Multiobjective Analysis
Dec 17, 2015
Multiobjective Analysis
An Example
Adam Miller is an independent consultant. Two year’s ago he signed a lease for office space. The lease is about to expire and he needs to decide whether to renew it or move to a new location. Adam defines five overriding objectives that he needs his office to fulfill: a short commute, good access to clients, good office services, sufficient space and low cost.
Consequence Table
Alternatives
Objectives Parkway Lombard Baranov Montana Pierpoint
Commute (min.)
45 25 20 25 30
Cust. Access (%)
50 80 70 85 75
Office Services
A B C A C
Office Size (sq. ft.)
800 700 500 950 700
Monthly Cost ($)
1850 1700 1500 1900 1750
Ranking Table
Alternatives
Objectives Parkway Lombard Baranov Montana Pierpoint
Commute (min.)
5 2 (tie) 1 2 (tie) 4
Cust. Access (%)
5 2 4 1 3
Office Services
1 (tie) 3 4 (tie) 1 (tie) 4 (tie)
Office Size (sq. ft.)
2 3 (tie) 5 1 3 (tie)
Monthly Cost ($)
4 2 1 5 3
Eliminating “Dominated” Alternatives
Dominance – If alternative A is better than alternative B on some objectives and no worse than B on all other objectives, B can be eliminated from consideration.
Example – Lombard Dominates Pierpoint
Eliminating “Dominated” Alternatives
Practical Dominance – If alternative A is better than alternative B on some objectives and no worse than B on all but one objective, B may be eliminated from consideration.
Example – Except for cost Montana dominates Parkway. Miller believes that the advantages of Montana justify the extra cost so that Montana dominates Parkway.
Updated Consequence Table
Alternatives
Objectives Lombard Baranov Montana
Commute (min.) 25 20 25
Cust. Access (%) 80 70 85
Office Services B C A
Office Size (sq. ft.) 700 500 950
Monthly Cost ($) 1700 1500 1900
“Even Swaps”
If every alternative for a given objective is rated equally you can eliminate that objective
Even Swaps is a way to adjust the values of different alternatives’ objectives in order to make them equivalent.
Even Swaps
First, determine the change necessary to cancel out an objective.
Second, assess what change in another objective would compensate for the needed change.
Third, make the even swap.
Even Swap
Alternatives
Objectives Lombard Baranov Montana
Commute (min.) 25 20 → 25 25
Cust. Access (%) 80 70 → 78 85
Office Services B C A
Office Size (sq. ft.) 700 500 950
Monthly Cost ($) 1700 1500 1900
Even Swap
Alternatives
Objectives Lombard Baranov Montana
Cust. Access (%) 80 78 85
Office Services B C → B A → B
Office Size (sq. ft.) 700 500 950
Monthly Cost ($) 1700 1500 → 1700 1900 → 1800
Dominance
Alternatives
Objectives Lombard Baranov Montana
Cust. Access (%) 80 78 85
Office Size (sq. ft.) 700 500 950
Monthly Cost ($) 1700 1700 1800
Even Swap
Objectives Lombard Montana
Cust. Access (%) 80 85
Office Size (sq. ft.) 700 → 950 950
Monthly Cost ($) 1700 → 1950 1800
Dominance
Objectives Lombard Montana
Cust. Access (%) 80 85
Monthly Cost ($) 1950 1800
Conclusion
Montana location is the final choice.
Multiobjective Value Analysis A procedure for ranking alternatives and
selecting the most preferred Appropriate for multiple conflicting
objectives and no uncertainty about the outcome of each alternative.
The Value Function Approach Specify decision alternatives and objectives Evaluate objectives for each alternative
A Multiobjective ExampleA prospective home buyer has visited four open houses in Medfield over the weekend. Some details on the four houses are presented in the following table.
A Multiobjective Example
Price
No. of bedrooms
No. of bathrms.
Style
$389,900 3 1.5 Ranch
$530,000 4 2 Colonial
$549,900 5 3
Garrison Colonial
$599,000 4 2.5 Colonial
The Value Function Approach Determine a value function which combines
the multiple objectives into a single measure of the overall value of each alternative.
The simplest form of this function is a simple weighted sum of functions over each individual objective.
The Value Function ApproachThis functional form:
requires specificatuion of
the objectives
... the weights
and the single objective value
functions
v x x x w v x w v x w v x
x x
w w
v x v x
m m m m
m
m
m m
( , ,... ) ( ) ( ) ... ( )
,...
( ),... ( )
1 2 1 1 1 2 2
1
1
1 1
The Value Function ApproachEstimating the single objective value functions Price - price ranges from roughly $300,000 to
$600,000 dollars with lower amounts being preferred.
Suppose that a decrease in price from $600,000 to $450,000 will increase value by the same amount as would a decrease in price from $450,000 to $300,000.
The Value Function Approach This implies that over the range $300,000 to
$600,00 the value function for price is linear and the value for each price alternative can be found by linear interpolation.
First set v1(389,900)=1 and v1(599,000)=0.
Then
The Value Function ApproachPrice = $530,000
530 000 389 900
599 000 389 900
530 000 1
0 1
67530 000 1
133 530 000
1
1
1
, ,
, ,
( , )
.( , )
. ( , )
v
v
v
The Value Function ApproachPrice = $549,900
549 900 389 900
599 000 389 900
549 900 1
0 1
77549 900 1
123 549 900
1
1
1
, ,
, ,
( , )
.( , )
. ( , )
v
v
v
The Value Function Approach Number of bedrooms - the number of bedrooms
for the four alternatives is 3, 4 or 5 with more bedrooms preferred to fewer.
Thus v2(5)=1 and v2(3)=0.
Suppose the increase in value in going from 3 to 4 bedrooms is twice the increase in value in going from 4 to 5 bedrooms.
The Value Function Approach Then if the value increase in going from 4 to 5
bedrooms is x, the value increase in going from 3 bedrooms to 4 is 2x.
And since the value increase in going from 3 bedrooms to 5 is 1, 2x+x=1.
Thus x=1/3 and finally the v2(4)=0+2(1/3) =.67
The Value Function Approach Number of bathrooms - The number of bathrooms for the
four alternatives are 1.5, 2, 2.5, and 3 with more bathrooms being preferred to fewer bathrooms.
Thus v3(3)=1 and v3(1.5)=0.
Suppose that the increase in value in going from 1.5 to 2 bathrooms is small and about equal to the increase in value in going from 2.5 to 3 bathrooms. The increase in value in going from 2 to 2.5 bathrooms is more significant and is about twice this value.
The Value Function Approach Then, the value increase in going from 1.5 to 2
bathrooms is x. The value increase in going from 2 to 2.5 bathrooms is 2x. And the value increase in going from 2.5 to 3 bathrooms is also x.
The sum of the value increases x+2x+x=1 and x=1/4.
So, v3(2)=0+x=0+1/4=.25, and v3(2.5)=0+x+2x=0+1/4+2/4=.75
The Value Function Approach Style - there are three house styles available:
Ranch, Colonial and Garrison Colonial. Suppose that Colonial, is most preferred, Ranch is
least preferred and the value of Garrison Colonial is about mid-value.
Then v4(Colonial)=1, v4(Garrison Colonial)=.5 and v4(Ranch)=0
A Multiobjective Example
Price
No. of bedrooms
No. of bathrms.
Style
$389,900 (1)
3 (0)
1.5 (0)
Ranch (0)
$530,000 (.33)
4 (.67)
2 (.25)
Colonial (1)
$549,900 (.23)
5 (1)
3 (1)
Garrison (.5)
$599,000 (0)
4 (.67)
2.5 (.75)
Colonial (1)
The Value Function ApproachDetermine the weights Consider the value increase that would result
from swinging each alternative (one at a time) from its worst value to its best value (e.g.. the value increase from swinging price from $599,000 to $389,900).
Determine which swing results in the largest value increase, the next largest, etc..
The Value Function Approach Suppose going from a Ranch to a Colonial results
in the largest value increase, going from 3 to 5 bedrooms the second largest, going from 1.5 bathrooms to 3 bathrooms the next largest and swinging price from $599,000 to $389,900 results in the smallest value increase.
The Value Function Approach Set the smallest value increase equal to w and set
each other value increase as a multiple of w. Suppose the bathroom swing is twice as valuable
as the price swing, the style swing is 3 times as valuable as the price swing and the bedroom swing falls about half way in between these two.
The Value Function Approach Since the single objective value functions are
scaled from 0 to 1 the weight for any objective is equal to its value increase for swinging from worst to best.
And because we would like the multiobjective value function to be scaled from 0 to 1, the weights should sum to 1.
The Value Function ApproachThen for price
for bedrooms
for bathrooms
w w
w w
w w
1
2
3
25
2
.
The Value Function Approachand for style
Now set
or
and
w w
w w w w
w
w
4 3
25 2 3 1
85 1
12
.
.
.
The Value Function ApproachFinally
and
w w
w w
w w
w w
1
2
3
4
12
25 29
2 24
3 35
.
. .
.
.
The Value Function ApproachDetermine the overall value of each alternativeCompute the weighted sum of the single objective
values for each alternative. Rank the alternatives from high to low.
A Multiobjective Example
Price
No. of Bedrms
No. of Bathrms
Style
WeightedSum
$389900 (1)
3 (0)
1.5 (0)
Ranch (0)
(.12)
$530000 (.33)
4 (.67)
2 (.25)
Colonial (1)
(.64)
$549900 (.23)
5 (1)
3 (1)
Garrison (.5)
(.73)
$599000 (0)
4 (.67)
2.5 (.75)
Colonial (1)
.(72)
.12 .29 .24 .35
The Value Function Approach The weighted sums provide a ranking of the
alternatives. The most preferred alternative has the highest sum.
The “ideal“ alternative would have a value of 1. The value for any alternative tell us how close it is to the theoretical ideal.