Professional Documents
Culture Documents
CBA 2.
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
CBA 2. The Time Value of Money
Overview of Lecture
Simple Interest
P: Principal amount
n: Number of interest periods
i: Interest rate
I: Interest earned
Interest and principal become due at the end of n.
I = Pni
The interest is proportional to the length of time
the principal amount was borrowed.
CBA 2. The Time Value of Money
Compound Interest
Interest is payable at the end of each interest period.
If the interest is not paid, the borrower is charged interest
on the total amount owed (principal plus interest).
Example: $1,000 is borrowed for two years at 6%
(compounded). A single payment will be made at the end of
the second year.
Amount owed at the beginning of year 2: $1,060
Amount owed at the end of year 2:
$1,060x1.06 = $1,000x(1.06)2 = $1,123.60
For simple interest, the amount owed at the end of year 2
would be: $1,000 + 1,000x2x0.06 = $1,120.00
CBA 2. The Time Value of Money
$100
$100
$100
$1,100
P (present value)
F
(future value)
A
(annual value
or annuity)
P(1+i)
P(1+i)2
(F/P, i, n) = (1+i)n
F= P(1+i)n
1
(P / F , i,n ) =
n
(1 + i )
The reciprocal of the single-payment compound
amount factor.
Discount rate: i
$1 n years in the future is worth (P/F, i, n) now.
Equal-Payment-Series Compound-Amount
Factor
F
0
n-1
n
(1 + i )n 1
(F / A, i, n ) =
i
CBA 2. The Time Value of Money
10
Equal-Payment-Series Sinking-Fund
Factor
May be used to determine the payments A required
to accumulate a future amount F.
i
( A / F, i, n) =
(1 + i )n 1
Example.
We wish to deposit an amount A every 6 months
for 3 years so that well have $10,000 at the end of this period.
The interest rate is 5% per year.
n = 6 deposits
i = 2.5% per 6-month period
F=$10,000
(A/F, 0.025, 6) = 0.15655
A=$1,565.50
CBA 2. The Time Value of Money
11
Equal-Payment-Series Capital-Recovery
Factor (1)
An amount P is deposited now at an annual interest
rate i.
We will withdraw the principal plus the interest in
a series of equal annual amounts A over the next n
years.
The principal will be worth P(1+i)n (slide 8) at the
end of n years. This amount is to be recovered by
receiving A every year the sinking-factor
formula applies (slide 11)
CBA 2. The Time Value of Money
12
Equal-Payment-Series Capital-Recovery
Factor (2)
A = P(1 + i ) [
n
i
(1 + i ) 1
n
i(1 + i )n
( A / P, i, n) =
(1 + i)n 1
13
14
lim 1 +
m
i=
1 = er 1
m
15
(F/P, r, n) = ern
(1 + i ) 1
(F / A, i, n ) =
i
1
e
(F / A, r , n ) = r
e 1
i(1 + i )n
( A / P, i, n) =
(1 + i)n 1
1
e
( A / P, r, n) =
1 ern
rn
16
17
18
19
20
21
22
Solution:
1
$500
$500
500
500
1,000
P=
+
+
3
4
5
(1 + 0.07 ) (1 + 0.07 ) (1 + 0.07 )
23
500
3 x 0.07
500
4 x 0.07
1,000
5 x 0.07
24
Inflation
The purchasing power of money declines when the
prices increase.
This must be included in equivalence calculations.
A price index is the ratio of the price of a
commodity or service at some point in time to the
price at some earlier point.
The Consumer Price Index (CPI) represents the
change in prices of a market basket, that includes
clothing, food, utilities, and transportation.
The CPI measures the changes in retail prices to
maintain a fixed standard of living for the
average consumer.
CBA 2. The Time Value of Money
25
Consumer Price
Index (CPI)
(Annual Rate
of Inflation)
1967
1972
100.0
125.3
2.9%
3.3%
1977
181.5
6.5%
1980
246.8
13.5%
1985
322.2
3.6%
1990
391.4
5.4%
1995
456.5
2.8%
1999
497.6
1.9%
Figure by MIT OCW.
From Table 5.1 of Thuesen & Fabrycky, Engineering Economy, 7th Edition,
Prentice Hall, NJ, 2001.
CBA 2. The Time Value of Money
26
Inflation Rate
Annual inflation rate for year t+1:
CPI
t +1 CPI t
f t +1 =
CPIt
For many calculations, an average inflation rate is
sufficient.
CPIt(1+f)n = CPIt+n
Note: Thuesen & Fabrycky use f for the average rate.
CBA 2. The Time Value of Money
27
Example
The average inflation rate from 1967 to 1999 is given
by
100(1+f)32 = 497.6
f = 5.14%
28
Definitions
Market interest rate (or current-dollar interest rate) i: The
interest rate available in finance. Inflation impact is
included.
Inflation-free interest rate (or constant-dollar interest rate) i':
It represents the earning power of money with inflation
removed. It must be calculated.
Actual dollars:
The amount received or disbursed at any
point in time.
Constant dollars: The hypothetical amount received or
disbursed in terms of the purchasing power of dollars at
some base year.
29
30
Proof
Proof: At the base year (t=0), constant and actual dollars
coincide.
Let P be the present value. Then, n years from now,
F = (1+i)nP
F' = (1+ i')nP
F = (1+f)n F = (1+f)n (1+ i)nP
actual dollars
constant dollars
actual dollars
1+ i
i =
1
1+ f
'
31
32
1 + 0.08
'
1 + 0.055
1 = 3.35%
=
1 = 3.43% iOC =
1 + 0.045
1 + 0.02
CBA 2. The Time Value of Money
33
'
US / OC
1 + 0.08
=
1 = 5.88%
1 + 0.02
US inflation rate
(John lives there)
CBA 2. The Time Value of Money
34
CBA 3.
Overview
(Net) Present Worth or Value [(N)PW or
(N)PV] and Annual Equivalent (AE)
The Effect of Discount and Inflation Rates
Using PW and AE as Decision Criteria
The Internal Rate of Return as a Decision
Criterion
The Benefit-Cost Ratio as a Decision
Criterion
Comparing Projects with Unequal Lives
PW ( i ) = Ft (1 + i )t
t =0
t =n
PW (i ) = NPW (i ) = NPV (i ) = B t (1 + i ) Ct (1 + i )t
t =0
t =0
( P / A , i, n ) =
(1 + i) n 1
i(1 + i ) n
(1 + 0.0083) 36 1
0.0083(1 + 0.0083)
= 5,000 + 15,505 = $20,505
CBA 3. Bases for Comparison of Alternatives
36
]=
Annual Equivalent
The annual equivalent of receipts minus the annual
equivalent of disbursements (the annualized profit).
Any present worth can be converted to a series of
equal annual amounts by multiplying by (A/P, i, n).
n
i
(
1
i
)
+
AE(i ) = PW (i )( A / P, i , n ) = [ Ft (1 + i )t ][
]
n
(1 + i ) 1
t =0
n
if
Aj f Ak
if
or
B3
-$12,000
-1,200
-1,200
1,500
B4
-$15,000
-400
-400
3,000
AE on Total Investment
AE(10%)B3 = 12,000( A / P,10%,3) 1,200 + 2,700( A / F,10%,3) =
= 12,000
0.1x1.13
1, 200 + 2,700
0.1
1.1 1
1.1 1
= 12,000x0.402 1,200 + 2,700x0.302 = $5,209
3
10
End of Year
0
1
2
3
B3
-$12,000
-1,308
-1,426
1,943
CBA 3. Bases for Comparison of Alternatives
B4
-$15,000
-436
-475
3,885
11
+
=
2
3
1.1
1.1
1.1
= $12,908 < $12,870 = PW (10%)B 4
12
Year
0
1
2
3
B3
B4
-$12,000 -$15,000
-1,200
-400
-1,200
-400
1,500
3,000
B4 - B3
-$3,000
800
800
1,500
13
14
0 = PW ( i ) = Ft (1 + i* )t
*
t =0
15
Example
End of year t
0
1
2
3
4
5
Ft
-1,000
-800
500
500
500
1,200
0 = PW (i*) =
= 1000 800( P / F , i*,1) + 500( P / A , i*,4)( P / F , i*,1) +
+ 700( P / F , i*,5)
CBA 3. Bases for Comparison of Alternatives
16
Example (contd)
Trial and error:
For i* = 12%
For i* = 13%
PW(12) = $39
PW(13) = -$12
By interpolation:
39 0
i* = 12% + 1%
= 12.8%
39 ( 12)
17
PW (i)
$900
i* = 12.8%
0
10%
20%
18
Sufficient Conditions
1. F0 < 0 (The first nonzero cash flow is a disbursement)
2. The sequence F0, F1, F2, , Fn, has one change in
sign only.
3. PW(0) > 0 (sum of all receipts > sum of all
disbursements)
CBA 3. Bases for Comparison of Alternatives
19
Examples
Both B3 and B4 of slide 8 fail the third condition.
For B3, PW(0) = -$12,900 <0.
For the cash flow on slide 16, PW(0) = 900 > 0.
Consider the cash flow:
t
$
0
1
2
3
-1,000 +600 +600 +600
4
5
+600 -1,410
20
21
22
23
Example (1)
End of
Year
0
1 10
A0
0
0
A1
A2
A3
-$5,000
1,400
-8,000
1,900
-10,000
2,500
MARR = 15%
1. All satisfy the conditions on slide 19.
i*A A x10
2. Find
1
24
Example (2)
0 = -5,000 + 1,400(P/A, x10, 10)
(1 + x10 )10 1
0 = 5,000 + 1,400
10
x10 (1 + x10 )
i*A A x10 = 25% > 15%
1
25
Example (3)
Similarly, we solve
(1 + x 21 )10 1
0 = 3,000 + 500
10
x 21 (1 + x 21 )
i*A A x 21 = 10.5% < 15%
2
26
Example (4)
A1
A2
A2-A1
$12,000
$10,000
$8,000
$6,000
$4,000
0.1987
$2,000
0.2499
$0
0
0.1
0.2
0.3
0.4
0.5
($2,000)
($4,000)
0.1056
CBA 3. Bases for Comparison of Alternatives
27
Example (5)
To compare A3 to A1, we must solve
0 = -5,000 + 1,100(P/A, x31, 10)
(1 + x 31 )10 1
0 = 5,000 + 1,100
10
x 31 (1 + x 31 )
i*A A x 31 = 17.6% > 15%
3
28
Example (6)
A1
A3
A3-A1
$16,000
$14,000
$12,000
$10,000
$8,000
$6,000
0.2141
$4,000
0.2499
$2,000
$0
0
0.1
0.2
0.3
0.4
0.5
($2,000)
($4,000)
0.1768
($6,000)
29
x2 iA
*
then
(1 + x 2 )10 1
0 = 8,000 + 1,900
10
x 2 (1 + x 2 )
x 2 i*A = 19.9%
2
30
Similarly,
i A = 15%
i A = 25%
i A = 19.9%
i A 3 = 21.4%
31
A2
A3
$16,000
$14,000
$12,000
$10,000
$8,000
$6,000
0.2141
$4,000
0.2499
$2,000
$0
($2,000)
($4,000)
0.1
0.2
0.3
0.4
0.5
0.1987
($6,000)
32
33
34
(1 + 0.15)10 1
= 5,000 + 1,400
=
10
0.15(1 + 0.15)
= 2,026 > 0
35
PW (15)A A
2
(1 + 0.15)10 1
= 3,000 + 500
=
10
0.15(1 + 0.15)
= 490 < 0
36
(1 + 0.15)10 1
= 5,000 + 1,100
=
10
0.15(1 + 0.15)
= 521 > 0
37
Summary
The economic desirability of alternatives is determined by
examining their differences.
Use the PW (or, equivalently, the AE) criterion, as described
on slides 32 33.
PW- (or AE-)based results using total and incremental
investments are identical.
The IRR criterion (assuming that the conditions on slide 18
are satisfied) on the differences gives results consistent with
those of the PW and AE criteria.
The IRR criterion on total investments may not give results
consistent with those of the PW and AE criteria.
CBA 3. Bases for Comparison of Alternatives
38
Unequal Lives
39
Example (1)
Alternative
Initial Cost
$4,000
$16,000
$20,000
Annual Cost
$6,400
$1,400
$1,000
Lifetime
6 years
3 years
4 years
40
Example (2)
AE(A) = 6.4K + 4K (A/P, 7%, 6) 6.4K + 4K (0.21)
= $7,240
AE(B) = 1.4K + 16K(A/P, 7%, 3) 1.4K+16K (0.38)
= $7,480
AE(C) = 1K + 20K (A/P, 7%, 4) 1K+ 20K (0.3) =
= $7,000
41
16
1.4
1.4
1.4
1.4
17.4
CBA 3. Bases for Comparison of Alternatives
17.4
42
43
A
(in $000)
4
6.4
6.4
6.4
6.4
6.4
10.4
6.4
6.4
6.4
6.4
6.4
6.4
B
(in $000)
16
1.4
1.4
17.4
1.4
1.4
17.4
1.4
1.4
17.4
1.4
1.4
1.4
C
(in $000)
20
1
1
1
21
1
1
1
21
1
1
1
1
44
Present Worth
PW(A, 12 yrs) = 4K + 6.4K (P/A, 7%, 12)
+ 4K (P/F, 7%, 6) $57,500
PW(B, 12 yrs) = 16K + 1.4K (P/A, 7%, 12) + 16K (P/F, 7%, 3) +
16K (P/F, 7%, 6) + 16K (P/F, 7%, 9) $59,400
PW(C, 12 yrs) = 20K + 1K (P/A, 7%, 12)
+ 20K (P/F, 7%, 4) + 20K (P/F, 7%, 8) $55,600
These are PWs of costs over 12 years the lowest common
45
46
CBA 4.
Including Uncertainty
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
CBA 4. Including Uncertainty
Uncertainty
Methods
1. Scenario analysis
2. Adjustments of interest rates
3. Decision Theory
4. Simplified probabilistic models
CBA 4. Including Uncertainty
Scenario Analysis
Preparation and analysis of scenarios:
-- Optimistic or most favorable estimate
-- Most likely or best estimate or fair
estimate
-- Pessimistic or least favorable estimate
Interpretation is difficult without assignment of
probabilities to scenarios.
Benefit: Brings additional information into the
process.
CBA 4. Including Uncertainty
Example
A new machine is to be purchased for producing
units in a new manner.
Pessim. Fair Optim.
Annual number of units:
900 1,000 1,100
Savings per unit:
$50
55
60
Operating costs:
$2,000 1,600 1,200
etc.
PW:
Developing Scenarios
For each element in the problem, e.g., interest rate
and costs, define the three values.
We really dont know how conservative
(pessimistic) the final answer is.
People are bad processors of information.
Point estimates tend to cluster around the median
value. Possibility of displacement bias.
Extremes greater than the 75th or smaller than the
25th percentile are difficult to imagine.
Overconfidence.
CBA 4. Including Uncertainty
Example (contd)
The high rate of 30% is intended to cover
uncertainty.
If the annual income were $60,000, then the net
annual profit would be 60,000 - (30,272 + 28,600) =
= $1,128 and the venture would be accepted.
A high interest rate does not guarantee that all
uncertainties are accounted for. Its choice is
arbitrary.
10
11
Earnings (payoffs):
L1: $15,000/yr, old product,
L2: $30,000/yr, new product and the market is strong,
L3: -$10,000/yr, new product and the market is weak
Probabilities
0.4 = P(s1s2)
0.4 = P(s1w2)
0.2 = P(w1w2)
P(s1) = P(s1s2) + P(s1w2) = 0.8; P(w1) = 0.2; P(s2/s1) = 0.5; P(w2/w1) = 1.0
CBA 4. Including Uncertainty
12
Decision Tree
Decision
Options
States of Nature
P1
PW of Payoffs
P2
s2
s1
w1
$243.26K
w2
$96.94K
w2
-$81.09K
O
$121.61K
CBA 4. Including Uncertainty
13
10
= 44.51K
30
PW w 2 = 109 .74 x
10
= 36 .58 K
30
14
New Product
EMVN = 243.26x0.4 + 96.94x0.4 -81.09x0.2 = 119.86K
Decision
Stay with the old product?
15
16
Probabilistic Models
We have:
XT
X1
X2
PW[ X(T)] = X 0 +
+
+ .... +
2
(1 + i ) (1 + i )
(1 + i )T
(1)
17
Analysis
Computing the probability density function (pdf)
of PW is usually difficult in practice.
Try to compute the quantities E[PW], the
expected value of PW,
and
2
PW
18
E[Z] = a E[W] + b
2
Z
2 2
= a W
19
Let
Z = W1 + W2
E [ Z ] = E [W1 ] + E [W2 ]
2
Z
2
W1
2
+ W2
20
Let
Z = aW1 + bW2 + c
2Z = a 2 2W1 + b 2 2W2
21
22
Example (2)
The capacitor fails when the surge voltage, S, is
greater than the capacity, C.
S: rv with E[S] = 100, S = 15 volts
C: rv with E[C] = 130, C = 15 volts
Define a new rv D C S = aC + bS
Then, E[D] = 130 100 = 30 volts
and
C S = 21.21
+
volts
23
Example (3)
D is also normally distributed, therefore
Pd/sv(D < 0) = P( Z < -(30/21.21)) = P(Z < - 1.41) =
= P(Z > 1.41) = 0.5 0.42 = 0.08
RPRA 3, page 31, shows that Pd/sv = conditional
= P(Z >
24
Assuming Independence of Xj
X1
X2
XT
PWInd [ X(T)] = X 0 +
+
+ .... +
2
(1 + i ) (1 + i )
(1 + i )T
E[ X j ]
(1 + i ) j
j= 0
2j
(1 + i )2 j
j= 0
25
Example: Project A
T = 3 yrs.; i = 8%; Initial Cost = $10K
Probability, p
Net Benefits
t=1
t=2
t=3
0.10
$3K
$3K
$3K
0.25
$4K
$4K
$4K
0.30
$5K
$5K
$5K
0.25
$6K
$6K
$6K
0.10
1.00
$7K
$7K
$7K
26
Example (2)
Denote with X1, X2, and X3, the net benefits of
A in years 1, 2 and 3, respectively.
Note: Net benefits, X1, X2, and X3, do not have
to be identically distributed or symmetric or
discrete; these choices are made just to keep the
example simple.
Denote with Y the present worth of A, PW(A).
Then:
PWInd(A) = YInd = -10K + X1/(1.08) + X2/(1.08)2
+X3/(1.08)3
CBA 4. Including Uncertainty
27
Observations
Y is a random variable (takes more than one value
with different probabilities for any given
implementation of project A)
Value of Y will be determined by the values of the
combination of X1, X2, and X3 that will actually
materialize
Corresponding a priori probability of any value
of Y is equal to probability of that particular
combination of X1, X2 and X3.
CBA 4. Including Uncertainty
28
or,
X 1 $1,140 = X 2 = X 3
29
Independence: Calculations
Assume that the net benefits obtained from Project A in years 1,
2 and 3 are determined independently of one another.
This means the probability of the combination {X1 = 3, X2 = 6,
X3 = 4} is equal to
P(X1 = 3, X2 = 6, X3 = 4) = P(X1 =3)P(X2 = 6)P(X3 =4) =
= (0.1)(0.25)(0.25) = 0.00625
that is, with probability 0.00625, the r.v. Y, i.e., the PW of Project
A, will take on the value
PWInd = YInd = -10K + 3K/(1.08) + 6K/(1.08)2 + 4K/(1.08)3
$1,097.14
30
[1/(1.08)
[1/(1.08)3]2
=
[1/1.08]
Y
X1
X2
X3
,Ind
= (1,140)2[1/(1.08)2 + 1/(1.08)4 + + 1/(1.08)6] =
= (1,140)2(2.2225), or, Y ,Ind = (1,140)(2.2225)1/2 $1,700
CBA 4. Including Uncertainty
31
Conclusion
32
E[Y] = E[PW]
20
15
15
12
17
16
33
34
Probability of Loss
A1: P(PW < 0) = P(Z < -(20/15)) = P(Z < -1.33) =
= 0.09
best alternative
A2: P(PW < 0) = P(Z < -(5/7)) = P(Z < -0.71) =
= 0.24
A3: P(PW < 0) = P(Z < -(15/12)) = P(Z < -1.25) =
= 0.10
A4: P(PW < 0) = P(Z < -(17/16)) = P(Z < -1.06) =
= 0.14
CBA 4. Including Uncertainty
35
36
Complete Dependence of Xj
Once the net benefits, X1, for year 1 are known, we shall
also know exactly the net benefits for years 2 and 3.
Probability, p
Net Benefits
t=1
t=2
t=3
0.10
$3K
$3K
$3K
0.25
$4K
$4K
$4K
0.30
$5K
$5K
$5K
0.25
$6K
$6K
$6K
0.10
1.00
$7K
$7K
$7K
37
2
(1 + i) T
(1 + i) (1 + i)
2
2
=
Y ,Dep X [(P/A, 8, 3)]2 =(1,140)2[2.5771]2
Y ,Dep= (1,140)(2.5771)
$2,938
38
Comparison
X1
X2
XT
PWInd [ X(T)] = X 0 +
+
+ .... +
2
(1 + i ) (1 + i )
(1 + i )T
1
1
1
PWDep[X(T)] = X0 + X
+
+ .... +
2
T
(
1
i
)
+
(
1
i
)
(
1
i
)
+
+
39
Spring 2007
CBA 5. Evaluating Public Activities
General Welfare
The selection of public projects is not made on the basis of
profit, but, rather, on the basis of maximization of the
general (or, social) welfare of the citizens.
What is general welfare?
Pareto Improvement: At least one person is made better
off by the project and no one is made worse off.
Potential Pareto Improvement: Those who are better off
could compensate those who are worse off and still be
better off. This means that the aggregate benefits are
bigger than the aggregate costs.
No actual compensations need to occur.
CBA 5. Evaluating Public Activities
Distributional Inequity
The risks and benefits do not usually accrue to the
same individuals or to the same generation.
Taxation is one way to remedy inequities.
Ethical issues may arise. Can the impact of a
project on future generations (e.g., waste disposal,
climate change) be measured in monetary terms?
Executive Order 12898 on Environmental Justice:
No policy will result in disproportionately high
adverse human health and environmental effects
on low-income and minority populations
compared to the general population in affected
communities.
CBA 5. Evaluating Public Activities
Public:
(+)
(-)
Government:
Criterion:
Aj f Ak
6
t
t
t (1 + i )
t (1 + i )
See slide 4, CBA 3.
Criterion:
If NPV(i)Aj-Ak > 0 Aj f Ak
The following criteria are consistent when
mutually exclusive alternatives are compared:
NPV(i) and AE(i) on total or incremental investment.
Rate of return on incremental investment.
BC(i) on incremental investment.
CBA 5. Evaluating Public Activities
A0
0
0
A1
A2
-$5,000
1,400
-8,000
1,900
A3
-10,000
2,500
MARR = 15%
(1 + 0.15)10 1
1,400
= 7,026
10
0.15(1 + 0.15)
Therefore,
Similarly:
7,026
BC(15)A1 =
= 1.41
5,000
9,536
BC(15)A 2 =
= 1.19
8,000
12,547
BC(15)A 3 =
= 1.25
10
,
000
CBA 5. Evaluating Public Activities
10
5,521
BC(15)
= 1,100( P / A ,15,10) =
= 1.10
A 3 A1
5,000
11
12
Willingness-to-Pay
-- how much is one willing to pay to decrease
his/her probability of death or injury?
Human capital (or future earnings) with
appropriate adjustments -- many ethical
questions
Inference from societal decisions
CBA 5. Evaluating Public Activities
13
Notes
1,100,000
2,900,000
7,000,000
1,100,000 to
9,700,000
14
Proposed barrier
%New death rate:
4 per 108 vehicle miles
%Life:
30 years
%Cost:
$1.5x106 per mile
%Annual maintenance: $45,000 per mile
15
16
245,572
BC(7 ) =
= 1.48 > 1
165,844
The project should be accepted.
OMB (1992): It is a mistake to choose among mutually
exclusive alternatives by selecting the alternative with the
highest ratio of benefits to costs. An alternative with a
lower benefit-cost ratio than another may have the higher
net benefits.
CBA 5. Evaluating Public Activities
17
18
19
% EO 12991 is revoked.
%A regulatory analysis should be prepared for all
significant regulatory actions.
20
21
22
Discount Rates
The Office of Management and Budget
(1992) recommends a rate of 7%.
For sensitivity purposes, a calculation using
a 3% rate is also recommended, because the
NRC actions typically involve 30- to 60-year
time horizons.
23
Decision Making
Selecting the alternative with the largest net
present worth is consistent with obtaining the
largest societal gain from among the alternatives.
The benefit-cost ratio should be displayed but
should not be the basis for the decision.
OMB (1992): It is a mistake to choose among
mutually exclusive alternatives by selecting the
alternative with the highest ratio of benefits to
costs. An alternative with a lower benefit-cost
ratio than another may have the higher net
benefits.
CBA 5. Evaluating Public Activities
24
Spring 2007
Limitations of DA
The theory is for an individual decision maker.
This reduces considerably its applicability in
practice. (But, great normative tool.) In most cases
there is no satisfactory way to combine the utility
function of several people
As with all formal analysis, the results are no better
than the quality of the model and its supporting
assessments
The required inputs may not be easily obtainable
DA 1. The Multistage Decision Model
Manufacturing Example
Decision: To continue producing old product (O)
or convert to a new product (N).
The payoffs depend on the market conditions:
s: strong market for the new product
m: mild market for the new product
w: weak market for the new product
DA 1. The Multistage Decision Model
Earnings (payoffs):
L1: $150,000, old product,
P[L1/O] = 1.0
L2: $300,000, new product and the market is
strong,
P[s] = P[L2/N] = 0.3
L3: $100,000, new product and the market is mild,
P[m] = P[L3/N] = 0.5
L4: -$100,000, new product and the market is weak,
P[w] = P[L4/N] = 0.2
DA 1. The Multistage Decision Model
Payoffs
L2 $300K
Payoff
depends on
market
L3 $100K
L4 -$100K
O
L1 $150K
DA 1. The Multistage Decision Model
Decision Tree
Decision
Options
States of Nature
Payoffs
L2 $300K
P[s]= 0.3
P[m]= 0.5
L3 $100K
N
P[w]=0.2
L4 -$100K
P[s m w ] = 1
L1 $150K
DA 1. The Multistage Decision Model
N:
-$100
O:
$150
Choose O
Maximax Rule:
N:
$300
O:
$150
Choose N
DA 1. The Multistage Decision Model
10
11
s, $90
m, $50
L3 $100
EMV[N]=$120
w, -$20
L4 -$100
EMV[O]=$150
$150
L1 $150
Best Decision: O
DA 1. The Multistage Decision Model
12
A New Decision
The DM considers the possibility of commissioning
a survey to be able to better judge the future
market.
The survey costs $20,000.
There are now two decisions (multistage model):
The initial decision of whether to buy the survey
The terminal decision of whether to market the new
product.
DA 1. The Multistage Decision Model
13
P(s/L3) = 0.2
P(s/L4) = 0.0
Mild
P(m/L2) = 0.2
P(m/L3) = 0.6
P(m/L4) = 0.3
Weak
P(w/L2) = 0.0
1.0
P(w/L3) = 0.2
1.0
P(w/L4) = 0.7
1.0
14
Purchase
C
survey
Survey
response
"strong"
Survey
response
"mild"
Make old
product
Product earns*
C
300,000
D
Make new C
product
Make old
product
150,000
100,000
-100,000
150,000
300,000
Make new
C
product
100,000
-100,000
Survey
response
"weak"
Do not
purchase
survey
Make old
product
No new
information D
is obtained
150,000
300,000
Make new
C
product
Make old
product
100,000
-100,000
150,000
300,000
Make new
C
product
100,000
-100,000
15
New inputs
16
P (s / L j )P ( L j )
4
P(s / L j )P( L j )
2
j = 2, 3,4
DA 1. The Multistage Decision Model
17
0.3
0.5
0.2
1.0
P(s/ L2)=0.8
P(s/ L3)=0.2
P(s/ L4)=0.0
Product
0.24
0.10
0.00
0.34
Posterior
Probability
P(L2/s)=0.706
P(L3/s)=0.294
P(L4/s)=0.000
1.000
18
P(L2/m) = 0.143
P(L2/w) = 0.000
P(L3/m) = 0.714
P(L3/w) = 0.417
P(L4/m) = 0.143
1.000
P(L4/w) = 0.583
1.000
19
L1
L2
.706
.294
0
L1
.143
.714
.143
L2
L3
L1
0
.417
.583
L2
L3
L1
150,000
L2
.3
.5
.2
300,000
100,000
-100,000
.34
.42
.24
D
130,000
280,000
L3
80,000
L4 -120,000
130,000
280,000
80,000
L4 -120,000
130,000
280,000
80,000
L4 -120,000
L3
L4
20
21
221,300
221,300
.34
130,000
C
. 42
. 24
80,000
130,000
130,000
-36,600
150,000
150,000
L1
130,000
.706
.294
0
L2
L3
L4
280,000
80,000
-120,000
L1
130,000
.143
.714
.143
L2
L3
L1
0
.417
.583
L2
L3
L1
.3
.5
.2
L2
130,000
161,008
150,000
D
120,000
280,000
80,000
L4 -120,000
130,000
280,000
80,000
L4 -120,000
150,000
300,000
L3 100,000
L4 -100,000
Figure by MIT OCW.
22
Best decision
23
24
P(H1 |C1 )
H1
C1
P(C 1 )
D1
D2
P(C 2 )
P(C n )
C2
G2
Gk
Cn
D3
Dm
25
2.
3.
4.
26
6.
7.
27
Spring 2007
P(s/L3) = 0.2
P(s/L4) = 0.0
Mild
P(m/L2) = 0.2
P(m/L3) = 0.6
P(m/L4) = 0.3
Weak
P(w/L2) = 0.0
1.0
P(w/L3) = 0.2
1.0
P(w/L4) = 0.7
1.0
0.3
0.5
0.2
1.0
P(s/ L2)=0.8
P(s/ L3)=0.2
P(s/ L4)=0.0
Product
0.24
0.10
0.00
0.34
Posterior
Probability
P(L2/s)=0.706
P(L3/s)=0.294
P(L4/s)=0.000
1.000
P[L2 materializes/survey says L2] = 0.706, because the survey is not perfect.
DA 2. The Value of Perfect Information
P(CVsaysL 2 / Limaterializes)xP(Limaterializes)
2
1xP(L 2materializes)
=
=1
1xP(L 2materializes) + 0 + 0
P[L2 materializes/ CV says L2] = 1 regardless of the prior probability,
because the CV is perfect.
DA 2. The Value of Perfect Information
States of Nature
Payoffs
L2 $300K
s, 0.3
m, 0.5
L3 $100K
N
w, 0.2
1.0
DA 2. The Value of Perfect Information
L4 -$100K
L1 $150K
6
Modifications
In a decision tree, the order of the nodes is
chronological.
With perfect information, the uncertainty is resolved
before the decision is made (a chance node is
followed by a decision node).
The evaluation is done a priori (before the CV is
hired).
Therefore, the DM believes that the CV will predict L2
with probability 0.3, L3 with probability 0.5, and L4
with probability 0.2.
DA 2. The Value of Perfect Information
s, 0.3
m, 0.5
L3 $100K
N
=$150K
CV
w, 0.2
1.0
L4 -$100K
L1 $150K
=$195K
10
11
s, 0.3
m, 0.5
L3 $100K
N
=$150K
CV
w, 0.2
1.0
L4 -$100K
L1 $150K
=$195K
12
General Tree
If the DM faces uncertainty in a decision
(uncertainty nodes after the decision node), the
impact of perfect information will be evaluated by
redrawing the tree and reordering the decision and
chance nodes.
13
14
Spring 2007
Lotteries
A lottery is a probabilistic trial characterized by a set of
mutually exclusive and exhaustive possible outcomes C1, C2,
, Cm, with respective probabilities p1, p2, , pm.
L(C1, C2, , Cm; p1, p2, , pm)
Example:
$5
L:
0.4
$0
Preferences exist
For every pair of consequences Ci and Cj, a DM
will:
prefer Ci to Cj
Ci f Cj
Ci ~ Cj
Ci p Cj
Definition of C* and C*
Define:
C* a consequence that is at least as preferred as the
most preferred of C1 Cm C* f Ci for all i
C* a consequence that is at least as low in
preference as the least preferred of C1 Cm
C*p Ci for all i
C* and C* need not be included in C1 Cm
DA 3. The Axioms of Rational Behavior
The probabilities
The consequences
if p1 > p2,
be indifferent
if p1 = p2
prefer L2 over L1
if p1 < p2.
Notes on Axiom 2a
The indifference probability (or "preference
value") (Ci) is a measure of the preference of Ci on
a range of consequences from C* to C*.
This axiom provides the basis for the development
of the metric of utility (preference value).
From Axiom 1, the DM will prefer Ci for sure over
the lottery L (C*, C* ;p , 1 - p), if p < (Ci).
DA 3. The Axioms of Rational Behavior
Notes on Axiom 2b
10
11
12
C1
C2
Cm
C1
p1
p2
( C1 )
C*
1 ( C 1 )
C*
( C2 )
C*
1 ( C 2 )
C*
( Cm )
C*
1 ( C m )
C*
C2
pm
Cm
13
14
Summary of Axioms
Axiom 0: Preferences exist
Axiom 1: Two simple lotteries, each with same prize and
penalty: choose lottery with higher probability of prize
Axiom 2a: Quantification of preferences ("indifference
probability" or "preference value")
Axiom 2b: Quantification of uncertainty
Axiom 3: Transitivity of preferences
DA 3. The Axioms of Rational Behavior
15
16
DA 4. Introduction to Utility
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
DA 4. Introduction to Utility
Schematic representation
C1
p1
p2
p1
C2
p2
~
pm
pm
Cm
DA 4. Introduction to Utility
( C1 )
C*
1( C1 )
( C2 )
C*
1( C2 )
C*
( Cm )
C*
1( Cm )
C*
C*
Recall Axiom 1:
L1 = L (C*, C*; p1 , 1 - p1)
Then:
L1
f L2
L1 p L2
Given
and
if p1 > p2
if p1 < p2;
L2 ~ L1
if p1 = p2
p1
p2
( C1 )
C*
1( C1 )
( C2 )
C*
1( C2 )
C*
C*
C*
1P
pm
( Cm )
C*
1( Cm )
C*
C*
P=
of possible outcomes of L
DA 4. Introduction to Utility
P = p i ( C i )
i =1
Utility
The indifference probability or any positive linear
transformation of the form
U(x) = a (x) + b
a>0
DA 4. Introduction to Utility
10
Utility of outcomes
Let the utility of payoffs be
U(x) = 1.18 ln(x+5) - 1.29
-2 x 2 (x in $M)
[U(2) = 1, U(-2)= 0]
For L(2, -2; 0.5, 0.5)
EMV = 0
U(L) = 0.5x1 +0.5x0 = 0.5 = U(CE)
CE = -0.442M
DA 4. Introduction to Utility
11
DA 4. Introduction to Utility
12
L2 $300K
s, 0.3
m, 0.5
L3 $100K
N
w, 0.2
L4 -$100K
O
1.0
DA 4. Introduction to Utility
L1 $150K
13
0.30
0.50
L3 0.87
0.855
N
0.20
O
0.92
1.00
DA 4. Introduction to Utility
L4 0.60
L1 0.92
14
15
16
.949
(Ci)
.900
D
L1
0.90
.706
.294
0
L2
L3
L4
0.99
0.85
0.55
L1
0.90
.143
.714
.143
L2
L3
L4
0.99
0.85
0.55
L1
0.90
0
.417
.583
L2
L3
L4
0.99
0.85
0.55
L1
0.92
L2
.3
.5
.2
1.00
0.87
0.60
.949
.34
.900
.917
.42
.900
.827
.24
.900
.900
.675
.920
.920
.920
.855
L3
L4
DA 4. Introduction to Utility
17
Best Decision
DA 4. Introduction to Utility
18
DA 5. Risk Aversion
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
DA 5. Risk Aversion
a>0
and (C*) = 1
DA 5. Risk Aversion
Example
Suppose that it has been determined that
U(x) = ln(x+5)
for
C = 2
and
Let
Then,
1 = aln(7) + b
and 0 = aln(3) + b
to obtain a = 1.18 and b = -1.29
DA 5. Risk Aversion
DA 5. Risk Aversion
x1
p1
x1- BP
L'
pm
pm
xm
(0) = U(L') = p i ( xi BP )
xm- BP
DA 5. Risk Aversion
Example
(x) = 1.18 ln(x+5) 1.29
(0) = 0.61
DA 5. Risk Aversion
Example (contd)
ln[(6-BP) (5-BP)] = 3.22
(BP)2 11(BP) + 30 = exp(3.22) = 25.04
(BP)2 11(BP) + 4.96 = 0
BP = 0.47
DA 5. Risk Aversion
(certainty equivalent)
(0) = 0.6091
DA 5. Risk Aversion
Example
(0) = 0.6091
DA 5. Risk Aversion
Risk Aversion
The DM is always willing to sell any lottery for less
than its expected monetary value.
L(x1, x2; p1, p2)
EMV = p1 x1 + p2 x2
10
Example
L(1, 0; 0.5, 0.5)
EMV = $0.5 M
(0.5) = 0.6091
SP = $0.4771
11
C* = $2M, C* = -$2M
12
Decision Tree
p
-x
I
1-p
NI
-x
-$500,000
1-p
0.0
DA 5. Risk Aversion
13
Utilities
(I) = (-x)
14
Your Perspective
You are willing to pay up to $5,661 to insure
your house.
The expected loss is
10-2 x $500,000 = $5,000 < $5,661
15
DA 5. Risk Aversion
16
- 494,339
Accept I
5,661
1-p
Do not
Accept I
0.00
1-p
0.00
DA 5. Risk Aversion
17
18
19
Set
U(C*) = 0
and U(C*) = 1
2.
20
21
DA 5. Risk Aversion
22
Spring 2007
Consequences
Multiattribute Utility
u(x1, , xN)
Decision alternative Ai is preferred over alternative
Ak if and only if its expected utility is greater, i.e.,
Ei[u] > Ek[u]
Ai f Ak
1-p
( x1 ,..., x N )
4
Independence Assumptions
Finding the multiattribute utility function using the
preceding method is very burdensome.
Can we find a function f such that
u(x1, , xN) = f[u1(x1), , uN(xN)]
where ui(xi) is the utility function of attribute xi?
The answer is yes, if we can establish
independence among the attributes.
DA 6. Multiattribute Utility Theory
implies
(y, z ) f (y' , z )
0 ki 1
i = 1,2
0 ui ( xi ) 1
i = 1,2
10
11
Additive Independence
A stronger assumption than utility independence.
For two attributes, we must be indifferent between
1/2
1/2
( x1 , x 2 )
( x1' , x'2 )
1/2
1/2
( x1 , x'2 )
( x1' , x 2 )
12
u( x ) = k i u i ( x i )
1
where
0 ki 1
N
ki = 1
1
0 ui ( xi ) 1
Two attributes:
13
14
Spring 2007
DA 7 Decision Analysis in Practice
The Problem
Recommend a technology to remediate a
contaminated site.
The recommendation should reflect the views of a
number of stakeholders.
Multi-Attribute Utility Analysis (MAUA) and
deliberation are the bases for the recommendation.
DA 7 Decision Analysis in Practice
S te p 2
Is ea c h
d ec isio n
a lte r n a tiv e
a cc ep ta b le?
No
R e m o v e fr o m
fu rth er
c o n sid er a tio n
Y es
S te p 3
R a n k a n d a n a ly z e
d e cisio n
a lte r n a tiv es fo r
ea c h sta k e h o ld e r
(P R A , M A U T ,
o th er )
S tep 4
D elib e ra te a n d
c h o o se th e b est
a lte rn a tiv e
The Site
1.9 Acres
Disposal 1962 to 1981
Solvents, PCBs, metal acids, lab trash, misc. ebris
4 miles to nearest drinking - water well
3 miles to nearest spring
480 feet to water table
Network of vapor extraction wells to reduce TCE vapor
plume
Landfill-wide excavation to top 15 ft to remove shallow
primary sources of potential contamination
The Stakeholders
Decision Options
Means Objectives
Means objectives are not important in and by
themselves. They help to achieve the fundamental
objectives.
Examples:
Core damage in nuclear reactors
Water contamination
10
11
12
13
14
Stakeholder Changes
15
The decision is not made in general but for the specific problem.
DA 7 Decision Analysis in Practice
16
PI j = wi uij
PM
i =1
where
NPM
wi = 1
1
17
18
The Weights
Recall that
N
PI j = wi uij
PM
i =1
wi = 1
1
19
20
_______
2.
_______
3.
_______
4.
_______
5.
_______
6.
_______
7.
_______
8.
_______
9.
_______
10.
_______
3 weakly
5 strongly
9 absolutely
21
The Practice
People are not consistent in their assessments.
Redundant information is elicited.
Define the consistency index as
max n
CI =
n 1
22
Socioeconomic
Cultural
SH1
4 (8)
3 (11)
6 (4)
6 (4)
2 (34)
1 (39)
SH2
5 (2)
3 (14)
6 (2)
4 (6)
1 (38)
1 (38)
SH3
6 (2)
4 (7)
5 (4)
3 (8)
2 (39)
1 (40)
SH4
5 (5)
4 (8)
2 (25)
6 (4)
3 (17)
1 (41)
SH5
6 (3)
4 (10)
5 (4)
3 (11)
2 (20)
1 (52)
SH6
4(12)
6 (5)
3 (13)
5 (10)
2 (27)
1 (33)
23
Utilities
PM scales provide the means to measure the
degree of achievement of the objectives.
In this problem, each PM range is divided into
three intervals: good, moderate, and bad.
Utilities are developed using the AHP again, e.g.,
Good
Good
1
Moderate
1/3
Bad
1/6
Moderate Bad
3
6
1
3
1/3
1
DA 7 Decision Analysis in Practice
u
0.6
0.3
0.1
24
PM
25
POLLUTANT PATHWAYS
Atmospheric Transport
off
Infiltration
Over
flow
Evapotranspiration
Run
Resuspension
Precipitation
Deposition
Drinking Water
& Irrigation
Site Boundary
Leaching
Stream
Transport to Aquifer at
Retarded Velocities
26
27
STAKEHOLDERS
RAA
.094 (6)
.048 (6)
.071 (6)
.053 (6)
.050 (6)
.130 (5)
.205 (4)
.172 (3)
.154 (4)
.111 (5)
.091 (2)
.159 (2)
.216 (3)
.128 (4)
.177 (3)
.122 (3)
.091 (3)
.155 (3)
.183 (5)
.115 (5)
.179 (2)
.120 (4)
.082 (5)
.139 (6)
.223 (2)
.185 (2)
.132 (5)
.135 (1)
.107 (1)
.114 (4)
.258 (1)
.205 (1)
.181 (1)
.128 (2)
.089 (4)
.194 (1)
28
29
Stakeholder 3
0.2
0.18
Performance Index
0.16
0.14
0.12
0.1
0.08
0.06
F
30
31
32
Deliberation
33
34
35
Additional References
Accorsi, R., Apostolakis, G., and Zio, E., Prioritizing Stakeholder Concerns in
Environmental Risk Management, Journal of Risk Research, 2:11-29, 1999.
Accorsi, R., Zio, E., and Apostolakis, G.E., Developing Utility Functions for
Environmental Decision Making, Progress in Nuclear Energy, 34:387-411,
1999.
Apostolakis, G.E. and Pickett, S.E., Deliberation: Integrating Analytical Results
into Environmental Decisions Involving Multiple Stakeholders, Risk Analysis,
18:621-634, 1998.
Bonano, E.J., Apostolakis, G. E., Salter, P.F., Ghassemi, A., and Jennings, S.,
Application of Risk Assessment and Decision Analysis to the Evaluation,
Ranking, and Selection of Environmental Remediation Alternatives, Journal
of Hazardous Materials, 71:35-57, 2000.
Clemen, R.T., Making Hard Decisions, 2nd Edition, Belmont, California: Duxbury
Press, 1996
36
References (contd)
Edwards, W., von Winterfeldt, D., and Moody, D.L., Simplicity in Decision Analysis: An
Example and a Discussion, in: Decision Making, Edited by D.E. Bell, H. Raiffa, and A.
Tversky, Cambridge University Press, UK, 1988.
Gregory, R. and Keeney, R.L., Creating Policy Alternatives Using Stakeholder values,
Management Science, 40:1035-1048, 1994.
Hughes, W.R., "Deriving Utilities Using the Analytic Hierarchy Process, Socio-Econ.
Plann. Sci., 20:393-395, 1986.
Keeney, R.L. and von Winterfeldt, D., Managing Nuclear Waste from Power Plants, Risk
Analysis, 14:107-130, 1994.
National Research Council, Understanding Risk: Informing Decisions in a Democratic
Society. National Academy Press, Washington, D.C., 1996.
Susskind, L. and Cruikshank, J., Breaking the Impasse, Harvard University Press,
Cambridge, MA, 1987.
DA 7 Decision Analysis in Practice
37
Introduction
Risk-Benefit Tradeoffs
Introduction
Introduction
Train Accidents
Train Doors
Detrain
Escalators!
From: Vincent Ho, The Applications of Quantitative Risk Assessment in the Railway
Industry, INER presentation, December 2004
Introduction
ERBA
Make informed decisions concerning engineering
projects that include important elements of
technological or financial risk.
The quantification of uncertainty is a key element.
Three main topics to be covered:
Reliability and Probabilistic Risk Assessment
(RPRA) (includes a review of elementary
probability theory)
Decision Analysis (DA)
Cost-Benefit Analysis (CBA) (except in ESD.721)
Introduction
Level of Coverage
Broad Survey
Emphasis on:
standard terminology
fundamental issues
Introduction
RPRA 1.
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
Event Definition
Event: A statement that can be true or false.
It may rain tonight is not an event.
According to our current state of knowledge, we
may say that an event E is TRUE, FALSE, or
POSSIBLE (UNCERTAIN).
Eventually, E will be either TRUE or FALSE.
RPRA 1. The Logic of Certainty
True
Event
False
Possible
Venn Diagrams
Sample Space: The set of all possible outcomes of an
experiment. Each elementary outcome is represented by a
sample point.
Failure Time {0, }
Venn Diagram
Indicator Variables
1,If
is T
If E
is F
Xj =
0,
Important Note: Xk = X,
k: 1, 2,
S
___
Venn Diagram
A B = C
X C = 1 (1 X A )(1 X B )
X
C
A
A
RPRA 1. The Logic of Certainty
B
6
A B = C
XC = X AX B
XC X j
C
A B =
Simple Systems
Reliability Block Diagram for the Series System
1
....
failure: X = 1
System
Failure
(1 X j ) C X j
N
success :
Y = Yj
1
...
X = X j
1
N
Y = CYj
i+1
TOP
N
i+1
Event-Tree Analysis
IE
BARRIER 1
BARRIER 2
1 (OK)
SUCCESS
2 (R1)
3 (R2)
FAILURE
10
Fault-Tree Analysis
Reliability Block Diagram for the 2-out-of-3 System
2/3
C
RPRA 1. The Logic of Certainty
11
X T = 1 (1 Y1 )(1 Y2 )
= 1 (1 X A X B X C ){1 [1 (1 Z1 )(1 Z 2 )(1 Z 3 )]}
= 1 (1 X A X B X C ){1 [1 (1 X A X B )(1 X B X C )(1 X C X A )]}
X T = 1 (1 X A X B )(1 X B X C )(1 X C X A )
RPRA 1. The Logic of Certainty
12
13
F a ilu r e
M 1 = X A X B,
M2 = X B XC ,, M3 = XC X A
X T = C M j 1 (1 M 1 ) (1 M 2 ) (1 M 3 ) =
1
= 1 (1 X A X B )(1 X B X C )(1 X C X A)
RPRA 1. The Logic of Certainty
14
XT = 1 (1 M i ) C M i
Sum-of-Products Form:
N
N 1 N
XT = Mi
i =1
N +1 N
M i M j + ... + (1) M i
i =1 j =i +1
i =1
15
16
3
5
17
Causes of Failure
1.
2.
3.
P r im a r y
F a ilu r e
S e c o n d a ry
F a ilu r e
C om m and
F a ilu r e
18
T2
Fuel
Source
P1
P2
Control Valve
V1
Pump Train 1
Emergency
Diesel
Engine
Control Valve
V2
Pump Train 2
Electric
Power
Source, E
Control
System, C
Cooling
System,
CO
19
OR Gate
INTERMEDIATE
EVENT, A
AND Gate
INCOMPLETELY
DEVELOPED
EVENT, B
2
Transfer in
from Sheet 2
A1
A2
Basic
Event
A1
Basic
Event
A2
20
LOSS OF FUEL
FLOW , T
LOSS OF TRAIN
1
LOSS OF TRAIN
2
E1
MECHANICAL M
LOSS OF TRAIN 2
2
Loss of
Electricity
E
P2
T2
Loss of
Electricity
E
Loss of
Control
C
Loss of
Cooling
E2
Loss of
Control
C
Loss of
Cooling
CO
V2
MECHANICAL
LOSS OF TRAIN M 1
1
CO
T1
P1
V1
21
E
Fails
C
Fails
CO
Fails
Pumping
Branches Fail
Train 1 Fails
T1
Fails to
Supply
Fuel
P1
Fails to
Pump
Fuel
Train 2 Fails
V1
Fails
Closed
T2
Fails to
Supply
Fuel
P2
Fails to
Pump
Fuel
V2
Fails
Closed
22
Development of T1
Tank T1
Failure to
Supply Fuel
Tank is Intact
But Em pty
and Undetected
Tank (and
Supply Pipe)
is Not Intact
Supply Pipe
is Plugged
Tank is Em pty
Tank is
Em ptied
Inadvertantly
(hum an
error)
Tank is
Em ptied
in Use and
Not Refilled
Tank
Drain
Valve is
Left O pen
Fuel
Level
Detection
Fails
Hum an
Action
Sludge
Buildup
M issile
Im pact
Induced
Failure
Internal
Fire/Explosion
Induced
Failure
Corrosion
Fatique
Induced
Failure
Faulty
M anufacture
& Control
Program
23
Any combination of
an element of
C
plus
CO
T2, Tank
P2, Pump
V2, Valve
Control System
or
Electric Power
Source
or
Cooling System
24
25
26
RPRA 2.
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
RPRA 2. Elements of Probability Theory
Relative-frequency interpretation
Imagine a large number n of repetitions of the
experiment of which A is a possible outcome.
It is postulated that:
k
n
k
lim P( A )
n n
EE = S
___
Venn Diagram
P(E) = 1 P(E)
RPRA 2. Elements of Probability Theory
K + ( 1)
N +1
N
P I A i
1
Rare-Event Approximation:
N N
P U A i P(A i )
1 i =1
RPRA 2. Elements of Probability Theory
Union (contd)
For two events: P(AB) = P(A) + P(B) P(AB)
{1, 2, 3, 4, 5, 6}
(discrete)
X T = M i M i M j + ... + (1)
i =1
P( X T ) =
i =1 j= i +1
N 1 N
N +1
N +1
i =1 j = i +1
Mi
i =1
P Mi
i=1
Rare-event approximation:
P(XT ) P (Mi )
1
N1 N
N+1 N
i =1
i =1 j=i +1
P(XT ) P (Mi )
1
P( X T ) P( M i )
i =1
P(Mi )
i =1
P( M i M j )
i =1 j= i +1
10
Conditional probability
P(AB )
P(A B )
P(B )
P(AB ) = P(A B )P(B ) = P(B / A )P(A )
11
M 1 = X1 X 2 X 3
M2 = X2 X3 X4
M 3 = X3 X 4 X 1
M4 = X1 X2 X4
12
P(XT = 1) 4q3
13
The events, Hi, i = 1...N, are mutually exclusive and exhaustive, i.e.,
HiHj= , for ij, Hi = S, the sample space.
H3
H1
E
H2
P( E) = P( E / H i )P(H i )
1
14
P(H i / E) =
P ( E / H i )P ( H i )
P(E)
15
Bayes Theorem
Likelihood of the
Evidence
P (H i E ) =
P(E H i )P(H i )
Prior
Probability
P(E H i )P(H i )
1
Posterior
Probability
16
Suppose that you are on a TV game show and the host has offered you
what's behind any one of three doors. You are told that behind one of
the doors is a Ferrari, but behind each of the other two doors is a Yugo.
You select door A.
At this time, the host opens up door B and reveals a Yugo. He offers you
a deal. You can keep door A or you can trade it for door C.
17
Bayes Theorem
18
P (E A )P (A )
=
P (E A )P (A ) + P (E B )P (B ) + P (E C )P (C )
But
P(E/B) = 0
P(E/C) = 1
19
p
x=
1+ p
Therefore
20
Since
P(A/E) = 1/2
offer any advantage.
21
Random Variables
Random Variable: A function that maps sample points onto the real
line.
Example:
For a die
S = {1,2,3,4,5,6}
22
Events
3.6
-
4 5
{X -62} =
23
Sample Spaces
The SS for the die is an example of a discrete sample space and X is a discrete
random variable (DRV).
Example:
{T t} = {failure occurs before t}
24
Properties:
1. F(x) is a non-decreasing function of x.
2. F(-) = 0
3. F() = 1
25
F(x)
1
1/
6
1
26
P(X = x i ) p i
xi x
normalization
F ( 2 .3 ) = P ( 1 2 ) = p i =
1
1 1 1
+ =
6 6 3
27
dF (x )
f (x ) =
dx
P( S ) = F( ) =
F( x ) = f (s )ds
f (s)ds = 1
normalization
28
f (x ) = kx 2 ,
for
f (x ) = 0,
0x1
otherwise
is a pdf.
Answer:
2
kx
dx = 1
k=3
29
F(x)
1
0.875
0.67
0.75
0.42
1
f(x)
3
1
0.75
0.875
30
Moments
Expected (or mean, or average) value
xf (x )dx
E[X ] m
x jp j
j
CRV
DRV
2
(
)
x
m
f (x )dx
2
2
E (X m ) =
(x j m )2 p j
j
RPRA 2. Elements of Probability Theory
CRV
DRV
31
Percentiles
F(xm) = 0.50
f (x )dx =
32
Example
1
m = 3x 3 dx = 0.75
0
3
F (x m ) = x m
= 0. 5
x 03.05 = 0.05
x 03.95 = 0.95
= 0.194
x m = 0.79
x 0.05 = 0.37
x 0.95 = 0.98
33
RPRA 3.
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
RPRA 3. Probability Distributions in RPRA
Overview
We need models for:
The probability that a component will run for a period of time given a
successful start.
Failure to start
Start with an experiment that can have only two outcomes: success
and failure or {0, 1} with probabilities p and q, respectively.
N k
Pr[ X = k ] = q (1 q )N k
k
N
N!
k k! (N k )!
RPRA 3. Probability Distributions in RPRA
N k
N k
(
1
q
)
=1
k =0 k
m N k
N k F(m )
P[at most m failures] = q (1 q )
k=0 k
CDF
Pr[k ] = e
k! 1*2**(k-1)*k
( t )
0! = 1
k!
m = t
= t
8
A component fails due to "shocks" that occur, on the average, once every
100 hours. What is the probability of exactly one replacement in 100
hours? Of no replacement?
t = 10-2*100 = 1
Pr[1 repl.] = e-t = e-1 = 0.37 = Pr[no replacement]
Expected number of replacements: 1
2
1
1
e
1
=
= 0.185
Pr[ 2repl] = e
2! 2
10
t
F(t ) = 1 exp h(s )ds .
11
I
0
II
t1
III
t2
Infant Mortality
II
Useful Life
III
Aging (Wear-out)
12
F( t ) = 1 e
>0 t>0
(failure density)
R(t ) = e
F( t ) t
m=
1
=
13
System Logic.
XT = (XAXB+XBXC+XCXA) - 2XAXBXC
14
q( t ) = 1 e t F( t )
System Unreliability:
with = 5x10 4
hr 1
FT ( t ) = 3(1 e t ) 2 2(1 e t ) 3
15
t = 0.36
q( t ) = 1 e 0.36 = 0.30
Indeed,
FT(720) 3x0.362 2x0.363 = 0.295 > 0.216
RPRA 3. Probability Distributions in RPRA
16
MTTF = R (t )dt
0
Proof
dR
)dt = tdR =
MTTF = tf (t )dt = t(
dt
0
0
0
17
and
R(t ) 0
faster
than
18
MTTF = e
0
1
dt =
19
YT = Yj
1
MTBF =
1
1
M system
20
Fs ( t ) = (1 e t ) 2
R s ( t ) = 1 Fs ( t ) = 2e t e 2t
2 1
3
=
MTTF =
2 2
21
1
The MTTF for a single exponential component is:
22
0.0035
b=0.8
b=1.5
0.003
b=1.0
b=2.0
0.0025
0.002
0.0015
0.001
0.0005
0
b b 1
200
400
600
800
1000
h ( t ) = b t
R(t ) = e
(t )
For b = 1 the
exponential distribution.
23
A Simple Calculation
The average rate of loss of electric power in a city is 0.08 per year. A
hospital has an emergency diesel generator whose probability of
starting successfully given loss of power is 0.95.
i.
What is the rate of occurrence of blackouts at this hospital?
Loss of power
=0.08 yr-1
b = 0.004 yr
p = 0.95
ok = 0.076 yr
-1
-1
24
-3
per year.
ii. What is the probability that the hospital will have no blackouts in a period of
five years? Exactly one blackout? At least one blackout?
b t = 0 . 004 x 5 = 0 . 02
Use the Poisson distribution:
P(no blackouts in 5 yrs) = exp(-0.02) = 0.9802
P(exactly one blackout in 5 yrs) = (0.02)x0.9802 = 0.0196
P(at least one blackout in 5 yrs) = P(1 or 2 or 3) =
= 1 - P(no blackouts in 5 yrs) = 1 0.9802 = 0.0198
RPRA 3. Probability Distributions in RPRA
25
f (x) =
1
(x )
exp[
]
2
2
2
< x<
0<<
Mean:
< <
Standard Deviation:
RPRA 3. Probability Distributions in RPRA
26
Standard Normal
Distribution:
X
Z=
1
z2
( z ) =
exp[ ]
2
2
27
0
0
0.0398
0.0793
0.1179
0.1554
0.1915
0.2257
0.258
0.2881
0.3159
0.01
0.004
0.0438
0.0832
0.1217
0.1591
0.195
0.2291
0.2611
0.291
0.3186
0.02
0.008
0.0478
0.0871
0.1255
0.1628
0.1985
0.2324
0.2642
0.2939
0.3212
0.03
0.012
0.0517
0.091
0.1293
0.1664
0.2019
0.2357
0.2673
0.2967
0.3238
0.04
0.016
0.0557
0.0948
0.1331
0.17
0.2054
0.2389
0.2704
0.2995
0.3264
0.05
0.0199
0.0596
0.0987
0.1368
0.1736
0.2088
0.2422
0.2734
0.3023
0.3289
0.06
0.0239
0.0636
0.1026
0.1406
0.1772
0.2123
0.2454
0.2764
0.3051
0.3315
0.07
0.0279
0.0675
0.1064
0.1443
0.1808
0.2157
0.2486
0.2794
0.3078
0.334
0.08
0.0319
0.0714
0.1103
0.148
0.1844
0.219
0.2517
0.2823
0.3106
0.3365
0.09
0.0359
0.0753
0.1141
0.1517
0.1879
0.2224
0.2549
0.2852
0.3133
0.3389
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
0.3413
0.3643
0.3849
0.4032
0.4192
0.4332
0.4452
0.4554
0.4641
0.4713
0.3438
0.3665
0.3869
0.4049
0.4207
0.4345
0.4463
0.4564
0.4649
0.4719
0.3461
0.3686
0.3888
0.4066
0.4222
0.4357
0.4474
0.4573
0.4656
0.4726
0.3485
0.3708
0.3907
0.4082
0.4236
0.437
0.4484
0.4582
0.4664
0.4732
0.3508
0.3729
0.3925
0.4099
0.4251
0.4382
0.4495
0.4591
0.4671
0.4738
0.3531
0.3749
0.3944
0.4115
0.4265
0.4394
0.4505
0.4599
0.4678
0.4744
0.3554
0.377
0.3962
0.4131
0.4279
0.4406
0.4515
0.4608
0.4686
0.475
0.3577
0.379
0.398
0.4147
0.4292
0.4418
0.4525
0.4616
0.4693
0.4756
0.3599
0.381
0.3997
0.4162
0.4306
0.4429
0.4535
0.4625
0.4699
0.4761
0.3621
0.383
0.4015
0.4177
0.4319
0.4441
0.4545
0.4633
0.4706
0.4767
2
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
0.4772
0.4821
0.4861
0.4893
0.4918
0.4938
0.4953
0.4965
0.4974
0.4981
0.4778
0.4826
0.4864
0.4896
0.492
0.494
0.4955
0.4966
0.4975
0.4982
0.4783
0.483
0.4868
0.4898
0.4922
0.4941
0.4956
0.4967
0.4976
0.4982
0.4788
0.4834
0.4871
0.4901
0.4925
0.4943
0.4957
0.4968
0.4977
0.4983
0.4793
0.4838
0.4875
0.4904
0.4927
0.4945
0.4959
0.4969
0.4977
0.4984
0.4798
0.4842
0.4878
0.4906
0.4929
0.4946
0.496
0.497
0.4978
0.4984
0.4803
0.4846
0.4881
0.4909
0.4931
0.4948
0.4961
0.4971
0.4979
0.4985
0.4808
0.485
0.4884
0.4911
0.4932
0.4949
0.4962
0.4972
0.4979
0.4985
0.4812
0.4854
0.4887
0.4913
0.4934
0.4951
0.4963
0.4973
0.498
0.4986
0.4817
0.4857
0.489
0.4916
0.4936
0.4952
0.4964
0.4974
0.4981
0.4986
0.4987
0.4987
0.4987
0.4988
0.4988
0.4989
0.4989
0.4989
0.499
0.499
28
= 10,000 hr (MTTF)
= 1,000 hr
11,000 10,000
Z=
=1
1,000
29
An Example
A capacitor is placed across a power source. Assume
that surge voltages occur on the line at a rate of one
per month and they are normally distributed with a
mean value of 100 volts and a standard deviation of
15 volts. The breakdown voltage of the capacitor is
130 volts.
30
An Example (2)
i. Find the mean time to failure (MTTF) for this capacitor.
sv = 1 per month
Pd/sv = conditional probability of damage given a surge voltage
= P (surge voltage>130 volts/surge voltage) =
130 100
) = P(Z > 2) =
= P(Z >
15
= 1 P ( Z < 2 ) = 1 0 . 9772 = 0 . 0228
31
An Example (3)
Therefore, the rate of damaging surge voltages is
d = sv xPd / sv = 1x 0.0228 = 2.28x10 2
(month)1
1
2.28x10
= 43.86
months
32
An Example (4)
33
Observations
34
*t
35
2
2
2
()
36
2
mean : m = exp +
2
median : 50 = e
95 = e +1.645
05 = e 1.645
Error Factor:
95 50
95
EF =
=
=
50 05
05
RPRA 3. Probability Distributions in RPRA
37
Y ln
is a normal variable with parameters
(mean) and
(standard deviation).
RPRA 3. Probability Distributions in RPRA
38
But, Y ln
95 = e
+1.645
ln 95 = + 1.645
as in slide 37
39
= -6.91
Median:
and
=1.40
50 = exp(-6.91) 10-3
Mean:
95th percentile:
5th percentile:
Error Factor:
40
RPRA 4.
Availability
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
RPRA 4. Availability
Definitions
Unavailability:
q(t) = Pr[down at t]
Availability:
q(t) + a(t) = 1
RPRA 4. Availability
Unattended components
q(t) = F(t) = Pr[T < t]
Example:
RPRA 4. Availability
RPRA 4. Availability
q=
=
, Note : q for < 0.1
1 1 +
+
RPRA 4. Availability
5
Example
2
1
A
B
3
MTTF (hrs)
MTTR (hrs)
1
2
3
800
600
600
8
15
15
RPRA 4. Availability
Example (2)
What is the reliability of the system for one
month assuming that no repair is available?
Step 1: System Logic
Minimal path sets:
{ Y1, Y2}
{ Y1 , Y3 }
Example (3)
Step 2: Reliability of the system in terms of component
reliabilities:
RS = R1(R2 + R3 - R2R3)
Component reliabilities:
R1 = exp(-720/800) = 0.407
R2 = R3 = exp(-720/600) = 0.301
Therefore, RS = 0.208
RPRA 4. Availability
Example (4)
What is the availability of the system assuming that the
repair process starts immediately upon detection of
failure?
Structure function for success
YS = Y1(Y2 + Y3 - Y2 Y3)
Availability of the system in terms of component
availabilities:
AS = a1(a2 + a3 - a2a3)
RPRA 4. Availability
Example (5)
Component availabilities:
a1 = 1 (8/800) = 0.990
a2 = a3 = 1 (15/600) = 0.975.
Therefore, AS = 0.989.
RPRA 4. Availability
10
Note
RPRA 4. Availability
11
Note (2)
The unreliability of the system for one month is:
FS = F1 + F2F3 - F1F2F3
where:
F1 = 1 - exp(-720/800) = 0.593
F2 = F3 = 1 - exp(-720/600) = 0.699.
Thus,
FS = 0.792 = 1 - 0.208 = 1 - RS
RPRA 4. Availability
12
13
RPRA 5.
Data Analysis
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
Statistical Inference
Theoretical Model
Evidence
Failure distribution,
e.g., f (t ) = e t
Sample, e.g.,
{t1,,tn}
Random Samples
The observed values are independent and the
underlying distribution is constant.
1 n
t = ti
n 1
Sample mean:
Sample variance:
n
1
2
s2 =
(
t
t
)
i
(n 1) 1
1
=t
229.9
1
t = (10.2 + 54 + 23.3 + 41.2 + 73.2 + 28) =
= 38.32
6
6
MTTF = 38.32
hrs;
RPRA 5. Data Analysis
1
= 0.026
38.32
hr 1
4
2.032
= 0.508
( 5 1)
s = 0.713 =
s2 =
{3 eqs in 7 years}
r
=
t
3
= = 0.43
7
yr 1
6
{k 1s in n trials}
k
q=
n
3
{3 failures to start in 17 tests} q = = 0.176
17
RPRA 5. Data Analysis
T = t i + ( n k )t 0
T = t i + ( n r )t r
k
T
or =
r
T
Example
Sample: 15 components are tested and the test
is terminated when the 6th failure occurs.
The observed failure times are:
{10.2, 23.3, 28.0, 41.2, 54.0, 73.2} hrs
The total operational time is:
T = 10.2 + 23.3 + 28 + 41.2 + 54 + 73.2 + (15 6)73.2 = 888.7
Therefore
6
= 6.75x10 3
888.7
hr 1
Bayesian Methods
Recall Bayes Theorem (slide 16, RPRA 2):
Likelihood of the
Evidence
P (H i E ) =
P(E H i )P(H i )
Prior
Probability
P(E H i )P(H i )
1
Posterior
Probability
10
11
12
R ( t ) = R ( t / )( )d
13
Communication of Epistemic
Uncertainties: The discrete case
Suppose that P( = 10-2) = 0.4 and P( = 10-3) = 0.6
Then, P(e-0.001t) = 0.6
and
P(e-0.01t) = 0.4
exp(-0.001t)
0.4
exp(-0.01t)
14
Communication of Epistemic
Uncertainties: The continuous case
15
Risk Curves
95th Percentile
1.0E-04
Mean
Probability of Exceedence
Median
5th Percentile
1.0E-05
1.0E-06
1.0E-07
1.0E-08
1.0E-09
10
100
Public Acute Fatalities
1,000
16
17
(ln )2
1
( ) =
exp
2
2
2
18
Mechanical Hardware
Component/Primary
Failure Modes
Assessed Values
Lower Bound Upper Bound
Pumps
Failure to start, Qd:
3 x 10-4/d
3 x 10-3/d
3 x 10-6/hr
3 x 10-4/hr
3 x 10-4/d
3 x 10-3/d
3 x 10 /d
3 x 10-4/d
3 x 10-4/d
3 x 10-3/d
Plug, Qd:
3 x 10-5/d
3 x 10-4/d
1 x 10-4/d
1 x 10-3/d
Plug, Qd:
3 x 10-5/d
3 x 10-4/d
Check
Failure to open, Qd:
3 x 10-5/d
3 x 10-4/d
3 x 10-6/d
3 x 10-5/d
Manual
Plug, Qd:
3 x 10-5/d
3 x 10-4/d
3 x 10-11/hr
3 x 10-8/hr
3 x 10-12/hr
3 x 10-9/hr
1 x 10-4/d
1 x 10-3/d
Failure to run, o:
(Normal Environments)
Valves
Motor Operated
Failure to operate, Qd:
Plug, Qd:
-5
Solenoid Operated
Air Operated
Relief
Pipe
Plug/rupture
Clutches
Mechanical
Failure to engage/disengage
19
Electrical Hardware
Component/Primary
Failure Modes
Assessed Values
Lower Bound Upper Bound
Electrical Clutches
Failure to operate, Qd:
1 x 10-4/d
1 x 10-3/d
1 x 10-4/d
1 x 10-3/d
3 x 10-6/hr
3 x 10-5/hr
3 x 10-7/hr
3 x 10-6/hr
3 x 10-5/d
3 x 10-4/d
3 x 10-4/d
3 x 10-3/d
1 x 10-4/d
1 x 10-3/d
3 x 10-5/d
3 x 10-4/d
3 x 10-5/d
3 x 10-4/d
3 x 10-6/d
3 x 10-5/d
1 x 10-6/hr
1 x 10-5/hr
3 x 10-7/hr
3 x 10-5/hr
1 x 10-2/d
3 x 10-4/hr
1 x 10-1/d
3 x 10-2/hr
1 x 10-7/hr
1 x 10-5/hr
Motors
Failure to start, Qd:
Failure to run
(Normal Environments), o:
Transformers
Open/shorts, o:
Relays
Failure to energize, Qd:
Circuit Breaker
Failure to transfer, Qd:
Limit Switches
Failure to operate, Qd:
Torque Switches
Failure to operate, Qd:
Pressure Switches
Failure to operate, Qd:
Manual Switches
Failure to operate, Qd:
Instrumentation
Failure to operate, o:
20
Example
Lognormal prior distribution with median and
95th percentile given as:
50 = exp( ) = 3x10 3 hr 1
95 = exp( + 1.645 ) = 3x10 2 hr 1
Then
= 5.81 , = 1.40
2
E[ ] = exp( + ) = 8x10 3 hr 1
2
05 = exp( 1.645 ) = 3x10 4 hr 1
RPRA 5. Data Analysis
21
22
23
Prior (
) and posterior (
distributions
0.008
0.007
Probability
0.006
0.005
0.004
0.003
0.002
0.001
0.000
1e-4
1e-3
1e-2
1e-1
24
Prior
distr.
95th
Mean
(hr-1)
(hr-1)
8.0x10-3 3.0x10-2
Median
5th
(hr-1)
(hr-1)
3x10-3 3.0x10-3
9x10-4
1.5x10-4
25
Selected References
26
RPRA 6.
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
Objectives
Identify accident scenarios.
Rank these scenarios according to their
probabilities of occurrence.
Rank systems, structures, and components
according to their contribution to various risk
metrics.
RPRA 6. Probabilistic Risk Assessment
PRA Steps
Define end states
Identify initiating events
Develop event and fault trees
Quantify
PWRs keep water under pressure so that it heats, but does not boil. Water from
the reactor and the water in the steam generator that is turned into steam never
mix. In this way, most of the radioactivity stays in the reactor area.
Courtesy NRC
http://www.nrc.gov/reading-rm/basic-ref/students/animated-pwr.html
MLD Development
Begin with a top event that is an end state.
The top levels are typically functional.
Develop into lower levels of subsystem and
component failures.
Stop when every level below the stopping level
has the same consequence as the level above it.
RPRA 6. Probabilistic Risk Assessment
Excessive
Offsite
Release
Excessive
Release of
Core Material
Excessive
Core Damage
Insufficient
Reactivity
Control
Insufficient
Core-heat
Removal
Insufficient
RCS Inventory
Control
Excessive
Release of
Non-Core Material
RCS pressure
Boundary
Failure
Insufficient
RCS Heat
Removal
Insufficient
RCS Pressure
Control
Conditional
Containment
Failure
Insufficient
Isolation
Insufficient
Pressure &
Temperature
Control
Insufficient
Combustible
Gas Control
NPP:
Initiating Events
Transients
Loss of offsite power
Turbine trip
others
Loss-of-coolant accidents (LOCAs)
Small LOCA
Medium LOCA
Large LOCA
RPRA 6. Probabilistic Risk Assessment
10
Secondary
Heat Removal
Bleed
& Feed
Recirc.
Core
OK
OK
PDSi
PDSj
11
Human Performance
The operators must decide to perform feed &
bleed.
Water is fed into the reactor vessel by the
high-pressure system and is bled out through
relief valves into the containment. Very costly
to clean up.
Must be initiated within about 30 minutes of
losing secondary cooling (a thermal-hydraulic
calculation).
RPRA 6. Probabilistic Risk Assessment
12
13
Reasons Categories
Unsafe acts
Unintended action
Slip
Lapse
Mistake
Intended violation
J. Reason, Human Error, Cambridge University Press, 1990
14
Latent Conditions
Weaknesses that exist within a system that
create contexts for human error beyond the
scope of individual psychology.
They have been found to be significant
contributors to incidents.
Incidents are usually a combination of
hardware failures and human errors (latent
and active).
RPRA 6. Probabilistic Risk Assessment
15
Reasons Model
Line
Fallible
Management
Psychological
Unsafe
Decisions
Deficiencies
Precursors
Acts
16
Errors of commission
Errors of omission
Median
EF
3x10-3
10-3
17
Post-IE Errors
Models still being developed.
Typically, they include detailed task analyses,
identification of performance shaping factors
(PSFs), and the subjective assessment of
probabilities.
PSFs:
System design, facility safety culture,
organizational factors, stress level, others.
18
Risk Models
IE2
AA
BB
CC
DD
END-STATE-NAMES
OK
2 T => 4
TRAN1
LOV
4 T => 5
TRAN2
LOC
LOV
AA
A1
A2
BB
B-GATE1
B-GATE2
B-GATE3
EVENT-B1
B-GATE4
EVENT-B2
EVENT-B3
EVENT-B4
B-GATE5
EVENT-B6
EVENT-B5
B-GATE6
EVENT-B7
EVENT-B8
B-GATE7
EVENT-B9
EVENT-B10
EVENT-B11
19
System Analysis
20
Component B2
System Logic
Failure
Probability
21
Example (contd)
22
Dependencies
Some dependencies are modeled explicitly, e.g.,
fires, missiles, earthquakes.
After the explicit modeling, there is a class of
causes of failure that are treated as a group.
They are called common-cause failures.
Special Issue on Dependent Failure Analysis, Reliability Engineering and
System Safety, vol. 34, no. 3, 1991.
23
24
Component Failure
25
26
Calculating Probabilities
Using the rare event approximation, the system
failure probability of a two-out-of-three system
is given by
P(S) = P(AI) P(BI) + P(AI) P(CI) + P(BI) P(CI) +
P(CAB) + P(CAC) + P(CBC) + P(CABC)
27
28
Component B 1
Common Cause Failure
(1 )
Component B 2
Independent Failure
Component B 2
Common Cause Failure
(1 )
B 1*
B 12
B 2*
B 1*
B 2*
B 12
Fault Tree
29
0.18
0.16
0.14
Average
0.12
0.1
0.08
0.06
0.04
0.02
EN
G
IP
IS
S
EL
TR
R
PW
TO
EA
C
R
ER
BR
AK
ER
S
M
AT
O
BW SAF
T
ET
O
O
R
R
R
Y/
SA
S
V
AL
FE RE
VE
TY LIE
F
S
/R
EL PU
M
IE
P
F
VA S
LV
R
H
ES
R
P
C
U
O
M
N
PS
SI
T
SP
PU
R
AY MP
S
PU
AF
M
PS
SW W
/C PU
M
C
PS
W
PU
M
PS
30
31
Data
A1- A1-PRI-VLV-FC
The epistemic
distribution is
Lognormal.
4.500E-003 5.900E+000
Error factor
32
Data Analysis
The process of collecting and analyzing
information in order to estimate the
parameters of the epistemic PRA models.
Typical quantities of interest are:
33
Sources of Information
Ideally parameters of PRA models of a specific system
should be estimated based on test and/or operational
data of that system.
Often, however, the analysis has to rely on a number of
other sources and types of information as the quantity
or availability of system-specific data are insufficient.
In such cases surrogate data, generic information, or
expert judgment are used directly or in combination
with (limited) system-specific data.
34
Data Sources
Generic
System-specific
Maintenance Logs
Test Logs
Operation Records
RPRA 6. Probabilistic Risk Assessment
35
Data Needs
The type of data needed varies depending on the type
of event and their specific parametric representation
Probabilities typically require
Event Counts (e.g., Number of Failure)
Exposure, or Success Data (e.g., Total Operating
Time)
Other parameters may require only one type of data
Maintenance/Repair Duration
Counts of Multiple Failures (CCFs)
RPRA 6. Probabilistic Risk Assessment
36
Bayesian Estimation
Two main steps:
The first step involves using available information
fit a subjective, or prior, distribution to a
parameter, such as a failure rate. The uncertainties
in the parameter values are expressed in the prior
distribution.
The second step involves using additional or new
data to update an existing prior distribution using
Bayes' Theorem.
RPRA 6. Probabilistic Risk Assessment
37
L(E / )( )
' ( / E ) =
L(E / )( )d
38
39
40
Risk Management
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
Risk Acceptance
UNACCEPTABLE REGION
TOLERABLE REGION
Frequency (Events/Year)
1/10
Explosions
1/100
Dam Failures
1/1,000
Chlorine Releases
Air Crashes Persons
on Ground
1/10,000
100 Nuclear
Power Plants
1/100,000
1/1,000,000
1/10,000,000
10
100
1,000
10,000
100,000
1,000,000
Fatalities
Safety Goals
A lifetime cancer risk of less than 10-4 for the most
exposed person is acceptable. (US EPA)
A lifetime cancer risk of less than 10-6 for the average
person is acceptable. (US EPA)
Individual probability of death greater than 10-3 per year
for workers and 10-4 per year for the public is
unacceptable. (UK Health and Safety Executive)
Individual probability of death less than 10-6 per year is
broadly acceptable. (UK Health and Safety Executive)
RPRA 7. Risk Management
10
11
x =
Q
j
d jQ x d , j
12
Case study
Zone designator/
Scenario
3. Fire on Floor
Damaging
Control Cable 10
Feet Above the
Floor and Failing
All Control
Instrumentation
Capability
Frequency,
Events Per Year
x
y
Qd/j
Qx/d,j
Qy/x,d,j
1.0
1.0
4.6-8
7.9-6
4.2-4
4.6-8
7.9-6
4.2-4
1.1-7
1.3-5
3.7-4
0.32
0.62
0.90
7.1-5
7.1-5
1.2-4
0.62
5.5-8
4.7-6
1.0-4
5.5-8
4.7.6
1.0-4
1.2-7
8.4-6
1.6-4
0.20
0.55
0.87
2.4-5
2.4-5
4.2-5
1.0
0.57
1.0
3.0-10
7.3-8
3.3-6
<1.0-10
7.3-9
5.9-7
5.3-7
3.3-5
5.0-4
0.12
0.45
0.80
2.5-4
5.0-3
1.0-1
0.02
0.1
0.5
1.9-6
3.0-7
1.5-4
0.48
2.6-2
0.16
-8
13
4.2E-5;
14
15
16
Analytical results
Option
Description
Percentile
th
x
Events per
year
Reduction
Factor
-6
y
Events per
year
Reduction
Factor
-6
1.5 10
-5
2.6 10
-4
9.7 10
5
th
50
th
95
2.2 10
-5
3.0 10
-3
1.1 10
1.0 x 10
-7
5.9 10
-6
9.1 10
-4
2.3 10
1.0
-5
9.6 10
-7
2.1 10
-6
7.4 10
-4
2.1 10
1.0
Mean
th
5
th
50
th
95
Mean
th
5
th
50
th
95
3.9 10-5
-6
1.6 10
-6
8.8 10
-5
9.9 10
2.6
3.3 10
-7
3.6 10
-6
3.4 10
-5
9.2 10
-5
2.9
Mean
th
5
th
50
th
95
1.9 10
-6
1.7 10
-6
7.1 10
-5
4.8 10
-5
5.3
1.2 10
-7
5.6 10
-6
3.0 10
-5
2.6 10
-5
8.0
Mean
1.4 x 10
-5
7.1
6.9 10
-6
14.0
Base Case
Fire Barriers
Self
Contained
Charging
Pump
Alternate
Power
Source
-4
17
Deliberation
18