You are on page 1of 174

Probability for Computer Scientists

This material is provided for the educational use of students in CSE2400 at FIT. No further use or reproduction is permitted. Copyright G.A.Marin, 2008, All rights reserved.

Applied Statistics: Probability

1-1

Permutations and Combinations


Suppose that we have n objects O1 , O2 ,..., On . A permutation of order k is an "ordered" selection of k of these for 1 k n. A combination of order k is an "unordered" selection of k of these. Common notation: P ( n, k ) or Pkn = n(n 1)
n k

(n k + 1) = n k

n nk C ( n, k ) = C = = . k k! Example: Given the 5 letters a,b,c,d,e how many ways can we list 3 of the 5 when order is important? Answer: P35 =53 =5*4*3=60. Note that each choice of 3 letters (such as a,c,e) results in 6 different results: ace, aec, cae, cea, eac, eca... Example: Given the 5 letters above how many ways can we choose 3 of the 5 when order is NOT important? 5 53 5*4*3 = 10. In this case we have the 60 that result when we care about order Answer: = = 3 3! 3*2*1 divided by 6 (the number of orderings of 3 fixed letters).
Applied Statistics: Probability 1-2

Definition
n k = n (n 1) (n k + 1) for any positive integer n and for integers k such that 1 k n. This symbol is pronouced "n to the k falling." Examples: 6 3 = 6 5 4 = 120. 36 is not defined (for our purposes). 55 = 5! n nk Again: P = n and = . k k!
n k k

Applied Statistics: Probability

1-3

Permutations of Multiple Types


The number of permutations of n = n1 + n2 + n! . n1 !n2 ! nr ! + nr objects of which n1 are of one type, n2 are of a second type, ..., and nr are of an rth type is

Example: Suppose we have 2 red buttons, 3 white buttons, and 4 blue buttons. How many different orderings (permutations) are there? 9! = 1260. 2!3!4!

Answer:

Applied Statistics: Probability

1-4

Try These
There are 12 marbles in an urn. 8 are white and 4 are red. The white marbles are numbered w1,w2,...,w8 and the red ones are numbered r1,r2,r3,r4. For (a) - (d): Without looking into the urn you draw out 5 marbles. (a) How many unique choices can you get if order matters? 12 5 = 95, 040 12 (b) How many unique choices can you get if order does not matter? = 792 5 (c) How many ways can you choose 3 white marbles and 2 red marbles if order matters? You will fill 5 "slots" by drawing. First determine which 5 two slots (positions) will be occupied by 2 red marbles: = 10. Next 2 multiply by orderings of 3 white and 2 red: 10i8 3 i4 2 = 40,320. (d) How many ways can you choose 3 white marbles and 2 red marbles if 8 4 order does not matter? = 336 3 2 (e) How many marbles must you draw to be sure of getting two red ones? 10
Applied Statistics: Probability 1-5

Complex Combinations
How many ways are there to create a full house (3-of-akind plus a pair) using a standard deck of 52 playing cards?

13 4 12 4 = 13i4i12i6 = 3, 744. 1 3 1 2

(choose denomination)x(choose 3 of 4 of given denomination)x(choose one of the remaining denominations)x(choose 2 of 4 of this second denomination).
This follows from the multiplication principle (Theorem 2.3.1 in text).
Applied Statistics: Probability 1-6

Try these
n n Suppose = . What is n ? 11 7 18 18 Suppose = . What is r ? r r 2

Applied Statistics: Probability

1-7

Examples*
Consider a machining operation in which a piece of sheet metal needs two identical diameter holes drilled and two identical size notches cut. We denote a drilling operation as d and a notching operation as n. In determining a schedule for a machine shop, we might be interested in the number of different possible sequences of the four operations. The number of possible sequences for two drilling operations and two notching operations is 4!

2!2!

=6

The six sequences are easily summarized: ddnn, dndn, dnnd, nddn, ndnd, nndd.

*Applied Statistics and Probability for Engineers, Douglas C. Montgomery, George C. Runger, John Wiley & Sons, Inc. 2006
Applied Statistics: Probability 1-8

Example*
A printed circuit board has eight different locations in which a component can be placed. If five identical components are to be placed on the board, how many different designs are possible? Each design is a subset of the eight locations that are to contain the components. The number of possible designs is, therefore,

8 8 83 8*7 *6 = = 56. = = 5 3 3! 3* 2*1


*Applied Statistics and Probability for Engineers, Douglas C. Montgomery, George C. Runger, John Wiley & Sons, Inc. 2006 Applied Statistics: Probability 1-9

Sample Space
Definition: The totality of the possible outcomes of a random experiment is called the Sample Space, .
Finite
Outcome from one roll of one die = {1, 2,3, 4,5, 6} .

Countable
The number of attempts until a message is transmitted successfully when the probability of success on any one attempt is p = {1, 2,3, 4,5, 6,...} =
+

Continuous (We begin with the discrete cases.)


The time (in seconds) until a lightbulb burns out = {t : t 0} , where is the set of all real numbers.

Applied Statistics: Probability

1-10

Events
Definition: An event is a collection of points from the sample space. Example: the result of one throw of die is odd. We use sets to describe events.
From the die example let the set of "even" outcomes be E = {2, 4, 6} . Let the set of "odd" outcomes be O = {1,3,5} .

If is finite or countable, then a simple event is an event that contains only one point from the sample space. For the die example the simple events are S1 = {1} , S 2 = {2} ,..., S6 = {6} . Suppose we toss a coin until first Head appears. What are the simple events? Unless stated otherwise, ALL SUBSETS of a sample space are included as possible events. (Generally we will not be interested in most of these, and many events will have probability zero.)
Applied Statistics: Probability 1-11

Describe the sample space and events


Each of 3 machine parts is classified as either above or below spec.
At least one part is below spec.

An order for an automobile can specify either an automatic or standard transmission, premium or standard stereo, V6 or V8 engine, leather or cloth interior, and colors: red, blue, black, green, white.
Orders have premium stereo, leather interior, and a V8 engine.
Applied Statistics: Probability 1-12

Describe: sample space and events


The number of hours of normal use of a lightbulb.
Lightbulbs that last between 1500 and 1800 hours.

The individual weights of automobiles crossing a bridge measured in tons to nearest hundredth of a ton.
Autos crossing that weigh more than 3,000 pounds.

A message is transmitted repeatedly until transmission is successful.


Those messages transmitted 3 or fewer times.

Applied Statistics: Probability

1-13

Operations on Events
Because the sample space is a set, , and any event is a subset A , we form new events from existing events by using the usual set theory operations.
A B Both A and B occur. A B At least one of A or B occurs. A A does not occur. S A = S A S occurs and A does not occur. the empty set (a set that contains no elements). A B = A and B are "mutually exclusive." A B Every element of A is an element of B, or, if A occurs, B occurs. Review Venn diagrams (in text).

Applied Statistics: Probability

1-14

Example
Four bits are transmitted over a digital communications channel. Each bit is either distorted or received without distortion. Let Ai denote the event that the ith bit is distorted, i = 1, 2,3, 4. (a) Describe the sample space. (b) What is the event A1 ? (c) What is the event A1 A2? (c) What is the event A1 A2? (d) What is the event A1 ?
Applied Statistics: Probability 1-15

Venn Diagrams Identify the following events:


A B

(a ) A (b) A B (c ) ( A B ) C (d )

( B C ) ( A B ) C

(e)

Applied Statistics: Probability

1-16

Mutually Exclusive & Collectively Exhaustive


A collection of events A1 , A2 ,... is said to be mutually exclusive if

Ai Aj =
i

if i j

Ai = Aj if i = j.

A collection of events is collectively exhaustive if Ai = .

A collection of events forms a partition of if they are mutually exclusive and collectively exhaustive. A collection of mutually exclusive events forms a partition of an event E if A = E.
i i

Applied Statistics: Probability

1-17

Partition of

An 1
An

A1 A2

The sets Ai are "events." No two of them intersect (mutually exclusive) and their union covers the entire sample space.
Applied Statistics: Probability

1-18

Probability measure
We use a probability measure to represent the relative likelihood that a random event will occur. The probability of an event A is denoted P( A). Axioms:
A 1. For every event A , P ( A ) 0. A 2. P ( ) = 1. A 3. If A and B are m utually exclusive, then P (A B )= P (A )+ P (B ). A 4. If the events A1 , A2 , ... are m utually exclusive, then P An = n =1

P(A
n =1

).
Applied Statistics: Probability 1-19

Theorem:
Given a sample space, , a "well-defined" collection of events, F, and a probability measure, P, defined on these events then the following hold: (a) P ( ) = 0. (b) P [ A] = 1 P A , A F. (c) P [ A B ] = P [ A] + P [ B ] P [ A B ] , A, B F. (d) A B P [ A] P [ B ] , A, B F.
You must know these and be able to use them to solve problems. Dont worry about proving them.

Applied Statistics: Probability

1-20

Applying the Theorem


We roll 1 die and obtain one of the numbers 1 through 6 with equal probability. (a) What is the probability that we obtain a 7? The event we want is ; thus, the probability is 0. (b) What is the probability that we do NOT get a 1? 5 The event we want is {1} or {1} , and P {1} = 1 P {1} = . 6 (c) What is the probability that we get a 1 or a 3? 1 1 1 + = . 6 6 3 (d) If E = {1, 4,5, 6} , and E G , what might the event G be? P {1} {3} = P {1} + P {3} = G = {1, 2, 4,5, 6} , G = {1,3, 4,5, 6} , G = {1, 2,3, 4,5, 6} , or G = E. Note that in all of these cases P [ E ] P [G ] .
Applied Statistics: Probability

1-21

Assigning Discrete Probabilities


When there are exactly n possible outcomes of an experiment, x1 , x2 ,..., xn then the assigned probabilities, p ( xi ) , i = 1, 2,...n, must satisfy the following: (1) 0 p ( xi ) 1, i = 1, 2,..., n. (2)

p ( x ) = 1.
i =1 i

1 If all of the outcomes have equal probability, then each p ( xi ) = ; thus, the n 1 probability of any particular outcome on the roll of a fair die is . 6 Suppose, however, we have a biased die and the probability of a 4 is 3 times more likely than the probability of any other outcome. This implies that p (1) = p ( 2 ) = p ( 3) = p ( 5 ) = p ( 6 ) = a (for example) and p ( 4 ) = 3a. 1 1 3 It follows that 8a = 1 a = . Thus, p ( i ) = , i 4, and p (4) = . 8 8 8
Applied Statistics: Probability 1-22

Complex Combinations
How many ways are there to create a full house (3-of-akind plus a pair) using a standard deck of 52 playing cards?

13 4 12 4 = 13i4i12i6 = 3, 744. 1 3 1 2

(choose denomination)x(choose 3 of 4 of given denomination)x(choose one of the remaining denominations)x(choose 2 of 4 of this second denomination).
This follows from the multiplication principle (Theorem 2.3.1 in text).
Applied Statistics: Probability 1-23

What is the probability of a full house?


In discrete problems we interpret probability as a ratio: number of successful outcomes successes . = total number of outcomes. successes + failures In this case the number of successful outcomes is the number of ways to get a full house (3,744). The total number of outcomes is: 52 525 52i51i50i49i48 = = 2,598,960. = 5i4i3i2i1 5 5! 3,744 = 0.00144 Thus, the probability of getting a full house is 2,598,960 This is an example of a hypergeometric distribution; we'll study this soon.
A full house happens about once in every 694 hands! This is why people invented wild cards.
Applied Statistics: Probability 1-24

Conditional Probability
The conditional probability of A given that B has occurred is
P( A B) , provided P ( B ) 0. P( A | B) = P( B)

Q1: What is the probability of obtaining a total of 8 when rolling two dice? Q2: Suppose you roll two dice that you cannot see. Someone tells you that the sum is greater than 6. What is the probability that the sum is 8?
Applied Statistics: Probability 1-25

Dice Problem
Let A be the event of getting 8 on the roll of two dice. Let B be the event that the sum of the two dice is greater than 6. The first question is Find P( A). Here is the sample space: (1,1) (1,2 )(1,3)(1, 4 )(1,5 ) (1, 6 ) (2,1) ( 2,2 )( 2,3)( 2, 4 )( 2,5 )( 2, 6 ) (3,1) ( 3,2 )( 3,3)( 3, 4 )( 3,5 )( 3, 6 ) (4,1) ( 4,2 )( 4,3)( 4, 4 )( 4,5 )( 4, 6 ) (5,1) ( 5,2 )( 5,3)( 5, 4 )( 5,5 )( 5, 6 ) (6,1) ( 6,2 )( 6,3)( 6, 4 )( 6,5 )( 6, 6 )
Sum=8.

P( A) =

5 . 36

because we are assuming that each outcome pair has the same probability, 1/36.

Applied Statistics: Probability

1-26

Dice Problem (conditional)


Here we roll the dice and learn that the sum is greater than 6. Let B represent the event that the sum is greater than 6. With this knowledge the sample space becomes the following:

(1, 6 ) ( 2,5)( 2, 6 ) ( 3, 4 )( 3,5)( 3, 6 ) ( 4,3)( 4, 4 )( 4,5)( 4, 6 ) ( 5,2 )( 5,3)( 5, 4 )( 5,5)( 5, 6 ) (6,1) ( 6,2 )( 6,3)( 6, 4 )( 6,5 )( 6, 6 )

It follows that P( A | B ) =

5 . 21

Alternatively, by definition of conditional probability, we have P( A B ) P( A) P( A | B) = = because A B = A. P( B) P( B) Note well! 5 P ( A) 36 5 = = . Furthermore, 21 21 P( B) Sothe definition makes sense! 36

Applied Statistics: Probability

1-27

Try this.
A university has 600 freshmen, 500 sophomores, and 400 juniors. 80 of the freshmen, 60 of the sophomores, and 50 of the juniors are Computer Science majors. For this problem assume there are NO seniors. What is the probability that a student, selected at random, is a freshman or a CS major (or both)? If a student is a CS major, what is the probability he/she is a sophomore?

Applied Statistics: Probability

1-28

Use these steps to solve previous slide


What is the sample space? 2. What are the events (subsets) of interest? 3. What are the probabilities of the events of interest? 4. What is the answer to the problem?
1.

Applied Statistics: Probability

1-29

Alternate Form
We have seen that the conditional probability of event A given that event A B has occurred is: P( A | B) = P(P()B) , provided P ( B ) 0. B
Clearly this implies that P( A B) = P( A | B) P( B). This is referred to as the "multiplication rule," and holds even when P ( B ) = 0. Notice that we could also write P( A B) = P( B | A) P( A). Both these equations always hold for any two events. But there is a special case where the conditional probabilities above are not needed.
Note: memorize these conditional probability equations TODAY. They are extremely important.

Applied Statistics: Probability

1-30

Independent Events
Two events A and B are independent iff the probability P AB =P APB ( ) ( ) ( ). Example (dice)

Q1: If one die is rolled twice, is the probability of getting a 3 on the first roll independent of the probability of getting a 3 on the second roll? Q2: If one die is rolled twice, is the probability that their sum is greater than 5 independent of the probability that the first roll produces a 1?
Applied Statistics: Probability

1-31

Dice sample spaces (Q1)


The sample space associated with one roll of a die: 1 = 1, 2,3, 4,5, 6. Unless otherwise stated we assume the die is fair so that the probability of 1 any one of the simple events is . The sample space associated with two 6 rolls of one die (or with one roll of a pair of dice): (1,1) (1,2 )(1,3)(1, 4 )(1,5 )(1, 6 ) (2,1) ( 2,2 )( 2,3)( 2, 4 )( 2,5 )( 2, 6 ) (3,1) ( 3,2 )( 3,3)( 3, 4 ) ( 3, 5 )( 3, 6 ) (4,1) ( 4,2 )( 4,3)( 4, 4 )( 4,5 )( 4, 6 ) (5,1) ( 5,2 )( 5,3)( 5, 4 )( 5,5 )( 5, 6 ) (6,1) ( 6,2 )( 6,3)( 6, 4 )( 6,5 )( 6, 6 ) 1 1 Clearly P(3,3) = . P(3 on first roll) = P(3 on second roll) = . 36 6 1 1 1 Because = , the two events are independent. 6 6 36
Applied Statistics: Probability 1-32

Dice: Q2
The probability of getting a 1 on the first die is 1 . Let G5 be the event that 6 the sum of the two dice is greater than 5 and F1 be the event that the first roll

produces a 1. The sample space is:


(1,1) (1,2 ) (1,3)(1, 4 )(1,5 )(1, 6 ) (2,1) ( 2,2 )( 2,3)( 2, 4 )( 2,5 )( 2, 6 ) (3,1) ( 3,2 )( 3,3)( 3, 4 )( 3,5 )( 3, 6 ) (4,1) ( 4,2 )( 4,3)( 4, 4 )( 4,5 )( 4, 6 ) (5,1) ( 5,2 )( 5,3)( 5, 4 )( 5,5 )( 5, 6 ) (6,1) ( 6,2 )( 6,3)( 6, 4 )( 6,5 )( 6, 6 )

F1 G 5

P [ F 1 G 5] =

2 1 = . 36 18 26 13 P [ G 5] = = . 36 18 1 P [ F1] = . 6

P [ F 1 G 5] =

1 1 13 13 P [ F1] P [G5] = = . 18 6 18 108

Thus, these two events are NOT independent.


Applied Statistics: Probability 1-33

Practice Quiz 1 Explain your work as you have been taught in class.
1. A university has 600 freshmen, 500 sophomores, and 400 juniors. 80 of the freshmen, 60 of the sophomores, and 50 of the juniors are Computer Science majors. For this problem assume there are NO seniors. If a student is a CS major, what is the probability that he/she is a Junior?
i

1 2. Evaluate . i =3 4

3. What is the probability of drawing 2 pairs in a draw of 5 cards from a standard deck of 52 cards? (A pair is two cards of the same denomination such as two aces, two sixes, or two kings.)

Applied Statistics: Probability

1-34

Multiplication and Total Probability Rules*

Multiplication Rule

*This slide from Applied Statistics and Probability for Engineers,3rd Ed ,by Douglas C. Montgomery and George C. Runger, John Wiley & Sons, Inc. 2006
Applied Statistics: Probability 1-35

Multiplication and Total Probability Rules*

*This slide from Applied Statistics and Probability for Engineers,3rd Ed ,by Douglas C. Montgomery and George C. Runger, John Wiley & Songs, Inc. 2006
Applied Statistics: Probability

1-36

Multiplication and Total Probability Rules*


Total Probability Rule

Partitioning an event into two mutually exclusive subsets.

Partitioning an event into several mutually exclusive subsets.

*This slide from Applied Statistics and Probability for Engineers,3rd Ed ,by Douglas C. Montgomery and George C. Runger, John Wiley & Sons, Inc. 2006
Applied Statistics: Probability 1-37

Problem 2-97a
A batch of 25 injection-molded parts contains 5 that have suffered excessive shrinkage. If two parts are selected at random, and without replacement, what is the probability that the second part selected is one with excessive shrinkage?
S={pairs (f,s) of first-selected, secondselected taken from 25 total with 5 defects} SD={second selected (no replace) is a defect} FD={first selected is a defect} FN={first selected is not a defect}.

Applied Statistics: Probability

1-38

Problem Solution
We seek P[SD]=P[SD FD]+P[SD FN] This becomes P[ SD | FN ]P[ FN ] + P[ SD | FD]P[ FD]
5 4 4 5 1 1 1 = * + * = + = = 0.2. 24 5 24 25 6 30 5

Applied Statistics: Probability

1-39

Multiplication and Total Probability Rules*

Total Probability Rule (multiple events)

*This slide from Applied Statistics and Probability for Engineers,3rd Ed ,by Douglas C. Montgomery and George C. Runger, John Wiley & Sons, Inc. 2006
Applied Statistics: Probability 1-40

Total Probability Example


A semiconductor manufacturer has the following data regarding the effect of contaminants on the probability that chips fail.

Probability of Failure 0.1 0.01 0.001

Level of Contamination High Medium Low

In a particular production run 20% of the chips have high-level, 30% have medium-level, and 50% have low-level contamination. What is the probability that one of the resulting chips fails?

Applied Statistics: Probability

1-41

Bernoulli Trials
Consider an experiment that has two possible outcomes, success and failure. Let the probability of success be p and the probability of failure be q where p+q=1. Now consider the compound experiment consisting of a sequence of n independent repetitions of this experiment. Such a sequence is known as a sequence of Bernoulli Trials. The probability of obtaining exactly k successes in a sequence of n Bernoulli trials is the binomial probability

p(k ) = ( n ) p k q n k . k

Note that the sum of the probabilities

p(k ) = 1. Thus they


k

are said to form a probability distribution.


Applied Statistics: Probability 1-42

When we take a discrete or countable sample space = {s1 , s2 ,...} and assign probabilities to each of the possible simple events: P({s1}) = p1 , P({s2 }) = p2 ,..., we have created a probability distribution. (Think that you have "distributed" all of the probability over all possible events.) As an example, if I toss a coin 1 1 one time then P( H ) = and P (T ) = represents a probability distribution. 2 2 The single coin toss distribution also is an example of a Bernoulli trial because it has only two possible outcomes (generally called "success" or "failure).

Probability Distribution

Applied Statistics: Probability

1-43

The binomial probabilities are defined by p( k ) = ( n ) p k q n k , where k

Binomial Probability Distribution

p is the probability of success and q is the probability of failure in n Bernoulli trials. Suppose we toss a coin 10 times and we want the total number of heads. Then p = P [ H ] , q = P [T ] , n = 10. Using the above formula we obtain the probabilities:
3.00E-01 2.50E-01 2.00E-01 1.50E-01 1.00E-01 5.00E-02 0.00E+00 0 1 2 3 4 5 6 7 8 9 10 probabilities

Binomial n=10 p=0.5

Applied Statistics: Probability

1-44

Regarding Parameters
Notice that the binomial distribution is completely defined by the formula for its probabilities, p (k ) = ( n ) p k q n k , and by it "parameters" p and n. k The binomial probability equation never changes so we regard a binomial distribution as being defined by its parameters. This is typical of all probability distributions (using their own parameters, of course). One of the problems we often face in statistics is estimating the parameters after collecting data that we know (or believe) comes from a particular probability distribution (such as the p and n for the binomial). Alternatively, we may choose to estimate "statistics" such as mean and variance that are functions of these parameters. We'll get to this, after we consider random variables and the the continuous sample space.

Applied Statistics: Probability

1-45

Example (from Trivedi*)


Consider a binary communication channel transmitting coded words of n bits each. Assume that the probability of successful transmission of a single bit is p and that the probability of an error is q = 1 p. Assume also that the code is capable of correcting up to e errors, where e 0. If we assume that the transmission of successive bits is independent, then the probability of successful word transmission is: Pw = P [ e or fewer errors in n trials] n i n i = q p . i =0 i Notice that a "success for the Binomial distribution" means getting an error, which has probability q.
e

*Probability and Statistics with Reliability, Queuing and Computer Science Applications, 2nd Ed, Kishor S. Trivedi, J. Wiley & Sons, NY 2002
Applied Statistics: Probability 1-46

Example
A communications network is being shared by 100 workstations. Time is divided into intervals that are 100 ms long. One and only one workstation may transmit during one of these time intervals. When a workstation is ready to transmit, it will wait until the beginning of the next 100ms time interval before attempting to transmit. If more than one workstation is ready at that moment, a collision occurs; and each of the k ready workstations waits a random amount of time before trying again. If k = 1, then transmission is successful. Suppose the probability of a workstation being ready to transmit is p. Show how probability of collision varies as p varies between 0 and 0.1.

Applied Statistics: Probability

1-47

Practice Quiz 2
A partial deck of playing cards (fewer than 52 cards) contains some spades, hearts, diamonds, and clubs (NOT 13 of each suit). If a card is drawn at random, then the probability that it is a spade is 0.2. We write this as P[Spade]=0.2. Similarly, P[Heart]=0.3, P[Diamond]=0.25, P[Club]=0.25. Each of the 4 suits has some number of face cards (King, Queen, Jack). If the drawn card is a spade, the probability is 0.25 that it is a face card. If it is a heart, the probability is 0.25 that it is a face card. If it is a diamond, the probability is 0.2 that it is a face card. If it is a club, the probability is 0.1, that it is a face card.
1. 2. 3.

What is the probability that the randomly drawn card is a face card? What is the probability that the card is a Heart and a face card? If the card is a face card, what is the probability that it is a spade?

Applied Statistics: Probability

1-48

Discrete Random Variables


G. A. Marin

Applied Statistics: Probability

1-49

Review of function
Defn: A function is a set of ordered pairs such that no two pairs have the same first element (unless they also have the same second element). Example: g = (1,2 ) , 3, 5 , ( 5,12 ) defines a function, g , whose "domain" consists of the real numbers 1,3,5 and whose "range" consists of the numbers 2, 5,12. All functions are said to "map" values in their domain to values in their range. Example: f ( x) = x 2 + 5. Here a function is defined using a formula. This actually implies the the function is f =

{( x, x

+ 5 ) : x is a real number .

Notice the following: (a) The function has a "name." Here that name is f . (b) The implied domain of the function includes all real numbers, x, that can be plugged into the formula. In this case that includes all real no's. (c) Every number x in the domain (all reals) is "mapped" to the number x 2 +5. Thus f (1) = 6, f (5) = 30, f ( ) = 2 +5. (d) Sometimes we write this as 1 6, -5 30, 2 +5. (e) The range of f is { x : x 5} .
Applied Statistics: Probability 1-50

Random Variable
Definition: A random variable X on a sample space is a function that assigns a real number x to each sample point s . The inverse image of X ( s) is the set of all points in that the random variable X maps to the value x . It is denoted Ax = {s | X ( s ) = x}.
discrete X discrete continuous X continuous

Applied Statistics: Probability

1-51

Random Variable

= {s1 , s2 , s3 ,...}

= ( , )

We write X ( s ) = x, where s , and x .


We define Ax as the set of all points in that "map" into the value x . Sometimes we write Ax = X ( x ) and state that Ax is the "inverse image" of the value x under the random variable X . For discrete random variables, we then define the probability of the value x to equal P [ Ax ] .

p X ( x) = P [ Ax ] .

OR we may be given a discrete (continuous later) random variable, a description of the values it can produce and the probability of each value. For example, For k = 1, 2,..., n, P [ X = k ] = pk . In this case we need not know what the underlying experiment really is.
Applied Statistics: Probability 1-52

The role of a random variable


Experiment 1: Roll 1 fair die and determine the outcome. Experiment 2: Spin an arrow that lands with equal probability on one of the numbers 1 through 6. Experiment 3: You have 6 cards numbered 1 through 6. Shuffle them and draw one at random. Replace the card and reshuffle to repeat.
Notice that wed represent the sample space of each of these as {1,2,3,4,5,6} usually without drawing dice or arrows or cards, but the sample spaces really include dice, arrows, cards.

For each probability distribution 1 let pi = for i = 1, 2,...6. 6

The importance of the random variable is that it lets us deal with such an experimental setup without thinking dice, arrows, or cards. We say: Let X be a random variable such that takes on the discrete values 1,2,3,4,5,6. Its probability mass function is given as: the probability that X=i is 1/6. We write this as

p X (i ) =

1 for i=1,2,...,6. 6

Applied Statistics: Probability

1-53

Probability Mass Function


If X is a discrete random variable, then its probability mass function (pmf) is given by: p X ( x ) = P ( X = x ) = P ( Ax ) =
s Ax

P ( s ).

The pmf satisfies the following properties: (p1) 0 p X ( x ) 1 for all values, x , such that P[ X = x ] is defined. (p2) If X is a discrete random var iable then

p
i

( xi ) = 1, where the set { x1 , x2 ,...} includes all real

numbers, x , such that p X ( x ) 0.


Note: you cannot define a pmf without first defining a random variable. You can, however, define a probability distribution directly on a sample space with no random variable defined.

Applied Statistics: Probability

1-54

Discrete RV Example 1
Let the sample space represent all possible outcomes of a roll of one die; thus, = {1, 2,3, 4,5, 6} . We define the random variable X on this sample space as 1 follows: X (i ) = 0 if i = 1, 2 . Because the probability of rolling a 1 or 2 if i = 3, 4, 5, 6 if i = 1 . X has if i = 0 1 and 3

1 3 1 is , we define X's probability mass function as p X ( i ) = 3 2 3

a Bernoulli distribution. Alternatively, we could just write that p X (1) = 2 p X ( 0 ) = , or we could define the pmf using a table: 3

pmf of X Value Prob 0 2/3 1 1/3


Applied Statistics: Probability 1-55

Discrete RV Example 2
A die is tossed until the occurrence of the first 6. Let the random variable X = k if the first 6 occurs on the kth roll for integer k > 0. What is the probability mass function (pmf) for X ? In order for the first 6 to occur on the 5th toss, for example, we must have the event AAAA6 occur where A means any result other than 6. Clearly, these represent a sequence of 5 Bernoulli trials where success = 6 and failure = 1 through 5. Each trial is independent; thus, the probability 5 1 625 of this particular result is = = 0.08. Similarly, the probability 6 6 7776 5 of the first 6 on the kth roll is 6 5 p X (k ) = 6
k 1 k 1 4

1 . This defines the pmf, 6

1 . This is a particular instance of the geometric distribution. 6


Applied Statistics: Probability 1-56

Useful die illustrations


(1) Roll a die once and the probability of getting any one number (choose one of six) is 1/6. 1 The uniform distribution for X : p X ( k ) = , k = 1, 2,..., 6. 6 (2) Roll a die n times and count the number of times, k , that you get, say, a 2. This is n 1 5 k 6 6
k nk

n 1 5 . The binomial distribution for X : p X ( k ) = k 6 6


k

nk

, k = 0,1,...n.

(3) Roll a die once, twice, ... until you get, say, a 2 for the first time. Suppose that the first 5 time you get the 2 is on the kth roll. The probability of this is 6 5 distribution for X : p X ( k ) = 6
k 1 k 1

1 . The geometric 6

1 , k = 1, 2,.... 6

Applied Statistics: Probability

1-57

Probability of Sets & Intervals


For a discrete RV, X , and any set of real numbers, A, we can write: P( X A) = pX ( xi ).
xi A

If A = (a, b), we write: P( X A) = P(a<X <b). If A = (a, b], we write: P( X A) = P(a<X b), etc.
For any real number x the probability that the random variable X takes a value in the interval (, x] is especially important and is denoted as: FX (x) = P( < X x) = P( X x) = pX (t), where the last equality holds
t x

only for discrete RVs X . The function F is called the cummulative distribution function (or just the distribution function) of X .
Applied Statistics: Probability

1-58

Simple cdf example


Let X be a random variable with pmf given by: 1 2 5 p X (1) = , p X ( 2 ) = , and p X ( 3) = . 8 8 8 Then the cdf, FX , is given by: 0 1 8 FX ( x) = 3 8 1 for x < 1 for 1 x < 2 for 2 x < 3 for x 3.
NOTICE that F simply adds up the probability mass function's range values as it gets to them (starting from - and moving towards +). F starts at 0 (for a discrete random variable like this one) and adds up the probabilities until it ends at 1. The function F is defined for ALL REAL NUMBERS. (Its domain is all reals.) The range of F is always between 0 and 1. The "meaning" of F is that "FX ( x) is the probability that X x." For a discrete random variable the graph of FX is a step function.

Applied Statistics: Probability

1-59

Cumulative Distribution Function Properties


Important: P(a < X b) = FX (b) FX (a). (F1) 0 FX (x) 1. (F2) F(x) is an increasing function of x. (F3) lim F(x) = 0 and lim F(x) =1.
x- x+

(F4) For discrete X that has positive probability only at the values x1, x2... F has a positive jump at xi equal to pX (xi ) and takes a constant value in the interval [xi1, xi ). Thus, it graphs a step function. Cumulative distribution functions of discrete RVs grow only by jumps, and cumulative distribution functions of continuous RVs have no jumps. A RV is said to be of mixed type if it has continuous intervals plus jumps.
Applied Statistics: Probability 1-60

Bernoulli Distribution
The RV, X, is Bernoulli (or has a Bernoulli distribution) if its pmf is given by p0 = pX (0) = q and p1 = pX (1) = p where p + q =1. The corresponding CDF is given by:

F(x) =

1 for x1.

0 for x<0 q for 0x<1

Example: Roll a die once. Let X=1 if the result is 1 or 2. Let X=0 otherwise. This is a Bernoulli trial with p=1/3 and q=2/3.

Applied Statistics: Probability

1-61

Bernoulli pmf
Bernoulli Distribution p=0.5 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 Bernoulli Distribution p=0.5

Applied Statistics: Probability

1-62

Bernoulli cdf p=0.5

0 for x < 0 Write as: F ( x ) = 0.5 for 0 x < 1 1 otherwise. Notice that the cdf is defined for all real numbers, x.
Applied Statistics: Probability 1-63

Discrete Uniform Distribution


Let X be a random variable that can take any of n values x1 ,x2 ,...,xn 1 . The RV X is said to have a Discrete Uniform n Distribution, and has pmf given by: with equal probability p X ( xi ) =

for i =1,2,..., n 0 otherwise.


1 n

If we let X take on the integer values 1,2,...,n, then its distribution function is given by 0 for x < 1 x 1 x FX ( x) = = for 1 x n n i =1 n 1 for x > n.
Applied Statistics: Probability 1-64

Discrete Uniform pmf n=10


Discrete Uniform pmf
0.12 0.1 0.08 0.06 0.04 0.02 0 1 2 3 4 5 6 7 8 9 10
Discrete Unifrom n=10

Applied Statistics: Probability

1-65

Discrete Uniform cdf n=10


0 for x < 1 x x FX ( x) = p X ( xi ) = for 1 x 10 10 i =1 1 otherwise.

Applied Statistics: Probability

1-66

Binomial Distribution
Let Yn denote the number of successes in n Bernoulli trials. The pmf of Yn is given by: pk = P(Yn = k ) = PYn (k ) =
distribution if 0 for t < 0 t n i P[Yn t ] = FYn (t ) = ( i ) p (1 p)ni for 0 t n Example: Toss a coin 10 i =0 times and count the total 1 for t > 0. number of heads. This is
binomial with p=0.5.
Applied Statistics: Probability 1-67

( )p
n k

(1 p )nk

for 0 k n , k an integer,

otherwise.

The random variable Yn is said to have a binomial

Probability Mass Function

Binomial n=10 p=0.5


3.00E-01 2.50E-01 2.00E-01 1.50E-01 1.00E-01 5.00E-02 0.00E+00 0 1 2 3 4 5 6 7 8 9 10 probabilities

Applied Statistics: Probability

1-68

Binomial cdf n=10 p=0.5

0 for t < 0 t 10 i P[Yn t ] = FYn (t ) = ( i ) p (1 p )10i for 0 t 10 i =0 1 otherwise.


Applied Statistics: Probability 1-69

Geometric Distribution
Consider any arbitrary sequence of Bernoulli trials and let Z be the number of trials up to and including the first success. Z is said to have a geometric distribution with pmf given by pZ (i ) = q i 1 p for i = 1, 2,.... and probabilities p + q = 1 This is p well-defined because pq = = 1. 1 q i =1 The distribution function of Z is given by
i-1

Example: See the previous example concerning rolling 1 die until a 6 occurs.
Applied Statistics: Probability 1-70

0 for t < 1 FZ (t ) = t t p (1 p )i 1 = 1 (1 p ) for t 1. i =1

Geometric pmf p=0.5


Geometric pmf Example
0.6 0.5 0.4 0.3 0.2 0.1 0 1 2 3 4 5 6 7 8 9 10 geom p=0.5

Applied Statistics: Probability

1-71

Geometric cdf p=0.5

0 for t < 1 FZ (t ) = t p (1 p)i 1 = 1 (1 p ) t for t > 1. i =1


Applied Statistics: Probability 1-72

Poisson Distribution
A random variable, X t , has a Poisson Distribution with parameter >0 if its pmf is given by: ( t ) k (e t ) P( X t = k ) = for k = 0,1,.... and t 0. (A distinct RV for each t.) k! NOTE: The Poisson is typically used to model the number of jobs arriving during time t in a time-share system, the arrival of calls at a switchboard, the arrival of messages at a terminal, etc. The parameter is then interpreted as an arrival rate "per unit time." That is, if t is in seconds, then must be the average arrivals per second. (In our text the parameter is given as .) The cumulative distribution function is: 0 for x < 0 FX t ( x ) = x ( t ) k (e t ) for x 0. k! k =0 Notice that in mathematical notation does not typically appear on the left-hand side even though the function is unspecified without it.
Applied Statistics: Probability 1-73

Packet Arrival Example


---- Packet Arrivals ---X1 X2 X3 X4

X1 X2

XN

Each of the random variables

XN

( ) k (e ) has a Poisson distribution; thus, P( X i = k ) = for i = 1, 2,..., n. k!

If Yt represents the total number of arrivals during any time t , then ( t ) k (e t ) P (Yt = k ) = , per the previous slide. k!
Applied Statistics: Probability

1-74

Poisson pmf (x,3,1)


Poisson pmf
0.25 0.2 0.15 Poisson alpha=3 0.1 0.05 0 0 1 2 3 4 5 6 7 8 9 10

Applied Statistics: Probability

1-75

Poisson cdf = 3 and t = 1.

Applied Statistics: Probability

1-76

Poisson Example
Connections arrive at a switch at a rate of 11 per ms. The arrival distribution is Poisson. (a) What is the probability that exactly 11 calls arrive in one ms? (b) What is the probability that exactly 100 calls arrive in 10 ms? (c) What is the probability that the number of calls arriving in 2 ms is greater that 7 and less than or equal to 10?

Let X t be the random variable giving the number of arrivals during t ms. We know that X t has a Poisson distribution, which implies that P [ X t = k ] = arrival rate is 11 ; thus, P [ X t = k ] =

( t )

(11t )

ms

k!

( e ) with t in ms.
11t

k!

(e ) .
t

The

(a) Probability of exactly 11 arrivals in one ms is P [ X 1 = 11] = (b) Probability of 100 calls in 10 ms is P [ X 10 = 100] = (c) P [ 7 < X 2 10] =
k =8 10

(11)
100

11

(1110 )

(e

11!

( e ) = 0.119.
11

1110

(11 2 )

(e

11 2

k!

) = 22

100!

) = 0.025.

e 22

1 22 222 228 794 + + = 22 = 0.003. 8! 9! 10! e 10!


Applied Statistics: Probability 1-77

Summary for Discrete Random Variable, X


To define a pmf when X takes integer values write the following: n k nk "The pmf is p X ( k ) = an expression often involving k , such as ( p ) (1 p ) ." k
This means the probability X = k .

Be sure to specify all possible values of k , such as "for integers k = 1, 2,..., n." Be sure to use the values of other parameters (p, n, ...) that are correct for this particular problem. To define a cdf write the following: 0 x "The cdf is: FX ( x ) = ( the expression for p X ( k ) ) k =1 1
This means the probability X x.

for k < whatever min for 1 x n for k > n


Applied Statistics: Probability 1-78

"

Practice Quiz 3
A college student phones his girlfriend once each night for three nights. The 1 anytime he calls. Suppose that the random 3 variable X equals the number of nights (out of three) that he is able to reach her. 1. What is the pmf for X ? What is the name of X ' s distribution? probability that he reaches her is 2. What is the cdf for X ?

3. Now suppose that this student will phone once each night until the first night that he is able to reach his girlfriend. Let Y be a random variable that equals the number of nights that it takes him to reach her for the first time. (For example, Y = 2 if she doesn't answer the first night but does answer the second night.) What is the pmf for Y ? What is the name of Y ' s distribution?

Applied Statistics: Probability

1-79

Suppose values of X are not integers.


You may have to list each possible value and its probability. For example, suppose that 1 1 1 1 X = with probability , X = 1 with probability , X = with probability . 2 6 3 2 You can define the pmf in a table: This mean the
probability X = x.

x
1 2

p X ( x)
1 6
1 3
1 2
Applied Statistics: Probability 1-80

The cdf for this example


0 1 FX ( x) = 6 1 This means the 2 probability X x. 1 1 for x < 2 1 for x < 1 2 for 1 x < for x .

Applied Statistics: Probability

1-81

Mean and Variance of a Discrete Random Variable


Definition

Working formula: Var ( X ) = E ( X 2 ) E 2 ( X ).


Applied Statistics: Probability 1-82

Mean and Variance of a Discrete Random Variable

Figure 3-5 A probability distribution can be viewed as a loading with the mean equal to the balance point. Parts (a) and (b) illustrate equal means, but Part (a) illustrates a larger variance.
Applied Statistics: Probability 1-83

Mean and Variance of a Discrete Random Variable

Figure 3-6 The probability distribution illustrated in Parts (a) and (b) differ even though they have equal means and equal variances.
Applied Statistics: Probability 1-84

Example 3-11

Applied Statistics: Probability

1-85

Properties of E(X) and Var(X)


If X and Y are discrete random variables and a and b are real numbers, then * E (aX ) = aE ( X ) * E ( X + Y ) = E ( X ) + E (Y ) and E ( aX + bY ) = aE ( X ) + bE (Y ) * Var ( aX ) = a 2Var ( X ) * Var ( X + Y ) = Var ( X ) + Var (Y ), only when X and Y are independent. and Var ( aX + bY ) = a 2Var ( X ) + b 2Var (Y ) , only when X and Y are independent.

Continuing the example from previous slide: E (5 X ) = 5(12.5) = 62.5 Var (5 X ) = 25(1.85) = 46.25
Applied Statistics: Probability

1-86

Expected Value vs Average


Suppose we take 10 playing cards numbered Ace, 2,3,4,5,6,7,8,9,10 and arrange them randomly, face-down, on a table. If we choose one at random (and then replace and reshuffle), it is equally likely that we get any value between 1 and 10. If the random variable V gives the value obtained, then V has a discrete uniform distribution on the integers 1 through 10. If we were asked, "What is the average face value of 1+2+3+4+5+6+7+8+9+10 these 10 cards?", we would compute = 5.5. If we're 10 1 1 1 asked "What is the expected value of V?", we find 1( 10 ) + 2 ( 10 ) + + 10 ( 10 ) = 5.5. In fact, for any random variable with discrete uniform distribution (like our "die") the expected value is the same as the average of all possible values. This is NOT TRUE for other distributions. If we actually perform the experiment by drawing 10 times, we are not likely to get each value exactly once. For example, we might draw 1,5,2,6,8,9,7,2,3,6. The average of these outcomes is 4.9 - NOT 5.5. The more times we repeat the draw, the closer we are likely to get to the expected average of 5.5. So you might think of the expected value of a random variable as the value expected from averaging many outcomes.
Applied Statistics: Probability 1-87

Binomial Mean and Variance


The mean of a binomial distribution with parameter p is
n n k n k nk E ( X ) = k p (1 p ) = k p (1 p ) n k . Let m = k 1 to get k =0 k k =1 k n

n m +1 n n 1 m = (m + 1) = (m + 1) p (1 p ) p m +1 (1 p ) n 1 m ( m + 1)! m =0 m =0 m + 1
n 1 n 1

m +1

(n 1) m m = np p (1 p ) n 1 m = np. m! m=0 Similar work will show that Var(X ) = np(1 p), but there are much easier ways to show this.
n 1

Applied Statistics: Probability

1-88

Exercise
The interactive computer system at Gnu Glue has 20 communication lines to the central computer system. The lines operate independently and the probability that any particular line is in use is 0.6. What is the probability that 10 or more lines are in use? What is the expected number of lines in use? What is the standard deviation of lines in use?

Applied Statistics: Probability

1-89

Binomial Revisited
Recall that the Binomial RV X = X 1 + X 2 + ... + X n where each X i has a Bernoulli distribution and is mutually independent with the others. Because E ( X i ) = p, it follows trivially that E ( X ) = np. Because of independence we can write that Var ( X ) = Var ( X i ) also. Var ( X i ) = E ( X i2 ) E 2 ( X ) = p p 2 = p(1 p ).
i =1 n

It follows simply that Var ( X ) = np (1 p ).

Applied Statistics: Probability

1-90

Poisson Mean and Variance


The mean of a Poisson distribution with parameter > 0 is

k
k =0

k e
k!

= e
k =1

k 1
(k 1)!

= e e = . = e k
k =1

Similarly, E ( X 2 ) = k
k =0

k e 2
k!
k =1

k 1
(k 1)! +

= e [ (k 1) = 2e
k =2

] (k 1)! k=1 (k 1)! + e


k=1

k 1

k 1

k 2
(k 2)!

k 1
(k 1)!

= 2 + .

It follows that Var(X ) = E ( X 2 ) E 2 ( X ) = 2 + 2 = . NOTE : The mean and variance of the Poisson random variable, X t is t (or t ).

Applied Statistics: Probability

1-91

Exercise
Suppose it has been determined that the number of inquiries that arrive per second at the central computer system can be described by a Poisson random variable with an average rate of 10 messages per second. What is the probability that no inquiries arrive in a 1-second period? What is the probability that 15 or fewer inquiries arrive in a 1-second period? What are the mean and variance of the number of arrivals in 1 second?
Applied Statistics: Probability

1-92

Geometric Mean and Variance


E ( X ) = kp (1 p)
k =1 k 1

= p k (1 p ) k 1.
k =1 k

1 1 ( p) = k (1 p ) k 1 = 2 . Write s ( p ) = (1 p ) = . Then s p p k =0 k =1

1 1 Thus, E ( X ) = p 2 = . p p Homework: Use similar technique to show that Var ( X ) = (1 p) . 2 p

Applied Statistics: Probability

1-93

Discrete Uniform Distribution


Let X be a random variable with a discrete uniform distribution on the integers n ( n + 1) n + 1 k 1 n 1, 2,..., n. Then E ( X ) = = k = . = n k =1 2n 2 k =1 n
n

k 2 1 n 2 n ( n + 1)( 2n + 1) ( n + 1)( 2n + 1) = Similarly, E ( X ) = = k = . n k =1 n (6) 6 k =1 n


n 2

( n + 1)( 2n + 1) ( n + 1) Therefore, Var ( X ) =

2 ( n + 1)( 2n + 1) 3 ( n + 1) = 6 4 12 12 ( n + 1)( 4n + 2 3n 3) = ( n + 1)( n 1) = n2 1 . = 12 12 12


2

Example: Let X be uniformly distributed on 1,2,...10. Then E ( X ) = Var ( X ) = 99 33 = . 12 4

11 and 2
1-94

Applied Statistics: Probability

Hypergeometric Distribution
Suppose that a set of n objects includes k objects of type 1 (successes?) and n k objects of type 0 (failures perhaps?). A sample of size m is selected from the n objects "without replacement," where m n (and k n). Let X be the random variable that denotes the number of type 1 objects in the sample. Then X is said to be a hypergeometric random variable and its pdf is given by: k n k i m i for i = max 0, m + k n to min k , m { } { } . pX ( i ) = n m 0 otherwise Values of i (examples): (1) n = 20, k = 5 (type 1), m = 3 (sample size) i = 0,1,...,3 (2) Same as (1) but m = 7 i = 0,1,...5 (3) Same as (1) but m = 17 i = 2,3...,5.
Applied Statistics: Probability 1-95

Text problem 3-101


A company employs 800 men under the age of 55. Suppose that 30% carry a marker on the male chromosome that indicates an increased risk for high blood pressure. (a) If 10 men in the company are tested for the marker in this chromosome, what is the probability that exactly 1 man has the marker? Answer: Notice that this is certainly sampling without replacement. (We don't put the first man back into the pool before we draw the second one.) Let X be the number of men that have the marker in a sample of size 10. X is hypergeo 240 560 1 9 metric. Thus, p X (1) = = 0.12. 800 10
Applied Statistics: Probability

1-96

Text problem 3-101 Continued


(b) If 10 men are tested for the marker, what is the probability that more than 1 has the marker? Answer: Out of 10 the number with the marker can be 0,1,2,...,10 (because 240 560 i 10 i i = 0,1,...,10 a total of 240 have the marker). Thus, p X (i ) = 800 10 0 otherwise. The answer is either p X (i ) or 1 p X (i ) = 0.852.
i=2 i =0 10 1

Applied Statistics: Probability

1-97

Mean and Variance of Hypergeometric


If X is a hypergeometric random variable with parameters n (total objects), k (number of type 1 objects), and m (sample size), then = E ( X ) = mp and

2 = var( X ) = mp (1 p )
objects in the total).

k nm , where p = (the proportion of type 1 n 1 n

240 Example: In the previous problem E ( X ) = 10 = 3 and 800 240 240 800 10 Var(X ) = 10 1 = 2.076. 800 800 799

Applied Statistics: Probability

1-98

Continuous Random Variables and Moments of Random Variables


G. A. Marin For educational purposes only. No further distribution authorized.

Applied Statistics: Probability

1-99

Continuous Cumulative Distribution Function


The CDF FX of a random variable X is defined to be the function FX ( x) = P( X x), < x < . The subscript is dropped if there is no abiguity.

A continuous random variable is characterized by a distribution function that is a continuous function of x for all x . If the distribution function has a derivative at all except, possibly, a finite number of points, then the random variable is said to be absolutely continuous. Example: 0, x < 0 FX ( x) = x, 0 x < 1 1, x 1. This means the
probability X x.
Applied Statistics: Probability 1-100

Properties of CDF
* 0 F ( x) 1, < x < * F ( x) is an increasing function of x. *

lim F ( x) = 0 and lim F ( x) = 1.


x - x +

Applied Statistics: Probability 1-101

Probability Density Function


dF ( x) is called the dx probability density function (pdf) of X . Thus, discrete random variables have a For a continuous (differentiable) random variable, X , f ( x) = probability mass function and continuous random variables have a probability density function. The cumulative distribution function is used in both cases (or in "mixed" cases). We obtain the distribution function from the density function through integration: This means NOTHING P(X x) = F(x) =
-

f (t )dt ,

< x < .

except through integration.

The pdf satisfies the following properties: (1) f ( x) 0 for all x.

(2)

f ( x)dx = 1.

Note: in most of our problems The pdf will be defined piecewise.

Applied Statistics: Probability 1-102

Example
The probability density function f is given as: xk2 for x > 2 f ( x) = 0 otherwise. What is the value of k ? What is the corresponding cdf?

Applied Statistics: Probability 1-103

Probabilities on Intervals and cdf


Suppose X is a continuous RV with pdf given by f and cdf given by F . This implies that for any real number x, P [ X x ] = F ( x) = implies that P [ a X b ] = F (b) F (a ) P [ a<X b ] = F (b) F (a ) P [ a X < b ] = F (b) F (a ) P [ a < X < b ] = F (b) F (a ). All 4 cases hold because, for a continuous random variable X , P [ X = a ] = P [ X = b ] = 0. That is, the probability of any particular value is zero in this case. Note that it is traditional to write F ( x) = P [ X x ] . If the distribution is continuous, it is also true that F ( x) = P [ X < x ] .
Applied Statistics: Probability 1-104

f (t )dt. It also

Exponential Distribution
A random variable has an exponential distribution if for some >0 its distribution function is given by: 1 e x , if 0 x < F ( x) = otherwise. 0 It follows that its pdf is given by: e x , if x 0 f ( x) = otherwise. 0
Note that in most problems the parameter represents a "rate," such as a rate of arrivals or a rate of failures.

Examples of use: Interarrival times at a communication switch Service times at a server Time to failure or repair of a component.
Applied Statistics: Probability 1-105

Exponential pdf = 2

e x , if x 0 f ( x) = otherwise. 0
Note the values of pdf mean nothing except through integration.
Applied Statistics: Probability 1-106

Exponential cdf

=2

1 e x , if 0 x < F ( x) = otherwise. 0
Applied Statistics: Probability 1-107

Class Problem
Suppose that we stand at a mile marker on I-4 and watch cars pass. We notice that on the average 10 cars pass by us per minute and we're given that the time lapse between two consecutive cars has an exponential distribution. If we begin timing at the moment that one car passes by, what is the probability that we will have to wait more than 20 secs for the next car to pass?
Note 20 sec = 1/3 min.

1 1 Answer. Let W be waiting time in minutes. We seek P W > = 1 F ( ), 3 3 where F (t ) = 1 e t , is the exponential cdf. The average "rate" is =10; 1 10 thus, the answer is P W > = e 3 = 0.036 3

Applied Statistics: Probability 1-108

Simple Exercises
Use F in the previous problem to write:
Probability W<6 Probability W>6 Probability W<0 Probability W<-1 Probability 2<W<5 Probability W=1.

Applied Statistics: Probability 1-109

Memoryless Property
If X has an exponential distribution, x > 0, and t > 0, then we know that P( X x) = e
0 x y t+x

dy and P ( X t + x) =

e y dy.
t+x

P ( X t + x ) ( X > t ) = Thus, P ( X t + x | X > t ) = P( X > t )

e y dy
.

e y dy

e t (1 e x ) = = 1 e x = P( X x). e t This is why we don't replace lightbulbs until they fail. (Would you like it if waiting time at your doctor's office was exponentially distributed?)

Applied Statistics: Probability 1-110

Exponential/Poisson Relationship
Show that the time between adjacent arrivals of a Poisson Process has an exponential distribution.

Hint: If N t denotes the number of arrivals during time t and N t has a Poisson distribution, then the probability that waiting time to the next event is greater than t is P[W > t ], where W is waiting time, and 0 arrivals during time t. P[W > t ] = P[ N t = 0].
| t | |

4 arrivals during time t.

Applied Statistics: Probability 1-111

Properties of Gamma Function


On the next slide we introduce the gamma distributions, which is a family of distributions that includes the exponential distribution. That definition incorporates something called the gamma function, which is defined as ( ) = x 1e x dx, > 0.
0

Using integration by parts one can show ( ) = ( 1) ( 1) for > 1. Because (1) = 1, it follows that (n) = (n -1)(n -1) = ... = (n -1)! when n is a positive integer. Note also that ( 1 ) = . Also, it is well known that 2 ( )

x e

-1 x

dx =

, for > 0 and > 0. We shall refer to the last equation

as the "gamma integration formula."


Applied Statistics: Probability 1-112

Example
Evaluate

x 2 e 4 x dx.

Answer: Recall that

1 x

dx =

( )

In this case -1 = 2 = 3 while =4. It follows that

, for >0 and > 0.

xe

2 4 x

( 3) 2! 1 = . dx = 3 = 4 64 32

Applied Statistics: Probability 1-113

Gamma Distribution
A random variable with pdf given by
t 1e t f (t ) = , > 0, t > 0, > 0 ( )
Is said to have a Gamma distribution with parameters and and we write X GAM( , ). The parameter is called the shape parameter and the parameter is called the scale parameter. For =1 the gamma becomes
identical to the exponential distribution. NOTE: If a sequence of random variables X 1 , X 2 ,..., X k are mutually independent and identically distributed as GAM( , ), then their sum has a GAM( ,k ) distribution.
Applied Statistics: Probability 1-114

Gamma Density:

g ( x, , )
scale, shape

Applied Statistics: Probability 1-115

Practice Quiz 4
The pdf of a random variable, X , is given by: for x < 0 0 3 x fX ( x) = for 0 x 4 64 for x > 4. 0 Answer each of the following and EXPLAIN (show your work). (a ) What is the cdf of X ? (Find it explicitly.) (b) What is P ( X > 2 ) ? (c) What is P( X > 2 | X > 1) ? BONUS: Use the gamma integration formula to evaluate the following integral:

x 3e 2 x dx
0

Applied Statistics: Probability 1-116

Mean and Variance of a Continuous Random Variable


Definition

Applied Statistics: Probability 1-117

Expected Value (continuous example)


0 3 x Let f X ( x ) = 64 0
4

for x < 0 for 0 x 4 By definition for x > 4.

x3 x 5 4 1024 16 E ( X ) = xf X ( x)dx = xi dx = = = . 64 5 64 0 5 64 5 0 x3 x 6 4 4096 32 E ( X 2 ) = x 2 f X ( x) dx = x 2 i dx = = = 64 6 64 0 6 64 3 0 32 16 = 0.427 Var ( X ) = E ( X ) E ( X ) = 3 5


2 2
Applied Statistics: Probability 1-118

Existence of E(X)
Continuing a previous example: Let X be a random variable with pdf given by: x22 for x > 2 f ( x) = 0 otherwise. 2 2 Notice that E ( X ) = x 2 dx = dx = 2 ln x 2 = . x x 2 2 Thus, well-defined random variables may not have finite means. Similarly, a random variable may have a finite mean and not have a finite 2nd moment or a finite kth moment (to be defined).

Applied Statistics: Probability 1-119

Mean and Variance of a Continuous Random Variable


Expected Value of a Function of a Continuous Random Variable

Applied Statistics: Probability 1-120

Other expected values


Let X be a random variable with pdf given by f . Let = E ( X ) = Then (a) E ( X 2 ) =

xf ( x)dx.

( is called the mean of X .)

x 2 f ( x)dx. ( x 2 5 x 2) f ( x)dx.

(b) E ( X 2 + 5 X 2) =

(c) E (sin X ) =

( sin x ) f ( x)dx.

(d) 2 = Var ( X ) = E ( X ) 2 =
Variance of X

( x ) 2 f ( x)dx.
Applied Statistics: Probability 1-121

Try these with/without Mathcad


0 for x < 0 Let X be a random variable with probability density f ( x ) = cos( x) for 0 x . 2 0 for x > 2 (a) Find the cdf of X . (b) Find E ( X ). (c) Find E (sin X ).

Applied Statistics: Probability 1-122

Exponential Mean and Variance


Let X be an exponential random variable with parameter . Then E ( X ) =

and Var( X ) =

Proof. Recall the gamma integration formula:

x 1e x dx =
0

( )

Then E ( X ) =

xf

( x)dx = x e
0 2 2

dx =

(2)

=
2

.
x

Similarly, Var(X ) = E ( X ) and E ( X ) = x e


0

dx =

(3)

Thus, Var(X ) =

Applied Statistics: Probability 1-123

Exponential Example
The lifetime of a particular brand of lightbulb is exponentially distributed. The "mean time to failure" (MTTF) is 2000 hours. If the lightbulb is installed at time t = 0, what is the probability that it fails at time t = 3000 hours? What is the probability that it fails in fewer than 3000 hours?

Applied Statistics: Probability 1-124

Gamma Mean and Variance


Recall that the pdf of a Gamma random variable is given by:

Think scale**

t 1e t f (t ) = , > 0, t > 0 ( ) shape

The expected value (mean) of a Gamma random variable is

The variance is 2 .
Note that when =1, the distribution is exponential.

Applied Statistics: Probability 1-125

Class Exercise
A gamma distribution has a mean of 1.5 and a variance of 0.75. Sketch the pdf.

= 1.5 and 2 = 0.75. First, solve for and . From the first 1.5 = 0.75. equation we get =1.5 . Substitute this in second equation to get 2 This gives =2 which implies =3. This is all we need to obtain pdf values.
We have

0.596

Scale=2 and Shape=3

Gam( x) 0.5

Important: Make sure that you can find the scale and shape parameters from a given mean and variance.

0 0

1 x

3 3
Applied Statistics: Probability 1-126

Continuous Uniform Distribution


A continuous random variable X is said to have a uniform distribution over the interval (a,b) if its density is given by: 1 ,a < x < b f ( x) = b a 0, otherwise. The distribution function is: x<a 0, xa F ( x) = , a x < b, b a x b. 1,

Applied Statistics: Probability 1-127

Continuous Uniform Density on (3,5)

Applied Statistics: Probability 1-128

Continuous Uniform cdf on (3,5)

Applied Statistics: Probability 1-129

Continuous Uniform Mean and Variance


Let X be a random variable having the continuous uniform distribution over the interval (a, b). It's density is, thus,
1 ba f ( x) = 0 b

for a < x < b otherwise. 1 b+a )= . 2 ba

1 E ( X ) = x b a dx = 1 (b 2 a 2 )( 2 a

b3 a 3 . It follows that Similarly, E ( X ) = ba


2 1 3

b3 a 3 b + a (b a ) 2 1 Var ( X ) = 3 . = ba 2 12 Recall that if X has a "discrete" uniform distribution on 1,2,...,n, then


2

n +1 n2 1 E( X ) = and Var (X ) = . 2 12
Applied Statistics: Probability 1-130

Example:
Suppose that the time that it takes to drive from the Orlando airport to FIT is uniformly distributed between one hour and one and a quarter hours. (a) What is the mean driving time? (b) What is the standard deviation of the driving time? (c) What is the probability that it will take less than one hour and 5 minutes to make the trip? (d) 80% of the time the trip will take less than ______ minutes?

Applied Statistics: Probability 1-131

Means and Variances (so far)


Distribution Bernoulli Binomial
Geometric Discrete Uniform

E(X)

Var(X)
p (1 p )

np
p n +1 2 1

np (1 p )
(1 p )
p2

n2 1 12

Poisson Exponential Gamma


Continuous Uniform

(or )
1

(or )
1

a+b 2

2 2 2 (b a ) 12

Applied Statistics: Probability 1-132

Practice Quiz 5
1. Suppose that the driving time between Orlando and Melbourne is uniformly distributed between one hour and one hour plus 15 minutes. (a) What is the variance of the driving time.? (b) Eighty percent of all drivers make the trip in fewer than x minutes. What is x ? 2. The random variable X has an exponential distribution. What is the pdf of X ? What is the mean of X ? What is the variance of X ? (Just write down what we known the mean and variance are. Do not derive them from the definition.) 3. The random variable X has a Binomial distribution and the total number of trials is 30. (a) If the probability of success on a single trial is 0.2, write the pmf for X . (b) If E ( X ) = 2.1, then what is the probability of success on a single trial?

Applied Statistics: Probability 1-133

Normal Distribution
Definition

Applied Statistics: Probability 1-134

Normal Distribution

Figure 4-10 Normal probability density functions for selected values of the parameters and 2.
Applied Statistics: Probability 1-135

Normal Distribution
Some useful results concerning the normal distribution

Applied Statistics: Probability 1-136

Normal Distribution
Definition : Standard Normal

Applied Statistics: Probability 1-137

Normal Distribution
Example 4-11

Figure 4-13 Standard normal probability density function.


Applied Statistics: Probability 1-138

Standard Normal Table


0.00 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 0.5000 0.5398 0.5793 0.6179 0.6554 0.6915 0.7257 0.7580 0.7881 0.8159 0.8413 0.8643 0.8849 0.9032 0.9192 0.9332 0.9452 0.9554 0.9641 0.9713 0.9772 0.9821 0.9861 0.9893 0.9918 0.9938 0.9953 0.9965 0.9974 0.9981 0.9987 0.01 0.5040 0.5438 0.5832 0.6217 0.6591 0.6950 0.7291 0.7611 0.7910 0.8186 0.8438 0.8665 0.8869 0.9049 0.9207 0.9345 0.9463 0.9564 0.9649 0.9719 0.9778 0.9826 0.9864 0.9896 0.9920 0.9940 0.9955 0.9966 0.9975 0.9982 0.9987 0.02 0.5080 0.5478 0.5871 0.6255 0.6628 0.6985 0.7324 0.7642 0.7939 0.8212 0.8461 0.8686 0.8888 0.9066 0.9222 0.9357 0.9474 0.9573 0.9656 0.9726 0.9783 0.9830 0.9868 0.9898 0.9922 0.9941 0.9956 0.9967 0.9976 0.9982 0.9987 0.03 0.5120 0.5517 0.5910 0.6293 0.6664 0.7019 0.7357 0.7673 0.7967 0.8238 0.8485 0.8708 0.8907 0.9082 0.9236 0.9370 0.9484 0.9582 0.9664 0.9732 0.9788 0.9834 0.9871 0.9901 0.9925 0.9943 0.9957 0.9968 0.9977 0.9983 0.9988 0.04 0.5160 0.5557 0.5948 0.6331 0.6700 0.7054 0.7389 0.7704 0.7995 0.8264 0.8508 0.8729 0.8925 0.9099 0.9251 0.9382 0.9495 0.9591 0.9671 0.9738 0.9793 0.9838 0.9875 0.9904 0.9927 0.9945 0.9959 0.9969 0.9977 0.9984 0.9988 0.05 0.5199 0.5596 0.5987 0.6368 0.6736 0.7088 0.7422 0.7734 0.8023 0.8289 0.8531 0.8749 0.8944 0.9115 0.9265 0.9394 0.9505 0.9599 0.9678 0.9744 0.9798 0.9842 0.9878 0.9906 0.9929 0.9946 0.9960 0.9970 0.9978 0.9984 0.9989 0.06 0.5239 0.5636 0.6026 0.6406 0.6772 0.7123 0.7454 0.7764 0.8051 0.8315 0.8554 0.8770 0.8962 0.9131 0.9279 0.9406 0.9515 0.9608 0.9686 0.9750 0.9803 0.9846 0.9881 0.9909 0.9931 0.9948 0.9961 0.9971 0.9979 0.9985 0.07 0.5279 0.5675 0.6064 0.6443 0.6808 0.7157 0.7486 0.7794 0.8078 0.8340 0.8577 0.8790 0.8980 0.9147 0.9292 0.9418 0.9525 0.9616 0.9693 0.9756 0.9808 0.9850 0.9884 0.9911 0.9932 0.9949 0.9962 0.9972 0.9979 0.9985 0.08 0.5319 0.5714 0.6103 0.6480 0.6844 0.7190 0.7517 0.7823 0.8106 0.8365 0.8599 0.8810 0.8997 0.9162 0.9306 0.9429 0.9535 0.9625 0.9699 0.9761 0.9812 0.9854 0.9887 0.9913 0.9934 0.9951 0.9963 0.9973 0.9980 0.9986 0.09 0.5359 0.5753 0.6141 0.6517 0.6879 0.7224 0.7549 0.7852 0.8133 0.8389 0.8621 0.8830 0.9015 0.9177 0.9319 0.9441 0.9545 0.9633 0.9706 0.9767 0.9817 0.9857 0.9890 0.9916 0.9936 0.9952 0.9964 0.9974 0.9981 0.9986

Applied Statistics: Probability 1-139


0.9989 0.9989 0.9990 0.9990

Normal Distribution
Standardizing

Applied Statistics: Probability 1-140

Normal Distribution
To Calculate Probability

Applied Statistics: Probability 1-141

Normal Distribution
Example 4-13

Applied Statistics: Probability 1-142

Normal Distribution
Example 4-14

Note: In MathCad we simply compute:

Mean and standard deviation

Applied Statistics: Probability 1-143

Normal Distribution
Example 4-14 (continued)

Applied Statistics: Probability 1-144

Normal Distribution
Example 4-14 (continued)

Figure 4-16 Determining the value of x to meet a specified probability. Applied Statistics: Probability

1-145

Recall Gamma Function


The gamma function (studied previously) is defined as: ( ) = x 1e x dx, > 0.
0

Using integration by parts one can show ( ) = ( 1) ( 1) for > 1. Because (1) = 1, it follows that (n) = (n-1)(n-1) = ... = (n-1)! Note also that (
1 2

) = .

Also, it is well known that

x e

-1 x

dx =

( )

We shall refer to the last equation as the "gamma integration formula."


Applied Statistics: Probability 1-146

Joint Probability Distributions


When two or more random variables naturally occur together or take values that seem to be related, it is common to consider their joint probability distributions. Suppose, for example, that X and Y are two random variables whose outcomes we wish to consider jointly. In a manner similar to single random variables we consider two cases. (1) X and Y are discrete. Here we use a joint pmf and joint cdf (tbd). (2) X and Y are continuous. Here we use a joint pdf and joint cdf (tbd).
Examples: (1) The location of a ship is given by its latitude and longitude. Suppose you are searching for a ship from its last known position. It is natural to consider the joint distribution of ( X , Y ), the ship's probable position. (2) Similarly an aircraft's position might be predicted using a joint distribution of (X , Y , Z ), where ( X , Y ) is as above and Z is altitude.
Applied Statistics: Probability 1-147

Discrete Random Variables


For discrete random variables X 1 , X 2 ,..., X n their joint probability mass function is given by: p X ( x ) = p X1 , X 2 ,... X n ( x1 , x2 ,..., xn ) = P ( X 1 = x1 , X 2 = x2 ,..., X n = xn ) If only two random variables are involved, we usually denote them as X and Y, instead of X 1 , and X 2 , and we write: p X ,Y ( x, y ) = P ( X = x, Y = y ). Note that the author writes the latter as f X ,Y ( x, y ) = P ( X = x, Y = y ).

Applied Statistics: Probability 1-148

Simple example
The joint distribution of ( X , Y ) is given as follows: 1 p X ,Y (1,1) = , 3 1 1 p X ,Y ( 2,1) = and p X ,Y ( 2, 2 ) = 6 6 1 1 1 p X ,Y ( 3,1) = , p X ,Y ( 3, 2 ) = , and p X ,Y ( 3,3) = . 9 9 9
Do the values of X and Y seem to "influence" each other?

Applied Statistics: Probability 1-149

Joint PMF and Independence


For discrete random variables X 1 , X 2 ,..., X n their joint probability mass function is given by: p X ( x ) = p X1 , X 2 ,... X n ( x1 , x2 ,..., xn ) = P ( X 1 = x1 , X 2 = x2 ,..., X n = xn )

The discrete random variables X 1 , X 2 ,..., X n are said to be mutually independent if their joint pmf can be written as: p X ( x ) = p X1 ( x1 ) p X 2 ( x2 )iii p X n ( xn ).

Applied Statistics: Probability 1-150

Problem:
Joint pmf for X 1 and X 2
X2 =1
X1 = 1 X1 = 2
X2 = 2
X2 = 3

1/12 1/6 1/12

1/6 1/4 1/12

1/12 1/12 0

X1 = 3
Find:

(a) P [ X 1 X 2 is even ] (b) P [ X 1 is odd ] (c) P [ X 1 1.5] (d) P [ X 2 is odd | X 1 is odd ]

Are X 1 and X 2 independent?


Applied Statistics: Probability 1-151

Marginal pmf
If X 1 , X 2 ,..., X n are discrete random variables with joint pmf p X1 , X 2 ,..., X n ( x1 , x2 ,..., xn ) , then p X i ( xi ) =
[ X i = xi ]

p X1 , X 2 ,..., X n ( x1 , x2 ,..., xn ), where the sum is over the points in the

range of ( X 1 , X 2 ,..., X n ) where X i = xi . The function p X i ( xi ) is called the marginal probability mass function for X i . Example: Using the previous slide the marginal pmf for X 1 is given by: 1 p X1 (1) = 3 1 p X1 ( 2 ) = 2 1 p X1 ( 3 ) = . 6 1 3

The notation above is difficult; however, notice that, in the example, to get p X1 (1) = you just sum over all the ( x1 , x2 ) pair values that have 1 as the value for X 1.

Applied Statistics: Probability 1-152

Joint cdf for Discrete RVs


If X 1 , X 2 ,..., X n are discrete random variables, then their joint cumulative distribution function is given by: FX1 , X 2 ,..., X n ( x1 , x2 ,...xn ) = P ( X 1 x1 , X 2 x2 ,..., X n xn ) . If there are only two random variables involved we usually write: FX ,Y ( x, y ) = P ( X x, Y y ) . Returning to the "Simple Example" we find, for example, 2 FX ,Y ( 2,3) = . 3

Applied Statistics: Probability 1-153

Skip Section 5-1.3


We will not cover the material on conditional probability distributions that is in this section of the text.

Applied Statistics: Probability 1-154

Double Integral:
z z=f(x,y)

f ( x, y)dA
R

b n( x)

a m( x)

f ( x, y )dydx

Iterated Integral y Y=n(x)

f(x,y)

Y=m(x) x=a

x,y

x=b

Applied Statistics: Probability 1-155

Example:
Evaluate the double integral

2 xydA where R is the region bounded by the


R
x Answer: 2 xydydx = x 2 ydydx = x y 2| 0 0 0 0 0 0 2 x2 2 x2 2
2

curve y = x 2 and the lines y = 0 and x = 2.


dx

x 6 2 32 = xx dx = | = . 6 0 3 0
4

2 0 0 0 0

Region R
1 x 2 3 3

Computes the volume under the surface z=2xy and above the region R.

x=2

Applied Statistics: Probability 1-156

Another Evaluation Approach


2 xydydx = x 2 ydy dx 0 0 0 0
2 x2 2 x2 x2

The inside integral is simply

2 ydy = y | = x 4 .
2 x2 0 0

Substitute this inside the brackets above to get x6 xx 4 dx = 6 0


2 2 0

32 = . 3

Applied Statistics: Probability 1-157

Evaluate the double integral y = x 2 and y = 2 x.


10 10 8
2

4 x 3 ydxdy where R is the region bounded by


R

6 4 2 0 0 0 0 1 x 2 3 3

2 x

Applied Statistics: Probability 1-158

Extra Practice
Evaluate the double integrals: 1. 2. 3. 4. 5. 6. 3 x 2 y 2 dxdy for R bounded by y = x, y = 2 x, x = 1.
R

10 x 3 y 4 dxdy for R bounded by y = x3 , y = 0, x = 1.


R

y 2 dxdy for R bounded by y = x 2 , x = 2 y 2 .


R

R R

xydxdy for R the triangle with vertices ( 0, 0 ) , (1,1) , ( 4,1) .

12 x 2 ydxdy for R bounded by x = 1 y 2 , x = 1 + y 2 , y = 0, y = 1. 3 ( y 2 + 4 ) dxdy for R bounded by x = 1, x = 4, y =


R

1 1 ,y= . x x

Applied Statistics: Probability 1-159

Problem 3
5 4

y1 ( x)3 y2 ( x) 2

The two curves are y1 = x 2


(-1.353,1.831) (1,1)

y2 = 2 x

0 2

0 x

Iterated integral:

1.353

2 x

y 2 dydx

Applied Statistics: Probability 1-160

Joint Distribution of Continuous RVs


The cumulative joint distribution of continuous random variables X and Y is defined by FX ,Y ( x, y ) = P ( X x, Y y ), < x < , < y < . Properties: * 0 F ( x, y ) 1 * F ( x, y ) is monotone increasing in BOTH variables * P(a < X b and c < Y d ) = F (b, d ) F (a, d ) F (b, c) + F (a, c) Note:

lim
y

FX ,Y ( x, y ) = FX ( x) is called the marginal cumulative distribution of X . FX ,Y ( x, y ) = FY ( y ) is called the marginal cumulative distribution of Y.

lim
x

Think of FX ( x ) = P [ X x ] = P [ X x Y < ] FX ,Y ( x, ) . Also: The marginal cdf is defined in the same manner for discrete random variables.

Applied Statistics: Probability 1-161

Joint Probability Density


If X and Y are both continuous random variables, there is often a function f such that F ( x, y ) =

f (u , v)dvdu. The function f is known as the joint

probability density function. Note that P(a < x b, c < y d ) = f ( x, y )dydx.


a c b d

Also, by definition of marginal distribution we know that FX ( x) =

f (u , v)dvdu.

The marginal pdf of X is f X ( x) =

f ( x, y ) dy. Similarly, the marginal pdf of Y

is fY ( y ) =

f ( x, y ) dx.
Applied Statistics: Probability 1-162

Example (using Mathcad):


The random variables X and Y have the following joint pdf: ( x + y ) ( 1 x 3) ( 0 y 5) 45 (You need only define the non-zero portion of f to Mathcad.)
3

f ( x, y ) :=

Checking:

1 45

( x + y ) dy d x = 1

0 x

The marginal cdf for X is FX( x) := or FX( x) ( x 1) ( x + 6) 18

1 45

( u + v ) dv du

for 1 x 3.

The marginal cdf for Y is FY( y ) := or FY( y ) y ( y + 4) 45

1 45

( u + v ) du dv

for 0 y 5.

Applied Statistics: Probability 1-163

Example: Joint Density of X,Y


1 for (x, y ) A Let f ( x, y ) = 6 where A is the triangle shown. 0 otherwise What is the probability that X 2 and Y 2?
y=3

y = ( x 1)
3 4

P [ X 2, Y 2] =
1

3 2 4 ( x 1)

1 6

dydx

A
x =1 x=5

= 1 y| 6
1

3 ( x 1) 4

1 dx = 8 ( x 1)dx 1

Also, answer is ratio of area of small triangle to area of A. 3


8

x2 2 1 1 = 8 x | = . 2 1 16
Applied Statistics: Probability 1-164

3 1 = 6 48 16 This only works because f is constant above area A. =

Marginal Density of X (previous example)


fX ( x) =

f X ,Y ( x, y ) dy, for 1 x 5, (The value of x is held constant, and and the value of y is "integrated out.")

3 ( x 1) 4

1 6

y 3 ( x 1) 1 4 dy = | = ( x 1) . 0 6 8

Notice that

x2 5 1 1 f X ( x ) dx = 8 ( x 1) dx = 8 x | 2 1 1 = 1 ( 25 5 ) ( 1 1) = 1 8 2 2
5

as required for a pdf.


Applied Statistics: Probability 1-165

Example:
The random variables X and Y have the joint density function xy exp 1 ( x 2 + y 2 ) for x > 0 and y > 0 2 f ( x, y ) = otherwise. 0 Find FY ( y ) , f X ( x ) , and F (1, 2). Solution : We begin by finding f X ( x ) = f X ,Y ( x, y ) dy.
0

2 f X ( x ) = xy exp 1 ( x 2 + y 2 ) dy = xe
0

x2 2

ye

y2 2

dy = xe

x2 2

By symmetry, fY ( y ) = ye

y2 2

, also. This is Y 's density function,


y t2 2 t2 y 2 0

and we can now find the cdf, FY ( y ) = te dt = e


0

= 1 e

y2 2

Applied Statistics: Probability 1-166

Finding F(1,2)
Note that in the region of integration, x>0 & y>0, values of y do not depend on values of x.

F (1, 2) FX ,Y (1, 2) = = xe
0 1 x2 2 2

0 0

xy exp 1 ( x 2 + y 2 ) dydx 2
1 x2 2

dx 0 ye dydx = 0 xe 0 x2 x2 1 = (1 e 2 ) xe 2 dx = (1 e 2 ) e 2 0
y2 2 2 1 = (1 e ) 1 e 2 0.340. 2

y2 e 2

Applied Statistics: Probability 1-167

2 1 2 xy exp 2 ( x + y ) for x > 0 and y > 0 Scatterplot of f ( x, y ) = otherwise. 0

Applied Statistics: Probability 1-168

Lessons to Learn
F(1,2)=P(X<1,Y<2) Integrate a pdf to get a cdf:
FX ( x) =

f X (t )dt.

Limit out the other variables of a joint cdf to get a marginal cdf.

lim

FX ,Y ( x, y ) = FX ( x)

Integrate out the other variables of the joint pdf to get a marginal pdf.
fX ( x) =

f X ,Y ( x, y )dy.
Applied Statistics: Probability 1-169

Bivariate Normal pdf

x := 0 y := 0 := 0

x := 1 y := 1

= 0.6

c :=

1 2 x y 1
2

Applied Statistics: Probability 1-170

Independent RVs
Two random variables, X and Y , are independent if FX ,Y ( x, y ) = FX ( x) FY ( y ) for < x < and < y < . If the corresponding density functions exist, this is equivalent to f X ( x, y ) = f X ( x) fY ( y ) for < x < and < y < . EXAMPLE: Let X and Y have joint pdf f ( x, y ) =

1 , x 2 + y 2 1

0, otherwise

Determine the marginal pdf's of X and Y . Are X and Y independent?

Applied Statistics: Probability 1-171

Solution (Independent RVs)


We're given: f ( x, y ) =

1 , x 2 + y 2 1

0, otherwise

; thus, the marginal density for X is 1 y|


1-x 2

f X ( x) =

1-x 2

f ( x, y )dy =

- 1-x 2

dy =

- 1-x 2

1-x 2 , 1 x 1.

Because of the symmetry the marginal density of Y is 2 1 y 2 , 1 y 1. fY ( y ) =

To check for independence notice that 1 2 2 4 f X ,Y (0, 0) = and f X (0) fY (0) = = .

Thus, X and Y are NOT independent.

Applied Statistics: Probability 1-172

Keeping Sums in the Family


A sum of independent normals is normal. A sum of independent Poissons is Poisson. A sum of independent exponentials is Erlang. A sum of independent gammas is gamma. A large sum of independent, identically distributed RVs is approximately normal.

Applied Statistics: Probability 1-173

Linearity of Expectation
Suppose that X and Y are random variables and that a and b are any two real numbers, then E (aX + bY ) = aE ( X ) + bE (Y ). In previous example find E (3 X + 2Y ).

Use this property to derive the working Formula for variance: 2 = E ( X 2 ) E 2 ( X ).


Also notice that Var (aX ) = a 2Var ( X ) and that Var (aX + bY ) = a 2Var ( X ) + b 2Var (Y ) if X and Y are independent.

Applied Statistics: Probability 1-174

You might also like