P. 1
da6

|Views: 0|Likes:
mit
mit

See more
See less

07/03/2013

pdf

text

original

# Engineering Risk Benefit Analysis

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82, ESD.72, ESD.721

DA 6. Multiattribute Utility Theory
George E. Apostolakis Massachusetts Institute of Technology

Spring 2007

DA 6. Multiattribute Utility Theory

1

xN). Multiattribute Utility Theory 2 . DA 6. • In general.Consequences • Not all the consequences are monetary. the consequences are described by a vector x ≡ (x1. …. for example. they may include the impact on health and safety of groups of stakeholders. • In risk management problems.

Multiattribute Utility Theory 3 .. xN) • Decision alternative Ai is preferred over alternative Ak if and only if its expected utility is greater. ….e. i. Ei[u] > Ek[u] ⇔ Ai f Ak DA 6.Multiattribute Utility • u(x1.

...... xN) Use the certainty-equivalent approach: p * ( x* . x N* ) 1 ( x1 .. …... x 1 N) 1-p ( x1* . Multiattribute Utility Theory .. x N ) 4 DA 6..Finding u(x1..

Multiattribute Utility Theory 5 . …. xN) = f[u1(x1). uN(xN)] where ui(xi) is the utility function of attribute xi? • The answer is “yes. • Can we find a function f such that u(x1.” if we can establish “independence” among the attributes.Independence Assumptions • Finding the multiattribute utility function using the preceding method is very burdensome. DA 6. … .

z ) f (y' . (y. z ) ∀z where y and y’ are two levels of y.e. if preferences for y levels do not depend on the level of z. z ) 0 ' 0 implies (y. DA 6. Multiattribute Utility Theory 6 ..Mutual Preferential Independence • Attribute Y is preferentially independent of attribute Z. i. z ) f (y .

• If you prefer “afternoon” to “morning” departure regardless of the price of the ticket. \$500) • If you prefer “afternoon” to “morning” departure regardless of the price of the ticket. then Y and Z are mutually preferentially independent. DA 6. Multiattribute Utility Theory 7 . afternoon) • Z: Ticket cost (\$300. and you prefer \$300 to \$500 regardless of the departure time. then Y is preferentially independent of Z. but the price depends on when you leave. but they are not mutually preferentially independent.Mutual Preferential Independence: Example • Y: Departure time (morning.

• For the previous example: The preference value of the lottery L(morning. except that the assessments are made with uncertainty present.Utility Independence (1) • It is similar to preferential independence. 0. • The CEs of lotteries on Y levels are independent of z. afternoon.5. 0. Multiattribute Utility Theory 8 .5) is independent of the price of the ticket. • Y is utility independent of Z if preferences over lotteries involving different levels of Y do not depend on a fixed level of Z. DA 6. It is a stronger assumption.

2 9 0 ≤ ui ( xi ) ≤ 1 DA 6. is u(x1. x2) = = k1 u1(x1) + k2 u2(x2) + (1.k1.k2) u1(x1) u2(x2) with 0 ≤ ki ≤ 1 i = 1. Multiattribute Utility Theory .2 i = 1.Utility Independence (2) • A form of the utility function for attributes X1 and X2 that are utility independent.

Utility Independence (3) • Fix the level of X2 at x'2 .k1.k2) u1(x1) u2( 2) ' ' x x =[k1+ (1. x'2 ) = ' ' x x =k1 u1(x1) + k2 u2( 2 ) + (1. ⇒ • Lotteries on X1 are independent of the level of X2. the preferences over levels of X1 are independent of the level of X2. DA 6.k2) u2( 2 )] u1(x1) + k2 u2( 2 ) • This is a linear transformation of u1(x1). • For another level of X2. therefore. then u(x1. we will get another linear transformation of u1(x1). Multiattribute Utility Theory 10 .k1.

they are mutually utility independent. • u(x1. x2) = g(x1) + h(x1)u2(x2) • X2 is utility independent of X1 but not vice versa. Multiattribute Utility Theory 11 . DA 6.Utility Independence (4) • When X1 and X2 are utility independent of each other.

Additive Independence • A stronger assumption than utility independence. • For two attributes. Multiattribute Utility Theory 12 . DA 6. x2 ) ' x1 and x1 are different levels of x1 We can get any pair of consequences with probability 0. x 2 ) 1/2 ( x1 .5. we must be indifferent between 1/2 ( x1 . the only difference is how the levels are combined. x'2 ) 1/2 ' ( x1 . x'2 ) 1/2 ' ( x1 .

Multiattribute Utility Theory 13 . x2) = k u1(x1) + (1-k) u2(x2) DA 6.Additive Utility Function u( x ) = ∑ k i u i ( x i ) 1 N where 0 ≤ ki ≤ 1 ∑ki = 1 1 N 0 ≤ ui ( xi ) ≤ 1 Two attributes: u(x1.

it should not matter what the other attribute’s level is. • For cases with no or little uncertainty. • Interaction among the attributes is not allowed. DA 6. • For complex problems. it could be a useful first-cut approximation. additive independence represents reasonably well people’s utilities.Additive Independence: Implications • When we assess the utility of one attribute. Multiattribute Utility Theory 14 .

scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->