Probability Distributions Probability Distributions

Chapter 5 Chapter 5
Levin & Rubin Levin & Rubin
Statistics Ior Management Statistics Ior Management
Probability Probability Basic concepts Basic concepts
Recap oI Preparatory classes: Recap oI Preparatory classes:
Three approaches to Probability
1. Classical
2. Based on Frequency distributions
3. Intuitive/Subjective
Throughout our program we will be working
with Irequency based approaches to
probability
Basic terminology Basic terminology
P(A): P(A): Probability oI the event A Probability oI the event A
occurring occurring
P(A or B): Probability oI event A or B P(A or B): Probability oI event A or B
occurring occurring
P(AB): P(AB): Probability oI both A & B occurring Probability oI both A & B occurring
P(A/B): P(A/B): Probability oI event A occurring Probability oI event A occurring
given that event B has occurred given that event B has occurred
We have also looked at Venn diagrams as a We have also looked at Venn diagrams as a
way oI representing probabilities oI events. way oI representing probabilities oI events.
Probability Rules Probability Rules
For single event situations: For single event situations:
II events A & B are mutually exclusive then II events A & B are mutually exclusive then
P(A or B) P(A) P(B) P(A or B) P(A) P(B)
II A & B are not mutually exclusive then II A & B are not mutually exclusive then
P(A or B) P(A) P(B) P(A or B) P(A) P(B) P(AB) P(AB)
For successive events situations: For successive events situations:
II A & B are statistically independent then II A & B are statistically independent then
Joint Probability Joint Probability P(AB) P(A) x P(B) P(AB) P(A) x P(B)
Conditional Probability Conditional Probability P(A/B) P(A) P(A/B) P(A)
II A & B are not statistically independent then II A & B are not statistically independent then
Conditional Probability Conditional Probability
P(A/B) P(AB)/P(B) P(A/B) P(AB)/P(B)
Joint Probability Joint Probability
P(AB) P(A/B) x P(B) or P(AB) P(A/B) x P(B) or
P(B/A) x P(A) P(B/A) x P(A)
Because Because P(AB) P(BA) P(AB) P(BA)
The conditional probability relationship is also The conditional probability relationship is also
called Bayes` Theorem called Bayes` Theorem
Probability Distributions Probability Distributions
What are Probability Distributions? What are Probability Distributions?
Frequency distributions are summaries of Frequency distributions are summaries of
the observed frequencies of the outcomes the observed frequencies of the outcomes
of an experiment. of an experiment.
A Probability distribution can be thought A Probability distribution can be thought
of as a theoretical frequency distribution of as a theoretical frequency distribution
that describes how outcomes are that describes how outcomes are expected expected
to vary. to vary.
Types oI Probability Distributions Types oI Probability Distributions
Probability distributions can be either continuous or discrete Probability distributions can be either continuous or discrete
A discrete prob. distribution can take on only a limited number A discrete prob. distribution can take on only a limited number
oI values that can be listed oI values that can be listed
Example: The number oI bad apples in a Example: The number oI bad apples in a
basket oI apples basket oI apples
A continuous distribution, on the other hand, can take on any A continuous distribution, on the other hand, can take on any
value within a range and hence cannot be listed value within a range and hence cannot be listed
Example: The eIIluent levels in water samples. Example: The eIIluent levels in water samples.
II the number oI samples in a discrete distribution are numerous II the number oI samples in a discrete distribution are numerous
and the outcomes are close to one another, then the discrete and the outcomes are close to one another, then the discrete
distribution can be approximated well by a continuous distribution can be approximated well by a continuous
distribution. (Recall the loom example). distribution. (Recall the loom example).
Random variables Random variables
A variable is random iI it takes on diIIerent A variable is random iI it takes on diIIerent
values as a result oI outcomes oI a random values as a result oI outcomes oI a random
experiment experiment
II the values that the variable is allowed to take II the values that the variable is allowed to take
are limited and can be listed, then it is a are limited and can be listed, then it is a
discrete random discrete random variable. variable.
owever, iI it is allowed to assume any value owever, iI it is allowed to assume any value
within a range, it is a within a range, it is a continuous random continuous random
variable variable
Expected Values OI Random Variables Expected Values OI Random Variables
The Expected Value oI a random variable is The Expected Value oI a random variable is
the product oI the value oI the random variable the product oI the value oI the random variable
Ior a particular outcome and the probability oI Ior a particular outcome and the probability oI
that particular outcome occurring. that particular outcome occurring.
The expected value can be viewed as the The expected value can be viewed as the
weighted mean oI the possible values oI the weighted mean oI the possible values oI the
random variable over all the outcomes. The random variable over all the outcomes. The
weighting Iactor is the probability associated weighting Iactor is the probability associated
with each outcome. with each outcome.
umerical Example umerical Example
Bill Johnson has bought a new VCR Ior $300. e has the Bill Johnson has bought a new VCR Ior $300. e has the
option oI buying an extended service warranty Ior 5 years Ior option oI buying an extended service warranty Ior 5 years Ior
$100. AIter talking to Iriends and reading various reports, Bill $100. AIter talking to Iriends and reading various reports, Bill
believes the Iollowing maintenance expenses could be incurred believes the Iollowing maintenance expenses could be incurred
during the next Iive years: during the next Iive years:
Find the expected value oI his maintenance costs Find the expected value oI his maintenance costs
Should Bill pay the $100 Ior the warranty? Should Bill pay the $100 Ior the warranty?
xpense xpense 00 S0 S0 100 100 1S0 1S0 200 200 2S0 2S0 300 300
Probability Probability 0.3S 0.3S 0.2S 0.2S 0.1S 0.1S 0.10 0.10 0.08 0.08 0.0S 0.0S 0.02 0.02
Solution Solution
The Expected value oI the maintenance costs The Expected value oI the maintenance costs
|0*0.35 50*0.25 100*0.15 150*0.10 |0*0.35 50*0.25 100*0.15 150*0.10
200*0.08 250*0.05 300*0.02| 200*0.08 250*0.05 300*0.02|
|0 12.50 15.00 15.00 16.00 12.50 |0 12.50 15.00 15.00 16.00 12.50
6.00| 6.00|
$77 $77
Since the expected value is less than $100, Since the expected value is less than $100,
Bill should not buy the warranty. Bill should not buy the warranty.
The Binomial Distribution The Binomial Distribution
The The Binomial Distribution Binomial Distribution is one oI the most widely is one oI the most widely
used discrete probability distributions. used discrete probability distributions.
The data generated Irom an experiment known as the The data generated Irom an experiment known as the
ernoulli process ernoulli process is described by the binomial is described by the binomial
distribution. distribution.
The Bernoulli process is best characterized by the The Bernoulli process is best characterized by the
repeated tossing oI a balanced coin Ior a pre repeated tossing oI a balanced coin Ior a pre- -
determined number oI times. determined number oI times.
The probabilities oI getting various combinations oI The probabilities oI getting various combinations oI
heads and tails Irom these tosses is described by the heads and tails Irom these tosses is described by the
binomial distribution. binomial distribution.
IdentiIying the Bernoulli Process IdentiIying the Bernoulli Process
Using the tossing oI the coin as the example: Using the tossing oI the coin as the example:
1. Each trial (each toss, in our example) has only 1. Each trial (each toss, in our example) has only
two possible outcomes: heads or tails, success or two possible outcomes: heads or tails, success or
Iailure, yes or no. Iailure, yes or no.
2. The probability oI the outcome oI any trial 2. The probability oI the outcome oI any trial
(toss) is Iixed over time. (With a Iair coin, the (toss) is Iixed over time. (With a Iair coin, the
probability oI either heads or tails remains 0.5 Ior probability oI either heads or tails remains 0.5 Ior
each toss, regardless oI the number oI times the coin each toss, regardless oI the number oI times the coin
has been tossed.) has been tossed.)
3. The trials are statistically independent: that is 3. The trials are statistically independent: that is
the outcome oI one toss does not aIIect the outcome the outcome oI one toss does not aIIect the outcome
oI any other toss. oI any other toss.
Formulae Ior the Binomial Distribution Formulae Ior the Binomial Distribution
Let n be the number oI trials oI the Let n be the number oI trials oI the
experiment. experiment.
Let p be the probability oI 'success in each Let p be the probability oI 'success in each
trial trial
ere, 'success is deIined as the outcome we ere, 'success is deIined as the outcome we
are interested in and is no way judgmental oI are interested in and is no way judgmental oI
the trial or the outcome the trial or the outcome
Let q be the probability oI Iailure: Let q be the probability oI Iailure:
q (1 q (1- -p) p)
Let r be the number oI successes desired Let r be the number oI successes desired
Then the Then the
Probability oI r successes in n trials is given by Probability oI r successes in n trials is given by
P (r) n!/ (r!*(n P (r) n!/ (r!*(n- -r)!)*p r)!)*p
r r
*q *q
nn- -r r
or or
nn
CC
r r
* p * p
r r
*q *q
nn- -r r
The mean oI a binomial distribution is given The mean oI a binomial distribution is given
by by
n*p n*p
The Standard deviation is given by The Standard deviation is given by
o o \n*p*q \n*p*q
umerical example umerical example
arry Ohme is in charge oI the electronics section oI arry Ohme is in charge oI the electronics section oI
a large department stores. e has noticed that the a large department stores. e has noticed that the
probability that a customer who is just browsing will probability that a customer who is just browsing will
buy something is 0.3. Suppose 15 customers browse buy something is 0.3. Suppose 15 customers browse
in the electronic section each hour: in the electronic section each hour:
What is the probability that What is the probability that at least one at least one browsing browsing
customer will buy something during a speciIied hour? customer will buy something during a speciIied hour?
What is the probability that What is the probability that at least Iour at least Iour ..buy.? ..buy.?
What is the probability that What is the probability that none oI them none oI them will buy? will buy?
What is the probability that What is the probability that no more than Iour no more than Iour buy? buy?
Solution Solution
We use the binomial probability tables to We use the binomial probability tables to
arrive at the probabilities: arrive at the probabilities:
The probability that The probability that at least one at least one customer buys customer buys
is the sum oI all the probabilities oI 1, 2, is the sum oI all the probabilities oI 1, 2,
3,.15 customers buy. This can be tedious to 3,.15 customers buy. This can be tedious to
calculate. calculate.
owever, the sum oI all the probabilities Irom owever, the sum oI all the probabilities Irom
0 to 15 customers buying should be 1. 0 to 15 customers buying should be 1.
So the above probability is equivalent to So the above probability is equivalent to
(1 (1 Probability oI no customer buying) Probability oI no customer buying)
This is a much easier approach since only one probability namely This is a much easier approach since only one probability namely
P(0) needs to be Iound P(0) needs to be Iound
Appling the Iormula Appling the Iormula
P (0) (15!/0!*15!) *(0.3) P (0) (15!/0!*15!) *(0.3)
00
*(0.7) *(0.7)
15 15
(0.7) (0.7)
15 15
0.00475 0.00475
Value Irom the table 0.0047 Value Irom the table 0.0047
So, the probability oI at least one customer buying is (1 So, the probability oI at least one customer buying is (1 0.0047) 0.0047)
0.9953 0.9953
The probability that none will buy is 0.0047 The probability that none will buy is 0.0047
The probability that no more than Iour will buy The probability that no more than Iour will buy
The sum oI the probabilities that exactly 0,1,2,3 & 4 The sum oI the probabilities that exactly 0,1,2,3 & 4 will will
buy buy
0.0047 0.0305 0.0916 0.1700 0.2186 0.0047 0.0305 0.0916 0.1700 0.2186
0.5154 0.5154
The probability that at least Iour will buy The probability that at least Iour will buy
(1 (1 The probability that at the most three will buy) The probability that at the most three will buy)
The probability that at the most three will buy is The probability that at the most three will buy is
0.0047 0.0305 0.0916 0.1700 0.2968 0.0047 0.0305 0.0916 0.1700 0.2968
So, the probability that at least Iour will buy (1 So, the probability that at least Iour will buy (1 0.2968) 0.7032 0.2968) 0.7032
The Poisson Distribution The Poisson Distribution
The second important discrete distribution that we The second important discrete distribution that we
will be looking at is the Poisson Distribution. will be looking at is the Poisson Distribution.
The Poisson distribution provides a suitable model The Poisson distribution provides a suitable model
Ior a situation where we can observe the number oI Ior a situation where we can observe the number oI
times an event (Iavorable) occurs; but where it is not times an event (Iavorable) occurs; but where it is not
meaningIul to speak oI the number oI times the event meaningIul to speak oI the number oI times the event
did not occur. did not occur.
(Compare this with the Binomial distribution where (Compare this with the Binomial distribution where
we keep track oI both occurrences.) we keep track oI both occurrences.)
This distribution is best used to describe occurrences This distribution is best used to describe occurrences
oI isolated events in a continuum over a speciIied oI isolated events in a continuum over a speciIied
interval oI time or space. interval oI time or space.
Usually, the probability oI success p is very low. Usually, the probability oI success p is very low.
Poisson Process Poisson Process
The characteristics oI a Poisson process are as below: The characteristics oI a Poisson process are as below:
1. The occurrences oI the events are independent. That is: 1. The occurrences oI the events are independent. That is:
Occurrence oI an event in an interval oI space or time has no Occurrence oI an event in an interval oI space or time has no
eIIect on the probability oI a second occurrence oI the event in eIIect on the probability oI a second occurrence oI the event in
the same or any other interval. the same or any other interval.
2. Theoretically an inIinite number oI occurrences oI the 2. Theoretically an inIinite number oI occurrences oI the
event must be possible in the interval. event must be possible in the interval. 4709,3430 4709,3430
4:9.420 4:9.420
3. The probability oI a single occurrence oI the event in a 3. The probability oI a single occurrence oI the event in a
given interval is proportional to the length oI the interval. given interval is proportional to the length oI the interval.
4. In any inIinitesimally small portion oI the interval the 4. In any inIinitesimally small portion oI the interval the
probability oI more than one occurrence oI the event is probability oI more than one occurrence oI the event is
negligible. negligible.
Formulae Ior the Poisson Distribution Formulae Ior the Poisson Distribution
We deIine the parameters that describe a Poisson We deIine the parameters that describe a Poisson
distribution as under: distribution as under:
(lambda): mean number oI occurrences per (lambda): mean number oI occurrences per
interval oI time interval oI time
x no oI occurrences we are interested in x no oI occurrences we are interested in
e base oI the natural logarithm system: 2.71828 e base oI the natural logarithm system: 2.71828
The probability oI exactly x occurrences in a Poisson The probability oI exactly x occurrences in a Poisson
process is calculated by the Iormula process is calculated by the Iormula
P(x) P(x)
xx
*e *e
- -
/x! /x!
The mean oI the distribution, The mean oI the distribution, &&
The standard deviation, The standard deviation, o o \ \
umerical Example umerical Example
Guy Ford, the production supervisor at a plant is Guy Ford, the production supervisor at a plant is
worried about an elderly employee`s ability to keep worried about an elderly employee`s ability to keep
up the minimum work pace. In addition to the normal up the minimum work pace. In addition to the normal
daily breaks, this employee stops Ior short rest daily breaks, this employee stops Ior short rest
periods an average oI 4.1 times per hour. The rest periods an average oI 4.1 times per hour. The rest
period is a Iairly consistent 3 minutes each time. period is a Iairly consistent 3 minutes each time.
Ford has decided that, apart Irom the normal Ford has decided that, apart Irom the normal
breaks, iI the probability oI the employee resting Ior breaks, iI the probability oI the employee resting Ior
12 minutes or more per hour is greater than 0.5, he 12 minutes or more per hour is greater than 0.5, he
will move the employee to a diIIerent job. will move the employee to a diIIerent job.
Should Ford move the employee? Should Ford move the employee?
Solution Solution
The problem talks oI occurrences oI outcomes (breaks) in a The problem talks oI occurrences oI outcomes (breaks) in a
Iixed interval oI time (1 hour). Iixed interval oI time (1 hour).
Since more than outcome is possible, it is not a Bernoulli Since more than outcome is possible, it is not a Bernoulli
process. process.
An average number oI occurrences Ior the interval oI time is An average number oI occurrences Ior the interval oI time is
given. given.
ence we should use the Poisson distribution. ence we should use the Poisson distribution.
ere ere 4.1; 4.1;
We need to choose the number oI occurrences that we are We need to choose the number oI occurrences that we are
interested in and Iind the probability. interested in and Iind the probability.
The supervisor wants to know iI the probability oI the The supervisor wants to know iI the probability oI the
employee resting Ior 12 minutes or more is _ 0.5 employee resting Ior 12 minutes or more is _ 0.5
The employee rests Ior 3 minutes during every break. So, he The employee rests Ior 3 minutes during every break. So, he
will rest Ior 12 minutes iI he takes 4 breaks. will rest Ior 12 minutes iI he takes 4 breaks.
ence, we need to Iind the probability oI his taking at the most ence, we need to Iind the probability oI his taking at the most
3 breaks, so that his total rest time at 9 minutes will be less 3 breaks, so that his total rest time at 9 minutes will be less
than 12 minutes than 12 minutes
So, the probability oI his taking at the most 3 breaks is the sum So, the probability oI his taking at the most 3 breaks is the sum
oI the probabilities oI his taking no breaks, one, two & three oI the probabilities oI his taking no breaks, one, two & three
breaks. breaks.
P(x P(x_ 3) P(x0) P(x1) P(x2) P(x3) _ 3) P(x0) P(x1) P(x2) P(x3)
From the table on Poisson probabilities we get the values as From the table on Poisson probabilities we get the values as
Iollows: Iollows:
P(0) 0.0166 P(0) 0.0166
P(1) 0.0679 P(1) 0.0679
P(2) 0.1393 P(2) 0.1393
P(3) 0.1904 P(3) 0.1904
Total 0.4142 Total 0.4142
ence his probability oI taking 4 or more breaks ence his probability oI taking 4 or more breaks
(1 (1 Probability oI taking at the most3 breaks) Probability oI taking at the most3 breaks)
(1 (1 0.4142) 0.5858 0.4142) 0.5858
Guy Ford should shiIt the employee to another job. Guy Ford should shiIt the employee to another job.
Approximating the Binomial Approximating the Binomial
distribution by the Poisson distribution distribution by the Poisson distribution
BeIore the advent oI computers, calculating BeIore the advent oI computers, calculating
probabilities Ior binomial distributions with large probabilities Ior binomial distributions with large
values Ior n was very tedious. values Ior n was very tedious.
ence distributions that were easier to handle were ence distributions that were easier to handle were
used to approximate the binomial values. used to approximate the binomial values.
In certain cases, the Poisson can be used to In certain cases, the Poisson can be used to
approximate the binomial. approximate the binomial.
en n is greater tan 20 and p is less tan or equal en n is greater tan 20 and p is less tan or equal
to 0.05, statisticians ave found tat te Poisson is a to 0.05, statisticians ave found tat te Poisson is a
good approximation for te binomial. good approximation for te binomial.
We substitute the mean oI the binomial (n*p) in We substitute the mean oI the binomial (n*p) in
place oI the mean oI the Poisson place oI the mean oI the Poisson
distribution( distribution(). ).
The substituted Iormula becomes The substituted Iormula becomes
P(x) (n*p) P(x) (n*p)
xx
*e *e
- -n*p n*p
/x! /x!
umerical Example umerical Example
In a hospital with 20 dialysis machines and the In a hospital with 20 dialysis machines and the
probability oI one oI them malIunctioning at 0.02, probability oI one oI them malIunctioning at 0.02,
what is the probability that exactly 3 machines will be what is the probability that exactly 3 machines will be
out oI service on the same day? out oI service on the same day?
The Binomial Approach The Binomial Approach
n 20, p 0.02, r 3: n 20, p 0.02, r 3:
From the table P(3) From the table P(3) 0.0065 0.0065
The Poisson Approach The Poisson Approach
n*p 20*0.02 0.4 n*p 20*0.02 0.4
From the table P(3) From the table P(3) 0.0072 0.0072
ypergeometric Distribution ypergeometric Distribution
The binomial distribution evaluates the probability oI The binomial distribution evaluates the probability oI
r successes in n trials where the probability oI success r successes in n trials where the probability oI success
in any trial is p. in any trial is p.
We assume that the trials are independent ie the We assume that the trials are independent ie the
result oI a particular trial does not inIluence the result oI a particular trial does not inIluence the
outcome oI the later trials and that outcome oI the later trials and that
The probability p is a constant throughout the entire The probability p is a constant throughout the entire
set oI experiments. set oI experiments.
This means when we are drawing a sample Irom a This means when we are drawing a sample Irom a
population to estimate a particular characteristic, we population to estimate a particular characteristic, we
assume that the population is inIinite and our sample assume that the population is inIinite and our sample
will not alter the probability p in the population. will not alter the probability p in the population.
This assumption need not always be true This assumption need not always be true
especially in cases where the population Irom especially in cases where the population Irom
where we draw the sample itselI is Iinite. where we draw the sample itselI is Iinite.
In such cases, the samples that we draw can In such cases, the samples that we draw can
aIIect the probability p Ior the later samples. aIIect the probability p Ior the later samples.
Under such circumstances we use the Under such circumstances we use the
hypergeometric distribution. hypergeometric distribution.
Mathematically speaking, when we draw a Mathematically speaking, when we draw a
sample Irom sample Irom a finite population witout a finite population witout
replacement, replacement, the probability Iunction Ior the the probability Iunction Ior the
occurrence oI a preIerred outcome is described occurrence oI a preIerred outcome is described
by the hypergeometric distribution. by the hypergeometric distribution.
Formulae oI the ypergeometric Formulae oI the ypergeometric
distribution distribution
We consider a Iinite population which We consider a Iinite population which
contains A number oI items which is our contains A number oI items which is our
preIerred outcome, Irom which we draw a preIerred outcome, Irom which we draw a
sample n Ior our study without replacement. sample n Ior our study without replacement.
We wish to evaluate the probability oI Iinding We wish to evaluate the probability oI Iinding
x Iavorable outcomes in this sample. x Iavorable outcomes in this sample.
This probability is given by This probability is given by
P(x) P(x)
A A
CC
xx
**
- -A A
CC
nn- -xx
/ /

CC
nn
The mean oI the hypergeometric distribution is The mean oI the hypergeometric distribution is
given by given by
n*A/ n*A/
And the standard deviation is given by And the standard deviation is given by
o o \nA( \nA(- -A)/ A)/
22
*\( *\(- -n)/( n)/(- -1) 1)
The Iactor The Iactor \( \(- -n)/( n)/(- -1) 1) is called the is called the
population correction Iactor that population correction Iactor that results Irom results Irom
sampling without replacement Irom a Iinite sampling without replacement Irom a Iinite
population population