making

• Where?
• Rules of thumb in non-verbal animals
• Bail decisions in London courts (Dhami, 2003)
• ? Patient allocation to coronary care unit
  (Green Mehr, 1997)
• Prescription of antibiotics to young children
  (Fischer et al. 2002)
• Parents choice of doctor when child is seriously
  ill (Scott, 2002)
• Physicians prescription of lipid-lowering drugs
  (Dhami Harris, 2001)
• Voting and evaluating political parties
  (Gigerenzer, 1982)
• Why?
• Fishers runaway theory of sexual selection
• Zahavis handicap principle
• Environmental structures
• Robustness
• Speed, frugality, and transparency

43
The Science of Heuristics

• The Adaptive Toolbox
• Ecological rationality
• Design
• Gigerenzer et al. (1999). Simple Heuristics That
  Make Us Smart. Oxford University Press.
• Gigerenzer Selten (Eds.) (2001). Bounded
  Rationality The Adaptive Toolbox. MIT Press.

44
Notes

• Start with if you open a textbook, then the
  term heuristic of of Greek origin
• Collective wisdom arising from individual
  ignorance
• - Researchers from Oxford University and from the
  Georgia Institute of Technology developed
  computer programs mimicking honey bee heuristics
  to solve the problem of allocate computers to
  different applications when internet traffic is
  highly unpredictable. Economist, April17, 2004,
  p. 78-9.
• Selten, in his Nobel Laureate speech, used the
  term repair program
• A widely shared conclusion among decision
  theorists neglecting attributes means neglecting
  information, thereby violating a central
  principle of good decision making.
• Duration 45-50 minutes