Transcript Slide - Mike Shor

Decision Making
With Many Options
Tibor Besedes
Cary Deck
Sudipta Sarangi
Mikhael Shor
October 2007
Motivation
 Life is full of choices
 Many important life decisions are made from an often
overwhelming number of options
 Mathematical truism:
max( X )  max( X  Y )
 Psychological perspective:
 Information Overload
 People “give up” when facing too many options
 Cognitive perspective:
 Brain “processing power” is limited
2
Evidence on Information Overload
 Fewer people join a 401(k) retirement plan
when more savings options are presented
 Iyengar, Jiang and Huberman 2004
 Physicians are less likely to prescribe any
drug when more drugs are available
 Redelmeier and Shafir 1995, Roswarski and Murray 2006
 Total amount of recycling decreases when
people are offered multiple recycling options.
 Greater “choice satisfaction” when choosing
among six Godiva chocolates than among 30
 Iyengar and Lepper 2000
 I don’t like long restaurant menus
3
Limitations of Prior Studies
 Past studies examine either satisfaction with
choice or whether a choice was made
Fewer choices made does not imply that
the average choice is worse
Satisfaction with choice does not imply
objectively good decision-making
We want to know whether a choice is optimal
4
Research Hypotheses
 When faced with a large set of options,
individuals make inefficient and suboptimal
decisions.
 Older individuals, will suffer a greater
deterioration of decision accuracy as decision
complexity increases.
 Reducing the complexity of the task makes
decision-making more efficient.
5
Motivating Example
 Medicare Part D
 Private insurers offer prescription drug plans
 A person may see
as many as 140 competing plans
6
Motivating Example
7
Motivating Example
8
Motivating Example
Motivating Example
10
Nature of Decisions
Available Plans
A
B
C
D
Physician Office Visit
Preventive Care
Urgent Care Service
Emergency Room Service
Hospital Expenses (inpatient)
Hospital Expenses (outpatient)
Diagnostic Services
11
Nature of Decisions
 There exist unknown future states of nature
 I’ll be healthy or sick. I’ll need what drug?
 States have associated probabilities
 Options “cover” some states but not others
 Choice is a maximization over states
 Simplified:
 Exactly one state is realized
 No “cost” of options
 If chosen option covers the state that occurs,
subject receives payment
12
Experimental Decision Problem
13
Experimental Design
 2 x 2 x 2 (+ 1) within-subject design
 Number of states: Either 6 or 10
 Number of options: Either 4 or 13 options
 Probability distribution
 10 state problems equivalent to 6 state problems
 Options “expanded” (all check marks preserved)
 For probability distribution 1:
 All states are rather likely
 Going from 4 options to 13 by introducing suboptimal options
 For probability distribution 2:
 Several states have very low probability
 4 to 13: one new option is much better (96% v. 71%)
14
Methodology
 Random order of
 Decision problems
 Options
 States
 125 subjects recruited online
 Paid $1 for every successful state, plus $3
 Collected demographics:
 Age, sex, education
 Dependent variables:
 Frequency of optimal decisions
 Efficiency of decisions (how suboptimal is suboptimal)
15
Results
Increasing options reduces frequency of optimal choice
Selection of optimal option
4 options
PDF 1
(all states likely, new options all bad)
6 attributes
41%
10 attributes
45%
PDF 2
(low prob. states, new option better)
6 attributes
50%
10 attributes
52%
13 options
29%
24%
50%
36%
16
Results
 How suboptimal are the choices?
 42% of all choices were the optimal option
 66% of all choices were within 10% of optimal
 Average efficiency loss was 13%
 Subjects were half-way between
optimal choice and random choice
17
Impact of Age: First Order Effects
Optimal decision-making decreases with age
Age Group
Frequency of
Optimal Choice
Nearly Optimal Choice
(within 10%)
Improvement over
random choice
Frequency of
“dominated” options
18-40
41-60
61+
52%
42%
31%
72%
66%
59%
61%
49%
36%
0%
5%
18%
18
Impact of Age: Second Order Effects
Decision complexity interacts with age
% optimal choice (most complex decision)
% optimal choice (least complex decision)
Age Group
Relative Frequency
18-40
41-60
61+
82%
66%
47%
18%
34%
53%
19
Results
Regressions
Chance of selecting optimal option:
Decreases with age
Increases with education
Does not depend on sex
Parameter magnitudes
11 years of age offsets an education category
3 education categories offsets having more options
20
Implications
 With reasonable a priori knowledge about optimal
options, presenting fewer options is better
 AT&T knows this, government does not
 For older people, fewer may be better even without
any a priori knowledge
 Best of any 4 better than random of 13
 Future investigation of choice presentations
 Default “suggested” options
 Break up big decisions into smaller ones
 Recommender systems
21