The success of success in cue search

1
The success of success in cue search

• Tim Rakow
• University of Essex
• Ben Newell, Nicola Weston David Shanks
• ELSE at University College London

2
The Take-The-Best Heuristic (TTB)

• Validity determines the search rule for TTB
  (Gigerenzer Goldstein, 1996, 1999)
• search through cues in descending order of
  validity
• stop search with the first cue that discriminates
• choose the object this (single) cue points to
• This strategy is surprisingly accurate!
• G G propose that TTB is psychologically
  plausible

3
TTB - two potential challenges

• Learning?
• How easy is it to learn and to access a hierarchy
  of cue validities?
• Efficiency?
• Would you always examine a highly valid cue if
  it hardly ever discriminated?

4
Using cues for inference

• Cue validity - the likelihood of a correct
  choice given the cue points to one option.
• Cue discrimination rate - the likelihood that a
  cue points to one object
• Cue success - the likelihood of a correct choice
  if only that cue is used
• (When all possible pairs of objects are
  considered)

5
Pros and Cons

• A cue with high validity (v) may be unhelpful
• IF its discrimination rate (d) is low
• Success (s) reflects both validity and DR
• s dv (1 - d)0.5

Proportion correct from occasions when the cue
discriminates
Proportion correct from occasions when the cue
doesnt discriminate (and you are forced to guess)
6
Some experiments

• In a task environment
• With an objective outcome criterion
• Where accuracy is important
• But information search has some cost
• Which order will best describe peoples search
  through cues? Descending order of
• Validity (as specified by TTB)
• Discrimination Rate (DR)
• Success

7
The share prediction task
8
Experiment 1 Structure (N 20)
64 learning trials All information free
128 test trials Costs 1p to obtain a pair of cue
values Earn 6p for a correct choice
Post-test rating Usefulness of each cue
9
Experiment 1 Task environment
Set odds for each cue, combine using Bayes
Theorem
Fix proportion of yes values for each cue
Follows from validity DR
Outcome on each trial determined (randomly)
according to the cue pattern and the programmed
odds for each cue
10
Experiment 1 Competing hypotheses
First cue (Acquired most)
Last cue (Acquired least)
11
Experiment 1 Results
12
Experiment 1 Findings

• It looks like
• Success is driving the search through cues
• But
• Are people learning different cue dimensions?
  (validity, DR, success?)
• Or, only a single cue weight?

13
Experiment 2 Structure (N 24)
64 learning trials
First test block 128 test trials
Post-test ratings Usefulness, validity, DR,
success (rank)
Validities equal group v .72 for all 4 cues 2nd
test block 64 trials
DR equal group d .50 for all 4 cues 2nd test
block 64 trials
14
Experiment 2 First Test Block
15
Estimates of Validity
16
Estimates of Discrimination Rate
17
Ranking of cues by success
18
Any effect of changing the environment?
DRs equal group can now focus solely on
validity Validities equal group can now focus
solely on DR
19
Experiment 2 Findings

• It still looks like
• Success is driving the search through cues
• Some ability to identify different cue dimensions
  (DR validity/success?)
• This ability also seen in other experiments
  (Rakow, Hinvest, Jackson, Palmer)
• Limited adaptation in search in relation to these
  (partially dissociated) features

20
Learning Efficiency

• Validity is hard to learn (when DRs vary)
• Its a conditional probability
• Cant simply compare frequencies of correct
  choices
• Need to encode relative frequencies
• Search behaviour does reflect both validity and
  discrimination
• Search order better described by success-order
  than by validity alone or by DR alone

21
The adaptivity of success-based searchA rational
analysis

• In a like-for-like comparison of
  validity-directed search, success-directed
  search, and DR-directed search, we compared 3
  one-reason decision strategies
• TTB (Take The Best)
• STS (Select The Successful)
• DTD (Draw The Discriminator)

22
Search through cues by validity is more accurate
Unless, something limits the scope of your
search
Search by Success
Search by Validity
23
More bangs for your buck with success when
information search is costly
Search by Success
Search by Validity
24
Adaptive Decision Making?

• As information costs or information constraints
  vary
• Which search order is best can change
• The optimal length of search will vary
• Will people adapt accordingly?

25
Recent Experiment
100 points for a correct choice 4 cost
conditions over 4 testing sessions Cost per cue
3 points, 10 points, 17 points, 24 points
26
Expected pay-off as a function of cue cost
27
Personal reward as a function of personal
information costs
Reward
Cost of information
28
The effect of varying cue cost
29
The stopping rule The effect of cue cost on
information purchase
30
Conclusions

• When searching through cues prior to choice,
  people are
• More likely to search by success than by validity
• Somewhat able to dissociate validity and DR
• Able to adapt to changes in information costs
• But adapt by truncating search rather than by
  adopting a different pattern of cue search