72 PDF

Engineering Risk Benefit Analysis
1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82, ESD.72
CBA 2.
The Time Value of Money
George E. Apostolakis
Massachusetts Institute of Technology
Spring 2007
CBA 2. The Time Value of Money
The Importance of Time

A dollar in hand now is worth more than a dollar
received in the future, because of its earning power,
i.e., it can be invested to generate income.
The purchasing power of money, i.e., the amount of
goods that a certain amount can buy, changes with
time also.
Objective: To develop methods for establishing
the equivalence of sums of money. It depends on the
amounts, the time of occurrence of the sums of
money, and the interest rate.
Overview of Lecture
The Basics of Interest Rates Simple and

Compound Interest
The Basic Discount Factors Present Value,
Future Value, Annual Value
Economic Equivalence and Net Present Value
Return to Interest Rates: Nominal and Effective
Rates
Inflation
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.
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
Cash Flows over Time

$1,000
$100
$100
$100
$1,100
Up arrow = we receive $; down arrow = we pay $

Amount borrowed: $1,000
Interest is paid at the end of each year at the rate of
10%.
The principal is due at the end of the fourth year.
The Basic Discount Factors
P (present value)
F
(future value)
A
(annual value
or annuity)
Single-Payment Compound-Amount Factor

P
1
P(1+i)
P(1+i)2
(F/P, i, n) = (1+i)n
F= P(1+i)n
A single payment is made after n periods.

The interest earned at the end of each period is
charged on the total amount owed (principal plus
interest).
$1 now is worth (F/P,i,n) at time n if invested at i%
Single-Payment Present-Worth Factor
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
Equal payments, A, occur at the end of each period.

We will get back (F/A, i, n) at the end of period n if funds
are invested at an interest rate i.
F = A + A(1+i) + A(1+i)2 ++ A(1+i)n-1
(1 + i )n 1
(F / A, i, n ) =
i
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
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)
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
Example: Your house mortgage is $300,000 for 30

years with an nominal annual rate of 7%. What is the
monthly payment?
n = 360 months
i = 0.583% per month
(A/P, 0.00583, 360) = 0.006650339
A = 300,000x 0.006650339 = $1,995.10 per month
13
Summary of the Formulas

Single-Payment Compound-Amount Factor
(F/P, i, n) = (1+i)n
Equal-Payment-Series Compound-Amount Factor
(1 + i )n 1
(F / A, i, n ) =
i
Equal-Payment-Series Capital-Recovery Factor

i(1 + i )n
( A / P, i, n) =
(1 + i)n 1
14
Continuous Compounding (1)

Suppose that interest is compounded a very
large number of times. Then, the effective
annual interest rate is
m
lim 1 +
m
i=
1 = er 1
m
where r is the nominal annual interest rate.
15
Continuous Compounding (2)

(F/P, i, n) = (1+i)n
(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
Nominal and Effective Interest Rates (1)

The nominal interest rate (or annual
percentage rate) is the annual rate without
the effect of any compounding.
The effective (actual) interest rate is the
annual rate taking into account the effect of
any compounding during the year.
Example:
A credit card advertises a nominal rate of 18%
compounded monthly. The actual rate is, then, (18/12) = 1.5%
per month. The effective annual rate is
(1.015)12 1 = 0.1956 or 19.56%
(if you do not pay anything each month)
17

The effective interest rate i depends on the
frequency of compounding.
Example: nominal interest rate r = 10%
Compounded annually: i = r = 10%

Compounded quarterly: i = (1+0.1/4)4 -1 = 10.38%
Compounded monthly: i = (1+0.1/12)12 -1 = 10.471%
Compounded weekly: i = (1+0.1/52)52 -1 = 10.506%
Compounded daily: i = (1+0.1/365)365 1 = 10.516%
Compounded continuously: i = e0.1 1 = 10.517%
18

In the formulas we introduced in earlier slides, i is the
effective interest rate for a given period and n is the
number of such periods.
Example:
You wish to buy a house and you can afford to
make a down payment of $50,000. Your monthly mortgage
payment cannot exceed $2,000. If 30-year loans are available
at 7.5% interest compounded monthly, what is the highest
price that you may consider?
19

Solution:
Lets use one month as the time period. Then, n = 360
months, and i = (7.5/12) = 0.625%. We know that
i(1 + i )n
( A / P, i, n) =
(1 + i)n 1
This yields (A/P, 0.00625, 360) = 0.00699
P x 0.00699 = (H - 50,000) x 0.00699 2,000
H (2,000/0.00699) + 50,000 = $336,123

20

Lets use one year as the time period. Then, n = 30 years,
and i = (1+0.00625)12 - 1 = 7.763%
Then, (A*/P, 0.07763, 30) = 0.0867
A* = 0.0867P per year
Your effective payment per year is

A* = $2,000x(F/A, 0.00625, 12) = $2,000x12.4212 = $24,842
P (24,842/0.0867) + 50,000 = $336,533 as before
Consistency between i and n will lead to identical solutions.
21
Economic Equivalence (1)

The formulas that we have developed establish
economic equivalence between P and F, an equalpayment series and F, and so on.
Example: Consider the following cash flow: You
will receive $500 at the end of years 3 and 4 and
$1,000 at the end of year 5. If the interest rate is
7%, what amount received at the present is
equivalent to this cash flow?
22

$1,000
Solution:
1
$500
$500
500
500
1,000
P=
+
+
3
4
5
(1 + 0.07 ) (1 + 0.07 ) (1 + 0.07 )
P = 408.15 + 381.45 + 712.99 = $1,502.59

23

If the interest is compounded continuously, the result
will be:
P=
500
3 x 0.07
500
4 x 0.07
1,000
5 x 0.07
P = 405.30 + 377.89 + 704.69 = $1,487.88
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.
25
CPI and Inflation

Year
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.
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.
27
Example
The average inflation rate from 1967 to 1999 is given
by
100(1+f)32 = 497.6
1+f = 4.9761/32 = 1.0514
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
Constant and Actual Dollars

(actual dollars) = (1+f)n (constant dollars)
(based on the purchasing power n years earlier)
Equivalence in terms of actual dollars: Use i.

Equivalence in terms of constant dollars: Use i'.
Relationship among i, i', and f:
1+ i
i =
1
1+ f
'
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
Example: Going to the Movies

1967 Ticket Price:
$1.25
1999 Ticket Price:
$8.50
Has there been a price increase above the rate of
inflation?
The average rate of inflation has been (slide 28):
f = 5.14%.
The actual rate of increase is
i = (8.5/1.25)1/32 -1 = 0.0617.
Therefore,
i' = [(1+0.0617)/(1+0.0514)] - 1 = 0.0098 1%
32
Example: Investments in Two Countries (1)

John has immigrated to the US where the inflation
rate is 2% while his brother Joe has stayed in the old
country where the inflation rate is 4.5%. The US
banks give an interest rate of 5.5% while those of the
old country give 8%.
1. What are the real interest rates in the two countries?
'
US
1 + 0.08
'
1 + 0.055
1 = 3.35%
=
1 = 3.43% iOC =
1 + 0.045
1 + 0.02
33
Example: Investments in Two Countries (2)

2. If John decides to invest in the Old Country, what
would his real interest rate be?
Interest rate of the OC
(John invests there)
'
US / OC
1 + 0.08
=
1 = 5.88%
1 + 0.02
US inflation rate
(John lives there)
34

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82, ESD.72
CBA 3.
Bases for Comparison of Alternatives

Spring 2007
CBA 3. Bases for Comparison of Alternatives
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
Present Worth (1)

Present Worth: The net equivalent amount at the
present that represents the difference between the
equivalent receipts and the equivalent
disbursements of an investment cash flow for a
selected interest rate i.
If Ft is the net cash flow at time t, then
n
PW ( i ) = Ft (1 + i )t
t =0
Present Worth (2)

Present worth may also refer to the present value of
receipts (benefits) or disbursements (costs). In this
case, the criterion for decision making is the Net
Present Worth.
We will follow the definition on slide 3 (Thuesen &
Fabrycky).
Present Worth, Net Present Worth, and Net

Present Value are equivalent terms.
t =n
t =n
PW (i ) = NPW (i ) = NPV (i ) = B t (1 + i ) Ct (1 + i )t
t =0
t =0
Present Worth: Example

You buy a car and you put down $5,000. Your payments
will be $500 per month for 3 years at a nominal interest rate
of 10%. Assuming monthly compounding, what is the
present price you are paying?
From CBA 2, Slide 14, we get
( P / A , i, n ) =
(1 + i) n 1
i(1 + i ) n
Here: A = $500/mo, i = 10/12 = 0.83%, n = 36 months

P = 5,000 + 500[
(1 + 0.0083) 36 1
0.0083(1 + 0.0083)
= 5,000 + 15,505 = $20,505
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
Another name is Net Annual Value (NAV).

For fixed i and n, AE and PW yield the same
results, i.e., the same ranking of alternatives.
PW and AE Criteria for Decision Making

An investment alternative Aj, j = 1,,n, is a decision
option representing a course of action.
Decision Criteria:
Aj f Ak
if
PW(i)Aj > PW(i)Ak
Aj f Ak
if
AE(i)Aj > AE(i)Ak
or
These criteria are using the total investment.

Example of Total Investment Comparisons

End of Year
0
1
2
3
B3
-$12,000
-1,200
-1,200
1,500
B4
-$15,000
-400
-400
3,000
The benefits from these alternatives are

identical. We must select one.
Assume that i = 10%.
Total Investment Comparisons (2)

1.12 1 1,500
PW (10%)B 3 = 12,000 1,200
+
=
2
3
0.1x1.1
1.1
1,500
= 12,000 1,200x1.736 +
= $12,956
1.331
PW(10%)B 4 = $13,440 < $12,956
Therefore, B3 should be preferred over B4.

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
AE(10%)B 4 = 15,000x0.402 400 + 3,400x0.302 =

= $5,403 < $5,209
Thus, B3 should be preferred over B4, just as before.
10
Impact of Inflation (1)

Suppose that the inflation rate is 9% and that the amounts
shown on slide 8 are in terms of constant dollars.
Converting them to actual dollars we get:
End of year 1: F = 1,200(F/P, 9, 1) = 1,200x1.091 = $1,308
End of year 2: 1,200x1.092 = 1,426
End of Year
0
1
2
3
B3
-$12,000
-1,308
-1,426
1,943
B4
-$15,000
-436
-475
3,885
11
Impact of Inflation (2)

Recalculate the PWs.
1,308 1,426 1,943
PW (10%)B 3 = 12,000
+
=
2
3
1.1
1.1
1.1
= $12,908 < $12,870 = PW (10%)B 4
Therefore, B4 is now preferred, while, in the case

without inflation, B3 was preferred (slide 9).
12
PW and AE on Incremental Investment

Derive the difference between the two alternatives.
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
If PW(i)B4-B3 > 0 or AE(i)B4-B3 > 0 B4 f B3

13
Example of Incremental Investment

PW (10%)B B =
4
= 3,000 + 800( P / A,10,2) + 1,500( P / F ,10,3) =

1.12 1 1,500
= 3,000 + 800
+
= $485 < 0
2
3
0.1x1.1
1.1
B3 should be accepted, just as in slide 9.

PW- or AE-based results using total and
incremental investments are identical.
14
Internal Rate of Return (IRR)

It is the interest rate for which the equivalent receipts of a
cash flow equal the equivalent disbursements.
It is the interest rate i* for which the present worth is zero.
n
0 = PW ( i ) = Ft (1 + i* )t
*
t =0
The IRR represents the percentage or rate earned on the

unrecovered balance of an investment such that the
payment schedule makes the unrecovered investment equal
to zero at the end of investment life.
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)
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
IRR and PW(i)

$
PW (i)
$900
i* = 12.8%
0
10%
20%
Interest rate (i)
IRR and its relationship to the present-worth amount.

Figure by MIT OCW.
Thuesen & Fabrycky, 9th Edition

18
Cash Flows with a Single IRR
The IRR is a useful concept when the shape of

PW(i) is like the one on Fig. 6.3 (slide 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)
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
The 2nd and 3rd conditions are violated.

Two IRRs are obtained: 1% and 18.5%
The IRR should not be used in decision making.
20
Minimum Attractive Rate of Return

(MARR)
MARR is a cut-off rate representing a yield on
investments that is considered minimally
acceptable.
The investor can always receive this rate (e.g., it
could be the bank rate).
It is determined by senior management.
If MARR is too high:
If MARR is too low:
Opportunities are lost.

Income is lost.
21
The Do-Nothing Alternative

The investor will do nothing about the proposed
alternatives and the funds will be placed in
investments that yield an IRR equal to the MARR.
PW(MARR)A0 = 0
AE(MARR)A0 = 0
For computational purposes, we may assume that
the cash flows for A0 alternative are all zero.
22
IRR on Incremental Investment

1. Make sure the cash flows satisfy the conditions on
slide 19.
2. List alternatives in ascending order based on
initial cost.
3. The current best alternative can be the Do
Nothing one.
4. Determine the differences between the
challenging alternative and the current best
alternative.
*
5. If i A A > MARR
Ak f A j
k
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
Therefore, A1 replaces A0 as the current best

alternative.
25
Example (3)
Similarly, we solve
0 = -3,000 + 500(P/A, x21, 10)
(1 + x 21 )10 1
0 = 3,000 + 500
10
x 21 (1 + x 21 )
i*A A x 21 = 10.5% < 15%
2
Therefore, A1 remains the current best alternative.

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
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
Therefore, A3 becomes the best solution.

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
IRR on Total Investment (1)

We can calculate the IRRs for the total cash flows
on slide 24.
Let
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%
The best alternative (i.e., the one having the

highest IRR) is A1, not A3 (slide 28).
31

A1
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
PW on Incremental Investment Revisited

We have already investigated this method on slide 14.
The major result was that:
PW- (or AE-)based results using total and
incremental investments are identical.
Approach
1. Make sure the cash flows satisfy the conditions on
slide 19.
2. List alternatives in ascending order based on
initial cost.
33
PW on Incremental Investment Revisited

(2)
3. The current best alternative can be the
Do Nothing one.
4. Determine the differences between the
challenging alternative (next highest
initial cost) and the current best
alternative.
5. If PW(MARR )A A > 0 Ak f A j
k
34
The Example Revisited (1)

Consider again the cash flows on slide 24.
PW(15)A1-A0 = -5,000 + 1,400(P/A, 15, 10)
PW(15)A A
1
(1 + 0.15)10 1
= 5,000 + 1,400
=
10
0.15(1 + 0.15)
= 2,026 > 0
Therefore, A1 becomes the current best alternative.

35

Similarly, PW(15)A2-A1 = -3,000 + 500(P/A, 15, 10)
PW (15)A A
2
(1 + 0.15)10 1
= 3,000 + 500
=
10
0.15(1 + 0.15)
= 490 < 0
Therefore, A1 remains the current best alternative.

36

Finally,
PW(15)A A
3
(1 + 0.15)10 1
= 5,000 + 1,100
=
10
0.15(1 + 0.15)
= 521 > 0
Therefore, A3 becomes the current and final best

alternative. This is the same as in slide 28, but not
the result on slide 31 (best: A1).
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.
38
Unequal Lives
All alternatives must be compared over the

same time span.
Assumptions are required to compare them
over the same study period or planning
horizon.
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
A, B, and C all fulfill the same objective, but for a

different number of years; select the least costly for i
= 7%
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
Repeating Cash Flows (1)

When sequences of cash flows are repeated, it is
only necessary to calculate the AE for the first
sequence to find the AE for all sequences.
For Alternative B, consider the extended B*:
0
16
1.4
1.4
1.4
1.4
17.4
17.4
42
Repeating Cash Flows (2)

AE(B*) = 1.4 + 16 [1 + (P/F, 0.07, 3)] (A/P, 0.07, 6)
= 1.4 + 16 x 1.816 x 0.21 = 1.4 + 6.10 = 7.50K
which is (approximately) the same as the original
value of 7.48K.
The AEs that we calculated in slide 41 represent the
following extended cash flows.
43
Extended Cash Flows

t (years)
0
1
2
3
4
5
6
7
8
9
10
11
12
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
multiple of the lifetimes.

Best alternative: C
45
Present Worth of Original Alternatives

PW(A, 6 yrs) = 4 + 6.4(P/A, 7%, 6) = 4 +(6.4)(4.76) =
= $34.46K
PW(B, 3 yrs) = 16 + 1.4 (P/A, 7%, 3) =
= 16 +(1.4)(2.62) = $19.68K
PW(C, 4 yrs) = 20 + 1 (P/A, 7%, 4) = 20 + 1(3.4) =
= $23.4K
These are the PWs of costs over the actual lifetimes.
Best alternative: B (not C). Incorrect.
46

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82, ESD.72
CBA 4.
Including Uncertainty
Spring 2007
CBA 4. Including Uncertainty
Uncertainty
Practically any CBA requires consideration of

uncertainty.
Most methodologies in use are ad hoc, due to
the intrinsic difficulty of the generalized
problem.
Methods
1. Scenario analysis
2. Adjustments of interest rates
3. Decision Theory
4. Simplified probabilistic models
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.
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:
-$49,000 22,000 120,000
What are we to do with such information?

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.
Forecasting Oil Prices

D. Draper, Assessment and propagation of model uncertainty, Journal
of the Royal Statist. Soc., B (1995) 45-97.
In 1980, 43 economists and energy experts forecast

the price of oil from 1981 to 2020 to aid in policy
planning.
They used 10 leading econometric models under
each of 12 scenarios embodying a variety of
assumptions about inputs, such as supply,
demand, and growth rates.
The Plausible Scenario

One scenario was termed as the plausible median
case. It represented the general trends to be
expected.
The 10 models were applied to the plausible
scenario.
Results for 1986:
Actual price: $13
Range of predictions: $27 to $ 51.

High Interest Rates

Justify choice of alternatives using a high interest
rate, e.g., 30%.
Example
Total annual income: $55,000
Capital cost:
$80,000
Annual capital recovery with return:
$80,000 (A/P, 30%, 6yrs) = $30,272
Annual operating cost $28,600
Net annual profit: 55,000 (30,272 + 28,600) =
= -$3,872
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
Decision Theory: 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
w: weak market for the new product
11
Manufacturing Example Payoffs
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
Demand and Probabilities:

Period 2 (5 yrs)
Period 1 (5 yrs)
s1
s2
s1
w2
w1
w2
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
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
13
Calculation of the Payoffs

(1 + 0.04 ) 5 1
PWs1 = 30 x ( P / A ,0.04 ,5 ) = 30 x
= 133 .52 K
5
0.04 x (1 + 0.04 )
PWs 2 = 30 x ( P / A , 0 .04 , 5 ) x ( P / F , 0 .04 , 5 ) =

(1 + 0.04 ) 5 1
1
= 30 x
x
= 109 .74 K
5
5
0 .04 x (1 + 0.04 ) (1 + 0 .04 )
PWs1 s 2 = 133 .52 + 109 .74 = 243 .26 K

PWw1 = 133.52x
10
= 44.51K
30
PW w 2 = 109 .74 x
10
= 36 .58 K
30
PWs1w 2 = 133.52 36.58 = 96.94K

PWw1w 2 = 44.51 36.58 = 81.09K
14
Calculation of the EMV

Old Product
(1 + 0.04 )10 1
= 121 .61K
PWO = 15 x ( P / A ,0.04 ,10 ) = 15 x
0.04 x (1 + 0.04 )10
New Product
EMVN = 243.26x0.4 + 96.94x0.4 -81.09x0.2 = 119.86K
Decision
Stay with the old product?
15
Calculation using Utilities

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]
U(243.26) = 0.665; U(121.61) = 0.637; U(96.94) = 0.632;
U(-81.09) = 0.590
Old Product
EU(O) = 0.637
New Product
EU(N) = 0.665x0.4 + 0.632x0.4 + 0.590x0.2 = 0.6368
Decision
Stay with the old product?
16
Probabilistic Models
We have:
XT
X1
X2
PW[ X(T)] = X 0 +
+
+ .... +
2
(1 + i ) (1 + i )
(1 + i )T
(1)
where X j B j C j are the net benefits in

year j.
All Xj are r.v.s, PW[X(T)] is a r.v.
Note that (1) is of the form:
Y = a0 X 0 + a1X 1 + a2 X 2 + ... + aT X T
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
i.e., the variance of PW.
18
Fundamental Relationships from

Probability Theory (1)
Let Z = aW+b
(Z and W are r.v.s, a and b constants)
E[Z] = a E[W] + b
2
Z
2 2
= a W
19

Let
Z = W1 + W2
(W1 , W2 : independent r.v.s)
E [ Z ] = E [W1 ] + E [W2 ]
2
Z
2
W1
2
+ W2
Note: Extends to any number of mutually

independent r.v.s.
20

Let
Z = aW1 + bW2 + c
E[Z] = aE[ W1 ] + bE[ W2 ] + c
2Z = a 2 2W1 + b 2 2W2
If W1 and W2 are normal, then Z is also normal.
We often assume that Z is normal even if W1 and

W2 are not.
21
Example: Reliability Physics

(RPRA 3, slide 30) 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.
Suppose that the breakdown voltage is also normally
distributed with standard deviation of 15 volts.
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
probability of damage given a surge voltage

= P(surge voltage>130 volts/surge voltage)
130 100
) = P(Z > 2) =
15
= 1 P ( Z < 2 ) = 1 0 . 9772 = 0 . 0228 < 0 . 08
= P(Z >
The uncertainty in the breakdown voltage increased the

failure probability.
24
Assuming Independence of Xj
X1
X2
XT
PWInd [ X(T)] = X 0 +
+
+ .... +
2
(1 + i ) (1 + i )
(1 + i )T
Xj ( j = 0, 1,........,T) are mutually independent

r.v.s with known E [ X j ] and X2 (= 2j )
E[PWInd { X(T)}] = E[YInd ] =
E[ X j ]
(1 + i ) j
j= 0
2PW ,Ind = 2Y ,Ind =
2j
(1 + i )2 j
j= 0
Note: Gaussian approximation for pdf of Y

may work well in this case.
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
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.
28
Expectation and Variance of Annual Net

Benefits
It is easy to determine the expected value and variance of each of X1,
X2, and X3, separately:
E[X1] = 0.1x3+0.25x4+0.3x5 +0.25x6+0.1x7 = $5,000
(Similarly, we have E[X2] = $5K and E[X3] = $5K. )
X2 1 = 0.1x(3-5)2 + 0.25x(4-5)2 + 0.3x(5-5)2 + 0.25x(6-5)2 +

+ 0.1x(7-5)2 = 1,300,000 (1,140)2
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
Independence: Calculations (2)

Note that Y can take on a total of 125 (= 555) different values
of independent outcomes in years 1, 2 and 3.
Using the expressions on slide 21 we get
E[YInd]= -10K+E[X1]/(1.08)+E[X2]/(1.08)2+E[X3]/(1.08)3 =
= -10K + (5K)[1/1.08 + 1/(1.08)2 + 1/(1.08)3]
= -10K+ (5K)(P/A, 0.08, 3) -10K + (5K)(2.5771) $2,885
2
2
2
2+ 2
2]2+
[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
31
Conclusion
Project A will, "on average," have a net

present value equal to about $2,885 and a
standard deviation of approximately
$1,700 around that average.
32
What can we do with this information?

Alternative
E[Y] = E[PW]
20
15
15
12
17
16
33
Possible Decision Criteria

Choose the alternative with the highest mean
value (A1).
Note how close A4 is and how large the standard
deviations are. Depending on the uncertainties,
A1 and A4 may be indistinguishable.
Minimize the probability of loss.
Assume normal distributions and find P(PW < 0).
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
35
Probability of Loss for Project A
PInd(PWInd < 0) = P(Z < -(2885/1700)) = P(Z < -1.7)

= 0.04
The fundamental assumption is that of
independence of the annual benefits.
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
Mean and Variance (Dependence)

1
1
1
....
+
+
+
PWDep [ X (T )] = X 0 + X
2
(1 + i) T
(1 + i) (1 + i)
PWDep (A) = YDep = -10K + X[1/(1.08) +

1/(1.08)2 +1/(1.08)3] = -10K + X(P/A, 8, 3)
From slide 15 we get
E[YDep] = -10K + E[X](P/A, 8, 3) $2,885, as

before
or
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
)
+
+
In both the independent and dependent cases the mean

values are the same ($2,885).
The standard deviation in the dependent case ($2,938) is
73% larger than that of the independent case ($1,700).
PDep(PW < 0) = P(Z < - (2885/2938)) = P(Z < -0.98) = 0.16
Compare with PInd(PW < 0) = 0.04 (slide 32)
These two cases are considered as bounding the problem.
39

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82, ESD.72
CBA 5. Evaluating Public Activities

Spring 2007
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.
Evaluating Public Activities

If all costs (disbenefits) and benefits can be
expressed in terms of money, then a decision
criterion such as the Net Present Value (Present
Worth) could be used.
Discounting of cost and benefit flows should be
applied, as appropriate.
Ethical issues may arise. Can the value of a park
be measured in monetary terms?
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.
Public Activities: Ethical Issues (1)

How should risk transfer between generations be
handled?
How much should we spend now to protect future
generations?
Future generations should not be discriminated
against.
Risks imposed on future generations should not
exceed current acceptable risk levels.
The Benefit-Cost Ratio

benefits to the public
BC(i ) =
cos t to the government
Public:
(+)
(-)
Government:
Criterion:
advantages, receipts, savings

disadvantages, disbursements,
losses
(+) disbursements, losses
(-)
savings, receipts
If BC(i)Aj-Ak > 1
Aj f Ak
6
Net Present Value (Present Worth)

Bt
Ct
NPV (i ) PW(i ) =
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.
Multiple Alternatives (1)
Example of slide 24, CBA 3

End of
Year
0
1 10
A0
0
0
A1
A2
-$5,000
1,400
-8,000
1,900
A3
-10,000
2,500
MARR = 15%

The present worth of the benefits for A1 is
(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

The best alternative (i.e., the one with the
highest CB ratio) is A1.
Recall (slide 28, CBA 3) that A3 was the best (PW
analysis).
7,026
Incremental Analysis BC(15)A1 A 0 =
= 1.41
5,000
2,510
BC(15) A 2 A1 = 500( P / A,15,10) =
= 0.83
3,000
A1 remains the best.

10
5,521
BC(15)
= 1,100( P / A ,15,10) =
= 1.10
A 3 A1
5,000
A3 becomes the final best

alternative, just as before.
11
Public Policy (1)

Should a median barrier be added to a highway?
Current situation:
% fatal accidents :
8 per 108 vehicle miles
% non-fatal accidents: 35 per fatal accident
% damaging accidents: 240 per fatal accident
% vehicle density:
10,000 per day
%interest rate:
7%
%cost of fatality:
$900,000 per person
%cost of injury:
$10,000 per person
%cost of damage:
$1,800 per damaging
accident
12
The Value of Life (1)
Essential in analyses and extremely difficult.
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
13
The Value of Life (2)

Value of Life
(2002 dollars)
Notes
1,100,000
Derived from future earnings

potential of average man.
2,900,000
Derived from Swedish road safety

data related to a willingness to pay to
improve traffic risk.
7,000,000
Derived from mortality risks related

to U.S. Government regulations.
1,100,000 to
9,700,000
Derived from analysis of jobs with

different wages and risks, , and from
direct questioning involving risk-money
tradeoffs in constructed markets.
14
Public Policy (2)
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
Public Policy (3)

Benefit-Cost Analysis
%Aggregate cost per fatal accident =
900,000 + 35x10,000 + 240x1,800 = $1,682,000
%Annual benefit per mile to the public:
(8 4)x10,000x 365x1,682,000
AE(7 )B =
= $242,572
8
10
%Annual cost per mile to the state:
0.07 x1.07 30
AE(7 )C = 1,500,000
+ 45,000 = $165,844
30
1.07 1
16
Public Policy (4)

Benefit-Cost Ratio:
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.
17
Social Discount Rate

The US Federal Government uses a 7% real
discount rate (Office of Management and Budget,
1992).
This rate approximates the pre-tax rate of return
on an investment in the private sector in recent
years.
For long periods of time, sensitivity studies are
required.
18
Regulatory Analysis (1)

Purpose:
To determine whether there is
adequate basis for imposing new
requirements.
Executive Order 12291 (President Reagan,
1981):
No actions by federal agencies should be
taken unless they result in a positive net
value to society.
19

Executive Order 12866 (President Clinton, 1993):
% EO 12991 is revoked.
%A regulatory analysis should be prepared for all
significant regulatory actions.
%These may result in a rule that may:

U
have an annual effect on the economy exceeding $100

million
U adversely affect jobs, the environment, public health
and safety
U seriously interfere with another agencys action
U raise novel legal or policy issues arising out of legal
mandates, or the Presidents priorities.
20
A regulatory analysis will not be required, if the

regulatory action is necessary to ensure that the
facility provides adequate protection to the health
and safety of the public and is in accord with the
common defense and security.
21
Value and Impact Evaluation

Values and impacts should be expressed on a
common basis, e.g., constant dollars from a
reference year.
A present-worth basis is normally used to allow
meaningful summations and comparisons.
Health effects must be expressed in monetary
terms.
Other impacts, e.g., land contamination, are
treated separately.
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.
24

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82,
ESD.72, ESD.721
DA 1. The Multistage Decision Model
Spring 2007
Why decision analysis?

A structured way for ranking decision options by:
1. Enumerating the immediate and later choices
available to the DM
2. Characterizing the relevant uncertainties
3. Quantifying the relative desirability of outcomes
4. Providing rules for ranking the decision options,
thus helping the DM to select the best one.
Value of formal analysis
Provides a systematic way to process large amounts

of information
Decision making process is explicit and enhances
communication
Provides formal rules for quantifying preferences
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
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
Manufacturing Example Payoffs
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
Building the decision tree

Decision
Options
Payoffs
L2 $300K
Payoff
depends on
market
L3 $100K
L4 -$100K
O
L1 $150K
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
Non-Probabilistic Decision Rules

Maximin Rule:
N:
Choose option with the largest smallest

payoff (risk averse DM).
-$100
O:
$150
Choose O
Maximax Rule:
N:
Choose option with the largest payoff

(risk taker).
$300
O:
$150
Choose N
Probabilistic Decision Rule

The maximin and maximax rules are incomplete
because they ignore uncertainties.
We include probabilities by taking expected values
of the payoffs (slide 31, RPRA 2).
Decision Rule: Maximize the expected monetary
value (EMV) of the earnings (payoffs).
In the decision tree, work from right to left and
compute expectations.
10
Calculation of the EMV
EMV[N] = 0.3x300 + 0.5x100 + 0.2x(-100) = $120K

EMV[O] = 1.0x150 = $150K
Option O has the largest EMV, therefore it should

be chosen.
11
Calculation of the EMV (contd)

L2 $300
s, $90
m, $50
L3 $100
EMV[N]=$120
w, -$20
L4 -$100
EMV[O]=$150
$150
L1 $150
Best Decision: O
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.
13
The survey results can be:

Strong
P(s/L2) = 0.8
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
The new decision tree

*Entries in earnings column
do not yet take account
of the cost of the survey.
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
Figure by MIT OCW.
15
New inputs
The earnings must be reduced by the survey cost of

$20K: L1 = $130K, L2 = $280K, L3 = $80K,
L4 = -$120K
The probabilities of the states of nature (the

probabilities of earnings) must also be updated to
reflect the survey findings.
16
Bayes Theorem (Slide 16, RPRA 2)

The mutually exclusive and exhaustive states are:
L2, L3, and L4
Evidence: Survey result is strong
P( L j / s ) =
P (s / L j )P ( L j )
4
P(s / L j )P( L j )
2
j = 2, 3,4
17
Calculations for survey result is s

Payoff Prior Likelihood
Prob.
L2
L3
L4
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
Results for m and w
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
The updated decision tree

s
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
Figure by MIT OCW.
20
Optimal terminal decisions

1. Solve backwards in time.
2. Determine the best solution at every terminal
node, conditional on the DMs being there.
3. Find the EMV for each terminal node and the
decision option that maximizes the EMV.
4. An arrow indicates the best decision for each
terminal node.
21
Decision tree solution

130,000
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
1. Assume that the DM makes the best decision at

each terminal node, if it is reached.
2. Find the EMV for the initial node.
3. Buy survey:
EMV = $161,008
Do not buy survey:
EMV = $150,000
4. Best initial decision:
Buy survey
23
Best terminal decision

If survey result is strong Market new product
(EMV = $221,200)
If survey result is mild
Market old product

(EMV = $130,000)
If survey result is weak Market old product

(EMV = $130,000)
24
General form of a decision tree

G1
P(H1 |C1 )
H1
C1
P(C 1 )
D1
D2
P(C 2 )
P(C n )
C2
G2
Gk
Cn
D3
Dm
GENERAL FORM OF A DECISION TREE
25
The multistage decision model

1.
Each stage consists of a decision node followed by a

chance node for each of the decision options available in
this stage.
2.
The DM must select one of the initial acts Aj.
3.
The Aj may be viewed as learning experiments

providing, at specified costs, opportunities for obtaining
partial or complete information about present
uncertainties.
4.
Following the probabilistic results a1, a2, , of the initial

decision, the DM must select the next decision B1, B2,
26
The multistage decision model (contd)

5.
Finally, a terminal chance node, b1, b2,, occurs.
6.
As a result of the initial decision Aw, its chance outcome

ax, the terminal decision By, and its chance outcome bz, the
total consequence C(Aw, ax, By, bz) is obtained.
7.
The solution proceeds sequentially backwards in time, i.e.,

we first identify the best decision at each terminal node,
then the best decision one stage earlier assuming that,
whatever chance event results from this stage, it will be
followed by the best of the available terminal decisions.
27

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82,
ESD.72, ESD.721
DA 2. The Value of Perfect Information

Spring 2007
Recall the evaluation of the survey results

(Slide 14, DA 1):
Strong
P(s/L2) = 0.8
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
Perfect Information (Clairvoyant)
A clairvoyant, CV, is always correct, i.e.,

P[CV says L2 /L2 materializes] = 1.0 = P[s/L2]
P[CV says L3 /L2 materializes] = 0.0 = P[m/L2]
P[CV says L4 /L2 materializes] = 0.0 = P[w/L2]
Receiving the CVs report removes all uncertainty.
Calculations for survey result is s or

survey says L2 (Slide 18, DA 1)
Payoff Prior Likelihood
Prob.
L2
L3
L4
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.
Bayes Theorem for the Clairvoyant

P[L2 materializes/ CV says L2] =
=
P(CVsaysL 2 / L 2materializes)xP(L 2materializes)
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.
The original decision tree

Decisions
States of Nature
Payoffs
L2 $300K
s, 0.3
m, 0.5
L3 $100K
N
w, 0.2
1.0
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.
Decision tree with a clairvoyant

L2 $300K
s, 0.3
m, 0.5
L3 $100K
N
=$150K
CV
w, 0.2
1.0
L4 -$100K
L1 $150K
=$195K
The value of alpha
EMV, if the terminal decision is to be

made with perfect information at no cost.
= 0.3x300 + 0.5x150 + 0.2x150 = $195K
The value of beta

What is the EMV without any information?
We solved this problem in DA 1 (original decision
tree).
EMV[no information] = $150K
:
EMV, if the terminal decision is to be made

without any opportunity to obtain additional
information.
Note:The chance node follows the decision node.
10
Expected Value of Perfect Information

(EVPI)
EVPI - = $195 - $150 = $45K
The EVPI is an upper bound on the amount the
DM would be willing to pay for additional
information.
The expected value of any information source must
be between zero and the EVPI. In DA 1, the cost of
the survey was $20K < EVPI.
11
Decision tree with a clairvoyant

L2 $300K
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.
The evaluation of perfect information is done a

priori. The DM has not yet consulted the
clairvoyant. The DM is considering whether to
actually do it.
13
Summary and Observations

We have developed single-attribute, multi-stage
sequential Decision Trees.
The model is useful to a single decision maker.
Decision Criterion: Maximize the EMV.
Maximizing the EMV is not the best decision
criterion.
14

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82,
ESD.72, ESD.721
DA 3. The Axioms of Rational Behavior

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:
L($5, $0; 0.6, 0.4)

0.6
$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
be indifferent between Ci and Cj

prefer Cj to Ci
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
The desirability of a lottery

It depends on:
The probabilities
The consequences
The persons present wealth, needs, and

attitude toward risk.
Axiom 1: Comparison of lotteries with

identical consequences
Given: L1 = L (C*, C* ;p1 , 1 - p1), L2 = L (C*, C*; p2 , 1 - p2)
and C*f Ci f C* for all i, then a Decision Maker will:
prefer L1 over L2
if p1 > p2,
be indifferent
if p1 = p2
prefer L2 over L1
if p1 < p2.
Given the same consequences, the DM prefers the lottery

with the higher probability of achieving the most desirable
consequence.
Axiom 2a: Quantification of preferences

For each Ci, the DM can specify a number (Ci),
with 0 (Ci) 1, such that the DM is indifferent
between:
possessing Ci with certainty
and
possessing the lottery L (C*, C*; (Ci), 1 -(Ci))
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).
Axiom 2b: Quantification of uncertainty

Let R be any event. For each R, the DM has a
quantity p(R), with 0 p(R) 1, such that the DM is
indifferent between
the lottery L (C* , C* ; p(R) , 1 - p(R))
a lottery as a result of which the DM will obtain C*
if R occurs and C* if R does not occur.
Notes on Axiom 2b
Judgmental probabilities exist for a rational DM.
This axiom provides the means for finding the

DMs probability of R. All the DM has to do is
adjust p(R) until he/she is indifferent between the
two lotteries.
10
Axiom 3: Transitivity of preferences
If C1 , C2 and C3 are consequences, then:

C1 ~ C2 and C2 ~ C3 implies C1 ~ C3
and
C1f C2 and C2 f C3 implies C1 f C3
11
Axiom 4: Substitution of consequences

If C1 ~ C2, then the DM is indifferent between two
decision problems which are identical except that
C1 in the first problem has been substituted by C2
in the second.
[If a DM is indifferent between two consequences,
the DM's solution to a decision problem cannot be
affected by substitution of one of these
consequences for the other.]
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
Axiom 5: Equivalence of preferences for

actual and conjectural situations
Let C1 and C2 be any two consequences which are possible if
only some chance event R occurs. After it is known that R
did indeed occur, the DM should have the same preference
between C1 and C2 that the DM had before (s)he knew
whether or not R occurred.
[A DM's preferences among consequences of a decision
should not be affected by knowledge as to whether (s)he
merely may or (s)he certainly will have to make that
decision.]
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
15
Summary of Axioms (contd)

Axiom 4: Substitution of consequences
Axiom 5: Equivalence of preferences for actual
and conjectural situations
A DM who satisfies these axioms is rational or

coherent.
16

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82,
ESD.72, ESD.721
DA 4. Introduction to Utility
Spring 2007
The Concept of utility

Utility of a consequence: A quantification of a
person's relative preference for that consequence
A simple extension of the indifference probability
(or "preference value") concept (Axiom 2a)
Utility function: Expresses a person's relative
preferences among a set of consequences (often
defined over a continuous range)
Axiom 2a: Quantification of preferences

For each Ci, the DM can specify a number (Ci),
with 0 (Ci) 1, such that the DM is indifferent
between
possessing Ci with certainty
and
possessing the lottery L (C*, C*; (Ci), 1 -(Ci))
Schematic representation
C1
p1
p2
p1
C2
p2
~
pm
pm
Cm
( C1 )
C*
1( C1 )
( C2 )
C*
1( C2 )
C*
( Cm )
C*
1( Cm )
C*
C*
Preference measures for lotteries
Having established the relative preferences, i.e., the (Ci ), i = 1,,m,

we can derive the relative preferences for lotteries.
Recall Axiom 1:
L1 = L (C*, C*; p1 , 1 - p1)
Then:
L1
f L2
L1 p L2
Given
and
L2 = L (C*, C*; p2 , 1 - p2)
if p1 > p2
if p1 < p2;
L2 ~ L1
if p1 = p2
The probabilities p1 and p2 are the preference measures of the two

lotteries.
Schematic representation (contd)
p1
p2
( C1 )
C*
1( C1 )
( C2 )
C*
1( C2 )
C*
P = p 1.(C 1 ) + ... + p m .(C m )
C*
C*
1P
pm
( Cm )
C*
1( Cm )
C*
C*
P=
p i.(C i) expected value of preference values

i=1
of possible outcomes of L
Preference value of a lottery

The DM is indifferent between the original lottery on
slide 4 and the last lottery on slide 6. The preference
value of the latter is
m
P = p i ( C i )
i =1
Conclusion:The preference value for a lottery is

the expectation of the preference values of the
possible outcomes of the lottery.
Utility
The indifference probability or any positive linear
transformation of the form
U(x) = a (x) + b
a>0
is said to be a utility function over the set of

consequences.
Convention:
x = 0 represents the present assets

of the DM.
Certainty Equivalent of a Lottery
The certainty equivalent (CE) of a lottery L is a

consequence such that the DM is indifferent
between the certainty equivalent and the lottery.
[Note: U(CE) = U(L)]
Risk Premium for a Lottery

The risk premium (RP) for a single-attribute lottery is
given by
RP = EMV - CE
For risk-averse individuals, RP is a positive quantity.
Often, in practice, RP also decreases as additional wealth is
acquired.
Example:
If a DM has a CE of $40 for the lottery
L($100, $0; 0.5, 0.5), then his RP is $50 - $40 = $10.
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
11
12
Marketing a new product
L2 $300K
s, 0.3
m, 0.5
L3 $100K
N
w, 0.2
L4 -$100K
O
1.0
L1 $150K
13
Decision tree with utilities

L2 1.0
0.30
0.50
L3 0.87
0.855
N
0.20
O
0.92
1.00
L4 0.60
L1 0.92
14
Choice of decision option

If the DM chooses to market the new product,
(s)he is choosing a lottery with preference value
0.855.
If the DM chooses to market the old product, (s)he
is choosing a lottery with preference value 0.920.
Since 0.920 > 0.855, the DM should market the
old product.
15
Decision Rule: Maximization of expected utility

Each possible consequence is replaced by its
utility.
Each decision option is a lottery whose preference
value is the expected utility of its outcomes.
Choose the option with the highest preference
value.
This is the result of accepting the axioms (unlike
the maximization of the EMV).
16
The survey problem

O
.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
Figure by MIT OCW.
17
Best Decision
Do not buy the survey and keep marketing the old

product.
The best decision using EMV was to buy the
survey and act according to its results, i.e., if the
result is strong, then market the new product,
otherwise market the old product.
18

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82,
ESD.72, ESD.721
DA 5. Risk Aversion
Spring 2007
DA 5. Risk Aversion
Calibration of utility functions

We can apply a positive linear transformation to a
utility function and get an equivalent utility
function.
(x) = a U(x) + b
a>0
A calibrated utility function is such that

(C*) = 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
-4.5 x 4.5 (in $ million)

C = 2
Let (x) = aln(x+5) + b.
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
The Buying Price for a Lottery

It is the purchase price at which the DM is indifferent
between the alternatives of buying the lottery and not
buying it.
Let (x) the DMs utility function with (0) the utility of his
present assets, i.e., before he buys the lottery.
p1
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
L(1, 0; 0.5, 0.5)
0.5 (1-BP) + 0.5 (0-BP) = 0.61
1.18 ln(6-BP) 1.29 + 1.18 ln(5-BP) 1.29 = 1.22

ln(6-BP) + ln(5-BP) = 3.22
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
(the other root is 10.53 and is rejected)
EMV = 1x0.5 + 0x0.5 = 0.5 > 0.47 risk aversion
DA 5. Risk Aversion
The Selling Price of a Lottery

SP = CE
(certainty equivalent)
We now interpret x = 0 to represent the DMs total assets

except of the lottery.
Utility of present wealth (situation) is (L).
(x) = 1.18 ln(x+5) 1.29
(0) = 0.6091
DA 5. Risk Aversion
Example
(x) = 1.18 ln(x+5) 1.29
(0) = 0.6091
L(1, 0; 0.5, 0.5)

(L) = 0.5 (1) + 0.5 (0) = 0.5x0.8243 + 0.5x0.6091
(L) = 0.7167 = (CE)
Using the figure on p. 4 we get
CE 0.4771 < EMV = 0.5
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
U(p1 x1 + p2 x2) = U(EMV) > p1 U(x1)+ p2 U(x2)

A DM with a concave utility for money will refuse a
monetarily fair bet and is said to be risk averse.
DA 5. Risk Aversion
10
Example
L(1, 0; 0.5, 0.5)
EMV = $0.5 M
0.5 (1) + 0.5 (0) = 0.7167
(0.5) = 0.6091
SP = $0.4771
Risk Premium EMV CE = 0.5 0.4771

RP = $0.0229 M
DA 5. Risk Aversion
11
Risk Aversion and Insurance

You own a house worth $500,000.
A fire (p = 10-2 per year) may destroy it
completely.
Your utility function is
(x) = 1.18ln(x+5) 1.29;
C* = $2M, C* = -$2M
How much premium would you be willing to pay?

DA 5. Risk Aversion
12
Decision Tree
p
-x
I
1-p
NI
-x
-$500,000
1-p
0.0
DA 5. Risk Aversion
13
Utilities
(I) = (-x)
(NI) = p (-0.5) + (1- p) (0.0)
(-0.5) = 1.18ln(4.5) 1.29 = 0.4848

(0.0) = 0.6091
(NI) = 0.4848 x 10-2 + (1- 10-2) x 0.6091 =
0.6078
(-x) = 0.6078 x = $5,661
DA 5. Risk Aversion
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
You are willing to pay more than the expected

loss because you are risk averse.
DA 5. Risk Aversion
15
The Companys Perspective
Why would the insurance company agree to

insure you?
The company may win $5,661 with
probability 0.99 or lose $494,339 with
probability 10-2.
DA 5. Risk Aversion
16
The Companys Decision Tree

p
- 494,339
Accept I
5,661
1-p
Do not
Accept I
0.00
1-p
0.00
DA 5. Risk Aversion
17
The Companys Perspective (2)

The company has $1 billion dollars in assets.
Its wealth is either $999,505,661 with
probability 10-2, or $1,000,005,661 with
probability 0.99.
For such small changes, the companys utility
of money is linear, i.e., the company makes
decisions using the EMV.
EMV = 5,661 x 0.99 - 494,339 x 10-2 = $661
DA 5. Risk Aversion
18
The Companys Perspective (3)

The alternative is for the company to refuse
the premium, in which case the EMV is zero.
The company should agree to insure the
house.
Note:
The companys overhead expenses have
not been factored in.
DA 5. Risk Aversion
19
Assessment Using Certainty Equivalents

1.
Set
U(C*) = 0
and U(C*) = 1
2.
Consider the reference lottery

L1(C*, C*; 0.5, 0.5)
Its certainty equivalent is derived by solving

U(CE1) = 0.5 U(C*) + 0.5 U(C*) = 0.5
DA 5. Risk Aversion
20
Assessment Using Certainty Equivalents (2)

Thus, we have found a third point of the utility
function, that with utility 0.5.
3.
Repeat the process with a new reference lottery

L2(C*, CE1; 0.5, 0.5)
Solve
U(CE2) = 0.5 U(C*) + 0.5 U(CE1) = 0.75
to get a fourth point of the utility function, CE2.
DA 5. Risk Aversion
21
Assessment Using Certainty Equivalents (3)

4.
Repeat using L3(CE1, C*; 0.5, 0.5)
U(CE3) = 0.5 U(CE1) + 0.5 U(C*) = 0.25
to get a fifth point, and so on.
DA 5. Risk Aversion
22

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

Spring 2007
Consequences
Not all the consequences are monetary.

In risk management problems, for example, they
may include the impact on health and safety of
groups of stakeholders.
In general, the consequences are described by a
vector x (x1, , xN).
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
Finding u(x1, , xN)

Use the certainty-equivalent approach:
p
1-p
( x*1 ,..., x*N )

( x1* ,..., x N* )
( 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.
Mutual Preferential Independence

Attribute Y is preferentially independent of
attribute Z, if preferences for y levels do not
depend on the level of z, i.e.,
(y, z 0 ) f (y' , z 0 )
implies
(y, z ) f (y' , z )
where y and y are two levels of y.

Mutual Preferential Independence:

Example
Y: Departure time (morning, afternoon)
Z: Ticket cost ($300, $500)
If you prefer afternoon to morning departure
regardless of the price of the ticket, and you prefer $300 to
$500 regardless of the departure time, then Y and Z are
mutually preferentially independent.
If you prefer afternoon to morning departure
regardless of the price of the ticket, but the price depends on
when you leave, then Y is preferentially independent of Z,
but they are not mutually preferentially independent.
Utility Independence (1)

It is similar to preferential independence, except
that the assessments are made with uncertainty
present. It is a stronger assumption.
Y is utility independent of Z if preferences over
lotteries involving different levels of Y do not
depend on a fixed level of Z.
For the previous example: The preference value of
the lottery L(morning, afternoon; 0.5, 0.5) is
independent of the price of the ticket.
The CEs of lotteries on Y levels are independent of
z.

A form of the utility function for attributes
X1 and X2 that are utility independent, is
u(x1, x2) =
= k1 u1(x1) + k2 u2(x2) + (1- k1- k2) u1(x1) u2(x2)
with
0 ki 1
i = 1,2
0 ui ( xi ) 1
i = 1,2

Fix the level of X2 at x'2 , then
u(x1, x'2 ) =
=k1 u1(x1) + k2 u2( x'2 ) + (1- k1- k2) u1(x1) u2( x'2)
'
'
=[k1+ (1- k1- k2) u2( x 2 )] u1(x1) + k2 u2( x 2 )
This is a linear transformation of u1(x1), therefore,
the preferences over levels of X1 are independent of
the level of X2.
For another level of X2, we will get another linear
transformation of u1(x1).
Lotteries on X1 are independent of the level of X2.
10
When X1 and X2 are utility independent of each

other, they are mutually utility independent.
u(x1, x2) = g(x1) + h(x1)u2(x2)

X2 is utility independent of X1 but not vice versa.
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 )
x1' and x1 are different levels of x1

We can get any pair of consequences with probability 0.5; the only
difference is how the levels are combined.
12
Additive Utility Function

N
u( x ) = k i u i ( x i )
1
where
0 ki 1
N
ki = 1
1
0 ui ( xi ) 1
Two attributes:
u(x1, x2) = k u1(x1) + (1-k) u2(x2)

13
Additive Independence: Implications

When we assess the utility of one attribute, it
should not matter what the other attributes level
is.
Interaction among the attributes is not allowed.
For cases with no or little uncertainty, additive
independence represents reasonably well peoples
utilities.
For complex problems, it could be a useful first-cut
approximation.
14

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82, ESD.72
DA 7. Decision Analysis in Practice

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.
Involving the Stakeholders
Risk assessment can and should be used to

involve stakeholders and provide a mechanism for
the consideration of their cultural, socioeconomic,
historical, and religious values, in addition to the
risks to human health and the environment
associated with the contamination of DOE
facilities and their remediation.
National Research Council, Building Consensus, 1994
The Analytic-Deliberative Process

Analysis uses rigorous, replicable methods, evaluated
under the agreed protocols of an expert community - such
as those of disciplines in the natural, social, or decision
sciences, as well as mathematics, logic, and law - to arrive
at answers to factual questions.
Deliberation is any formal or informal process for
communication and collective consideration of issues.
National Research Council, Understanding Risk, 1996.
Analytic-Deliberative Decision Making

S te p 1
F o r m u la te d e cisio n
a lter n a tiv e s
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
Image removed due to copyright restrictions.

Please see Table 1 in Apostolakis, George E.,
and Pickett, Susan E. Deliberation: Integrating
Analytical Results into Environmental
Decisions Involving Multiple Stakeholders.
Risk Analysis 18 (1998): 621-634.
Decision Options

Risk Analysis 18 (1998): 621-634.
Fundamental and Means Objectives

A distinction is made between
those things that participants fundamentally care
about, such as environmental quality, and those
that matter only through their effect on these
fundamental concerns, such as waste disposal*
*Gregory & Keeney, Creating policy alternatives using stakeholder

values, Management Science, 40: 1035-1048, 1994.
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
The distinction is important. We should consider

only fundamental objectives (things we really care
about).
10
Structuring the Fundamental Objectives

The set of fundamental objectives should be:
Complete
As small as possible
Not redundant
Fundamental objectives can be organized into a
hierarchy in which the lower levels explain what is
meant by the higher levels.
Clemen, R.T., Making Hard Decisions, 2nd Edition, Belmont, California:
Duxbury Press, 1996
Sometimes this hierarchy is called a value tree.

11
The Value Tree: Impact Categories

Please see Fig. 5 in Bonano, E. J., et al.
Application of risk assessment and decision
analysis to the evaluation, ranking, and
selection of environmental remediation
alternatives. Journal of Hazardous Materials
71 (2000): 35-57.
12
The Value Tree: Objectives

71 (2000): 35-57.
13
The Value Tree: Performance Measures

(Attributes)

71 (2000): 35-57.
14
Stakeholder Changes
Most agreed with the tree on slide 14.

Some stakeholders placed long-term public
risks under the category environment.
15
Performance Measure Units and Ranges

Risk Analysis 18 (1998): 621-634.
The decision is not made in general but for the specific problem.
16
Additive Independence (1)

The Performance Measures are assumed to be
additive independent, so that the expected utility
(Performance Index) of the jth decision option is
N
PI j = wi uij
PM
i =1
where
NPM
wi = 1
1
17
Additive Independence (2)

When we assess the utility of one attribute, it should not
matter what the other attributes level is.
Interaction among the attributes is not allowed.
For cases with no or little uncertainty, additive
independence represents reasonably well peoples utilities.
For complex problems, it could be a useful first-cut
approximation.
Even if used only as an approximation, the additive utility
function takes us a long way toward understanding our
preferences and resolving a difficult situation. (Clemen)
Deliberation will follow the analysis.
18
The Weights
Recall that
N
PI j = wi uij
PM
i =1
The weights are scaling factors that sum to unity

NPM
wi = 1
1
They represent trade-offs between PMs. They can

be assessed directly or using structured
approaches. (Clemen)
19
The Analytic Hierarchy Process

Relative rankings of the objectives are
determined with respect to an overall goal
Pairwise comparisons are used to derive

weights representative of decision maker
concerns
T.L. Saaty, Fundamentals of Decision Making and Priority Theory with
The Analytic Hierarchy Process, RWS Publications, Pittsburgh, 2000.
20
An Example of Pairwise Comparisons

Relative Importance Assessment
Objective categories:
Compare the following with respect to the OVERALL DESIRABILITY objective
1.
Socioeconomic / Cultural vs. Life Cycle Cost
_______
2.
Programmatic vs. Environment
_______
3.
Life Cycle Cost vs. Human Health & Safety
_______
4.
Environment vs. Human Health & Safety
_______
5.
Environment vs. Life Cycle Cost
_______
6.
Socioeconomic / Cultural vs. Environment
_______
7.
Programmatic vs. Life Cycle Cost
_______
8.
Human Health & Safety vs. Socioeconomic / Cultural
_______
9.
Programmatic vs. Socioeconomic / Cultural
_______
10.
Human Health & Safety vs. Programmatic
_______
Key for the evaluation:

1 equally
3 weakly
5 strongly
7 demonstrably or very strongly
9 absolutely
Use even numbers to express compromise.
21
The Practice
People are not consistent in their assessments.
Redundant information is elicited.
Define the consistency index as
max n
CI =
n 1
If CI > 0.2, identify inconsistencies and inform

the assessor.
The assessor always approves the final weights.
The CI is for internal consistency only, not for
consistency among stakeholders.
22
STAKEHOLDER RANKINGS AND WEIGHTS

Category
Stakeholder
Programmatic Life Cycle

Cost
Socioeconomic
Cultural
Environment Human Health

& Safety
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)
Figure by MIT OCW.
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
u
0.6
0.3
0.1
24
Consistency (sanity or reality) Checks

Recall again that
PI j = w 1 u1 j + ... + w i u ij + ... + w k ukj + ... + w N u N
PM
PM
Suppose that for two PMs i and k the levels l

and m on the constructed scales are such that
m
=
wi u wk ukj
l
ij
Then, the decision maker should be indifferent

between these two levels.
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
Transport Through Aquifer at Retarded Velocities
Figure by MIT OCW.
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)
Performance Indices and RAA rankings for all stakeholders.

Figure by MIT OCW.
28
Preparing for Stakeholder Deliberation

Each stakeholder receives his/her numerical
results in advance.
The dominant drivers for the decision
choices are also reported. This concept is
borrowed from risk assessment.
Preliminary conclusions regarding all
stakeholders are drawn.
29
Stakeholder 3
0.2
0.18
Performance Index
0.16
0.14
0.12
0.1
0.08
0.06
F
Remedial Action Alternative

Figure by MIT OCW.
30

Risk Analysis 18 (1998): 621-634.
31

Risk Analysis 18 (1998): 621-634.
32
Deliberation

Please see Section 6.1 in Apostolakis, George
E., and Pickett, Susan E. Deliberation:
Integrating Analytical Results into
Environmental Decisions Involving Multiple
Stakeholders. Risk Analysis 18 (1998): 621634.
33

Please see Section 6.2 in Apostolakis, George
E., and Pickett, Susan E. Deliberation:
Integrating Analytical Results into
Environmental Decisions Involving Multiple
Stakeholders. Risk Analysis 18 (1998): 621634.
34

Risk Analysis 18 (1998): 621-634.
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.
37

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82
ESD.72J, ESD.721
Introduction
Spring 2007
Introduction
Risk-Benefit Tradeoffs
Introduction
Dealing with Uncertainty

Risk management of large technological
systems
Choosing among alternative decision options
Cost-benefit tradeoffs
Introduction
Sometimes the historical record is surprising

Typical Metro Railway Accidents
Staircases
Slips/trips
Train Accidents
Train Doors
Falls between train and

platform
Detrain
Falls inside train
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
strengths and weaknesses
Introduction

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82
ESD.72J, ESD.721
RPRA 1.
The Logic of Certainty
Spring 2007
RPRA 1. The Logic of Certainty
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.
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, }
Examples: Die {1,2,3,4,5,6}
A collection of sample points is an event.

S
Venn Diagram
Indicator Variables
1,If
is T
If E
is F
Xj =
0,
Important Note: Xk = X,
k: 1, 2,
S
___
Venn Diagram
Union (OR operation)
A B = C
X C = 1 (1 X A )(1 X B )
X
C
A
A
B
6
Intersection (AND operation)
A B = C
XC = X AX B
XC X j
C
Mutually Exclusive Events:
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
...
Reliability Block Diagram

for the Parallel System
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
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 )]}
Expanding and using Xk = X we get
X T = 1 (1 X A X B )(1 X B X C )(1 X C X A )
12
Cut sets and minimal cut sets
CUT SET: Any set of events (failures of components and

human actions) that cause system failure.
MINIMAL CUT SET: A cut set that does not contain

another cut set as a subset.
13
New fault tree:

S y s te m
F a ilu r e
Minimal cut sets:
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)
14
XT = (X1, X2,Xn) (X)

(X) is the structure or switching function.
It maps an n-dimensional vector of 0s and 1s onto 0 or 1.
Disjunctive Normal Form:
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
For the 2-out-of-3 System:

XT=1-(1-XAXB) (1-XBXC) (1-XCXA)
XT = (M1+M2+M3) - (M1M2+M2M3+M3M1) + M1M2M3
But,
M1M2 = XAXB2XC = XAXBXC
Therefore, the sum-of-products expression is:
XT = (XAXB+XBXC+XCXA) - 2XAXBXC
16
The Bridge Network

A
3
5
{X1X2}, {X3X4}, {X2X3X5}, {X1X4X5}

Disjunctive Normal Form:
XT=1-(1-X1X2)(1-X3X4)(1-X2X3X5)(1-X1X4X5)
Sum-of-Products Form:
XT = X1X2+ X3X4+ X2X3X5+ X1X4X5- X1X2 X3X4- X1X2X3X5- X1X2X4X5 -X2X3X4X5 - X1X3X4X5 + 2X1X2X3X4X5
17
Causes of Failure
1.
2.
3.
Primary failure ("hardware" failure)

Secondary failure (external, environmental)
"Command" failure (no input; no power)
N o O u tp u t fro m
C om ponent
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
Reliability Block Diagram for the Fuel-Supply

System
T1
Fuel
Source
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
Fault tree elements

TOP EVENT
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
Note: Its helpful to start the fault-tree development from the

output of the system (the top event) and work backwards.
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
A simpler fault tree

No Fuel is
Delivered
When Needed
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
Corrosion Induced Failure

Earthquake
Induced
Failure
M issile
Im pact
Induced
Failure
Internal
Fire/Explosion
Induced
Failure
Corrosion
Fatique
Induced
Failure
Faulty
M anufacture
& Control
Program
23
System min cut sets

T1, Tank
P1, Pump and of
V1, Valve
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
Examples of Initiating Events

Loss of Coolant
Transients
Human Error
Loss of Power
Fires
Airplane Crashes
Earthquakes
26

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82,
ESD.72, ESD.721
RPRA 2.
Elements of Probability Theory
Spring 2007
RPRA 2. Elements of Probability Theory
Probability: Axiomatic Formulation

The probability of an event A is a number that satisfies the
following axioms (Kolmogorov):
0 P(A) 1
P(certain event) = 1
For two mutually exclusive events A and B:
P(A or B) = P(A) + P(B)
Relative-frequency interpretation
Imagine a large number n of repetitions of the
experiment of which A is a possible outcome.
If A occurs k times, then its relative frequency is:
It is postulated that:
k
n
k
lim P( A )
n n
Degree-of-belief (Bayesian) interpretation

No need for identical trials.
The concept of likelihood is primitive, i.e., it is meaningful
to compare the likelihood of two events.
P(A) < P(B) simply means that the assessor judges B to be
more likely than A.
Subjective probabilities must be coherent, i.e., must satisfy
the mathematical theory of probability and must be
consistent with the assessors knowledge and beliefs.
Basic rules of probability: Negation

S
EE = S
___
Venn Diagram
P(E) + P(E) = P(S) = 1
P(E) = 1 P(E)
Basic rules of probability: Union

N 1 N
N N
P U A i = P (A i ) P (A i A j ) +
1
i =1
i =1 j= i + 1
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
Union (contd)
For two events: P(AB) = P(A) + P(B) P(AB)
For mutually exclusive events:

P(AB) = P(A) + P(B)
Example: Fair Die

Sample Space:
{1, 2, 3, 4, 5, 6}
(discrete)
Fair: The outcomes are equally likely (1/6).

P(even) = P(2 4 6) = (mutually exclusive)
Union of minimal cut sets

From RPRA 1, slide 15
N 1 N
X T = M i M i M j + ... + (1)
i =1
P( X T ) =
i =1 j= i +1
N 1 N
N +1
P(Mi ) P(MiM j ) + ... + ( 1)

i =1
N +1
i =1 j = i +1
Mi
i =1
P Mi
i=1
Rare-event approximation:
P(XT ) P (Mi )
1
Upper and lower bounds

P( X T ) =
N1 N
N+1 N
P(Mi ) P(Mi M j ) + ... + (1)
i =1
i =1 j=i +1
P(XT ) P (Mi )
1
P( X T ) P( M i )
i =1
P(Mi )
i =1
The first term, i.e., sum,

gives an upper bound.
N 1 N
P( M i M j )
i =1 j= i +1
The first two terms, give

a lower bound.
10
Conditional probability
P(AB )
P(A B )
P(B )
P(AB ) = P(A B )P(B ) = P(B / A )P(A )
For independent events:
P(AB) = P(A )P(B )

P(B / A ) = P(B )
Learning that A is true has no impact on our probability of B.

11
Example: 2-out-of-4 System

1
2
M 1 = X1 X 2 X 3
M2 = X2 X3 X4
M 3 = X3 X 4 X 1
M4 = X1 X2 X4
XT = 1 (1 M1) (1 M2) (1 M3) (1 M4)
XT = (X1 X2 X3 + X2 X3 X4 + X3 X4 X1 + X1 X2 X4) - 3X1 X2 X3X4

12
2-out-of-4 System (contd)

P(XT = 1) = P(X1 X2 X3 + X2 X3 X4 + X3 X4 X1 + X1 X2 X4)
3P(X1 X2 X3X4)
Assume that the components are independent and nominally
identical with failure probability q. Then,
P(XT = 1) = 4q3 3q4
Rare-event approximation:
P(XT = 1) 4q3
13
Updating probabilities (1)
The events, Hi, i = 1...N, are mutually exclusive and exhaustive, i.e.,
HiHj= , for ij, Hi = S, the sample space.
Their probabilities are P(Hi).
H3
H1
Given an event E, we can always write
E
H2
P( E) = P( E / H i )P(H i )
1
14
Updating probabilities (2)

Evidence E becomes available.
What are the new (updated) probabilities P(Hi/E)?
Start with the definition of conditional probabilities,
slide 11.
P ( EH i ) = P ( E / H i ) P ( H i ) = P ( H i / E ) P ( E )
P(H i / E) =
P ( E / H i )P ( H i )
P(E)
Using the expression on slide 14 for P(E), we get

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
Example: Lets Make A Deal
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.
What do you do?
17
Lets Make A Deal: Solution (1)
Setting up the problem in mathematical terms:

A = {The Ferrari is behind Door A}
B = {The Ferrari is behind Door B}
C = {The Ferrari is behind Door C}
The events A, B, C are mutually exclusive and exhaustive.
P(A) = P(B) = P(C) = 1/3
E = {The host opens door B and a Yugo is behind it}

What is P(A/E)?
Bayes Theorem
18

P (A E ) =
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
(A Yugo is behind door B).
P(E/C) = 1
(The host must open door B, if the

Ferrari is behind door C; he
cannot open door A under
any circumstances).
19

Let P(A/E) = x and P(E/A) = p
Bayes' theorem gives:
p
x=
1+ p
Therefore
For P(E/A) = p = 1/2 (the host opens door B randomly, if

the Ferrari is behind door A)
P(A/E) = x = 1/3 = P(A) (the evidence has had no
impact)
20

P(A/E) + P(C/E) = 1
Since
P(C/E) = 1 - P(A/E) = 2/3
The player should switch to door C
For P(E/A) = p = 1 (the host always opens door B, if the

Ferrari is behind door A)
P(A/E) = 1/2
offer any advantage.
P(C/E) = 1/2, switching to door C does not
21
Random Variables
Sample Space: The set of all possible outcomes of an experiment.
Random Variable: A function that maps sample points onto the real
line.
Example:
For the coin: S = {H, T} {0, 1}
For a die
S = {1,2,3,4,5,6}
22
Events
3.6
-
4 5
We say that {X x} is an event, where x is any number on

the real line.
For example (die experiment):
{X 3.6} = {1, 2, 3} {1 or 2 or 3}
{X 96} = S
(the certain event)
{X -62} =
(the impossible event)
23
Sample Spaces
The SS for the die is an example of a discrete sample space and X is a discrete
random variable (DRV).
A SS is discrete if it has a finite or countably infinite number of sample points.
A SS is continuous if it has an infinite (and uncountable) number of sample

points. The corresponding RV is a continuous random variable (CRV).
Example:
{T t} = {failure occurs before t}
24
Cumulative Distribution Function (CDF)
The cumulative distribution function (CDF) is

F(x) Pr[X x]
This is true for both DRV and CRV.
Properties:
1. F(x) is a non-decreasing function of x.
2. F(-) = 0
3. F() = 1
25
CDF for the Die Experiment
F(x)
1
1/
6
1
26
Probability Mass Function (pmf)
For DRV: probability mass

function
P(X = x i ) p i
F(x ) = p i , for all

P( S ) = p i = 1
xi x
normalization
Example: For the die, pi = 1/6 and p i = 1

2
F ( 2 .3 ) = P ( 1 2 ) = p i =
1
1 1 1
+ =
6 6 3
27
Probability Density Function (pdf)

f(x)dx = P{x < X < x+dx}
dF (x )
f (x ) =
dx
P( S ) = F( ) =
F( x ) = f (s )ds
f (s)ds = 1
normalization
28
Example of a pdf (1)

Determine k so that
f (x ) = kx 2 ,
for
f (x ) = 0,
0x1
otherwise
is a pdf.
Answer:
The normalization condition gives:

1
2
kx
dx = 1
k=3
29
Example of a pdf (2)

F( x) = x 3
F(x)
1
0.875
F(0.875) F(0.75) = 3x 2dx =
0.67
0.75
0.42
= 0.67 - 0.42 = 0.25 =
1
f(x)
3
= P{0.75 < X < 0.875}
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
Variance (standard deviation )
2
(
)
x
m
f (x )dx
2
2
E (X m ) =
(x j m )2 p j
j
CRV
DRV
31
Percentiles
Median: The value xm for which
F(xm) = 0.50
For CRV we define the 100 percentile as

that value of x for which
f (x )dx =
32
Example
1
m = 3x 3 dx = 0.75
0
= 3(x 0.75 ) x 2 dx = 0.0375

2
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

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82,
ESD.72, ESD.721
RPRA 3.
Probability Distributions in RPRA
Spring 2007
RPRA 3. Probability Distributions in RPRA
Overview
We need models for:
The probability that a component will start (fail) on demand.
The probability that a component will run for a period of time given a
successful start.
The impact of repair on these probabilities.
The frequency of initiating events.
Failure to start
P[failure to start on demand] q unavailability

P[successful start on demand] p availability
Requirement: p + q = 1
The Binomial Distribution (1)
Start with an experiment that can have only two outcomes: success
and failure or {0, 1} with probabilities p and q, respectively.
Consider N "trials," i.e., repetitions of this experiment with constant q.

These are called Bernoulli trials.
Define a new DRV: X = number of 1's in N trials
Sample space of X: {0,1,2,...,N}
What is the probability that there will be k 1s (failures) in N trials?

For the coin: Assume 3 trials. We are interested in 1 failure.
There are 3 such sequences: fss, sfs, ssf (mutually exclusive).
The probability of each is qp2.
If order is unimportant, the probability of 1 failure in 3 trials is 3qp2.
N k
Pr[ X = k ] = q (1 q )N k
k
This is the probability mass function of the Binomial Distribution.

It is the probability of exactly k failures in N demands.
The binomial coefficient is:
N
N!

k k! (N k )!

Mean number of failures: qN
Variance: q(1-q)N
Normalization:
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
Example: 2-out-of-3 system
We found in slide 16 of RPRA 1 that the structure function is (using min

cut sets):
The failure probability is P(failure) = P(XT=1) = 3q2 2q3
Using the binomial distribution:
Pr(system failure) = P[2 fail] + P[3 fail] = 3q2(1-q) + q3

= 3q2 2q3
Notes: 1. We have assumed nominally identical and independent

components in both cases.
2. If the components are not nominally identical or independent,
the structure function approach still works, but the binomial
distribution is not applicable. Why?
The Poisson Distribution

Used typically to model the occurrence of initiating events.
DRV: number of events in (0, t)
Rate is constant; the events are independent.
The probability of exactly k events in (0, t) is (pmf):
k
Pr[k ] = e
k! 1*2**(k-1)*k
( t )
0! = 1
k!
m = t
= t
8
Example of the Poisson Distribution
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
Pr[k2] = 0.37 + 0.37 + 0.185 = 0.925

Failure while running

T: the time to failure of a component.
F(t) = P[T < t]: failure distribution (unreliability)
R(t) 1-F(t) = P[t < T]: reliability
m: mean time to failure (MTTF)
f(t): failure density, f(t)dt = P{failure occurs between t and
t+dt} = P [t < T < t+dt]
10
The Hazard Function or Failure Rate

f (t )
f (t )
h(t )
=
R (t ) 1 F( t )
t
F(t ) = 1 exp h(s )ds .
The distinction between h(t) and f(t) :

f(t)dt: unconditional probability of failure in (t, t +dt),
f(t)dt = P [t < T < t+dt]
h(t)dt: conditional probability of failure in (t, t +dt) given that
the component has survived up to t.
h(t)dt = P [t < T < t+dt/{ t < T}]
11
The Bathtub Curve

h(t)
I
0
II
t1
III
t2
Infant Mortality
II
Useful Life
III
Aging (Wear-out)
12
The Exponential Distribution

f ( t ) = e t
F( t ) = 1 e
>0 t>0
(failure density)
R(t ) = e
h( t ) = constant (no memory; the only pdf with

this property) useful life on bathtub curve
F( t ) t
when t < 0.1
(another rare-event approximation)
Mean Time Between Failures :
m=
1
=
13
Example: 2-out-of-3 system

Each sensor has a MTTF equal to 2,000 hours. What is the
unreliability of the system for a period of 720 hours?
Step 1:
System Logic.
14
Example: 2-out-of-3 system (2)

Step 2: Probabilistic Analysis.
For nominally identical components:

P(XT) = 3q2 2q3
(slide 7 of this lecture)
But
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
Example: 2-out-of-3 system (3)

For a failure rate of 5x10-4 hr-1 and for t = 720 hrs
t = 0.36
q( t ) = 1 e 0.36 = 0.30
FT(720) = 3 x 0.302 - 2 x 0.303 = 0.216

Since
= 0.36 > 0.1
the rare-event approximation does not

apply.
Indeed,
FT(720) 3x0.362 2x0.363 = 0.295 > 0.216
16
A note on the calculation of the MTTF
MTTF = R (t )dt
0
Proof
dR
)dt = tdR =
MTTF = tf (t )dt = t(
dt
0
0
0
= tR + R(t )dt = R(t )dt

17
A note on the calculation of the MTTF

(cont.)
since
dF d(1 R )
dR
=
f (t ) =
=
dt
dt
dt
and
R(t ) 0
faster
than
18
MTTF Examples: Single Exponential

Component
R(t) = exp(-t)
MTTF = e
0
1
dt =
19
MTTF Examples: The Series System

M
Step 1: System Logic
YT = Yj
1
(minimal path sets)
Step 2: Probabilistic Analysis

p( t ) = e t P ( T > t )
P(YT = 1) = pM
but
P( YT = 1) = R S ( t ) = e ( M )t
MTBF =
The system is exponential
1
1
M system
20
MTTF Examples: 1-out-of-2 System

Step 1: System Logic XT = X1 X2 (slide 9 of RPRA 1)
Step 2: Probabilistic Analysis
Fs ( t ) = (1 e t ) 2
R s ( t ) = 1 Fs ( t ) = 2e t e 2t
2 1
3
=
MTTF =
2 2
(compare with the MTTF for

1
a single component, )
21
MTTF Examples: 2-out-of-3 System

Using the result for FT(t) on slide 15, we get
MTTF = R T (t )dt = [1 3(1 e t )2 + 2(1 e t )3 ]dt

1
1
5
+
=
MTTF =
2 3 6
1
The MTTF for a single exponential component is:
The 2-out-of-3 system is slightly worse.

22
The Weibull failure model

Weibull Hazard Rate Curves
Adjusting the value of b,

we can model any part of
the bathtub curve.
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
Diesel does not start

q = 0.05
b = 0.004 yr
p = 0.95
ok = 0.076 yr
-1
-1
24
A Simple Calculation (cont.)

b = 0.08 (power losses per year)x 0.05 (Diesel fails given a power loss) = 4x10
-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
25
The Normal (Gaussian) distribution
f (x) =
1
(x )
exp[
]
2
2
2
< x<
0<<
Mean:
< <
Standard Deviation:
26
The Normal (Gaussian) distribution (2)

Standard Normal Variable:
Standard Normal
Distribution:
X
Z=
1
z2
( z ) =
exp[ ]
2
2
27
Area under the Standard Normal Curve

(from 0 to a)
a
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
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
Example of the normal distribution
= 10,000 hr (MTTF)
= 1,000 hr
Pr [X > 11,000 hr] = Pr [Z > 1] = 0.50 0.34= 0.16
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
Equivalently, the capacitors failure time follows an exponential

distribution with the above rate.
The mean time between failures of the capacitor is
MTBF =
1
2.28x10
= 43.86
months
32
An Example (4)
ii. Find the capacitors reliability for a time period of three

months.
R(3 mos) = exp(- d 3) = exp(-2.28 10-2x3) = 0.934
33
Observations
Events (shocks) occur in time according to the Poisson distribution

[the losses of electric power on slide 24, the surge voltages on slide 30].
There is a conditional probability that a given shock will be lethal,

i.e., will fail the component. This conditional probability was given on
slide 24 as 0.05, while on slide 31 it was calculated from reliability
physics, i.e., from the normal distribution of the voltage (2.28x10-2).
We calculated the rate of lethal shocks as the product of the rate of

shocks times the conditional probability of failure. The occurrence of
lethal shocks is, then, modeled as a Poisson process with this rate.
34
The Poisson and Exponential Distributions
Let the rate of lethal shocks be * .
*t
P[no lethal shocks in (0, t)] = e

(slide 8, k=0)
The Poisson DRV is the number of lethal shocks.
The component will not fail as long as no lethal shocks

occur. So, we can also write
P(failure occurs after t) = P[T > t] = e t (slide 13)

The exponential CRV is T, the failure time.
35
The Lognormal Distribution

(ln )2
1
( ) =
exp
2
2
2
()
36
The Lognormal Distribution (2)
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
37
Relationship with the Normal Distribution
If is a lognormal variable with parameters and ,

then:
Y ln
is a normal variable with parameters
(mean) and
(standard deviation).
38
The 95th percentile

Since Y is a normal variable, its 95th percentile is
Y95 = + 1.645
But, Y ln
95 = e
+1.645
ln 95 = + 1.645
as in slide 37
39
The Lognormal Distribution: An example

Suppose that
= -6.91
Median:
and
=1.40
50 = exp(-6.91) 10-3
Mean:
m = exp( + 2/2) = 2.65x10-3
95th percentile:
95 = exp(-6.91 + 1.645x1.40) 10-2
5th percentile:
Error Factor:
05 = exp(-6.91 - 1.645x1.40) 10-4

EF = 10
40

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82,
ESD.72, ESD.721
RPRA 4.
Availability
Spring 2007
RPRA 4. Availability
Definitions
Unavailability:
q(t) = Pr[down at t]
Availability:
a(t) 1-q(t) = Pr[up at t]
q(t) + a(t) = 1
Unattended components
q(t) = F(t) = Pr[T < t]
Example:
2-out-of-3 system of exponential components

Q s ( t ) = Fs ( t ) = 3(1 e t ) 2 2(1 e t ) 3
Continuously monitored repairable

components
Continuously monitored repairable

components (2)
Average unavailability:
MTTR
q=
MTTF + MTTR
For the exponential failure distribution:

1
MTTF =
q=
=
, Note : q for < 0.1
1 1 +
+
5
Example
2
1
A
B
3
MTTF (hrs)
MTTR (hrs)
1
2
3
800
600
600
8
15
15
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 }
Structure function for success:

YS = 1 - (1 - Y1Y2)(1 - Y1Y3)
YS = Y1(Y2 + Y3 - Y2 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
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)
Example (5)
Component availabilities:
a1 = 1 (8/800) = 0.990
a2 = a3 = 1 (15/600) = 0.975.
Therefore, AS = 0.989.
10
Note
Minimal cut sets: X1 and X2X3

Structure function for failure:
XS = 1 - (1 - X1)(1 - X2X3)
X S = X 1 + X2 X 3 - X 1 X 2 X 3
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
12
PROBABILITY OF SYSTEM FAILURE

OR SUCCESS
1. Determine the structure function.
2. Express system (un)reliability or
(un)availability as a function of component
(un)reliabilities or (un)availabilities.
3. Determine component (un)reliabilities or
(un)availabilities.
13

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82,
ESD.72, ESD.721
RPRA 5.
Data Analysis
Spring 2007
RPRA 5. Data Analysis
Statistical Inference
Theoretical Model
Evidence
Failure distribution,
e.g., f (t ) = e t
Sample, e.g.,
{t1,,tn}
How do we estimate from the evidence?

How confident are we in this estimate?
Two methods:
Classical (frequentist) statistics
Bayesian statistics
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
The Method of Moments:

Exponential Distribution
Set the theoretical moments equal to the sample
moments and determine the values of the
parameters of the theoretical distribution.
Exponential distribution:
Sample:
1
=t
{10.2, 54.0, 23.3, 41.2, 73.2, 28.0} hrs
229.9
1
t = (10.2 + 54 + 23.3 + 41.2 + 73.2 + 28) =
= 38.32
6
6
MTTF = 38.32
hrs;
1
= 0.026
38.32
hr 1
4

Normal Distribution
Sample: {5.5, 4.7, 6.7, 5.6, 5.7}
x=
( 5.7 + 4.7 + 6.7 + 5.6 + 5.7 ) 28.4

=
= 5.68 =
5
5
( x i x ) 2 = ( 5.5 5.68) 2 + ... + ( 5.7 5.68)2 = 2.032

1
2.032
= 0.508
( 5 1)
s = 0.713 =
s2 =

Poisson Distribution
Sample: {r events in t}
Average number of events: r
t = r
{3 eqs in 7 years}
r
=
t
3
= = 0.43
7
yr 1
6

Binomial Distribution
Sample:
{k 1s in n trials}
Average number of 1s: k

qn = k
k
q=
n
3
{3 failures to start in 17 tests} q = = 0.176
17
Censored Samples and the Exponential

Distribution
Complete sample: All n components fail.
Censored sample: Sampling is terminated at time t0
(with k failures observed) or when the rth failure
occurs.
Define the total operational time as:
k
T = t i + ( n k )t 0
T = t i + ( n r )t r
It can be shown that:
k
T
or =
r
T
Valid for the exponential distribution only (no

memory).
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
Prior information can be utilized via the prior

distribution.
Evidence other than statistical can be accommodated
via the likelihood function.
10
The Model of the World

Deterministic, e.g., a mechanistic computer code
Probabilistic (Aleatory), e.g., R(t/ ) = exp(- t)
The MOW deals with observable quantities.

Both deterministic and aleatory models of the world have
assumptions and parameters.
How confident are we about the validity of these
assumptions and the numerical values of the
parameters?
11
The Epistemic Model

Uncertainties in assumptions are not handled routinely.
If necessary, sensitivity studies are performed.
The epistemic model deals with non-observable
quantities.
Parameter uncertainties are reflected on appropriate
probability distributions.
For the failure rate: () d = Pr(the failure rate has a
value in d about )
12
Unconditional (predictive) probability
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
R(t) = 0.6 e-0.001t + 0.4 e-0.01t

1.0
0.6
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
Figure by MIT OCW.
16
The Quantification of Judgment

Where does the epistemic distribution ( )
come from?
Both substantive and normative goodness are
required.
Direct assessments of parameters like failure
rates should be avoided.
A reasonable measure of central tendency to
estimate is the median.
Upper and lower percentiles can also be
estimated.
17
The lognormal distribution

It is very common to use the lognormal distribution as
the epistemic distribution of failure rates.
(ln )2
1
( ) =
exp
2
2
2
UB = 95= exp( + 1.645 )

LB = 05 = exp( - 1.645 )
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
Failure to operate, Qd:
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
Failure to open, Qd:
3 x 10-6/d
3 x 10-5/d
Manual
Plug, Qd:
3 x 10-5/d
3 x 10-4/d
< 3" diameter, o:
3 x 10-11/hr
3 x 10-8/hr
Table by MIT OCW.
> 3" diameter, o:
3 x 10-12/hr
3 x 10-9/hr
1 x 10-4/d
1 x 10-3/d
Adapted from Rasmussen, et al.

"The Reactor Safety Study."
WASH-1400, US Nuclear Regulatory
Commission, 1975.
Failure to run, o:
(Normal Environments)
Valves
Motor Operated
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
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
Table by MIT OCW.
1 x 10-2/d
3 x 10-4/hr
1 x 10-1/d
3 x 10-2/hr
Adapted from Rasmussen, et al.

"The Reactor Safety Study."
WASH-1400, US Nuclear Regulatory
Commission, 1975.
1 x 10-7/hr
1 x 10-5/hr
Motors
Failure to run
(Normal Environments), o:
Transformers
Open/shorts, o:
Relays
Failure to energize, Qd:
Circuit Breaker
Failure to transfer, Qd:
Limit Switches
Torque Switches
Pressure Switches
Manual Switches
Battery Power Supplies

Failure to provide
proper output, s:
a. All values are rounded to the nearest

half order of magnitude on the exponent.
b. Derived from averaged data on pumps,
combining standby and operate time.
c. Approximated from plugging that was
detected.
d. Derived from combined standby and
operate data.
e. Derived from standby test on batteries,
which does not include load.
Solid State Devices

Failure to function, o:
Diesels (complete plant)

Failure to run, o:
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
21
Updating Epistemic Distributions

Bayes Theorem allows us to incorporate new
evidence into the epistemic distribution.
L ( E / ) ( )
' ( / E) =
L(E / )( )d
22
Example of Bayesian updating of epistemic

distributions
Five components were tested for 100 hours each and no
failures were observed.
Since the reliability of each component is exp(-100 ),
the likelihood function is:
L(E/ ) = P(comp. 1 did not fail AND comp. 2 did not
fail AND comp. 5 did not fail) = exp(-100 ) x exp(100 ) xx exp(-100 ) = exp(-500 )
L(E/ ) = exp(-500 )
Note:
The classical statistics point estimate is zero since no
failures were observed.
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
Impact of the evidence
Prior
distr.
95th
Mean
(hr-1)
(hr-1)
8.0x10-3 3.0x10-2
Posterior 1.3x10-3 3.7x10-3

distr.
Median
5th
(hr-1)
(hr-1)
3x10-3 3.0x10-3
9x10-4
1.5x10-4
25
Selected References
Proceedings of Workshop on Model Uncertainty: Its Characterization and

Quantification, A. Mosleh, N. Siu, C. Smidts, and C. Lui, Eds., Center for
Reliability Engineering, University of Maryland, College Park, MD, 1995.
Reliability Engineering and System Safety, Special Issue on the Treatment

of Aleatory and Epistemic Uncertainty, J.C. Helton and D.E. Burmaster,
Guest Editors., vol. 54, Nos. 2-3, Elsevier Science, 1996.
Apostolakis, G., The Distinction between Aleatory and Epistemic

Uncertainties is Important: An Example from the Inclusion of Aging
Effects into PSA, Proceedings of PSA 99, International Topical Meeting
on Probabilistic Safety Assessment, pp. 135-142, Washington, DC, August
22 - 26, 1999, American Nuclear Society, La Grange Park, Illinois.
26

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82,
ESD.72, ESD.721
RPRA 6.
Probabilistic Risk Assessment
Spring 2007
RPRA 6. Probabilistic Risk Assessment
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.
PRA Steps
Define end states
Identify initiating events
Develop event and fault trees
Quantify
Example: Nuclear Power Plant
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
NPP End States

Various states of degradation of the
reactor core.
Release of radioactivity from the
containment.
Individual risk.
Numbers of early and latent deaths.
Number of injuries.
Land contamination.
The Master Logic Diagram (MLD)

Developed to identify Initiating Events in a
PRA.
Hierarchical depiction of ways in which system
perturbations can occur.
Good check for completeness.
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.
Nuclear Power Plant MLD
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
Event Sequence Diagrams and Event Trees
Two different ways of depicting the progression

of a scenario.
Logically, they are equivalent.
10
NPP: Loss-of-offsite-power event tree

LOOP
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).
12
J. Rasmussens Categories of Behavior

Skill-based behavior: Performance during acts that,
after a statement of intention, take place without
conscious control as smooth, automated, and highly
integrated patterns of behavior.
Rule-based behavior: Performance is consciously
controlled by a stored rule or procedure.
Knowledge-based behavior: Performance during
unfamiliar situations for which no rules for control are
available.
J. Rasmussen, Information Processing and Human-Machine Interaction,
North-Holland, 1986.
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).
15
Reasons Model
Line
Fallible
Management
Psychological
Unsafe
Decisions
Deficiencies
Precursors
Acts
16
Pre-IE (Routine) Actions
Errors of commission
Errors of omission
Median
EF
3x10-3
10-3
A.D. Swain and H.E. Guttmann, Handbook of Human Reliability Analysis

with Emphasis on Nuclear Power Plant Applications, Report NUREG/CR
1278, US Nuclear Regulatory Commission, 1983.
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
What components constitute the system?

How do the components and system operate?
How does the system interact with other systems?
What functions does the system perform?
How does the system fail? (NOTE: The event tree
determines the boundary conditions)
Hardware
Software
Human errors
What external events is the system susceptible to?

20
Dependent Failures: An Example

Component B1
B1 and B2 are identical

redundant components
Component B2
System Logic
XS = XA + XB1 XB2 - XA XB1 XB2
Failure
Probability
P(fail) = P(XA) + P(XB1 XB2 )

P(XA XB1 XB2 )
21
Example (contd)
In general, we cannot assume independent

failures of B1 and B2. This means that
P(XB1 XB2 ) P(XB1) P(XB2 )
How do we evaluate these dependencies?
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
Expanding the set of failure causes

The complete set of basic events involving component A
in a three-component system is:
AI
= Independent failure of component A.
CAB = Failure of components A and B (and not

C) from common causes.
CAC = Failure of components A and C (and not
B) from common causes.
CABC = Failure of components A, B, and C from
common causes.
24
Component Failure
The equivalent Boolean representation of total

failure of component A is
A = AI + CAB + CAC + CABC
or
XA = 1 (1- XI)(1- XAB)(1- XAC)(1- XABC)
25
Minimal Cut Sets
The minimal cut sets of the expanded fault tree

are:
{AI,BI}; {AI,CI}; {BI,CI}; {CAB}; {CAC};
{CBC}; {CABC}
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
The Beta-Factor Model

The -factor model assumes that
common-cause events always involve
failure of all components of a common
cause component group
It further assumes that
CCF
=
total
28
- Factor Model (contd)

Component B 1
Independent Failure
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*
Reliability Block Diagram
B 2*
B 12
Fault Tree
From Prof. A. Mosleh, University of Maryland. Lecture at MIT, March 2006.

Courtesy of A. Mosleh. Used with permission.
29
Generic Beta Factors

0.2
GENERIC BETA FACTOR
(MEAN VALUE)
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
From Prof. A. Mosleh, University of Maryland. Lecture at MIT, March 2006.

Courtesy of A. Mosleh. Used with permission.
30
Space Shuttle Orbiter Dependent Failure

Data Collection, Analysis, and Results
474 Space Shuttle orbiter in-flight anomaly
reports analyzed.
Data used to:
Determine frequency and types of dependent
failures, causes, and defenses associated with
spacecraft
Estimate a beta factor of 0.13.
31
Data
A1- A1-PRI-VLV-FC
The epistemic
distribution is
Lognormal.
4.500E-003 5.900E+000
Mean value of the

epistemic distribution
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:
Initiating Events Frequencies

Component Failure Frequencies
Component Test and Maintenance Unavailability
Common-Cause Failure Probabilities
Human Error Rates
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
IEEE Standard 500

Reliability Analysis Center
MIL-Std 217
Offshore Reliability Data Project
T-Book
System-specific
Maintenance Logs
Test Logs
Operation Records
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)
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.
37
Updating Epistemic Distributions
Bayes Theorem allows us to incorporate new

evidence into the epistemic distribution.
L(E / )( )
' ( / E ) =
L(E / )( )d
38
The Quantification of Judgment

Where does the epistemic distribution ()
come from?
Both substantive and normative goodnesses
are required.
Direct assessments of parameters like failure
rates should be avoided.
A reasonable measure of central tendency to
estimate is the median.
Upper and lower percentiles can also be
estimated.
39

Courtesy of K. Kiper. Used with permission.
40

1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82, ESD.72
RPRA 7.
Risk Management
Spring 2007
RPRA 7. Risk Management
The Risks We Accept

Annual Individual Occupational Risks
All industries
7x10-5
Coal Mining:
24x10-5
Fire fighting:
40x10-5
Police:
32x10-5
US President
1,900x105
Annual Public Risks
Total
870x10-5
Heart Disease
271x10-5
All cancers
200x10-5
Motor vehicles:
15x10-5
Source: Wilson & Crouch, Risk/Benefit Analysis, Harvard
University Press, 2001.
Risk Acceptance
Voluntary vs. involuntary risks.

Adequate protection of public health and safety.
Unacceptable vs. tolerable risks.
Acceptable vs. Tolerable Risks

(UK Health and Safety Executive)
Risk cannot be justified
save in extraordinary
circumstances
Increasing individual risks and societal concerns
UNACCEPTABLE REGION
Control measures must be

introduced for risk in this
region to drive residual risk
towards the broadly
acceptable region
TOLERABLE REGION
BROADLY ACCEPTABLE REGION
Level of residual risk

regarded as insignificant -further effort to reduce risk
not likely to be required
NASA Integrated Action Team

Definition
Acceptable Risk is the risk that is understood and agreed to
by the program/project, Governing Program Management
Council (GPMC), and customer sufficient to achieve defined
success criteria within the approved level of resources.
Each program/project is unique.
Acceptable risk is a result of a knowledge-based review
and decision process.
Management and stakeholders must concur in the risk
acceptance process.
Effective communication is essential to the understanding
of risk.
Assessment of acceptable risk must be a continuing process.
The Importance of Risk Communication

Total Man Caused
1
Air Crashes Total

Fires
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
Figure by MIT OCW.

Adapted from Rasmussen, et al. "The Reactor Safety Study." WASH-1400, US Nuclear Regulatory Commission, 1975.
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)
Quantitative Health Objectives for

Nuclear Power Plants (NRC)
The individual early fatality risk in the region
between the site boundary and 1 mile beyond this
boundary will be less than 5x10-7 per year (one
thousandth of the risk due to all other causes).
The individual latent cancer fatality risk in the
region between the site boundary and 10 miles
beyond this boundary will be less than 2x10-6 per
year (one thousandth of the risk due to all other
causes).
Involving the Stakeholders
Risk assessment can and should be used to involve

stakeholders and provide a mechanism for the
consideration of their cultural, socioeconomic,
historical, and religious values, in addition to the
risks to human health and the environment
associated with the contamination of DOE
facilities and their remediation.
National Research Council, Building Consensus, 1994
The Analytic-Deliberative Process

Analysis uses rigorous, replicable methods, evaluated
under the agreed protocols of an expert community - such
as those of disciplines in the natural, social, or decision
sciences, as well as mathematics, logic, and law - to arrive
at answers to factual questions.
Deliberation is any formal or informal process for
communication and collective consideration of issues.
National Research Council, Understanding Risk, 1996.
10
PRA Provides the Analysis
The dominant accident scenarios are the basis

for risk management.
The objectives are to meet the safety goals and to
optimize the design and operations.
11
Risk management options from the

scenarios: Fire risk
x =
Q
j
d jQ x d , j
x = contribution to consequence x from fires

j = Fire frequency in location j
Qd|j = Pr[cable damage given fire in j]
Qx|d,j = Pr[occurrence of x given cable damage]
12
Case study
Zone designator/
Scenario
1. Fire Under Cable

Trays Damaging
Power Cables to
System A
2. Fire in the Aisle

Damaging Power
Cables to
System A
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
NOTE: Exponential notation is indicated in abbreviated form; i.e., 4.6-8 = 4.6 10

x = jQd/j Qx/d,j
y = jQd/j Qx/d,j Qy/x,d,j.
-8
13
Controlling fire frequency j

1.2E-4;
4.2E-5;
1.5E-4 per year
They are the result of transient fuels.

Option:
Stringent administrative controls
Not a credible option. Controls are already

supposed to be stringent.
14
Controlling cable damage Qd/j

Construct a fire barrier between redundant
cable trays.
15
Controlling system response Qxd,j and Qyx,d,j
Install a self-contained charging pump (System A).

Connect power cable independent of fire zones
(System A).
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
Management sought crews views.

Other issues were considered.
Alternate power source was the chosen option.
model acts of unbelievable intelligence either.
18

72 PDF

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

72 PDF

Uploaded by

Copyright:

Available Formats

Engineering Risk Benefit Analysis

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

The Time Value of Money

The Importance of Time

The Basics of Interest Rates Simple and

Cash Flows over Time

Up arrow = we receive $; down arrow = we pay $

The Basic Discount Factors

CBA 2. The Time Value of Money

Single-Payment Compound-Amount Factor

A single payment is made after n periods.

Single-Payment Present-Worth Factor

CBA 2. The Time Value of Money

Equal payments, A, occur at the end of each period.

Example: Your house mortgage is $300,000 for 30

Summary of the Formulas

Equal-Payment-Series Capital-Recovery Factor

Continuous Compounding (1)

where r is the nominal annual interest rate.

CBA 2. The Time Value of Money

Continuous Compounding (2)

CBA 2. The Time Value of Money

Nominal and Effective Interest Rates (1)

Nominal and Effective Interest Rates (2)

Compounded annually: i = r = 10%

CBA 2. The Time Value of Money

Nominal and Effective Interest Rates (3)

Nominal and Effective Interest Rates (4)

This yields (A/P, 0.00625, 360) = 0.00699

P x 0.00699 = (H - 50,000) x 0.00699 2,000

H (2,000/0.00699) + 50,000 = $336,123

Nominal and Effective Interest Rates (5)

A* = 0.0867P per year

Your effective payment per year is

Economic Equivalence (1)

Economic Equivalence (2)

P = 408.15 + 381.45 + 712.99 = $1,502.59

Economic Equivalence (3)

P = 405.30 + 377.89 + 704.69 = $1,487.88

CBA 2. The Time Value of Money

CPI and Inflation

1+f = 4.9761/32 = 1.0514

CBA 2. The Time Value of Money

CBA 2. The Time Value of Money

Constant and Actual Dollars

Equivalence in terms of actual dollars: Use i.

CBA 2. The Time Value of Money

CBA 2. The Time Value of Money

Example: Going to the Movies

Example: Investments in Two Countries (1)

Example: Investments in Two Countries (2)

Engineering Risk Benefit Analysis

Bases for Comparison of Alternatives

CBA 3. Bases for Comparison of Alternatives

CBA 3. Bases for Comparison of Alternatives

Present Worth (1)

CBA 3. Bases for Comparison of Alternatives

Present Worth (2)

Present Worth, Net Present Worth, and Net

CBA 3. Bases for Comparison of Alternatives

Present Worth: Example

Here: A = $500/mo, i = 10/12 = 0.83%, n = 36 months