Probability and Statistical Distributions

Introduction

Business decisions are made under uncertainty. A retailer must estimate the probability that demand will exceed inventory; a bank must assess the probability of loan default; a manufacturer must quantify the risk that defects exceed an acceptable level; and a digital platform must predict the likelihood that a customer will churn. Probability provides the language for describing uncertainty, while statistical distributions provide structured models for the behaviour of uncertain quantities.

This chapter develops the foundations required to quantify business uncertainty and risk. It explains probability rules, conditional probability, Bayes’ theorem, random variables, expected value and variance, and important discrete and continuous distributions. The emphasis is not only on calculation but also on interpretation, assumptions and responsible application.

Learning Objectives

By the end of this chapter, you should be able to:

1. Explain probability concepts, sample spaces, events and probability rules.
2. Apply addition, multiplication and complement rules.
3. Distinguish independent events from mutually exclusive events.
4. Calculate and interpret conditional probabilities.
5. Apply Bayes’ theorem to update probabilities using evidence.
6. Distinguish discrete and continuous random variables.
7. Calculate expected value, variance and standard deviation.
8. Select and apply appropriate discrete and continuous probability distributions.
9. Interpret distribution parameters and recognise modelling limitations.
10. Use probability models to support business and managerial decisions.

1. Foundations of Probability

1.1 Experiments, outcomes and sample spaces

A random experiment is a process whose outcome cannot be predicted with certainty before it occurs, although the possible outcomes can be specified. Examples include selecting a customer at random, observing the number of daily complaints, or recording whether a loan applicant defaults.

An outcome is one possible result of an experiment. The sample space, denoted by S, is the set of all possible outcomes. An event is a subset of the sample space. For example, if a customer survey records satisfaction as low, medium or high, then S = {low, medium, high}, and the event “satisfied” might be {high}.

The probability of an event A is written as P(A). It satisfies the following axioms:

If A and B are mutually exclusive events, meaning they cannot occur together, then:

Here, ∪ means “A or B”. The symbol ∩ means “A and B”.

1.2 Complement rule

The complement of A, written Aᶜ, is the event that A does not occur. Because either A or Aᶜ must occur:

This rule is particularly useful when the probability of “at least one” event is easier to calculate through its complement, “none”.

1.3 Addition rule for non-exclusive events

Events may overlap. For any two events A and B:

The intersection is subtracted because it is counted twice when P(A) and P(B) are added.

Worked Example 1: Overlapping customer characteristics

A retailer finds that 45% of customers use its mobile application, 30% use a loyalty card and 15% use both. The probability that a randomly selected customer uses at least one of these services is:

P(app ∪ card) = 0.45 + 0.30 − 0.15 = 0.60.

Thus, the probability is 0.60, or 60%.

2. Conditional Probability and Dependence

Conditional probability measures the probability of one event given that another event has occurred. For events A and B, where P(B) > 0:

The notation P(A | B) is read as “the probability of A given B”. The condition changes the relevant reference population from all observations to those in B.

Rearranging gives the multiplication rule:

Equivalently:

Independence

Events A and B are independent if knowledge that one occurred does not change the probability of the other. Therefore:

and

Independence is an assumption that must be justified. Two events can be unrelated in one business context but dependent in another. For example, whether two machines fail may be approximately independent under separate power supplies, but dependent if both share a faulty electrical system.

Mutual exclusivity versus independence

Mutually exclusive events cannot occur together, so P(A ∩ B) = 0. If both events have positive probability, they cannot also be independent, because independence would require P(A ∩ B) = P(A)P(B) > 0. This distinction is frequently misunderstood.

Worked Example 2: Conditional default risk

Suppose 8% of all business loans default. Among loans with a weak credit score, 20% default. If 25% of loans have weak credit scores, then:

P(default ∩ weak score) = P(default | weak score)P(weak score)

= 0.20 × 0.25 = 0.05.

Thus, 5% of all loans are both weak-score and defaulting loans. The remaining defaulting loans are associated with other credit-score categories.

3. Bayes’ Theorem

Bayes’ theorem reverses a conditional probability. It is useful when evidence is observed and the analyst wants to infer the probability of a possible cause.

For events A and B, where P(B) > 0:

If A₁, A₂, …, Aₖ are mutually exclusive and collectively exhaustive possible causes, then the law of total probability gives:

Therefore:

The prior probability P(A) represents belief before observing evidence. The likelihood P(B | A) measures how probable the evidence is if A is true. The posterior probability P(A | B) is the updated belief after observing evidence.

Worked Example 3: Fraud-alert investigation

A payment processor estimates that 2% of transactions are genuinely fraudulent. Its alert system flags 90% of fraudulent transactions, but also flags 5% of legitimate transactions. If a transaction is flagged, what is the probability that it is actually fraudulent?

Let F denote fraud and T denote a flag.

P(F) = 0.02, P(T | F) = 0.90, P(Fᶜ) = 0.98, and P(T | Fᶜ) = 0.05.

First calculate the total probability of a flag:

P(T) = (0.90 × 0.02) + (0.05 × 0.98)

= 0.018 + 0.049 = 0.067.

Then:

P(F | T) = (0.90 × 0.02) ÷ 0.067 ≈ 0.2687.

Therefore, only approximately 26.9% of flagged transactions are fraudulent. The result illustrates the base-rate effect: even an apparently accurate detection system may generate many false positives when the underlying event is rare.

4. Random Variables

A random variable assigns a numerical value to each outcome of a random experiment. It is usually denoted by X. A random variable may represent daily sales, number of claims, delivery time or customer lifetime value.

4.1 Discrete random variables

A discrete random variable takes countable values, such as 0, 1, 2, … complaints. Its probability mass function, or PMF, is:

The PMF must satisfy:

and

4.2 Continuous random variables

A continuous random variable can take any value within an interval, such as delivery time or product weight. It is described by a probability density function, or PDF, f(x). Probabilities are areas under the density curve:

For a continuous variable, the probability of one exact value is zero:

This does not mean the value cannot occur. It means probability is assigned to intervals rather than individual points.

The cumulative distribution function, or CDF, is:

For a continuous variable:

5. Expected Value, Variance and Risk

5.1 Expected value

The expected value is the long-run average or probability-weighted mean of a random variable. For a discrete variable:

For a continuous variable:

The expected value need not be an outcome that can actually occur. For example, the expected number of claims might be 4.6 even though the number of claims must be an integer.

5.2 Variance and standard deviation

Variance measures dispersion around the expected value:

where μ = E(X). An equivalent computational formula is:

The standard deviation is:

Variance is expressed in squared units, whereas standard deviation has the same units as X and is usually easier to interpret.

For a linear transformation Y = a + bX:

Adding a constant changes the mean but not the variance; multiplying by b changes the standard deviation by |b|.

Worked Example 4: Expected profit

A product launch has the following possible outcomes:

The expected profit is:

E(X) = (80,000 × 0.30) + (30,000 × 0.50) + (−20,000 × 0.20)

= 24,000 + 15,000 − 4,000 = $35,000.

The expected profit is $35,000. This is not a promise of profit; it is a probability-weighted measure of long-run average outcome. A risk-sensitive manager would also calculate variance or evaluate downside outcomes.

6. Important Discrete Distributions

6.1 Bernoulli distribution

A Bernoulli random variable represents one trial with two outcomes, often coded as success = 1 and failure = 0. If P(X = 1) = p, then P(X = 0) = 1 − p.

Its probability mass function is:

Its mean and variance are:

Applications include whether a customer responds to an offer, whether a transaction is fraudulent and whether a machine fails during a shift.

6.2 Binomial distribution

The binomial distribution counts successes in n Bernoulli trials. It requires:

• a fixed number n of trials;
• two outcomes per trial;
• a constant success probability p;
• independence between trials.

If X ~ Binomial(n, p), then:

where C(n,x) = n! ÷ [x!(n − x)!]. Its mean and variance are:

6.3 Poisson distribution

The Poisson distribution models the number of events occurring in a fixed interval of time, space or exposure when events occur independently at a constant average rate λ. Its probability mass function is:

For a Poisson variable:

Examples include calls arriving at a service centre, insurance claims per day or website errors per hour. The constant-rate and independence assumptions should be assessed; demand may vary by hour, day or season, and events may cluster.

Worked Example 5: Call-centre demand

A call centre receives an average of 3 complaints per hour. Assuming a Poisson model, the probability of exactly 2 complaints in an hour is:

P(X = 2) = e⁻³ 3² ÷ 2! ≈ 0.2240.

Thus, the probability is approximately 22.4%.

6.4 Geometric distribution

The geometric distribution models the number of independent trials required to obtain the first success, when each trial has constant success probability p. Under the convention that X counts the trial on which the first success occurs:

Its expected value is 1 ÷ p and its variance is (1 − p) ÷ p². The convention must be stated because some texts define the geometric variable as the number of failures before the first success.

7. Important Continuous Distributions

7.1 Uniform distribution

A continuous uniform variable assigns equal density to all values between a and b, where a < b. Its density is:

Its mean and variance are:

The uniform model is appropriate only when equal likelihood across the interval is defensible.

7.2 Normal distribution

The normal distribution is continuous, symmetric and bell-shaped. It is determined by mean μ and variance σ², and is written X ~ N(μ, σ²). Its density is:

The standardised value, or z-score, is:

A z-score indicates how many standard deviations an observation lies above or below the mean. The empirical rule states that approximately 68%, 95% and 99.7% of observations lie within 1, 2 and 3 standard deviations of the mean, respectively, when the normal model is appropriate.

The normal distribution is often useful for measurement errors, aggregate demand and residuals, but it is not suitable for every business variable. Sales counts, waiting times and proportions may be skewed or bounded.

7.3 Exponential distribution

The exponential distribution models a non-negative waiting time until the next event in a Poisson process with rate λ > 0. Its density is:

Its mean and variance are:

It has the memoryless property:

This property may be unrealistic where waiting time depends on elapsed time, service congestion or scheduled arrivals.

7.4 Lognormal distribution

A positive variable X is lognormally distributed if ln(X) is normally distributed. The lognormal distribution is useful for variables such as income, transaction value and firm size that are positive and right-skewed. It cannot produce negative values and should not be confused with a normal distribution applied to untransformed data.

8. Selecting and Evaluating a Distribution

Distribution selection should begin with the nature of the variable and the business process, not with a desire to use a familiar formula.

Analysts should examine histograms, summary statistics, time plots and empirical frequencies. Goodness-of-fit tests and graphical diagnostics can help, but statistical fit alone does not validate the business assumptions. A model can fit historical data and still fail when pricing, policy, seasonality or customer behaviour changes.

9. Applied Case: Inventory Risk at Northstar Electronics

The following is an illustrative case.

Background

Northstar Electronics sells a popular accessory. Historical daily demand averages 40 units, with substantial variation. The operations manager must decide whether to stock 50 units each day. Unsold units cost $2 each to hold or discount, while a stockout has an estimated opportunity cost of $12 per unmet unit.

Problem

The manager wants to estimate the probability of a stockout and understand whether the expected cost justifies increasing inventory.

Analysis

Suppose daily demand X is approximated by a Poisson distribution with λ = 40. A stockout occurs when X > 50. Therefore:

P(stockout) = P(X > 50) = 1 − P(X ≤ 50).

The exact cumulative probability can be calculated using statistical software. A rough normal approximation may also be considered because λ is reasonably large. For a Poisson variable, μ = λ = 40 and σ = √40 ≈ 6.325. With a continuity correction:

P(X > 50) ≈ P(Y > 50.5), where Y ~ N(40, 40).

The corresponding z-score is:

z = (50.5 − 40) ÷ 6.325 ≈ 1.66.

The upper-tail probability beyond z = 1.66 is approximately 0.048. Thus, the estimated stockout probability is about 4.8% under the approximation.

Approach and decision considerations

The manager should not use the probability alone. Expected shortage cost requires the expected number of unmet units, E[(X − 50)⁺], where (u)⁺ means max(u, 0). Holding cost depends on expected leftover inventory, E[(50 − X)⁺]. These quantities can be estimated from historical data or simulated from the fitted distribution.

The model should also be checked for weekday effects, promotions, product substitutions, supplier constraints and changing demand. If demand differs substantially by day, separate distributions or a predictive model with explanatory variables may be more appropriate than one Poisson distribution.

Outcome

The initial model indicates that stocking 50 units creates a non-negligible but relatively infrequent stockout risk. Management may choose a higher stock level if customer retention is strategically important, or retain 50 units if holding costs and obsolescence are more serious. The final decision should compare expected cost, service-level objectives and the asymmetric consequences of shortage and excess inventory.

Learning points

1. Distribution models translate uncertainty into measurable risk.
2. A model parameter such as λ must have a clear operational interpretation.
3. Approximations require stated conditions and should be validated.
4. Expected cost and service risk may lead to different decisions.
5. Historical distributions may change when business policies change.

10. Technology, Governance and Ethical Use

Spreadsheet functions, statistical packages, simulation tools and programming languages can calculate distribution probabilities efficiently. Predictive analytics platforms also estimate distributions automatically. However, automation does not remove the need for human judgement.

Probability estimates used in lending, insurance, recruitment or fraud detection may affect individuals materially. Analysts should document data sources, assumptions, model versions and validation results. They should assess whether variables act as inappropriate proxies for protected characteristics, monitor disparate error rates and explain uncertainty to decision-makers. A probability is not a statement of certainty or personal worth; it is a model-based estimate conditional on data and assumptions.

Important governance practices include:

• distinguish observed frequencies from model-based probabilities;
• report confidence or uncertainty where appropriate;
• test models on out-of-sample data;
• monitor drift in rates and distributions;
• avoid treating correlation as causation;
• preserve audit trails for consequential decisions;
• communicate base rates and false-positive consequences;
• review models when operating conditions change.

Challenges and Limitations

Probability distributions simplify reality. Common limitations include non-independence, changing event rates, outliers, seasonality, censored observations, measurement error and selection bias. A normal model may underestimate extreme losses because its tails are too light. A Poisson model may underestimate variation when events are clustered; this is called overdispersion. A fitted distribution may also hide multiple customer segments with different behaviours.

Model risk is especially important in financial and operational decisions. Analysts should compare alternative models, conduct sensitivity analysis and examine consequences under adverse scenarios. The best distribution is not necessarily the one with the smallest historical error; it is the one whose assumptions, predictive performance and decision consequences are acceptable for the intended use.

Chapter Summary

Probability provides rules for quantifying uncertain events. The complement rule, addition rule and multiplication rule support calculations involving single and combined events. Conditional probability changes the reference population, while Bayes’ theorem updates a prior probability using evidence. Independence is a substantive assumption and must not be confused with mutual exclusivity.

Random variables convert uncertain outcomes into numerical quantities. Discrete variables use probability mass functions, whereas continuous variables use density functions and interval probabilities. Expected value measures a probability-weighted average; variance and standard deviation measure dispersion and risk.

The Bernoulli and binomial distributions model binary outcomes and counts of successes. The Poisson distribution models event counts over an interval, while the geometric distribution models trials until a first success. The uniform, normal, exponential and lognormal distributions model different continuous phenomena. Distribution selection must be based on the variable, process, data and assumptions. In business analytics, probability models support inventory, credit, fraud, service and investment decisions, but responsible use requires validation, transparency, monitoring and awareness of uncertainty.

Key Terms

Probability: A numerical measure between 0 and 1 describing the likelihood of an event.

Sample space: The set of all possible outcomes of an experiment.

Event: A subset of the sample space.

Complement: The event that a specified event does not occur.

Intersection: The event that two or more events occur together.

Union: The event that at least one of two or more events occurs.

Conditional probability: The probability of an event given that another event has occurred.

Independence: A condition in which occurrence of one event does not change the probability of another.

Bayes’ theorem: A rule for updating a probability using observed evidence.

Random variable: A numerical assignment to outcomes of a random experiment.

PMF: A probability mass function for a discrete random variable.

PDF: A probability density function for a continuous random variable.

CDF: A cumulative distribution function giving P(X ≤ x).

Expected value: The probability-weighted mean of a random variable.

Variance: The expected squared deviation from the mean.

Standard deviation: The square root of variance.

Bernoulli distribution: A distribution for one binary trial.

Binomial distribution: A distribution for successes in a fixed number of independent trials.

Poisson distribution: A distribution for event counts in a fixed interval at a constant average rate.

Normal distribution: A symmetric continuous distribution defined by its mean and variance.

Review Questions

Short Answer Questions

1. What is the difference between a sample space and an event?
2. State the complement rule and explain when it is useful.
3. What does conditional probability measure?
4. Distinguish a probability mass function from a probability density function.
5. Why can mutually exclusive events with positive probabilities not be independent?

Long Answer Questions

1. Explain the addition and multiplication rules of probability, using business examples.
2. Describe Bayes’ theorem and explain the roles of prior probability, likelihood and posterior probability.
3. Compare the Bernoulli, binomial and Poisson distributions, including their assumptions and applications.
4. Explain expected value, variance and standard deviation and discuss their relevance to business risk.
5. Discuss how an analyst should evaluate whether a normal or Poisson distribution is appropriate for a business dataset.

Analytical / Application Questions

1. A bank’s fraud system flags 8% of legitimate transactions and 92% of fraudulent transactions. If 1% of transactions are fraudulent, calculate the probability that a flagged transaction is genuinely fraudulent. Interpret the result for investigators.
2. A service centre receives an average of 5 calls per minute. Explain how a Poisson model could be used to estimate staffing risk, and identify at least three assumptions that should be checked.
3. An investment has possible returns of 18%, 6% and −12%, with probabilities 0.25, 0.55 and 0.20. Calculate the expected return and explain why the expected return alone is insufficient for an investment decision.

Multiple Choice Questions with Answers

Question 1

If P(A) = 0.40 and P(B) = 0.30, and A and B are mutually exclusive, what is P(A ∪ B)?

A. 0.10

B. 0.30

C. 0.70

D. 1.20

Correct Answer: C. 0.70

Question 2

Which expression correctly represents conditional probability?

A. P(A) ÷ P(B)

B. P(A ∩ B) ÷ P(B)

C. P(A ∪ B) ÷ P(B)

D. P(B) ÷ P(A ∩ B)

Correct Answer: B. P(A ∩ B) ÷ P(B)

Question 3

A disease affects 2% of a population. A test detects 95% of diseased cases and has a 10% false-positive rate. Approximately what is the probability that a person with a positive test has the disease?

A. 2.0%

B. 16.2%

C. 50.0%

D. 95.0%

Correct Answer: B. 16.2%

Question 4

Which condition is required for a binomial distribution?

A. The number of trials is unlimited

B. The success probability changes after every trial

C. Trials are independent and have a constant success probability

D. The variable must be continuous

Correct Answer: C. Trials are independent and have a constant success probability

Question 5

If X has a Poisson distribution with λ = 7, what are its mean and variance?

A. Mean 7 and variance 49

B. Mean 49 and variance 7

C. Mean 7 and variance 7

D. Mean 0 and variance 7

Correct Answer: C. Mean 7 and variance 7

Question 6

A random variable has outcomes 0 and 10 with probabilities 0.7 and 0.3. What is its expected value?

A. 0.3

B. 3

C. 7

D. 10

Correct Answer: B. 3

Question 7

For a continuous random variable, what is P(X = 4) under a standard continuous probability model?

A. 0

B. 0.25

C. 0.50

D. 1

Correct Answer: A. 0

Question 8

Which distribution is most directly associated with waiting time until the next event in a constant-rate Poisson process?

A. Binomial

B. Exponential

C. Bernoulli

D. Uniform discrete

Correct Answer: B. Exponential

Question 9

If Y = 5 + 2X, how does Var(Y) relate to Var(X)?

A. Var(Y) = Var(X) + 5

B. Var(Y) = 2Var(X)

C. Var(Y) = 4Var(X)

D. Var(Y) = 25 + Var(X)

Correct Answer: C. Var(Y) = 4Var(X)

Question 10

A z-score of −2 indicates that an observation is:

A. Two units above the mean

B. Two standard deviations below the mean

C. Twice the variance below the mean

D. At the median

Correct Answer: B. Two standard deviations below the mean