1 multi-attribute utility theory (maut) dr. yan liu department of biomedical, industrial & human...
TRANSCRIPT
1
Multi-Attribute Utility Theory (MAUT)
Dr. Yan Liu
Department of Biomedical, Industrial & Human Factors Engineering
Wright State University
2
Introduction
Conflicting Objectives and Tradeoffs in Decision Problems e.g. higher returns vs. lower risks in investment, better performance vs.
lower price of computer
Objectives with Incomparable Attribute Scales “Attribute” refers to the quantity used to measure the accomplishment of an
objective e.g. maximize profits vs. minimize impacts on environments
Multi-Attribute Decision Making (MADM) A study of methods and procedures that handle multiple attributes Usages
Identify a single most preferred alternative Rank alternatives Shortlist a limited number of alternatives for subsequent detailed appraisal Distinguish acceptable from unacceptable possibilities
3
Introduction (Cont.)
Types of MADM Techniques Multiattribute scoring model (in Chapter 4)
Covert attribute scales to comparable scales Assign weights to these attributes and then calculate the weighted average of each
consequence set as an overall score Compare alternatives using the overall score
Multi-Attribute Utility Theory (MAUT) Use utility functions to convert numerical attribute scales to utility unit scales Assign weights to these attributes and then calculate the weighted average of each
consequence set as an overall utility score Compare alternatives using the overall utility score
…
4
You want to buy a car with a long expected life span and a low price. You have narrowed down your choices to three alternatives: the Portalo (a relatively expensive sedan with a reputation for longevity) , the Norushi (renowned for its reliability), and the Standard Motors car (a relatively inexpensive domestic automobile). You have done some research and evaluated these three cars on both attributes, as follows.
Alternatives
Attributes Portalo NorushiStandard Motors
Price ($k) 17 10 8
Life Spans (Years)
12 9 6
Worst Best
WorstBest
Automobile Example
5
None of the cars is dominated
How much are you willing to pay to increase the life span of your car? (subjective judgment)
Start with the Standard Motor, the cheapest among the three alternativesPrefer Norushi to Standard if you are willing to pay $2k or more to increase the life span of your car by 3 years Prefer Portalo to Norushi if you are willing to pay extra $7k or more for an additional 3 years
Portalo
Norushi
Standard
Life Span
Price
6
Trading Off Conflicting Objectives
Need Systematic Techniques to Handle Any Decision Situation Efficiently Three or more objectives Objectives with incomparable attribute scales
Issues to be addressed Construct a quantitative model of preferences to compare alternatives Numerical weight must be assessed for each attribute
7
Additive Utility Function A Simplified Utility Model
Ignores interactions among attributes For a consequence set that has values x1, x2, …, xm on the attributes
of m objectives, its overall utility is computed as
m
iiii
mmmm
xk
xkxkxkxxx
1
22211121
)(U
)(U)(U)(U),,,(U
Ui(xi) – the utility function of the ith attribute
)1( 21 mkkk
1),,,(U0 21 mxxx
1)(U0 ii x
ki – the weight of the ith attribute
8
UPrice(Norushi) = UPrice(10000) = (10000 – 17000) / (8000 – 17000) = 0.78
Set UPrice(Standard) =UPrice(8000) = 1,
Utility Functions
UPrice(Portalo) = UPrice(17000) = 0
ULife(Norushi) = ULife(9) = (9 – 6) / (12 – 6) = 0.5
Alternatives
Utilities Portalo NorushiStandard Motors
UPrice 0 0.78 1ULife 1 0.5 0
)(U x
ii
i
xx
xx ix : the worst value of attribute Xi ; : the best value of Xi
ix
ULife(Portalo) = ULife(12) = 1, ULife(Standard) = ULife(6) = 0
Automobile Example (Cont.)
9
• Directly specify the ratio of the weights
Weight Assessment
e.g. kPrice= 2kLife
Because kPrice+ kLife =1, then kPrice=2/3 and kLife = 1/3
U(Norushi) = 2/3•UPrice(Norushi) + 1/3•ULife(Norushi) = 2/3(0.78) + 1/3(0.5) =0.69
U(Standard) = 2/3•UPrice(Standard) + 1/3•ULife(Standard) = 2/3(1) + 1/3(0) =2/3
U(Portalo) = 2/3•UPrice(Portalo) + 1/3•ULife(Portalo) = 2/3(0) + 1/3(1) =1/3
10
• Indirectly specify the tradeoffs between objectives
e.g. You are willing to pay up to $600 for an extra year of life span
U(Norushi) = 0.714•UPrice(Norushi) + 0.286•ULife(Norushi) = 0.7
U(Standard) = 0.714•UPrice(Standard) + 0.286•ULife(Standard) = 0.714
U(Portalo) = 0.714•UPrice(Portalo) + 0.286•ULife(Portalo) = 0.286
Suppose taking the Standard Motors as the base case. You are indifferent between paying $8000 for 6 years of life span and paying $8,600 for 7 years of life span
U($8,000, 6 Years) = U($8,600, 7 Years)kPrice•UPrice(8000) + kLife•ULife(6) = kPrice•UPrice(8600) + KLife•ULife(7)
UPrice(8600) = (8600-17000)/(8000-17000)= 0.933, ULife(7) = (7-6)/(12-6)=0.167
kPrice•1 + kLife•0 = kPrice•0.933 + kLife•0.167 0.067kPrice= 0.167kLife (Eq. 1)
Weight Assessment (Cont.)
Solve Eqs (1) and (2) kPrice= 0.714, kLife = 0.286
kPrice + kLife = 1 (Eq. 2)
11
Indifference Curve• Alternatives falling on the same indifference curve have the same utility• The decision maker is indifferent among these alternatives
Indifference Curves of the Automobile Example (Trade $600 for an additional year of life span)
Portalo
Norushi
Standard
Price($K)
0.7140.7 0.286Life Span(Year)
Utility
8.6
7
Hypothetical car
12
Assessing Weights Indirectly
Pricing Out Determine the marginal rate of substitution between one particular attribute
(usually monetary) and any other attribute Marginal rate of substitution is the rate at which one attribute can be used to
replace another (the slope of the indifference curves in additive utility function) e.g. One year of life span of a car is worth $600
Appropriate for additive utility function
In an additive utility function, marginal rate of substitution between attributes xi and xj, Mij, is:
|-|/
|-|/-+
-+
jjj
iii
xxk
xxkijM
kLife = 0.286, kPrice= 0.714
|k17$-k8$|/714.0|yr6-yr12|/286.0
=LPM = $0.6k per year = $600 per year
13
Assessing Weights Indirectly (Cont.)
Swing Weighting Can be used virtually in any weight-assessment situation Requires a thought process of comparing individual attributes directly by
imaging hypothetical outcomes Step One: Create a table in which the first row indicates the worst possible
consequence set (with the worst level on each attribute), and each of the succeeding rows “swings” one of the attributes from the worst to best
Step Two: Rank the consequence sets created in the above table Step Three: Assign a rating score to each consequence set Step Four: Calculate the weights from the rating scores
14
Attributes Swung from the Worst to best
Consequence Sets to Compare
Rank Rate Weight
(Benchmark)
Life Span
Price
6 years, $17,000
12 years, $17,000
6 years, $8,000
3
1
2
0
100
75 75/175=0.429
100/175=0.571
Automobile Example (Cont.)
KLife/kPrice = 75: 100
15
Drug Counseling Center Choice
The drug-free center is a private nonprofit contract center that provides counseling for clients sent to it by the city courts as a condition of their parole. It has just lost its lease and must relocate.
The director of the center has screened the spaces to which it might move. After the prescreening, 6 sites are chosen for serious evaluation. The director must, of course, satisfy the sponsor, the Probation Department, and the courts that the new location is appropriate and must take the needs and wishes of both employees and clients into account. But as a first cut, the director wishes simply to evaluate the sites on the basis of values and judgments of importance that make sense internally to the center.
After consulting the members of the center staff, the director constructs a fundamental objective hierarchy that expresses the value-relevant objectives and attributes for comparing alternative center locations. Since the purpose of the evaluation to compare quality, cost is omitted.
16
A: Good conditions for staff
B: Easy access for clients
C: Suitability of space for center’s function
D: Administrative convenience
Maximize Overall Satisfaction
Office size
Convenience of commuting
Office attractiveness
Office privacy
Parking space
Closeness to clients’ homesAccess to public transportation
No. and suitability of counseling rooms
Suitability of reception and waiting area
No. and suitability of conference rooms
Adequacy of space
Flexibility of space layout
(0.43) (0.24) (0.19) (0.14)(0.39)
(0.21)
(0.14)
(0.14)
(0.12)
(0.50)
(0.50)
(0.52)
(0.32)
(0.16)
(0.64)
(0.36)
Fundamental Objectives Hierarchy
A1
A2
A3
A4
A5
B1
B2
C1
C2
C3
D1
D2
17
AttributesSites
1 2 3 4 5 6
A
A1 90 50 10 50 10 40A2 50 30 100 10 5 30A3 30 80 70 10 85 80A4 90 30 40 10 35 50A5 10 60 30 10 100 50
BB1 30 30 0 50 90 30B2 70 70 95 50 10 70
CC1 10 80 5 50 90 50C2 60 50 10 10 90 50C3 50 40 50 10 95 30
DD1 10 70 50 90 50 60D2 0 40 50 95 10 40
Ratings of Six Sites In terms of Attributes Corresponding to the Lowest-Level Fundamental Objectives
18
AttributesSites
1 2 3 4 5 6
A (0.43)
A1(0.39) 1 0.50 0 0.50 0 0.38A2 (0.21) 0.47 0.26 1 0.05 0 0.26A3 (0.14) 0.27 0.93 0.80 0 1 0.93A4 (0.14) 1 0.25 0.38 0 0.31 0.50A5 (0.12) 0 0.56 0.22 0 1 0.44
B (0.24)B1(0.50) 0.33 0.33 0 0.56 1 0.33B2 (0.50) 0.71 0.71 1 0.47 0 0.71
C (0.19)C1 (0.52) 0.06 0.88 0 0.53 1 0.53C2 (0.32) 0.63 0.50 0 0 1 0.50C3 (0.16) 0.47 0.35 0.47 0 1 0.24
D (0.14)D1 (0.64) 0 0.75 0.50 1 0.50 0.63D2 (0.36) 0 0.42 0.53 1 0.11 0.42
Relative Weights of Attributes and Utilities of Six Sites In terms of Attributes
19
Calculate the overall utility using the additive utility function
U(site 1) = kA∙(kA1∙U1, A1+ kA2∙U1, A2+kA3∙U1, A3+kA4∙U1,,A4 + kA5∙U1, A5) + KB∙(KB1∙U1, B1 +kB2∙U1,B2) + kC∙(kC1∙U1, C1+ kC2∙U1, C2+kC3∙U1, C3) + kD∙(kD1∙U1,
D1+ kD2∙U1, D2) = 0.43∙(0.39∙1+ 0.21∙0.47+0.14∙0.27+0.14∙1 + 0.12∙0) + 0.24∙(0.5∙0.33 +0.5∙0.71) + 0.19∙(0.52∙0.06+ 0.32∙0.63+0.16∙0.47) + 0.14∙(0.64∙0+ 0.36∙0)
=0.470
U(site 2) = kA∙(kA1∙U2, A1+ kA2∙U2, A2+kA3∙U2, A3+kA4∙U2,,A4 + kA5∙U2, A5) + KB∙(KB1∙U2, B1 +kB2∙U2,B2) + kC∙(kC1∙U2, C1+ kC2∙U2, C2+kC3∙U2, C3) + kD∙(kD1∙U2,
D1+ kD2∙U2, D2) = 0.43∙(0.39∙0.5+ 0.21∙0.26+0.14∙0.93+0.14∙0.25 + 0.12∙0.56) + 0.24∙(0.5∙0.33 +0.5∙0.71) + 0.19∙(0.52∙0.88+ 0.32∙0.5+0.16∙0.35) + 0.14∙(0.64∙0.75+ 0.36∙0.42)
=0.549
Utility of site 1 w.r.t attribute A1 Expected utility of site 1 w.r.t attribute A
20
U(site 3) = 0.378
U(Site 4) = 0.404
U(Site 5) = 0.491
U(Site 6) = 0.488
In conclusion, because U(Site 2) is the highest, site 2 should be chosen