NOTES: BERNOULLI RANDOM VARIABLE AND ITS DISTRIBUTION

This transcript covers Module 2 of "Business Statistics for Entrepreneurs,"

featuring Ananth Krishnamurthy and students Tejas and Ashwini. The
session focuses on the Bernoulli random variable, which serves as the
fundamental basis for the binomial distribution.

1. Fundamentals of Discrete Random Variables

Before diving into Bernoulli specifics, the transcript recalls key properties
of discrete random variables:

 Values: They take only integer values.

 Notation: Capital X denotes the random variable, while small x

denotes the particular values it takes.

 Probability Mass Function (PMF): Denoted as f(x), it provides the

probability that X = x.

 Cumulative Distribution Function (CDF): Denoted as F(x), it

represents the probability that X takes a value less than or equal to
x.

 Expectation (E[X]): The sum of x. f(x) over all possible values.

 Variance (Var(X)): The sum of {x - E[X]}^2 f(x) over all

possible values.

2. The Bernoulli Random Variable

The Bernoulli random variable is a specific type of discrete random

variable used when an experiment has exactly two possible outcomes:
success or failure.

 Notation Shift: For clarity, the transcript uses Y for Bernoulli

variables, with f(y) as the PMF and F(y) as the CDF.

 Outcomes:

o Success (y = 1): Occurs with probability p.

o Failure (y = 0): Occurs with probability q, where q = 1 - p.

o Property: p + q = 1.
Formulas for Bernoulli Variables

 Expectation (E[Y]): 1* p + 0 *(1-p) = p

 Variance (Var(Y)): p(1-p)^2 + q(0-p)^2 =
pq( p + q ) = pq
 Standard Deviation: sq. root pq
3. Practical Examples and Applications

The transcript illustrates how various scenarios can be framed as Bernoulli

trials:

Success (y=1) Probability Failure (y=0)

Scenario
Definition (p) Definition

Coin Toss Heads p Tails

Sales Call Purchase made p No purchase

Customer Satisfied with

p Not satisfied
Survey service

Daily Store Demand Less than or

Demand > 10 p
Demand equal10

Unbiased Die Value \le 2 1/3 Value \ge 3

Biased Die Value \ge 5 1/6 Value \le 4

\le 1 defective
Pump Factory 0.65 \ge 2 defective pumps
pump

Airport Delays \le 2 late flights 0.6 \ge 3 late flights

Insurance
Purchase made 0.2 No purchase
Sales

"Star Day"
Sales > 10,000 0.3 Sales \le 10,000
Sales
4. Calculated Expectations and Variances

The students calculate values for specific business cases:

 Pump Factory: E[Y] = 0.65; Var(Y) = 0.23.

 Airport Delays: E[Y] = 0.6; Var(Y) = 0.24.

 Insurance Agent: E[Y] = 0.2; Var(Y) = 0.16; SD = 0.4 (implies 20%

success rate).

 Nashik Store: E[Y] = 0.3; Var(Y) = 0.21; SD = 0.46.

 NOTES: BINOMIAL RANDOM VARIABLES AND ITS
DISTRIBUTION

This session, part of "Business Statistics for Entrepreneurs," transitions

from the Bernoulli random variable to the Binomial random variable,
detailing its properties, mathematical foundations, and relationship to
Bernoulli trials.

1. The Binomial Experiment

A binomial random variable is derived from a multi-step experiment

known as a binomial experiment. It must satisfy four specific properties:

 Sequence of Trials: It consists of n identical and independent

trials.

 Two Outcomes: Each trial results in exactly two outcomes: success

or failure (making each trial a Bernoulli trial).

 Constant Probability: The probability of success (p) remains the

same for every trial.

 Independence: The outcome of one trial does not affect the

probability of success or failure in another.

 The Random Variable (Y): Counts the total number of successes

occurring in the n trials.

Key Insight: The binomial variable only cares about the count of
successes, not their order or where they occur in the sequence of trials.
2. Probability Mass Function (PMF)

The PMF, f(y), provides the probability that the random variable Y takes a
specific value y.

f(y) = \binom{n}{y} p^y (1-p)^{n-y} \quad \text{or} \quad \binom{n}

{y} p^y q^{n-y}

Components of the Formula:

 \binom{n}{y} (Combinations): Calculated as \frac{n!}{y!(n-

y)!}, this represents the number of different ways y successes can
be distributed across n trials.

 p^y: The probability of achieving exactly y successes.

 q^{n-y}: The probability that the remaining n - y trials are failures.

3. Examples of Binomial Experiments

The transcript builds on previous Bernoulli examples by applying them to

multiple trials:

4. Expectation and Variance

The binomial distribution's measures of center and spread are linear

combinations of the underlying Bernoulli trials.

Expectation (E[Y])

 Formula: E[Y] = n*p.

 Logic: Since the binomial variable is the sum of n independent

Bernoulli trials (each with expectation p), the total expectation is
n*p.

Variance (Var(Y))

 Formula: Var(Y) = n*p*q.

 Logic: Because the trials are independent, the total variance is the
sum of the variances of each individual trial (p*q).
Standard Deviation

 Formula: \sigma = \sqrt{n \cdot p \cdot q}.

 NOTES: EXAMPLES OF BINOMIAL DISTRIBUTIONS - I

This session demonstrates how to apply binomial distribution

concepts to practical problems using Microsoft Excel, focusing on
probability calculations and "inverse probability" scenarios.

1. Using MS Excel for Binomial Calculations

Creating an Excel table for binomial distributions simplifies

calculations, reduces errors, and allows for sensitivity analysis
through parameter changes or graphing.

The [Link] Function

The primary tool in Excel for these calculations is the function

=[Link](y, n, p, cumulative):

 y: The number of successes.

 n: The total number of trials.

 p: The probability of success in a single trial.

 Cumulative (False/True):

o False: Returns the Probability Mass Function (PMF), f(y), for

exactly y successes.

o True: Returns the Cumulative Distribution Function (CDF),

F(y), for the probability of y or fewer successes.

2. Example 1: The Biased Die

 This scenario involves 10 trials (n=10) where success is obtaining a
5 or 6 (p = 1/6) and failure is 4 or less (q = 5/6).

 Expectation (E[Y]): n \times p = 1.67.

 Variance (Var(Y)): n \times p \times q = 1.39.

 Standard Deviation: \sqrt{Var(Y)} = 1.18.

 Probability of \ge 3 successes: Calculated by summing individual

probabilities for y = 3 through y = 10, resulting in 0.225.

Inverse Probability Calculation

To find the success probability (p) required to make the probability

of \ge 3 successes at least 0.5:

 The current probability at p=1/6 is only 0.225.

 By manually varying the value in Excel or using Goal Seek, the

required success probability (p) is determined to be 0.259.

3. Example 2: Defective Pump Production

 Kedar Apte tracks production over 14 independent days (n=14).
Success is defined as a day with \le 1 defective pump (p = 0.65),
while failure is \ge 2 defects (q = 0.35).

 Expectation (E[Y]): 14 \times 0.65 = 9.1.

 Standard Deviation: 1.78.

 Probability of \ge 10 success days: Summing probabilities for y

= 10 to 14 yields 0.423.

Improving Success Probability

To increase the chance of having at least 10 success days to 0.5, the

underlying daily success probability (p) must be improved:

 At p = 0.66, the probability is 0.454.

 At p = 0.67, the probability is 0.486.

 At p = 0.68, the probability is 0.519.

 A target value of approximately p = 0.675 achieves the desired 0.5

threshold.
 NOTES: EXAMPLES OF BINOMIAL DISTRIBUTION - II

This transcript continues Module 2, exploring complex applications

of the binomial distribution across various business scenarios. The
focus is on correctly identifying parameters (n and p) based on
specific business questions.

1. Case Study: Airport Delays (Mangesh Nadkarni)

Mangesh uses a 30-day period (n = 30) to analyze flight delays.

Success is defined as a day with \le 2 late take-offs (p = 0.6, q =
0.4).

 Scenario A (At least 18 success days): P(Y \ge 18) results in a

probability of 0.578.

 Scenario B (At most 10 failure days): Having \le 10 failure days

is equivalent to having \ge 20 success days. P(Y \ge 20) = \
mathbf{0.291}.

 Scenario C (Range of success): P(15 \le Y \le 25) = \

mathbf{0.901}.
 2. Case Study: Insurance Sales (Priya Phule)

Priya makes 12 independent calls per day (n = 12) with a success

(purchase) probability p = 0.2.

 Expectation and Variation: She can expect an average of 2.4

successful calls per day with a standard deviation of 1.39.

 High Performance Goal: The probability of having \ge 6 successes

(a 50% conversion rate) is only 0.019.

 "Bad Day" Probability: The probability of having \le 3 successes

(converting \le 25\% of calls) is approximately 0.795 (80%).

3. Case Study: Retail "Star Days" (Baburao Phalke)

Baburao defines a "star day" as sales > 10,000, which has p = 0.3.
He observes this over a 10-day vacation (n = 10).

 Expected Star Days: n*p = 3

 At most 3 Star Days: P(Y<=3) = 0.65

 Range (3 to 6 Star Days): P(3 \le Y \le 6) = \mathbf{0.607}.

o Alternative Calculation: P(Y \le 6) - P(Y \le 2) = 0.989 - 0.233 =

0.606.
4. Advanced Parameter Identification (The "Divide and
Conquer" Approach)
The session emphasizes that while Excel calculations are
mechanical, the critical task is translating business context into the
correct n and p values.

Example: Refined Pump Defect Analysis (n=14)

New Event
Business Success Ne of Probabili
Question Definitio wp Interes ty
n t

Days
P(Y \ge
with NO 0 defects 0.4 0.308
7)
defects

At most
3 days
with \ge \ge 3 P(Y \le
0.15 0.988
3 defects 3)
defects
(n=7)

All 14
days \le 1 P(Y =
0.65 0.002
with \le defect 14)
1 defect

Example: Refined Airport Delay Analysis

  At least 15 days with \le 1 plane late (n=30): p becomes 0.3
(0.2 for 0 planes + 0.1 for 1 plane). P(Y \ge 15) = \mathbf{0.017}.

 At most 5 days with exactly 4 planes late (n=10): Success is

exactly 4 planes, so p = 0.2. P(Y \le 5) = \mathbf{0.944}.

 Exactly 15 days with \le 2 planes late (n=30): p = 0.6. P(Y =

15) = \mathbf{0.078}.

RECAP OF BINOMIAL DISTRIBUTION

This concluding segment of Module 2 summarizes the foundational

properties of the binomial distribution and discusses the practical validity
of its underlying assumptions.

1. Essential Properties of a Binomial Experiment

A binomial experiment is defined by a specific set of criteria that must be

met to use the binomial distribution models:

 Sequence of Trials: The experiment consists of multiple trials that

are both identical and independent.

 Binary Outcomes: Each individual trial results in exactly one of two

possible outcomes: success or failure.

 Constant Probability: The probability of success (p) remains

constant and does not change from one trial to the next.

 Counting Successes: The binomial random variable (Y) represents

the total count of successes occurring across the n trials.

The "IID" Concept: A binomial experiment is essentially a sequence of

Identically and Independently Distributed (IID) Bernoulli random
variables.

2. Deviations from Binomial Assumptions

In practice, the assumptions of independence and identical distribution

may not always hold true.

 Variable Probabilities: Success probabilities might change from

trial to trial in either a defined or undefined manner.
  Interdependence: The outcome of a current trial might be
influenced by or dependent on the outcome of previous trials.

 Consequence: If these assumptions are violated, the resulting

random variable is no longer binomial. In such cases, different types
of random variables must be defined based on which properties
were lost.

3. Practical Application and Modeling

While the assumptions may seem restrictive, the binomial distribution

remains a robust tool for real-world scenarios:

 Useful Abstractions: Models are abstractions of reality; while they

may not hold perfectly (e.g., in cases of coin tosses, defective
pumps, or airport delays), they provide a useful framework for
decision-making.

 Versatility: As seen in previous modules, the binomial model is

applicable to various fields, including manufacturing (defective
pumps), logistics (late take-offs), and retail (Saraf Bazaar "star
days").

 Alternative Models: If the binomial assumptions are too far

removed from reality, statisticians use other specialized random
variables to better model the situation.
 NOTES: POISSON RANDOM VARIABLE AND ITS DISTRIBUTION

This session introduces the Poisson distribution, a discrete

probability distribution used to estimate the frequency of specific
events over a defined interval of time or space.

1. Definition and Core Concepts

 Origins: Named after French mathematician Baron Siméon Denis

Poisson (1781–1840), who contributed significantly to statistics,
analytical mechanics, and physics.

 Function: Estimates the number of occurrences (y) of an event in a

specified interval (time or space).

 Possible Values: The random variable Y can take discrete integer

values from 0, 1, 2,… all the way to infinity.

 Applications:

o Time-based: Customers arriving at a location in 30 minutes

or phone calls reaching a call center in 15 minutes.

o Space-based: Counting defects in one kilometer of highway

or the number of keywords on a page of text.

2. Mathematical Framework

The Poisson distribution is unique because it is defined by a single

parameter, \Lambda (Capital Lambda).

Probability Mass Function (PMF)

The probability of a random variable Y taking a specific value y is

given by:

f(y) = \frac{e^{-\Lambda} \Lambda^y}{y!}

 e: A constant approximately equal to 2.71828.(Euler Constant)

 \Lambda: The expected number of events in the specified interval.

Expectation and Variance

One of the most powerful properties of the Poisson distribution is the

relationship between its mean and spread:

 Expectation (E[Y]): Equal to Lambda

 Variance (Var(Y)): Equal to\Lambda

 Standard Deviation: sqrt. Lambda

3. Key Properties of Poisson Variables

To be classified as a Poisson random variable, three main conditions

must be met:

1. Stationarity: The probability of occurrence is the same for any two

intervals of equal length (e.g., arrivals between 9 AM–10 AM have
the same probability as arrivals between 1 PM–2 PM).

2. Independence: The occurrence or non-occurrence of an event in

one interval has no bearing on another interval.

3. Mean Equals Variance: As established, E[Y] = Var(Y) = \Lambda.

 4. The Poisson Process and Rates

The Poisson Process is a series of random events occurring over

time (a stochastic process).

 Rate (\lambda): Small lambda represents the arrival rate (e.g., 10

customers per hour).

 Time Period (t): The specific duration of interest.

 Parameter Relationship: \Lambda = \lambda \cdot t.

Example Calculation: If the arrival rate (\lambda) is 10 per hour:

 For a 2-hour period (t=2), \Lambda = 10 \times 2 = 20.

 For a 30-minute period (t=0.5), \Lambda = 10 \times 0.5 = 5.

 NOTES: EXAMPLES OF POISSON DISTRIBUTION - I

This session focuses on the practical application of the Poisson

distribution using Microsoft Excel, highlighting its utility in business
scenarios like customer arrivals and social media engagement.

1. Excel Implementation for Poisson Distribution

While Excel lacks a single dedicated inbuilt function like [Link], a

probability table can be constructed using other existing functions.

Key Advantages of Excel Tables:

 Accuracy: Simplifies complex calculations and limits human error.

 Versatility: Allows for quick re-calculations across different

business examples.

 Sensitivity Analysis: Facilitates visualizing how changes in the

parameter \Lambda (Capital Lambda) affect probabilities and PMF
plots.

Formula Construction in Excel:

 Probability Mass Function (PMF): To calculate f(y) = \frac{e^{-\

Lambda} \Lambda^y}{y!}, use the formula: =EXP(-D3) * (D3^B6) /
FACT(B6).

o EXP(-D3) handles e^{-\Lambda}.

o D3^B6 handles \Lambda^y.

o FACT(B6) calculates the factorial of y.

 Cumulative Distribution Function (CDF): Calculated in a

separate column by summing the PMF of the current value of y with
the previous cumulative total.

 Tail Probabilities (P(Y \ge y)): Derived by subtracting the

preceding cumulative probability from the total probability of 1.

2. Case Study: Customer Arrivals (Sheetal Dhaba)

Sheetal Dhaba models customer arrivals per hour to address staffing

challenges. Arrivals are assumed to be independent and stationary,
making the Poisson distribution an appropriate model.

Parameter: \Lambda = 10 (average customers per hour).

 Probability Calculation/
Business Question
Statement Result

Exactly 8 customers in 1 0.113 (from cell

P(Y = 8)
hour C14)

More than 10 customers in 0.417 (from cell

P(Y \ge 11)
1 hour E17)

Exactly 4 customers in 30 P(Y = 4) with \Lambda 0.175 (from cell

mins =5 C10)

At most 10 customers in P(Y \le 10) with \ 0.986 (from cell

30 mins Lambda = 5 D16)

Critical Note on Intervals: When the time interval changes (e.g., from 1
hour to 30 minutes), the parameter \Lambda must be scaled
proportionally (e.g., from 10 to 5).

3. Case Study: Social Media Engagement (Sriram Pallikar)

Sriram models the number of "likes" on his posts to track engagement
over weeks and days.

Weekly Parameter: \Lambda = 14 (average likes per week). Daily

Parameter: \Lambda = 2 (since 14 likes / 7 days = 2).

Probability
Business Question Result
Statement

At least 10 likes in a week P(Y \ge 10) 0.891 (from cell E16)

Between 10 and 20 likes 0.843 (summing PMF

P(10 \le Y \le 20)
(inclusive) values)

P(Y \ge 5) with \

At least 5 likes in a day 0.053 (from cell E11)
Lambda = 2

P(Y = 0) with \
No likes on a given day 0.135 (from cell C6)
Lambda = 2

4. Summary of Key Learning Points

 Poisson as a Rate: Small \lambda represents the rate (e.g.,
10/hour), while Capital \Lambda is the average for the specific time t
being analyzed (\Lambda = \lambda \cdot t).

 Independent Intervals: The model assumes events in one interval

do not affect those in another.

 Tail Probability Precision: Users must be careful to distinguish

between "greater than" (>) and "greater than or equal to" (\ge)
when reading Excel tables to ensure they select the correct cell.
 NOTES: EXAMPLES OF POISSON DISTRIBUTION - II

This transcript highlights a classic business application of the Poisson

distribution: call center operations. By modeling incoming call traffic,
managers can better understand how randomness affects staffing and
customer service quality.

1. Case Study: [Link] Call Center (Mihir Joshi)

Mihir Joshi seeks to evaluate customer service levels based on current

staffing. To simplify the analysis, the following assumptions are made:

 Stationarity: The probability of calls arriving depends only on the

length of the time period, not the time of day (e.g., morning and
afternoon intervals are treated similarly).

 Independence: The number of calls received in one hour is

independent of the preceding hour.

Note: While these assumptions oversimplify reality, they provide a

starting point. More advanced models, such as non-stationary Poisson
processes, can be used later to account for time-varying traffic.

2. Parameter Scaling and Problem Solving

The primary challenge in these problems is scaling the average arrival

rate to match the specific time interval of interest.

Baseline Parameter: Average of 6 calls in a 15-minute interval.

 Interval of Scaled Probability
Business Question Result
Interest Mean (Λ) Target

0.18
Probability of receiving
5 minutes 6 \div 3 = 2 P(Y = 3) (approx.
3 calls
20%)

Probability of exactly
15 minutes 6 P(Y = 10) 0.041
10 calls

(6 \div 15) \
Probability of no calls
3 minutes times 3 = P(Y = 0) 0.301
during a stretch break
1.2

3. Practical Implications for Staffing

 Agent Fatigue: To reduce fatigue, Mihir encourages 3-minute

stretch breaks when agents have no calls waiting.

 Interruption Risk: The data suggests there is only a 30.1%

chance that an agent can take a 3-minute break without being
interrupted by a new call.

 Waiting Times: Periods of high random arrivals (like receiving 3

calls in 5 minutes) can significantly increase customer waiting times
and agent workload.
 NOTES: EXAMPLES OF POISSON DISTRIBUTION - II

This session concludes the discussion on discrete probability distributions,

focusing on advanced Poisson distribution applications in crowd
management, quality control (spatial Poisson), and cyber security. It also
summarizes the relationships between the Bernoulli, Binomial, and
Poisson models.

1. Crowd Management Case Study: Prozone Mall

Mansoor Khan, the shift manager at Prozone Mall in Aurangabad, uses the
Poisson distribution to understand visitor arrival patterns at entry gates to
prevent overcrowding.

QUESTION IS TO BE SOLVED FOR 20 TO 30 RATHER THAN 10 TO

20.

 Base Scenario: Arrivals average 180 visitors per hour.

o Interval Conversion: For a 10-minute window, the average (\

Lambda) is 180 / 6 = \mathbf{30}.

o Probability Analysis: The probability of having 10 to 20

people collect during a 10-minute period is 0.526.

 Sensitivity Analysis:

o Low Arrival Rate (150/hr): \Lambda = 25 for 10 minutes.

The probability of 10–20 arrivals increases to 0.73.

o High Arrival Rate (210/hr): \Lambda = 35 for 10 minutes.

The probability of 10–20 arrivals decreases to 0.225.
  Key Insight: Although it seems counterintuitive, as \Lambda shifts,
the "mass" of the symmetric distribution moves. When the mean
moves from 25 to 35, less probability is assigned to the fixed
window of 10–20 arrivals.

2. Spatial Poisson Distribution: Highway Quality Control

The Poisson distribution can model occurrences over space (length, area,
volume) rather than just time. Applications include defects in cables,
fabric misprints, potholes, or defects in silicon wafers.

Case: Maharashtra State Highway No. 3 (Solapur to Nanded)

Shivali Patil inspects resurfacing quality, assuming an average of 3
defects per 1 km stretch (\Lambda = 3).

 1 km Stretch Analysis:

o At least 3 defects: P(Y \ge 3) = \mathbf{0.577} (Tail

probability).

o No more than 5 defects: P(Y \le 5) = \mathbf{0.916}

(Cumulative probability).

 8 km Stretch Analysis:

o Parameter Scaling: New \Lambda = 3 \times 8 = \

mathbf{24}.

o No more than 40 defects: P(Y \le 40) = \mathbf{0.999}.

o At least 20 defects: P(Y \ge 20) = \mathbf{0.82}.

  Common Error: Avoid "compounding" probabilities (0.916^8). This
incorrectly assumes success requires exactly \le 5 defects in every
individual kilometer, whereas the total 8 km stretch allows for some
kilometers to be worse as long as the overall count stays low.

4. Text Analysis: Cyber Security Monitoring

Nitin Kulkarni at the NIA uses Poisson to monitor "flags" (sensitive words)
in WhatsApp conversations to identify security threats.

 Single Conversation (\Lambda = 5):

o Between 2 and 8 flags (inclusive): P(2 \le Y \le 8) = \

mathbf{0.891}.

o More than 10 flags: P(Y \ge 11) = \mathbf{0.032}.

 Two Successive Conversations (\Lambda = 5 \times 2 = \

mathbf{10}):

o Between 4 and 16 flags: P(4 \le Y \le 16) = \mathbf{0.963}.

o Less than 4 flags: P(Y \le 3) = \mathbf{0.01}.

4. Summary of Discrete Distributions

The course reviewed three fundamental discrete distributions:

Distributi
Use Case Key Parameters Note
on

Bernoulli Single trial, 2 outcomes p (prob. of Basis for others

Distributi
Use Case Key Parameters Note
on

success)

Binomial n independent trials n, p Sum of Bernoulli

Occurrences over \Lambda (Mean = Related to

Poisson
time/space Var) Binomial

Key Commonalities:

 PMF (f(x)): Crucial for determining probabilities of specific

outcomes.

 CDF (F(x)): Provides the probability of taking a value less than or

equal to x.

 Calculations: Expectations (E[X]) and Variances (Var(X)) are all

derived from the PMF.

The Poisson-Binomial Connection:

If the probability of success (p) in a Binomial distribution is very small, the

variance (np(1-p)) becomes very close to the mean (np). In this state, the
Binomial and Poisson distributions look very similar.