Author Richard Serfozo

Basic Probability Problems

May 20, 2003
Springer
Berlin Heidelberg NewYork
Hong Kong London
Milan Paris Tokyo
1
Elementary Concepts
The subject of applied probability appears to be rather fragmented be-
cause problems involving randomness arise in many dierent contexts and
they often require the use of ad hoc mathematical techniques. The subject
has more underlying structure, however, than meets the eye. The set of
notes before you describes this structure via a number of basic probability
problems.
These notes are intended as a supplement to an introductory prob-
ability textbook that describes the methodology and notation. The basic
problems presented here should be viewed as a follow-on to elementary,
motivating examples. A reader just learning probability should aim at mas-
tering these basic problems in the sense of being able to recognize them in
various settings and solve them by carry out the required analysis and com-
putations. A good way to study each problem is to create and solve one or
more examples of the problem analogous to those below. By creating new
examples this way, one will actually own a piece of the subject as well as
understanding it.
1.1 Probabilities of Events
A description of events and their probabilities requires a framework for rep-
resenting events of interest in terms of a random experiment. Our textbook
tells us how to describe such an experiment in terms of a set of outcomes
called a sample space, and to represent events as subsets of . A basic
problem in this regard is as follows.
Problem 1 Sample Spaces and Events as Sets Given a verbal de-
scription of a random experiment and certain events of interest, represent
the sample space as a set , and identify the events as subsets of . There
may be several natural choices for ; select the simplest one that contains
enough information to describe the events of interest. Be able to describe a
2 1 Elementary Concepts
complicated event that may consist of intersections, unions or complements
of other events.
Example 1.1 Parts exiting an assembly line are inspected and categorized
as being good or bad (defective). In a batch of 10 parts, an inspector is
interested in the following events:
A = The number of bad parts is 0.
B = The number of bad parts is 1.
C = The number of good parts is 5.
The random experiment consists of observing a batch of ten parts. Since
each of the preceding events can be described by the number of bad parts
in the batch, we can represent an elementary outcome as the number of
bad parts in the batch. Accordingly, we represent the sample space as =
{0, 1, . . . , 10}. Then the events above are represented respectively as
A = {0}, B = {0, 1}, C = {0, 1, . . . , 4}.
Example 1.2 In the preceding context, suppose the inspector is interested
in the following events associated with the rst ve parts:
A = The rst two parts are good and the last three are bad.
B = Three good parts are produced before two bad ones are.
These events cannot be described as above by the number of bad parts, but
are describable by the sample space
= {(x
1
x
2
x
5
) : x
i
= 0 or 1},
where 1 means a good part and 0 means a bad part. Then A = {(11000)}
and
B = {(11100), (01110), (10111), (10110), (11011), (11010),
(11101), (11100), (11110), (11111)}.
Upon dening a random experiment, the next step is to assign proba-
bilities to the outcomes and use these to compute probabilities of events.
Problem 2 Elementary Probability Computations Suppose you are
considering an experiment with a probability space that is discrete (it
is nite or countably innite). A probability measure P for such a discrete
space is dened by specifying the probability P() for each elementary out-
come such that

P() = 1. Methods for doing this are discussed

below. Then you should be able to compute the probability of any event A,
which is
P(A) =

A
P().
Example 2.1 Consider the random experiment that describes the number
of re alarms that a certain station responds to in a week. Assume that it
is known, based on historical data (via the frequency method discussed
1.1 Probabilities of Events 3
shortly), that the probabilities of exactly 0, 1, 2, 3, 4, 5 alarm responses are
.10, .15, .20, .25, .20, .10, respectively. (These numbers sum to 1.) Then the
event A that the number of alarm responses is at least 3 is
P(A) =

A
P() = .25 +.20 +.10 = .55.
Similarly, P( less than 2 alarm responses ) = .10 +.15 = .25.
The preceding examples were for nite or countable sample spaces.
Many problems, however, involve sample spaces that are uncountable? For
instance, suppose that one is interested in the probability that a worker
completes a task within a certain time, say within 30 minutes, where the
time may be anywhere in the interval [0, 60], and the time scale is in min-
utes. Then the sample space of this experiment of recording task completion
times is = [0, 60]. This space has an uncountable number of points. We
refer to the task completion time as a continuous variable, since it takes
values in an interval. Other examples of continuous variables are tempera-
tures, weights, volumes, stresses, pressures, and locations in the plane or a
space, where one is interested their values over intervals or rectangles.
For continuous variables, we use the following general denition of a
probability measure. A probability measure P on a sample space is a function
that assigns a number P(A) to each event A in such that the following
properties are satised:
0 P(A) 1, P() = 1,
and, for any collection of disjoint events A
1
, . . . , A
n
,
P(
n
k=1
A
k
) =
n

k=1
P(A
k
).
This denition is a generalization of the one above for countable state
spaces.
Problem 3 Probabilities for Uncountable Outcomes For an exper-
iment with an uncountable sample space and a probability measure on it,
compute the probability of an event.
Example 3.1 A aw is detected in an underground cable of length 4000
feet. The aw is equally likely to be located anywhere, which means that the
probability of the aw lying in any region A [0, 4000] is P(A) = (length
of A)/4000. This rule obviously satises the denition of a probability. Find
the probabilities of the following events:
B = the aw is located within 200 feet of the center of the cable.
C = the aw is located within 100 feet of either end of the cable.
By the rule for dening P in this experiment, it follows that
4 1 Elementary Concepts
P(B) = (Length of the interval [1800, 2200])/4000
= 400/4000 = .1
P(C) = (Length of the region [0, 100] [3900, 4000])/4000
= 200/4000 = .05.
The next problem concerns how one obtains probabilities for random
experiments.
Problem 4 Assignment of Probabilities Select probabilities for the
events of an experiment by one of the following methods:
Frequency Method Repeat the experiment many times and set P(A)
equal to the percent of the outcomes in which A occurred (the relative fre-
quency of A).
Logical Method Use logic of the experiment to determine the probabili-
ties.
Subjective Method Use intuition or a best guess of an individual (or
group of persons) to assign the probabilities.
Later we will study some standard probability distributions (e.g. the
binomial or normal distributions) that occur frequently in nature and would
be appropriate for any of the preceding methods. For most of our work, the
probability measure or distribution will be specied and our focus will be
on computations with them. For applications that you encounter in other
courses or in practice, you may have to select the appropriate probability
measure. The frequency and logical methods are used the most. The fre-
quency method is feasible when there is historical data available, or when
one can perform the experiment many times (at a reasonable cost).
The subjective method is used as a last resort when the other two ap-
proaches are not practical. The subjective method is sometimes made rig-
orous by using utility functions of ones preferences for certain quantities
and deducing the probabilities from the functions (this is not treated in
standard courses).
Example 4.1 A company is interested in forecasting the quantity of a
product that it will sell in a month. The aim is to nd the probabilities of
selling 0, 1, . . . , 5 units of the product; no more than 5 per month have ever
been sold (typical of some expensive products). The company has historical
data of the quantities sold in the last 24 months. The demands during
this period have been stable (no seasonal variations), and it is assumed
that the demand during a month does not aect the demands in other
months. Therefore, we can consider the quantities sold x
1
, x
2
, . . . , x
24
in
the 24 months as 24 samples of the random experiment of the quantity
sold per month of the product. Let p
k
denote the probability of selling
k units, where k = 0, . . . , 5. Then by the frequency method, we assign (or
estimate) these probabilities as
p
k
= (number of x
i
s that equal k)/24.
1.1 Probabilities of Events 5
This is the percentage of x
i
s that equal k. Specically, suppose the quan-
tities sold are
3, 4, 3, 0, 4, 5, 2, 2, 2, 4, 1, 1, 0, 2, 4, 5, 4, 4, 1, 2, 1, 1, 3, 4.
Then
p
0
= 2/24 = 1/12, p
1
= 5/24, p
2
= 5/24,
p
3
= 1/8, p
4
= 7/24, p
5
= 1/12.
These would be be probabilities the company would use for forecasting. For
instance, the probability of selling at least 4 units in the next month would
be p
4
+p
5
= 3/8.
Here are some examples of the logical method.
Example 4.2 Suppose an experiment has n outcomes. If we say that each
outcome is equally likely, then it follows logically from the denition of a
probability measure that the probability of each outcome must be 1/n.
Because of this, the probability of an event A is
P(A) = |A|/||,
where |A| denotes the number of outcomes (elements or points) in the set
A, and || is the total number of outcomes, which we assumed was n.
Example 4.3 Three companies are bidding on a contract. The relative
qualities of the companies are such that Company A is twice as good as
Company B, and Company B is three times as good as Company C. What
is the probability of each company winning the contract? This is an ex-
periment in which the elementary outcomes are A, B, C that the respective
companies win the contract. The logic of this setting says that
P(A) = 2P(B), P(B) = 3P(C),
where P(A) + P(B) + P(C) = 1. Thus we have three equations in three
unknowns and the solution is clearly
P(A) = .6 P(B) = .3 P(C) = .1.
Example 4.4 A system fails when a defect occurs in one of 9 subsystems
labeled 1, . . . , 9 (only one defect occurs at a time). Let p
i
denote the prob-
ability that the defect is in subsystem i. Suppose that each of the subsys-
tems 4, 5, 6 is twice as likely to contain the defect as any one of the other
subsystems. This information tells us that p
i
= 2p
j
for i {4, 5, 6} and
j {4, 5, 6} (for instance p
4
= 2p
9
). From these equalities, it follows that
p
i
=
_
1/6 for i = 4, 5, 6
1/12 otherwise.
6 1 Elementary Concepts
The main property dening a probability P for an experiment with
outcome space is as follows: If an event A is the union of disjoint events
A
1
, . . . , A
n
, that is, A = A
1
A
2
. . . A
n
, then
P(A) =
n

i=1
P(A
i
).
We also know that P() = 1 and 0 P(A) 1 for any event A .
These relations lead to further properties of P. Here are a few:
P(A
c
) = 1 P(A), P(A B) = P(A) +P(B) P(A B),
P(A) = P(A B
c
) +P(A B), P(A) P(B), for A B
Let us see how these properties are used.
Problem 5 Algebra of Events and Probabilities If the probabilities of
certain events are known, how can one use these to compute the probabilities
of other events? Use the algebra of sets and basic properties of probabilities.
Example 5.1 A certain product was found to have two types of minor
defects. The probability that an item of the product has only a type 1
defect is .2, and the probability that it has only a type 1 defect is .2. Also,
the probability that it has both defects is .1. Find the probabilities of the
following events:
A = An item has either a type 1 defect or a type 2 defect.
B = An item does not have either of the defects.
C = An item has defect 1, but not defect 2.
D = An item has exactly one of the two defects.
Solution. Let D
1
, D
2
denote the events that the item has defects 1, 2,
respectively. We know that P(D
1
) = .2, P(D
2
) = .3 and P(D
1
D
2
) = .1.
Then, by the denitions of the events and the properties of probabilities, it
follows that
P(A) = P(D
1
D
2
) = P(D
1
) +P(D
2
) P(D
1
D
2
) = .2 +.3 .1 = .4.
P(B) = P((D
1
D
2
)
c
) = 1 P(D
1
D
2
) = 1 .4 = .6.
P(C) = P(D
1
D
c
2
) = P(D
1
) P(D
1
D
2
) = .2 .1 = .1
P(D) = P(B) P(D
1
D
2
) = .4 .1 = .3.
The last expression uses the fact that B is the union of the disjoint events
D and D
1
D
2
.
We now discuss random experiments with equally likely outcomes. Al-
though these experiments have a simple probability structure, they cover
many diverse applications involving permutations and combinations of
things. The main diculty is in deciding which counting principles to use
to determine the numbers of outcomes in the events and sample space.
1.1 Probabilities of Events 7
Problem 6 Probabilities Involving Equally Likely Outcomes If an
experiment has equally likely outcomes, how does one evaluate the probabil-
ity of an event? In this case, the probability of an event A is
P(A) = |A|/||,
where |A| denotes the number of outcomes (elements or points) in the set A.
We can evaluate these numbers by using the standard counting principles
described below (or by direct enumeration if a set is small).
The following are two fundamental counting principles, which are also
used in other engineering problems not involving randomness.
6.1 Multiplication Principle Consider a job consisting of performing
k tasks, where the numbers of ways of performing the tasks are n
1
, . . . , n
k
,
respectively. Then the number of ways in which the job can be done is
n
1
n
k
.
This principle is used in counting ordered samples, which are k-tuples
of the form (x
1
, . . . , x
k
). Namely, suppose there are n
1
choices for x
1
; n
2
choices for x
2
; . . . ; and n
k
choices for x
k
. Then the number of k-tuples
(x
1
, . . . , x
k
) is the product n
1
n
k
.
Two important special cases of the multiplication principle are as fol-
lows.
Permutations: The number of permutations of n items is
n! = n(n 1)(n 2) 2 1.
This is also the number of n-dimensional vectors that can be formed without
replacement from a population of size n.
Sampling Without Replacement: The number of k-dimensional vectors
that can be formed without replacement from a population of size n is
n(n 1)(n 2) (n k + 1) = n!/(n k)!.
These counting techniques have all been for counting numbers of vectors
in sets. We not switch to counting subsets instead of vectors.
6.2 Combinatorial Principle: The number of subsets of size k that can
be formed from a set of size n is the binomial coecient
_
n
k
_
=
n!
k!(n k)!
=
n(n 1) (n k + 1)!
k!
.
This n choose k quantity is also called the number of combinations of n
things taken k at a time.
8 1 Elementary Concepts
This principle is used for counting unordered samples, which are subsets
of the form {x
1
, . . . x
k
}. The number of such subsets that can be formed
from n distinct objects is
_
n
k
_
. Keep in mind the dierence between a k-tuple
and a set with k elements. For instance,
{1, 2, 3} = {1, 3, 2} = {2, 1, 3}, but (1, 2, 3) = (1, 3, 2) = (2, 1, 3).
The following are four examples of combinatorial probability problems
that use the counting principles. Essentially all the combinatorial problems
we will cover will be of these forms, so study them carefully.
The setting of the examples will be a production facility that oper-
ates as follows. Five types of parts labeled 1, . . . , 5 are produced for com-
panies X, Y, Z. Each part is identied by its type and the company for
which it is destined (e.g., part 3X is a type 3 part for company X). A
single production run consists of one part of each type for each company:
1X, . . . , 5X, 1Y, . . . , 5Y, 1Z, . . . , 5Z. The total number of parts (by the mul-
tiplication principle) is therefore 15. The following examples refer to this
lot (or population) of 15 parts.
Example 6.1 Ordered Sampling with Replacement. Suppose a random
sample of four parts is chosen one at a time with replacement from the
lot of 15. (For instance, the parts may be inspected and replaced at four
stages in the production.) Find the probability of the events:
A = The rst 2 parts are type 5 parts.
B = The rst and last parts are of type 1, 2 or 3.
Solution. These events involve the order of the parts, and so dene the
sample space as all 4-tuples of parts chosen with replacement from 15
(e.g. an outcome is (3Y, 5X, 1Z, 5X)). Since each sample is equally likely,
P(A) = |A|/|| = 3
2
15
2
/15
4
= 1/25.
Here || = 15
4
since consists of 4-tuples and there are 15 choices for
each of the four entries. Also, |A| = 3
2
15
2
, since A consists of 4-tuples
in which there are 3 choices for rst two entries and 15 choices for the last
two entries. Similarly,
P(B) = |B|/|| = 9
2
15
2
/15
4
= 9/25.
Here B consists of 4-tuples in which there are 9 choices for the rst entry,
15 choices for the second and third entries, and 9 choices for the fourth
entry.
Example 6.2 Ordered Sampling Without Replacement. Suppose in the
preceding setting that a random sample of four parts is chosen one at a
time without replacement from the lot of 15 parts. (For instance, an in-
spector analyzes four parts at once, or a destructive test is done on four
parts.) Find the probabilities of the events:
1.1 Probabilities of Events 9
A = All four parts are of type 1, 2, or 3.
B = The rst and last parts are for company Z.
Solution. Since event B depends on the order of the parts, we dene the
sample space as all 4-tuples of parts chosen from the lot of 15 without
replacement. Each sample is equally likely, and so
P(A) = |A|/|| =
9 8 7 6
15 14 13 12
= 6/65.
Here there are 15, 14, 13, 12 choices, respectively, for the four entries of an
outcome in , and there are 9, 8, 7, 6 choices, respectively for the four entries
of an outcome in A. Similarly,
P(B) = |B|/|| =
5 4 13 12
15 14 13 12
= 2/21.
Here there are 5 and 4 choices for rst and last entries of an outcome in B,
and there are 13 and 12 choices for the middle entries.
Example 6.3 Unordered Sampling. In the preceding setting, suppose four
parts are randomly chosen without replacement from the lot of 15. Find
the probabilities of the events:
A = All four parts are of type 1 or 2.
B = Two parts are for X and two parts are for Y .
C = One part is for X, one part is for Y and 2 parts are for Z.
Solution. Since these events do not depend on the order in which the parts
are chosen, we dene the sample space as all subsets of four parts chosen
from the lot of 15. Again, each sample is equally likely, and so
P(A) = |A|/|| =
_
6
4
_
/
_
15
4
_
= 1/91.
Here || is the number of subsets of size 4 that one can choose from the set
of 15 parts, and |A| is the number of subsets of size 4 that one can choose
from the set of 6 parts of type 1 or 2. Similarly,
P(B) = |B|/|| =
_
5
2
__
5
2
_
/
_
15
4
_
= 20/273.
Here B consists of two subevents: two parts are for X (which has (
5
2
) pos-
sibilities) and two parts are for Y (which has (
5
2
) possibilities). Then the
number of outcomes in B (by the multiplication principle) is the product
of these two binomial coecients. Using similar reasoning,
P(C) = |C|/|| = 5 5
_
5
2
_
/
_
15
4
_
= 50/273.
10 1 Elementary Concepts
Example 6.4 Permutation Problem. Suppose the lot of 15 parts described
above are randomly put on a conveyor to go into a storage area. Find the
probabilities of the events:
A = The rst three parts are for company X.
B = The last four parts are of type 1 or 2.
Solution. Dene the sample space as the set of all permutations of the 15
parts. It therefore has 15! outcomes. Since all outcomes are equally likely,
P(A) = |A|/|| = (5 4 3 12!)/15! = 2/91.
Here |A| is the number of permutations of 15 parts with an X-part in the
rst 3 entries (there are 5 4 3 choices for this) and the number of choices
for the remaining entries is 12!. Similarly,
P(B) = |B|/|| = (6 5 4 3 11!)/15! = 1/91.
2
Conditional Probabilities and Independent
Events
Many applications involve probabilities conditioned that some special event
has occurred. Assuming that an event occurs (or will occur), we have to
evaluate probabilities of events dierently than if we did not know the event
occurred. The new probabilities are conditional probabilities.
Problem 7 Conditional Probabilities From Initial Probability Mea-
sure Knowing the probability measure of an experiment, how does one com-
pute the conditional probability of an event B given that A occurs? Use the
denition
P(B|A) = P(A B)/P(A),
provided P(A) > 0.
Example 7.1 An item is randomly drawn from a set of items
{a, b, c, d, 1, 2, 3, I, II, III, V I, V }.
Find the probability that item c is drawn, given that it is known (somehow)
that the item drawn is one of the rst 6 in the set.
Solution. Let A denote the event that the item drawn is one of the rst 6
in the set, and let B denote the event that item c is drawn. Since each item
is equally likely, P(A) = 1/2. Then
P(B|A) = P(A B)/P(A)
= P(c is drawn from {a, b, c, d, 1, 2})/P(A)
= (1/6)/(1/2) = 1/3.
Instead of denoting events by A, B, etc., we could have expressed the events
in words. For instance,
P(item drawn is II|item drawn is a roman numeral)
=
P(II is drawn from {I, II, III, V I, V })
P(item drawn is a roman numeral)
= (1/5)/(5/12) = 12/25.
12 2 Conditional Probabilities and Independent Events
Example 7.2 A customer promises to place an order sometime in the next
10 months for a product you produce. You have to plan production for the
next year, and so you decide to come up with educated guesses (maybe
based on the customers hunches or past behavior) for the probabilities
p
1
, p
2
, . . . , p
10
that the customer places the order in months 1, . . . , 10, re-
spectively. These probabilities sum to one since the order is assured to
be placed. The following are typical conditional probabilities that one can
obtain from the preceding probabilities (which are supposedly known).
To express events in a convenient manner, we let X represent the month
in which the order is placed. The conditional probability that the order is
placed in month 4, given that it is placed after month 3, is
P(X = 4|X 4) = P({X = 4} {X 4})/P(X 4)
= p
4
/(p
4
+ +p
10
).
The conditional probability that the order is placed after month 7, given
that it is placed after month 2, is
P(X 8|X 3) = P({X 8} {X 3})/P(X 3)
= (p
8
+p
9
+p
10
)/(p
3
+ +p
10
).
The preceding example involves obtaining a conditional probability in
terms of regular probabilities. We will now discuss more interesting uses of
conditional probabilities in which conditional probabilities are specied (or
they are easy to obtain) and we want to nd regular probabilities of events
or other conditional probabilities.
Problem 8 Conditional Probabilities for Sequence of Events Com-
pute the probability of the intersection of the events A
1
. . . A
n
in terms of
conditional probabilities P(A
i
|A
1
A
i1
) of the event A
i
given the
past events A
1
, . . . , A
i1
. Use the multiplication formula
P(A
1
A
n
) = P(A
1
)P(A
2
|A
1
) P(A
n
|A
1
A
n1
).
Example 8.1 In a manufacturing plant, there are 10, 15 and 20 jobs of
types I, II and III, respectively, waiting to be processed on a machine (e.g.
these may be physically dierent parts, or orders for one type of product
for dierent customers). The machine operator randomly selects parts one
at a time for processing (i.e. each part is equally likely to be selected at any
time). What is the probability that the rst 4 jobs processed are of types
II, II, III, II and in this exact order?
Solution. Let A denote the event of interest. We can consider this event
as A = A
1
A
2
A
3
A
4
, where A
1
= job 1 is type II, A
2
= job 2 is
type II, A
3
= job 3 is type III, and A
4
= job 4 is type II. Then by the
multiplicative property of conditional probabilities,
2 Conditional Probabilities and Independent Events 13
P(A) = P(A
1
A
2
A
3
A
4
)
= P(A
1
)P(A
2
|A
1
)P(A
3
|A
1
A
2
)P(A
4
|A
1
A
2
A
3
)
= (15/45)(14/44)(20/43)(13/42) = 0.072.
Here 15/45 is the probability of chosing one of the 15 type II jobs from the
lot of 45 (each job is equally likely). Next, P(A
2
|A
1
) = 14/44, since this is
the probability of choosing one of the remaining 14 type II jobs from the
lot ot 44 after one type II job was already chosen for job 1. The 20/43 is
the probability of choosing one job from the 20 type III jobs when there
are 43 jobs left. Finally, 13/42 is the conditional probability of choosing a
type II job as job 4 given that there were already two type II jobs chosen
from the 15 (and so 13 remain).
Another approach to solving this problem is to treat it as a counting
problem. Namely, there are 45 44 43 42 ways of selecting an ordered
sample of four jobs from the 45 jobs, and the number of ways that A can
happen is 15 14 20 13. Thus
P(A) =
15 14 20 13
45 44 43 42
= .072.
Example 8.2 Consider a manufacturing system in which parts travel
through a network of work stations as shown in Figure 2.1. The p
jk
s are
the conditional probabilities that a typical part traverses the arcs. For in-
stance, p
35
is the conditional probability of going from station 3 to station 5
(3 5) given that the traveler is at station 3. The possible departures from
each station are assigned probabilities that sum to one (e.g., p
35
+p
36
= 1).
m m
m m
- -
- -
1
?
m m

>
Z
Z
Z
Z~
2
-
?
3 6
5
4

>
6
p
24
p
54
p56
p23
p35
p36
p
12
p
13
Fig. 2.1. Traveling in a Network
One is typically interested in the probability that a part takes a specic
route, or that the part goes through a certain sector of the network (or
graph). For instance, using the multiplication rule for conditional probabil-
ities, the probability of taking the route 1 3 5 4 is
P(1 3 5 4) = P(1 3)P(3 5|1 3)P(5 4|1 3 3 5)
= p
13
p
35
p
54
.
14 2 Conditional Probabilities and Independent Events
Next, consider the event A that a part goes through station 4. Clearly, A
is the union of three dierent routes that go through 4. Then reasoning as
above, we have
P(A) = P(1 2 4) +P(1 3 5 4) +P(1 2 3 5 4)
= p
12
p
24
+p
13
p
35
p
54
+p
12
p
23
p
35
p
54
.
In the same way, one can evaluate the probability that a part goes
through a certain sector of the network by taking the sum of all the prob-
abilities of the routes that go through that sector. As an exercise, nd the
probability that a part goes through 3 6. Also, consider another such
event and compute its probability. Similar networks or graphs are also used
to model a job or major task that consists of randomly evolving subtasks.
Each node represents a subtask and the arcs leading out of a node lead to
the next subtask that is chosen or determined by the probabilities on the
arcs.
Intuitively, two events A and B are independent if knowing that A
occurs does not tell us anything about B, and vice versa. The mathematical
denition is
P(A B) = P(A)P(B).
We now describe three types of problems dealing with independent events.
Problem 9 Independent Events (i) Determine whether events A and
B are independent or dependent. Their independence can be established by
verifying the denition P(A B) = P(A)P(B). Another approach is to
verify P(A|B) = P(A).
(ii) Establish the independence or dependence of a collection of events
A
1
, . . . , A
n
. They are independent if
P(A
i1
A
i
k
) = P(A
i1
) P(A
i
k
)
for any subsequence i
1
, . . . , i
k
of 1, . . . , n. If any one of these equalities is
not true, then the events are dependent.
(ii) Be able to identify when to use these product formulas for independent
events that may arise from physically independent phenomena.
Example 9.1 Suppose that a part is randomly selected from a set of parts
{S
1
, M
1
, L
1
, S
2
, M
2
, L
2
}
which represents small, medium and large parts (S, M, L) from two com-
panies 1, 2. Consider the events
A = the part is S or M = {S
1
, M
1
, S
2
, M
2
}
B = the part is M or L = {M
1
, L
1
, M
2
, L
2
}
C = the part is S = {S
1
, S
2
}
D = the part is from company 1 = {S
1
, M
1
, L
1
}
2 Conditional Probabilities and Independent Events 15
Describe whether or not these events are independent.
Solution. Since each part is equally likely to be chosen,
P(A) = 4/6, P(B) = 4/6, P(C) = 2/6, P(D) = 3/6.
Now
P(A B) = P({M
1
, M
2
}) = 2/6 = P(A)P(B).
Thus, A and B are dependent. Next,
P(C D) = P({S
1
}) = 1/6 = P(C)P(D).
Thus, C and D are independent. Note that all of the events A, B, C and D
are not independent since we already know that A and B are dependent.
However, are B and C independent? How about A and C?
The next example illustrates the evaluation of a probability of intersec-
tion of independent events by multiplying probabilities of the events.
Example 9.2 A project consists of three independent tasks and the prob-
abilities of these tasks being completed on time are .90, .80 and .95, respec-
tively. Find the probabilities that:
(i) All three tasks will be completed on time.
(ii) The rst two tasks will be completed on time and the third one will
not.
(iii) Either the rst or last task will be completed on time.
Solution. Let C
i
denote the event that task i is completed on time. The
physical independence of the tasks translates into the independence of the
events C
1
, C
2
, C
3
. Then by the denition of independent events, the rst
two probabilities are:
P(C
1
C
2
C
3
) = (.9)(.8)(.95) = 0.684.
P(C
1
C
2
C
c
3
) = (.9)(.8)(.05) = 0.036.
In the last equation we are using the property that independence of events
also implies the independence of their complementary events. The third
probability of interest is
P(C
1
C
3
) = P(C
1
) +P(C
3
) P(C
1
C
3
)
= .9 +.95 (.9)(.95) = 0.995.
Problem 10 Two-Stage Experiments The outcome of an event A in-
volves a rst stage in which one of the events B
1
, . . . , B
n
leads to the event
A at the second stage. Compute the probability of A by one of the following
formulas
16 2 Conditional Probabilities and Independent Events
P(A) =
n

k=1
P(A B
k
),
P(A) =
n

k=1
P(A|B
k
)P(B
k
).
The choice of formula would depend on what probabilities are specied (or
easily obtainable).
Example 10.1 The probabilities that a student in this class (such as you)
will expend a high, medium or low amount of eort in studying are .50, .30
and .10, respectively. Given that the student expends a high, medium or
low amount of eort, the respective conditional probabilities of getting an
A in this courses are .90, .40 and .05. Find the probability that the student
(you) will get an A in the course. Let B
1
, B
2
, B
3
denote the events of the
student expending a high, medium and low amount of eort. Let A denote
the event the student gets an A in the course. Then
P(A) =
3

k=1
P(A|B
k
)P(B
k
) = .90 .50 +.40 .30 +.05 .10 = .585
This is too low for comfort. If I were you, I would study hard and then your
probability of geting an A would be .90. Do it.
Problem 11 Bayes Formula Compute the conditional probability of an
event by Bayes formula (provided this approach is appropriate). Namely, if
events B
1
, . . . , B
n
form a partition of the sample space and you know the
probabilities of these events and the conditional probabilities P(A|B
i
), then
P(B
i
|A) =
P(A|B
i
)P(B
i
)

n
k=1
P(A|B
k
)P(B
k
)
.
One can usually identify when to use Bayes formula since the given
information includes conditional probabilities like P(A|B
i
) and one is asked
to nd a reverse conditional probability P(B
i
|A). An interpretation of
Bayes is that A happens and we want to nd the conditional probability
P(B
i
|A) that B
i
is the cause of A.
Example 11.1 The members of a consulting rm rent cars from 3 rental
companies: 60 percent from company 1, 30 percent from company 2, and
10 percent from company 3. The past statistics show that 9 percent of the
cars from company 1 need a tune-up, 20 percent of the cars from company
2 need a tune-up, and 6 percent of the cars from company 3 need a tune-up.
If a rental car delivered to the rm needs a tune-up, what is the probability
that it came from company 2?
Solution. Let A denote the event that the car needs a tune-up, and B
1
,
2 Conditional Probabilities and Independent Events 17
B
2
, and B
3
, respectively, denote the events that the car comes from rental
companies 1, 2, or 3. Then our assumptions are that
P(B
1
) = 0.6, P(B
2
) = 0.3, P(B
3
) = 0.1,
P(A|B
1
) = 0.09, P(A|B
2
) = 0.2, P(A|B
3
) = 0.06.
Therefore, by Bayes formula, we have
P(B
2
|A) =
(0.3)(0.2)
(0.6)(0.09) + (0.3)(0.2) + (0.1)(0.06)
= 0.5.
3
Random Variables and Their Distribution
Functions
Up to this point, we have focused on stating events in words and comput-
ing their probabilities by considering these events as subsets of the sample
space of all outcomes of an experiment. This is a micro-level description
of experiments, which is often not needed when focusing on measurable
quantities. Hereafter, we will consider events stated in terms of random
quantities called random variables. The probabilities associated with these
random variables will be represented by functions called distribution func-
tions. This is a higher-level method of describing randomness that does
not require the micro-level notation associated with a sample space. This
style of using random variables and distribution functions is standard for
technical reports and articles involving probability and statistics.
3.1 Discrete Random Variables
Problem 12 Dening Random Variables Given an experiment and a
random variable X of interest (a real-valued quantity), specify the random
variable X() as a function of the elementary outcomes . When X is a
discrete random variable and the probabilities P() are known (or you can
derive them), obtain the probability function
p(x) = P(X = x) =

:X()=x
P().
This just involves summing the probabilities of all the elementary outcomes
for which the random variable equals the exact value x.
There are experiments in which a random variable of interest is the same
as the outcome of the experiment. Here is an example.
Example 12.1 A trucking company is monitoring the amount of gasoline
that each of its trucks uses. A quantity of interest for a particular truck
20 3 Random Variables and Their Distribution Functions
is the number of gallons X it uses on a particular 200 mile trip. Then for
the experiment that records the gasoline consumption on this trip in integer
values, a sample space of outcomes might be = {1, 2, . . . , 50}. This tacitly
assumes that no more than 50 gallons is ever needed. Note that, for any
outcome , the random variable X() is the same as (e.g. X(42) = 42).
In other words, X is the identity function on . In such a case, one usually
denotes a generic outcome by the letter x. Then the probability function
is simply
p(x) = P(X = x) = P(x), x = 1, 2, . . . , 50.
That is, the probability function p(x) of X is the same as the probability
measure P(x) no extra work is needed to obtain p(x). The assumptions
in this experiment did not mention anything about the probability measure
P on the outcomes. One way of getting this would be to use the frequency
approach, provided that one could repeat the trip about 20 times under
essentially the same conditions.
Quite often, a random variable is not the same as the outcome of an
experiment, and there may be more than one random variable associated
with the experiment. In these cases, a little thought is needed to compute
the probability function P(X = x).
Example 12.2 A salesperson wants to assess future sales of a slow-moving
item in the next two months. In the rst month the probabilities of selling
1, 2, 3, 4 units of the item are .30, .35, .15, .20, respectively. In the second
month, the probabilities of selling 0, 1, 2, 3 units are .25, .30, .40, .05,
respectively, and the sales in one month have no bearing on the sales in the
other month. The salesperson is interested in the total number of sales in the
two months. Describe this as a random variable and specify its probability
function.
Solution. A typical outcome s of this random experiment is a pair, such as
(3, 1), that represents the sales of 3 and 1 units in the respective months.
For this outcome, the total number of sales in the two months is 4, and so
X((3, 1)) = 4. The probability of this outcome is P((3, 1)) = (.15)(.30) =
0.045, which follows by the independence of the two events of selling 3 and
1 units in the two months. The values of X() and P() for all outcomes
are listed below:
Outcome X() P() Outcome X() P()
(1, 0) 1 .075 (3, 0) 3 .0375
(1, 1) 2 .09 (3, 1) 4 .045
(1, 2) 3 .12 (3, 2) 5 .060
(1, 3) 4 .015 (3, 3) 6 .0075
(2, 0) 2 .0875 (4, 0) 4 .050
(2, 1) 3 .105 (4, 1) 5 .06
(2, 2) 4 .14 (4, 2) 6 .080
(2, 3) 5 .0175 (4, 3) 7 .01
3.1 Discrete Random Variables 21
From this table, we see that the possible values of X are 1, . . . , 7. Since
we know the probabilities of the elementary outcomes, we can compute the
probability function of X by
p(x) =

:X()=x
P(), x = 1, . . . , 7,
This function is as follows:
x 1 2 3 4 5 6 7
p(x) .075 .1775 .2625 .25 .1375 .0875 .01
It is common to have several random variables of interest associated
with an experiment. For instance, in the last example, one might have been
interested in the numbers X
1
, X
2
of units sold in the two months, or in
the random variable Y that represents the minimum of X
1
and X
2
. The
probability functions of these can be obtained by the procedure we used for
X (do this as an exercise).
The preceding example shows how to dene random variables from in-
formation of an experiment. This micro-level example is instructive for un-
derstanding random variables, but for practical problems this level of detail
is not needed. The standard approach is to start at the macro-level with
probability functions or distributions and not bother with the underlying
probability space. We rst discuss discrete random variables and later dis-
cuss continuous random variables.
Problem 13 Probabilities of Discrete Random Variables Compute
probabilities of interest for a discrete random variable X from its probability
function p(x) = P(X = x) or its distribution F(x) = P(X x). The basic
formulas are
P(X B) =

xB
p(x)
P(a < X b) = F(b) F(a).
Example 13.1 Suppose the number of emergency 911 telephone calls be-
tween 3 AM and 4 AM to a certain center is a discrete random variable X
with probability function
P(X = x) = e
3
3
x
/x!, x = 0, 1, 2, . . .
This is a Poisson probability function, which we will study in depth later.
The probability of X taking on a single value is obtained by evaluating the
preceding function at that value. For instance,
P(X = 0) = e
3
= .049, P(X = 3) = e
3
3
3
/3! = .224
For more general probabilities one would like to use the distribution func-
tion
22 3 Random Variables and Their Distribution Functions
F(x) = P(X x) =
x

n=0
e
3
3
n
/n!.
In this case, this sum does not reduce to a simple formula, and so we
will compute probabilities using the probability function P(X = x). For
instance,
P(2 < X 5) =
5

x=3
e
3
3
x
/x! = .797
P(2 X 5) =
5

x=2
e
3
3
x
/x! = .977.
Note that x = 2 is not included in the rst sum.
One may be interested in the probability that X is greater than some
value x. This is given by the innite summation
P(X > x) =
5

n=x+1
e
3
3
n
/n!.
However, it is easier to compute this via the complement formula (a nite
summation)
P(X > x) = 1 P(X x) =
x

n=0
e
3
3
n
/n!.
For instance,
P(X 1) = 1 P(X = 0) = 1 e
3
= .951
Also, once some of these probabilities are known, one can compute related
ones, e.g., knowing P(2 < X 5) = .797, we have
P(2 < X or X > 5) = 1 P(2 X 5) = .203.
In addition, one can compute conditional probabilities just as we did
for events. For instance,
P(X 7|X 3) = P(X 7, X 3)/P(X 3)
= P(3 X 7)/P(X 3)
=
_
7

k=3
e
3
3
k
/k!
_
/
_
1
2

k=0
e
3
3
k
/k!
_
= .979
P(X 7|X 3) = P(X 7, X 3)/P(X 3) = 0.
In many instances, the probability function and probability distribution of
a discrete random variable are not represented by formulas, but are given by
tables. Here is an example of computing probabilities for a random variable
from a table of its distribution function.
3.2 Continuous Random Variables 23
Example 13.2 Consider the random variable X that represents the num-
ber of people who are hospitalized or die in a single head-on collision on an
interstate highway in a year. The distribution of such random variables are
typically obtained from historical data. Without getting into the statistical
aspects involved, let us suppose that the distribution of X is as follows.
x 0 1 2 3 4 5 6 7 8
F(x) .250 .580 .784 .922 .972 .985 .991 .998 1.000
Probabilities of X can be obtained by simply reading them from this
table. For instance
P(X = 0) = F(0) = .25
P(X > 2) = 1 F(2) = 1 .784 = .216
P(3 X 6) = P(2 < X 6) = F(6) F(2)
= .991 .784 = .207.
Note that the last probability requires a little more care since we are dealing
with the cumulative distribution F(x). A typical conditional probability is
P(X 5|X > 2) = P(2 < X 5)/P(X > 2) =
F(5) F(2)
1 F(1)
= .452.
If the probabilities for all values of X were of interest, one would evaluate
the probability function
p(x) = P(X = x) = F(x) F(x 1), x = 1, . . . , 8.
A table of this function is as follows.
x 0 1 2 3 4 5 6 7 8
p(x) .250 .330 .204 .138 .050 .013 .006 .007 .002
3.2 Continuous Random Variables
In many instances, one is interested in random quantities such as time,
distance, pressure, volume that take values in some interval of the real
numbers. Such continuous quantities are the focus of this section.
A random variable X is continuous if there is a nonnegative function
f(x), called the density of X, such that
P(a X b) =
_
b
a
f(x) dx, for any a b. (3.1)
The density f must necessarily satisfy
24 3 Random Variables and Their Distribution Functions
_

f(x) dx = 1.
Note that the distribution F(x) = P(X x) of X can be written as
F(x) =
_
x

f(y) dy, for any x (, ). (3.2)

From equations (3.1) and (3.2), we have
P(a X b) = F(b) F(a). (3.3)
Taking the derivative of both sides of (3.2), we see that
F

(x) = f(x).
An important property of a continuous random variable X is that the prob-
ability that it equals a specic value x is 0. That is,
P(X = x) = P(x X x) =
_
x
x
f(y) dy = 0.
Consequently,
P(a < X b) = P(a X b) = P(a < X < b) = P(a X < b).
Problem 14 Probabilities of Continuous Random Variables Com-
pute probabilities of interest for a continuous random variable X from its
density function f(x) by formula (3.1), or from its distribution function
F(x) by formula (3.3). If f(x) is known, compute F by (3.2). If F(x) is
known, compute f by f(x) = F

(x).
Example 14.1 Suppose X is a continuous random variable with the den-
sity function
f(x) = ce
(3x)
, 3 x 6,
and f(x) = 0 elsewhere. Find the following:
(a) The value of c. (b) P(X 5). (c) P(X 3).
(d) P(4 X 5.5). (e) P(4 < X 7). (f) P(X 5|X 3.5).
Solution. (a) To be a density function, f should satisfy
_

f(x) dx = 1.
In this case,
1 =
_
6
3
ce
(3x)
dx = ce
3
_
6
3
e
x
dx = c(1 e
3
).
Therefore,
c = (1 e
3
)
1
= 1.0524.
The probabilities in (b)(f) can be determined by the probability distri-
bution of X, and so we will determine it rst. Clearly, F(x) = 0 for x < 3,
F(x) = 1 for x > 6, and
3.3 Percentiles of Distributions 25
F(x) = 1.0524
_
x
3
e
(3y)
dy = 1.0524[1 e
(3x)
], 3 x 6.
Then the probabilities in (b)(f) are as follows.
P(X 5) = F(5) = 1.0524(1 e
2
) = .91
P(X 3) = F(3) = 0
P(4 X 5.5) = F(5.5) F(4) = 1.0524[e
1
e
2.5
] = .301
P(4 < X 7) = F(7) F(4) = 1 F(4) = .335
P(X 5|X > 4) = P(4 < X 5)/P(X > 4)
=
F(5) F(4)
1 F(4)
= .7305.
3.3 Percentiles of Distributions
The problems above concern nding probabilities that a random variable
takes on certain values (e.g., nd the probability that a certain tennis
players serve will be over 125 miles per hour). We now consider the re-
verse problem of nding values of a random variable that are obtainable
with a certain probability. For example, nd the speed at which the tennis
player can serve at least 90% of the time. That is, if X is the speed of the
players serve, then the speed x at which the tennis player can serve at least
90% of the time is the smallest value of x such that
F(x) = P(X x) .90.
This value is called the 90th percentile of X (or of F).
In general, the (100p)th percentile of a random variable X (or of its
distribution) is the smallest value of x such that
F(x) = P(X x) p. (3.4)
When X is continuous, this inequality reduces to an equality because F(x) is
continuous in x. The 50th percentile is called the median of X. In addition
to the median, one may be interested in the 75th, 90th, 95th, or 99th
percentiles.
Problem 15 Finding Percentiles of a Distribution For a random
variable with a given probability distribution, nd its (100p)th percentile
by nding the smallest x that satises (3.4).
Example 15.1 Consider the random variable X in Example 13.2 repre-
senting the number of hospitalized or dead persons in a head-on collision.
From the table of its distribution, one can see, for example, that its 90th
percentile is 3. (Of course, 3 is also the 91th or 92th percentile.) Similarly,
the 99th percentile is 6.
26 3 Random Variables and Their Distribution Functions
Example 15.2 Consider the continuous random variable X in Exam-
ple 14.1 with distribution
F(x) =
1 e
(3x)
1 e
3
3 x 6.
Find its 90th percentile.
Solution. Since X is continuous, its 90th percentile is the value of x that
satises
F(x) =
1 e
(3x)
1 e
3
= .90.
Solving this equation by a hand calculator yields x = 4.93. Alternatively,
the equation, after a few manipulations, is the same as
e
(3x)
= 1 .9(1 e
3
).
Taking the natural logarithm of both sides of this equation yields
3 x = ln(1 .9(1 e
3
)).
Therefore, the 90th percentile is
x = 3 ln(.1 +.9e
3
) = 4.93.
Note that the same derivation applies with .9 replaced by a general
probability p. Therefore,
The (100p)th percentile of X = 3 ln(1 p +pe
3
).
In particular, for p = .5, we get
Median of X = 3 ln(.5 +.5e
3
) = 3.64.
3.4 Means and Variances of Distributions
The probability distribution of a random variable contains all the detailed
information about it. For some purposes, however, it is adequate to sum-
marize this information by the mean and variance of the random variable.
The mean or expected value of a random variable X is dened by
EX =

x
xp(x) if X is discrete
EX =
_

xf(x) dx if X is continuous.
It is sometimes customary to denote the mean EX by the greek letter .
The mean is viewed as a one-number macro measure of X. Its importance
3.4 Means and Variances of Distributions 27
stems from the following property. Suppose that one repeats the experiment
associated with X several times, say n times (n 25 would be preferable),
and observes the values x
1
, . . . , x
n
for X. Then the average of these values
is
1
n
n

k=1
x
k
.
This sample average of X would be very close to the mean of X. This is
why the word average is sometimes used loosely in referring to the mean.
For instance, suppose that X denotes the number of airline tickets that
a certain agent sells in one day. During the last 10 days, the number of
tickets the agent sold was 6, 5, 7, 2, 4, 6, 7, 3, 4, 4. Then the average number
sold in these n = 10 days is
1
10
10

k=1
x
k
= 4.8.
This value would be close to EX. Note that this information was obtained
without knowing anything about the distribution of X.
Suppose that X is a random variable with mean = EX. The variance
of X is dened by
VarX =

x
(x )
2
p(x), if X is discrete
VarX =
_

(x )
2
f(x) dx, if X is continuous
The standard convention is to write
2
= VarX. The square root of the
variance is called the standard deviation. The variance is a macro mea-
sure of the spread of the probability distribution away from the mean.
Alternative formulas that are easier for computations are
VarX =

x
x
2
p(x)
2
, if X is discrete
VarX =
_

x
2
f(x) dx
2
, if X is continuous
In physics or mechanics, the mean is called the rst moment of the
distribution about the value 0, and the variance is the second moment of
the distribution about . We will consider means and variances only when
they are nite (there are random variables whose means and variances are
innite or dont exist).
Problem 16 Mean and Variance of a Random Variable Find the
mean and variance of a random variable from its probability function or
density by the formulas above.
28 3 Random Variables and Their Distribution Functions
The next two examples show that nding a mean and variance is rather
mechanical in that it only involves computing sums or integrals.
Example 16.1 Consider the number of hurricanes X that hit the east
coast of the United States in one year. Suppose that its probability function
is as follows.
x 0 1 2 3 4 5 6 7
p(x) .05 .15 .22 .26 .14 .08 .07 .03
The expected number of hurricanes in a year is
EX = 0(.05) + 1(.15) + 2(.22) + 3(.26) + 4(.14)
+5(.08) + 6(.07) + 7(.03) = 2.96.
Thus, one would expect about 3 hurricanes per year. For the variance,

x
x
2
p(x) = 0
2
(.05) + 1
2
(.15) + 2
2
(.22) + 3
2
(.26) + 4
2
(.14)
+5
2
(.08) + 6
2
(.07) + 7
2
(.03) = 11.39.
Then
VarX = 11.39 (2.96)
2
= 4.154.
Example 16.2 Suppose that X is a continuous random variable with den-
sity function
f(x) = 3x
2
, 0 x 1,
and f(x) = 0 elsewhere. Its expected value is
EX =
_

xf(x) dx = 3
_
1
0
x
3
dx = 3/4.
Its variance is
VarX =
_

x
2
f(x) dx
2
= 3
_
1
0
x
4
dx (3/4)
2
= 3/80.
We now discuss an important practical use of the variance. Here is a
motivating problem. Suppose that the lifetime of an automobile battery is
a random variable X with an unknown probability distribution, but it is
known that its mean is = 5 years and its standard deviation is = .2
years Using only and , can we nd a value v such that at least 90%
of battery lifetimes will be in the time interval [ v, + v]? That is,
P( v X + v) .90. In other words, the lifetime of a typical
battery is 5 years plus or minus v years, with a 90% assurance.
We will solve such problems by using the following Chebyshev inequality:
P( k X +k) 1 k
2
.
3.4 Means and Variances of Distributions 29
Here k is any positive quantity. This inequality justies the following 2k-
sigma rule:
The probability that the random variable X will fall in the interval [
k, +k] (or X will be within k of the mean) is at least p = 1 k
2
.
An important case is k = 3 and p = 1 3
2
= .89. This yields the
6-sigma rule, which says that at least 89 percent (essentially 90 percent) of
the time, X will fall in the interval [ 3, +3]. This rule is frequently
used in quality control and in total quality management.
Another nice case is k = 2 and p = 1 2
2
= .75. This yields the
4-sigma rule, which says that at least 75 percent of the time, X will fall in
[ 2, + 2].
Problem 17 Chebyshev Inequality and the 2k Rule Suppose that
we do not know the distribution of a random variable X, but we do know
its mean and standard deviation . Find a value k such that the random
variable X will fall in the interval [k, +k] with a probability at least
p. The 2k-sigma rule says the solution is k = 1/

1 p.
Example 17.1 A stock broker tells a client that the price of a certain
stock next week is expected to be $55 per share, with a standard deviation
of $1.5. However, the broker does not know the form of the distribution.
Find an interval centered at the mean in which the stock price will fall with
about a 80 percent chance.
Solution. The 2k-sigma rule says to set k = 1/

.20 = 2.24. Then there is

at least a 80 percent chance the price will fall in the interval
[ k, +k] = [55 3.36, 55 + 3.36, ] = [51.64, 58.36].
The 6-sigma rule can also be used to get a crude approximation for the
variance as follows.
Example 17.2 The time it takes for you to do a task (like the next home-
work set) is a random variable X with a mean of 60 minutes. You are
condent (with about a .9 probability) that you can nish the task be-
tween 50 and 70 minutes. Find a good approximation for the variance of
X.
Solution. Note that the times 50 and 70 were chosen equally distant from
the mean = 60, and our assumption was that X will be in the interval
[50, 70] with at least a 89 percent chance. Then the 6-sigma rule says that
[50, 70] = [60 3, 60 + 3].
This implies that 3 = 10. Thus, = 10/3 would be a good approximation.