Multi-Criteria Decision Making and Measuring Utilities

advertisement
Preferences and Decision-Making
Decision Making and Risk, Spring 2006: Session 7
Decision Making Framework
Outcomes
Known Outcomes
Unknown Outcomes
Outcome probabilities
Distribution
Outcome A1
Option A
Consequences/Payoff
Portfolio
Payoff Portfolio A1
p(A1)
Outcome A2
Payoff Portfolio A2
p(A2)
Decision
Problem
Outcome B1
p(B1)
Option B
Outcome B2
p(B2)
Decision Problem
Discovering the right
decision problem.
Payoff Portfolio B1
Payoff Portfolio B2
Simple consequences
$ metric
Complex consequences
Revenues
Costs
Learning
Turnover
Morale
Comp. Response
Future Options
Tech
Market
Facilities
Alternatives
Known Options
Unknown Options
Deferred Decision
Integrating payoffs to
determine overall utility
Central Logic in Decision Making

Two key questions in regard to any decision:

What are the consequences of the options?


In other words, what will happen with each alternative?
What is our preference for those consequences?

In other words, do we know what we want among the various
consequences that can occur?
The Health Screening Preferences

The goal of this questionnaire is to understand people’s
preferences for generic screening and diagnostic tests.

Tests vary on:







Accuracy
Frequency
Invasiveness
Time commitment from you, the patient
Pain and discomfort
Exposure to radiation
Please fill out the questionnaire provided.
Stated Versus Revealed
40.0
38
35.0
30.0
Normed Relative Preference
30.0
28
25.0
Stated
19
20.0
Revealed
17.1
16.4
14.0
15.0
11.1
9.9
10.0
4.7
5.0
4.7
4.7
0.0
Accuracy
Frequency
Invasiveness
Time Commitment
Attribute
Pain and Discomfort
Exposure to Radiation
Conjoint Analysis
Decision Options as Attribute Bundles

Each option has multiple attributes

Processor Speed, RAM, Screen Size, Price

Decision is a function of what is more important.

Problem?

What is not important?
Assessing Preferences

Stated Preference

What is important to you?



Independent importance scores.
Relative importance scores.
Revealed Preference


Forced tradeoffs.
More realistic.
Example
When making the decision to buy a laptop computer, how important on a scale from 5
(very important) to 1 (not so important) is:
Price
__
Processor Speed
__
Screen Size
__
RAM
__
Drives
__
Now, please rate each attribute offering on a scale from 5 (acceptable) to 1(not acceptable)
Price:
$1800
__
$1200
__
$1000
__
Screen Size:
17”
15”
14.1”
__
__
__
Example
Instead…
Consider the following 3 models. Please rank these from 3 (Most Preferred) to 1
(Least Preferred):
….
1. $1800, 17”, 3GHz, 256MB, DVD/CD
_____
2. $1200, 15”, 2.8GHz, 512MB, DVD-RW
_____
3. $1500, 15”, 3 GHz, 512 MB, DVD-RW
_____
Managerial Questions







Focus on 2.8 GHz or 3.0 GHz?
What drives customer preferences?
What if we increased screen size but reduced screen
resolution?
How do customers trade-off attributes?
What would be the market-share?
What if we offered a DVD-RW for $120 more?
What if we removed “free shipping” and offered to upgrade
the RAM?
Conjoint Analysis

Conjoint Analysis is a versatile marketing technique that
can provide valuable information, enables us to answer
all the questions that were listed earlier.

Conjoint Analysis is popular because it is a less
expensive and more flexible method than concept
testing.


Superior diagnosticity
Parallels real-world decisions
Uses of Conjoint







Concept Optimization.
Quantifying impact of change in product design.
Volume forecasting: for categories that can be
described fully by components.
Measuring Brand Equity.
Quantifying price sensitivity.
Estimating interactions in “menu” choices with a
survey.
Quantifying lifetime value of a customer.
A brief overview
Input:

Rankings/ratings of attribute bundles
Output:



relative importance of attributes.
“what-if ” simulations of hypothetical attribute bundles.
estimates of market share, volume, and attribute sensitivity.
Process

part-worths, utilities
Assumptions in Conjoint

Product is a bundle of attributes

Attributes are “describable”

Customers are able to rate/rank

Rating/ranking is an indicator of underlying utility
How Conjoint Works

Assume CPU and screen size are two attributes of consequence
in a notebook computer.

Assume three CPUs:




2.8 GHz
3.0 GHz
3.4 GHz
Assume two screen sizes:


14.1”
15”
Rank Ordering Combinations
Screen Size
CPU
14.1”
15”
2.8 GHz
6
4
3 GHz
3
2
3.4 GHz
5
1
Generating Utilities
Screen Size
CPU
14.1”
15”
Average
2.8 GHz
0
2
1
3.0 GHz
3
4
3.5
3.4 GHz
1
5
3
Average
1.33
3.66
Determining Relevant Attributes

Physical Attributes

Performance Benefit

Psychological positioning
Stimulus Representation

Full-profile

all relevant attributes are presented jointly for each product



more realistic from product presentation point of view
less realistic and more complex from consumer decision point of
view
Partial profile


subset of attributes
subset varies over the exercise until stable utilities are
estimated
Response Type

Paired comparison

Choose one profile over the other




3.4 GHz CPU with 14.1” screen vs.
3.0 GHz CPU with 15” screen
Complexity increases with number of attributes
Ranking

Rank the set of attribute bundles in order of preference.

Can be very complicated if number of attribute bundles increase.
Response Criterion

Preference


useful for market share predictions
Purchase likelihood

useful for market size estimation
Analyzing Output

Aggregate analysis



Homogeniety of sample
Importance of each level of attribute
Importance of each attribute based on range of importance scores for the
various levels

Caveat, misspecification of attribute level can artificially inflate attribute
importance.

Segmentation analysis

Scenario simulations


First or maximum choice rule
Share of preference rule
Overview of the Conjoint Process







Develop a list of attributes to describe the product.
Identify an experimental design to select product profiles.
Develop selected product profiles into stimuli and collect respondents’
evaluations (ratings, rankings, choices).
Decompose these evaluations into part worths or utilities for each attribute
level.
Report marginal utility curves or aggregate attribute importance data.
Run simulations (using utilities) to estimate share for benchmark product or
other products of interest.
Segmentation analysis based on the utilities.
Data Analysis: Simulations





Simulations attempt to predict choices based on utilities.
Specify a competitive scenario of brands available and describe
them in terms of attributes.
For every respondent, calculate the total utility of competing
brands.
Select a choice rule to apply these utilities (usually the maximum
choice rule).
Count the choices to estimate how many respondents would
select each brand.
Data Analysis: Simulation Rules





All conjoint simulation rules accept the rating scale you use as a direct
measure of utility.
A number of choice rules are available and the maximum utility choice rule
has the best track record.
Maximum utility choice rule: consumer chooses with certainty the option
offering the highest total utility.
Probabilistic choice rules: respondents have a non-zero probability of choice
for all brands available, related to the magnitude of utility each offers.
Simplest probabilistic choice rule is the attraction type rule:
ProbProfile X = UtilityProfile X
ΣUtilitiesAll Profiles in the Scenario
Conjoint Caveats

Products as attribute bundles

Researcher preselects important attributes

Ratings are meaningful

Attributes are actionable
Download