Introduction to Probability & Statistics

Lecture - 7-8

Prof. Pritam Ranjan

Popular named continuous R.V.s
• Uniform −→ Beta
• Exponential −→ Gamma
• Weibull
• Pareto

• Normal
• Students’ t
• Chi-square
• Fisher’s F
..
.

• https://en.wikipedia.org/wiki/Category:Continuous distributions
• Continuous Univariate Distributions- Johnson, Kotz &
Balakrishnan

Pritam Ranjan / OM&QT 1

Example

• A packaging line consistently packages 200 cartons per hour. The package
weight varies uniformly between 18.2 lbs - 20.4 lbs. The customer requires
<20.0 lbs for ergonomic reasons. Find the probability that a package exceeds
the customer requirement.

Pritam Ranjan / OM&QT 2

Important remarks
• Random number generator on your calculator are supposedly
uniform (actually they are quasi random)

• Basis for generating numerous other random numbers

– discrete
– continuous

• Probability integral transformation

– Let X be a random variable and Y = FX (X ) such that FX (·) is
invertible, then Y ∼ Unif (0, 1)

Pritam Ranjan / OM&QT 3

Uniform - example
• Suppose X ∼ U[0, 1] and

1 x <p
Y =
0 otherwise.
Find the p.m.f. of Y .

Pritam Ranjan / OM&QT 4

Example - HW
• A stick of length 1 is split at a point U that is uniformly
distributed over (0, 1). Determine the p.d.f. of the longer piece.

Pritam Ranjan / OM&QT 5

Examples - Customer Service

• The waiting time between customer arrivals at a bank has a mean time between
arrivals of three minutes. If a customer has just arrived, what is the probability
that another customer will not arrive for at least two minutes?

Pritam Ranjan / OM&QT 6

Interesting property - example
• Let X be the time between two successive arrivals at the branch managers’
desk of the IIM branch of SBI with E(X ) = 10 mins. Given that no one visited
the branch manager in the last 30 mins, what is the probability of waiting
additional 10 minutes before the next arrival?

Pritam Ranjan / OM&QT 7

Reliability
• Laptop computers produced by a company have an average life of 38.36
months. [Assuming a reasonable distribution for the life of this laptop]

(a) find the probability that a computer will fail within 12 months.

(b) If the company wants not more than 5% of the computers to fail during the
warranty period, what should be the warranty period?

Pritam Ranjan / OM&QT 8

 Normal distribution

Pritam Ranjan / OM&QT 9

Normal distribution
• Notation: X ∼ N(µ, σ 2 )
• standard normal: Z ∼ N(0, 1), i.e., µ = 0, σ = 1

• pdf (  2

√ 1 exp − 12 (x−µ)
σ2
x ∈ (−∞, ∞)
fX (x) = 2πσ
0 otherwise.

• cdf

Notation: φ() and Φ() for the pdf and cdf of standard normal

Pritam Ranjan / OM&QT 10

Normal distribution
• Expectation

• Variance

A few important remarks:

• The most popular distribution −→ Why?
..
.

Pritam Ranjan / OM&QT 11

Popularity of normal – not normal !!
A few important remarks:
• The most popular distribution −→ Why? → CLT
If X1 , ..., X
√n are iid random variables with mean µ and variance
σ 2 , then n(X̄ − µ) ∼ N(0, σ 2 ).

• Its pdf is referred to as the bell-curve

• Basis for several distributions (e.g., Students’-t, χ2 , F , etc.)

• Also known as: Gaussian distribution, white noise distribution

• Been extended to: Gaussian process, Brownian motion, Wiener

process, Ornstein-Uhlenbeck process, etc.

• Ideal for representing ”measurement error” or ”replication error”

Pritam Ranjan / OM&QT 12
Standard normal prob. values
• If Z ∼ N(0, 1), find the following probabilities:

→ P(Z ≤ 0)

→ P(Z < 0)

→ P(Z < 5)

→ P(Z < −5)

→ P(−1 < Z < 1)

→ P(Z > 1)

Pritam Ranjan / OM&QT 13

General normal prob. values
Important transformation rule:
• If X ∼ N(µ, σ 2 ), then

(X − µ)
Z = ∼ N(0, 1)
σ

• e.g.,

Pritam Ranjan / OM&QT 14

Zα values
• If X ∼ N(0, 1), then Zα is basically the 100(1 − α)%-percentile
of the normal values, i.e.,
P(X ≥ Zα ) = α
OR
P(X ≤ Zα ) = 1 − α

Pritam Ranjan / OM&QT 15

Example - Stock market
• Based on the research of ”Ibbotson Associates, a Chicago investment firm, and
Prof. Jeremy Siegel of the Wharton School of the University of Pennsylvania”,
the average return on large-company stocks since 1920 has been 10.5% per
year and the standard deviation has been 4.75%. Assuming a normal
distribution for stock returns (and that the trend will continue this year), what is
the probability that a large-company stock you’ve just bought will loose money in
1 year?

Pritam Ranjan / OM&QT 16

Example - Forecasting
• Data accumulated by the National Climatic Data Center shows
that the average wind speed in miles per hour for St. Louis,
Missouri, is 9.7. Suppose wind speed measurements are
normally distributed for a given geographic location. If 22.45% of
the time the wind speed measurements are more than 11.6
miles per hour, what is the standard deviation of wind speed in
St. Louis?

Pritam Ranjan / OM&QT 17

Normal approximation
• Central limit theorem (CLT)

If X1 , ..., Xn are iid r.v.s with mean µ and variance σ 2 , then

X̄ ∼ N(µ, σ 2 /n),
OR,
(X̄ − µ)
p ∼ N(0, 1),
σ 2 /n
OR,
√
n(X̄ − µ)
−→ N(0, 1).
σ
(http://onlinestatbook.com/stat sim/sampling dist/)

• Several more powerful similar results are available

Pritam Ranjan / OM&QT 18

Binomial to Normal approximation
• Binomial distribution
Pn
If X ∼ bin(n, p), then X = i=1 Yi , where Yi ’s are iid Ber (p).

• Apply CLT on these n iid Yi ’s.

Pritam Ranjan / OM&QT 19

Binomial to Normal approximation
• We can compute binomial probability using normal table.

Bin(n, p) ≈ N(np, npq)

• Example: If X ∼ bin(15, 0.5) compute P(X ≤ 6)

Pritam Ranjan / OM&QT 20

Binomial to Normal approximation
• Example: If X ∼ bin(15, 0.5) compute P(X ≤ 6)

Pritam Ranjan / OM&QT 21

Continuity correction
• If X ∼ bin(n, p), then
 
a + 0.5 − np
P(X ≤ a) ≈ P(X ≤ a + 0.5) ≈ P Z ≤ √
npq
• Recompute P(X ≤ 6)

Pritam Ranjan / OM&QT 22

A few remarks
• Continuity correction should be done for P(X ≤ a) and NOT for
P(X < a), P(X > a) or P(X ≥ a).

• Binomial approximation works well if both np ≥ 10 and nq ≥ 10

Pritam Ranjan / OM&QT 23

Example
• One study on managers’ satisfaction with management tools
reveals that 59% of all managers use self-directed work teams
as a management tool. Suppose 70 managers selected
randomly in the US are interviewed. What is the probability that
fewer than 35 use self-directed work teams as a management
tool?

Pritam Ranjan / OM&QT 24

CLT example

• On a large computing system, a job queue will run large jobs in an orderly
fashion. The time (in minutes) needed to complete each job X follows an
exponential distribution with mean = 10 . Suppose 40 jobs are currently waiting
in the queue, and the completion times of these 40 jobs are independent.
Calculate the probability that the total time taken by the 40 jobs exceeds 500
minutes.

Pritam Ranjan / OM&QT 25

Example - CLT
• Suppose the length of a randomly selected song on Radio
Mirchi has an average length of µ = 200 seconds and standard
deviation σ = 30 seconds. I listen to songs at random and
record the length of each song. How many songs (n) will I need
to listen to so that there is a 90% chance that their average
length X̄ will be between 190 and 210 seconds?

Pritam Ranjan / OM&QT 26

Application - Value at Risk
• When a business venture involves chances of large losses, a
measure of risk that many companies use is the value at risk.

• The amount of loss at the 5th percentile of the distribution of

profit.

E.g., Suppose the average profit of a Pfizer stock is $10.46 with s = 7.45.
What is the Value at Risk?

Pritam Ranjan / OM&QT 27

Relation with other distributions
• Normal to Cauchy
1
f (x) =
π(1 + x 2 )
• Normal to χ2
1
f (x) = √ exp(−x/2) · x n/2−1
2n Γ(n/2)

• Normal and χ2 to t
−(n+1)/2
x2

Γ((n + 1)/2)
f (x) = √ 1+
nπΓ(n/2) n

• χ2 to F

Pritam Ranjan / OM&QT 28

Homework

• Practice problems: Chapter 4 (Normal distribution)

• Next topic: Joint distribution

Pritam Ranjan / OM&QT 29