Probability and Statistical Distributions

Introduction

Business decisions are made under uncertainty. A manager may not know next month’s demand, whether a customer will default, how long a service interaction will take, or whether a new product will succeed. Probability provides a formal language for describing uncertainty, while statistical distributions provide models for the behavior of uncertain quantities. Together, they support forecasting, risk analysis, inventory planning, credit assessment, quality control, project management, and managerial decision-making.

Probability is not a guarantee about an individual event. It is a numerical representation of uncertainty based on assumptions, evidence, or both. A probability model is useful only when its assumptions are made explicit and its outputs are interpreted in context.

Learning Objectives

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

1. Define experiments, outcomes, events, sample spaces, and probability.
2. Apply the addition, multiplication, complement, and conditional-probability rules.
3. Use Bayes’ theorem to update beliefs when new evidence becomes available.
4. Distinguish discrete and continuous random variables.
5. Calculate and interpret expected value, variance, and standard deviation.
6. Select and apply common discrete and continuous probability distributions.
7. Evaluate assumptions, limitations, and ethical implications in business applications.

1. Foundations of Probability

A random experiment is a process whose outcome cannot be known with certainty in advance. Examples include selecting an invoice for audit, observing whether a loan applicant defaults, and recording the number of website purchases in an hour.

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 randomly selected order is either on time or late, the sample space is S = {on time, late}; the event “late” contains one outcome.

A probability function P assigns a number between 0 and 1 to an event:

Here, A denotes an event. P(A) = 0 indicates impossibility, and P(A) = 1 indicates certainty. If outcomes are mutually exclusive and collectively exhaustive, their probabilities sum to 1.

Interpretations of probability

The classical interpretation assigns equal probabilities to equally likely outcomes. The empirical or frequentist interpretation views probability as a long-run relative frequency. The subjective interpretation represents a rational degree of belief based on available information. Business analytics commonly combines empirical data with managerial judgment; the resulting probability should therefore be documented as either estimated, model-based, or judgmental.

Fundamental rules

The complement rule is:

Aᶜ means that event A does not occur. This rule is especially useful when the probability of “at least one” occurrence is easier to calculate through “none.”

For two events A and B, the general addition rule is:

The symbol ∪ means “A or B, or both,” and ∩ means “A and B.” The intersection is subtracted because it is counted twice in P(A) + P(B). If A and B are mutually exclusive, P(A ∩ B) = 0, so P(A ∪ B) = P(A) + P(B).

Conditional probability and independence

Conditional probability measures the probability of A given that B has occurred:

The notation P(A | B) must not be confused with P(A). The condition changes the information set and may change the probability.

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

when P(B) > 0. Equivalent forms include:

Independence is an assumption, not a synonym for “different.” Two events can concern different business processes yet still be correlated through a common economic factor. Conversely, mutually exclusive events with positive probabilities cannot be independent, because occurrence of one prevents occurrence of the other.

Worked example 1: Addition and conditional probability

A retailer finds that 60% of customers use a mobile application, 35% use a loyalty card, and 20% use both. Let M denote mobile-app use and L denote loyalty-card use.

The probability that a customer uses at least one of the two services is:

P(M ∪ L) = 0.60 + 0.35 − 0.20 = 0.75.

The probability that a loyalty-card user also uses the mobile application is:

P(M | L) = P(M ∩ L) / P(L) = 0.20 / 0.35 ≈ 0.5714.

Thus, approximately 57.14% of loyalty-card users use the application. The denominator is 0.35 because the relevant population is loyalty-card users, not all customers.

2. Bayes’ Theorem and Updating Beliefs

Bayes’ theorem reverses a conditional probability. It is particularly important when managers observe an indicator and want to infer an underlying condition, such as inferring fraud from an alert or default risk from a credit score.

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

If A and Aᶜ partition the sample space, the denominator can be expanded as:

Therefore:

P(A) is the prior probability, P(B | A) is the likelihood, and P(A | B) is the posterior probability. The formula demonstrates why a highly accurate-looking signal may still produce many false positives when the underlying condition is rare.

Worked example 2: Fraud screening

Suppose 1% of transactions are genuinely fraudulent. A screening system flags 95% of fraudulent transactions and flags 4% of legitimate transactions. Let F represent fraud and + represent a flag.

P(F) = 0.01, P(+ | F) = 0.95, P(Fᶜ) = 0.99, and P(+ | Fᶜ) = 0.04.

The total probability of a flag is:

P(+) = (0.95)(0.01) + (0.04)(0.99) = 0.0095 + 0.0396 = 0.0491.

The probability that a flagged transaction is actually fraudulent is:

P(F | +) = (0.95 × 0.01) / 0.0491 ≈ 0.1935.

Only about 19.35% of flagged transactions are fraudulent. This does not mean the system is useless. It means that a flag should initiate investigation rather than automatically trigger rejection or punishment. Base rates, investigation costs, customer harm, and appeal procedures must be considered.

3. Random Variables

A random variable assigns a numerical value to each outcome of a random experiment. A random variable may be discrete or continuous.

A discrete random variable takes a finite or countably infinite set of values, such as the number of complaints received in a day. Its probability mass function (PMF) is:

A valid PMF must satisfy p(x) ≥ 0 for every x and ∑p(x) = 1.

A continuous random variable can take any value within an interval, such as delivery time or transaction value. Its probability density function (PDF), f(x), satisfies:

A PDF must be nonnegative and have total area 1. For a continuous variable, the probability of one exact value is zero: P(X = x) = 0. Probabilities concern intervals or ranges, not isolated points.

The cumulative distribution function (CDF) is:

For discrete variables, the CDF is a step function. For continuous variables, it is generally smooth, although not every continuous distribution has a simple closed-form expression.

4. Expected Value, Variance, and Risk

The expected value is the probability-weighted average of possible outcomes. For a discrete random variable:

For a continuous random variable:

The expected value need not be an outcome that can actually occur. It is a long-run average or model-implied center, not a prediction of a particular observation.

For a function g(X), the discrete form of the law of the unconscious statistician is:

and the continuous form is E[g(X)] = ∫g(x)f(x)dx, when the expectation exists.

Variance measures dispersion around the mean μ = E(X):

An equivalent computational form is:

The standard deviation is:

Variance is expressed in squared units, whereas standard deviation has the same units as X. For a linear transformation Y = a + bX:

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

For two random variables X and Y:

If X and Y are independent, Cov(X,Y) = 0, and the variances add. Independence is sufficient but not necessary for zero covariance in some settings; uncorrelated variables need not be independent in general.

Worked example 3: Expected profit

A product produces a profit of $80,000 with probability 0.30, $30,000 with probability 0.50, and a loss of $40,000 with probability 0.20. Let X be profit.

E(X) = (80,000)(0.30) + (30,000)(0.50) + (−40,000)(0.20)

E(X) = 24,000 + 15,000 − 8,000 = $31,000.

The expected profit is $31,000. A risk-neutral decision-maker comparing this option with a certain $25,000 would prefer the option on expected value alone. A risk-averse decision-maker would also examine variance, downside exposure, liquidity, and the organization’s ability to absorb loss.

5. Common Discrete Distributions

Bernoulli distribution

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

Examples include whether a customer renews, whether a payment is late, or whether a machine passes inspection. The model assumes a clearly defined binary outcome and a stable probability p for the relevant population.

Binomial distribution

A binomial variable counts successes in n Bernoulli trials with constant success probability p. The standard assumptions are: fixed n, two outcomes per trial, constant p, and independence of trials. Its PMF is:

where C(n,k) = n!/[k!(n−k)!] counts the possible arrangements. The mean and variance are:

Sampling without replacement can sometimes be approximated by a binomial model when the sample is small relative to the population, but the exact model may be hypergeometric.

Poisson distribution

The Poisson distribution models the number of events in a fixed interval when events occur independently at a constant average rate λ. Its PMF is:

Its mean and variance are both λ. The interval may be time, distance, area, or volume, provided the rate is defined consistently. Examples include calls per hour, defects per metre, or arrivals per minute. The assumption of a constant rate may fail during peak periods, promotions, or seasonal cycles.

Geometric distribution

The geometric distribution models the number of trials until the first success, assuming independent Bernoulli trials with constant success probability p. Under the convention that X = 1 means success on the first trial:

Its expected value is 1/p and its variance is (1 − p)/p². It has the memoryless property, which may be unrealistic when customer behavior or learning changes over time.

6. Common Continuous Distributions

Uniform distribution

If all values between a and b are equally likely, X has a continuous uniform distribution. Its density is 1/(b − a) for a ≤ x ≤ b, and zero otherwise. The mean and variance are:

The uniform model is useful when only minimum and maximum values are credible and no interior value is favored. It should not be used merely because it is mathematically convenient.

Normal distribution

The normal distribution is continuous, symmetric, and bell-shaped. It is described by mean μ and standard deviation σ > 0. Its density is:

The standardized variable is:

If X is normal, Z follows a standard normal distribution with mean 0 and standard deviation 1. Standardization permits the use of normal probability tables or software. Approximately 68%, 95%, and 99.7% of observations lie within 1, 2, and 3 standard deviations of the mean, respectively; these are approximations, not universal guarantees for non-normal data.

The normal model often arises as an approximation to a sum or average through the central limit theorem. The theorem does not say that every individual variable is normal. It states, under suitable conditions, that standardized sample means tend toward a normal distribution as sample size increases. Dependence, extreme skewness, heavy tails, and small samples can weaken the approximation.

Exponential distribution

The exponential distribution models a nonnegative waiting time when events occur according to a Poisson process with constant rate λ > 0. Its density is:

Its mean and variance are 1/λ and 1/λ². It is memoryless: the probability of waiting an additional period does not depend on how long one has already waited. This property is often a strong assumption in service systems.

Lognormal distribution

A positive variable X is lognormally distributed if ln(X) is normally distributed. Lognormal models are useful for quantities such as transaction values, income, and project durations when multiplicative effects produce right skewness. The mean is sensitive to the parameters of the underlying logarithmic distribution, so analysts should not interpret the median as the arithmetic mean.

7. Choosing and Assessing a Distribution

Distribution selection should begin with the business-generating process, not with a preferred formula. The analyst should ask:

• What is being counted or measured?
• Is the variable discrete, continuous, bounded, or nonnegative?
• Is there a fixed number of trials or a variable number of opportunities?
• Are observations independent, and is the rate stable?
• Does the data show skewness, seasonality, dependence, or outliers?
• What decision will the model support, and how costly are errors?

A model can be evaluated using histograms, empirical CDFs, Q–Q plots, summary statistics, likelihood-based methods, and goodness-of-fit tests. Statistical fit alone is insufficient: a distribution may fit historical data but fail after a policy change, product launch, recession, or change in measurement system. Out-of-sample validation and sensitivity analysis are essential.

Parameter uncertainty should be reported. For example, an estimated Poisson rate is not known with certainty, and a probability derived from a small sample may be unstable. Scenario analysis, bootstrap methods, Bayesian modeling, and simulation can quantify this additional uncertainty.

8. Detailed Applied Case: Capacity Planning for a Service Centre

Background

This illustrative case concerns a digital bank planning staffing for its customer-support centre. Management wants to estimate the risk that more than 12 support requests arrive in a 10-minute interval.

Problem and assumptions

Historical data suggest an average of 9 requests per 10 minutes. As an initial model, arrivals are represented by X ~ Poisson(λ = 9). This assumes a constant arrival rate and independent arrivals within the interval. The manager recognizes that marketing campaigns and payroll days may violate these assumptions.

The probability of more than 12 arrivals is:

P(X > 12) = 1 − P(X ≤ 12)

Using the Poisson PMF, software or a verified probability table gives approximately P(X ≤ 12) = 0.8758. Hence:

P(X > 12) ≈ 1 − 0.8758 = 0.1242.

The estimated probability is about 12.42% for a randomly selected interval under the model.

Analysis and decision

A 12.42% overload probability may be unacceptable if waiting-time commitments are strict. Management could add a flexible agent, introduce asynchronous support, or use a triage system. However, the decision should not be based on this probability alone. The cost of extra staffing, the cost of customer delay, service-level targets, and the relationship between arrivals and service times must be incorporated.

If arrivals are materially higher during certain hours, a single λ = 9 model averages across unlike periods and may underestimate peak risk. A nonstationary model with hour-specific rates, or a queueing simulation that includes service times and abandonment, would be more appropriate. The model should be recalibrated after operational changes.

Outcome and learning points

The Poisson model provides a transparent starting point and a measurable overload risk. Its value lies not in pretending to predict every interval, but in structuring a decision and exposing assumptions. The case illustrates four principles: use a distribution that matches the event-generating process; distinguish an average rate from a peak rate; translate probability into cost and service consequences; and validate the model after implementation.

9. Managerial, Ethical, and Technological Implications

Probability models influence credit approvals, insurance prices, fraud investigations, workforce schedules, and customer targeting. A mathematically correct probability can still support an unfair decision if the data reflect discrimination, unequal measurement, or historical exclusion. Managers should examine protected-group impacts, proxy variables, false-positive and false-negative costs, and the availability of appeal or human review.

Probability outputs should not be presented as certainties. A statement such as “the applicant has a 30% default probability” describes a model estimate for a relevant population, not proof that a particular applicant will default. Communication should include the reference population, time horizon, data period, model version, and uncertainty.

AI systems frequently use probabilistic classification, Bayesian updating, simulation, and predictive distributions. Governance should include data-quality controls, access restrictions, monitoring for drift, documentation of assumptions, reproducibility, and independent validation. Automated decisions require special attention to explainability, privacy, security, and regulatory obligations. Synthetic data and simulation can support testing, but they do not automatically eliminate bias or guarantee confidentiality.

10. Best Practices and Common Errors

1. Define the event and population before assigning a probability.
2. Separate prior probabilities, conditional probabilities, and posterior probabilities.
3. Check whether independence is justified rather than assumed.
4. Keep units consistent, particularly for Poisson rates and waiting times.
5. Distinguish a probability distribution from a sample distribution and from an empirical histogram.
6. Report both central tendency and dispersion.
7. Validate models using current and out-of-sample data.
8. Test sensitivity to plausible parameter changes.
9. Communicate assumptions and limitations with the result.
10. Avoid interpreting expected value as a guaranteed outcome.

Common errors include adding probabilities of overlapping events without subtracting their intersection, confusing P(A | B) with P(B | A), using a normal model for strongly bounded or skewed data without justification, and treating a fitted distribution as an established fact. Another frequent error is ignoring base rates, as illustrated by the fraud-screening example.

Chapter Summary

Probability formalizes uncertainty through events, sample spaces, and probability rules. The complement and addition rules support event calculations, while conditional probability changes the reference population. Bayes’ theorem combines prior beliefs and new evidence to produce posterior probabilities and is essential for interpreting screening signals.

Random variables translate uncertain outcomes into numerical form. Discrete variables are described by probability mass functions; continuous variables are described by density functions and interval probabilities. Expected value summarizes a probability-weighted center, while variance and standard deviation describe dispersion and risk. Linear transformations and covariance rules help analyze combined business outcomes.

Bernoulli and binomial models describe binary trials and counts of successes; Poisson models event counts over an interval; and geometric models trials until a first success. Uniform, normal, exponential, and lognormal distributions provide useful continuous models under different assumptions. Distribution selection must be guided by the data-generating process, validated empirically, and evaluated for parameter uncertainty and model risk.

In business, probability is a decision-support tool rather than a promise. Responsible use requires transparent assumptions, appropriate validation, ethical review, monitoring for drift, and communication that distinguishes model estimates from certainty.

Key Terms

Probability: A numerical representation of 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 does not occur, written Aᶜ.

Mutually exclusive events: Events that cannot occur together.

Conditional probability: The probability of one event given another event.

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

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

Prior probability: Probability assigned before considering new evidence.

Posterior probability: Updated probability after considering evidence.

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

Probability mass function: Function giving probabilities for values of a discrete variable.

Probability density function: Function whose area over an interval gives a continuous probability.

Expected value: Probability-weighted average of a random variable.

Variance: Expected squared deviation from the mean.

Standard deviation: Square root of variance.

Bernoulli distribution: Distribution for one binary trial.

Binomial distribution: Distribution of successes in a fixed number of independent Bernoulli trials.

Poisson distribution: Distribution of event counts in a fixed interval at a constant average rate.

Normal distribution: Symmetric continuous distribution characterized by a mean and standard deviation.

Review Questions

Short Answer Questions

1. What is the difference between a sample space and an event?
2. State the general addition rule for two events and explain why the intersection is subtracted.
3. What does independence mean in probability?
4. Distinguish a discrete random variable from a continuous random variable.
5. Why is expected value not necessarily an outcome that will be observed?

Long Answer Questions

1. Explain Bayes’ theorem and discuss the importance of base rates in business screening systems.
2. Compare the Bernoulli, binomial, geometric, and Poisson distributions, including their assumptions and business applications.
3. Explain expected value, variance, and standard deviation and discuss why a manager should examine all three.
4. Describe how an analyst should select, validate, and monitor a probability distribution for operational planning.
5. Discuss ethical and governance issues that arise when probability models are used in automated business decisions.

Analytical / Application Questions

1. A company estimates that 3% of invoices contain an error. An audit tool detects 90% of erroneous invoices and flags 5% of correct invoices. Calculate the probability that a flagged invoice actually contains an error and interpret the result.
2. A product has a 0.4 probability of generating a $50,000 contribution and a 0.6 probability of generating a $10,000 contribution. Calculate expected contribution and explain why a risk analysis may require more than the expected value.
3. A call centre receives an average of 6 calls per five minutes. Explain whether a Poisson model is appropriate, identify assumptions that should be investigated, and describe how the model could inform staffing.

Multiple Choice Questions with Answers

Question 1

A retailer reports P(A) = 0.60, P(B) = 0.35, and P(A ∩ B) = 0.20. What is P(A ∪ B)?

A. 0.15

B. 0.55

C. 0.75

D. 1.15

Correct Answer: C. 0.75

Question 2

If P(A ∩ B) = 0.18 and P(B) = 0.30, what is P(A | B)?

A. 0.06

B. 0.12

C. 0.60

D. 1.67

Correct Answer: C. 0.60

Question 3

A fraud rate is 1%, detection is 95% for fraud, and the false-positive rate is 4%. Approximately what proportion of flagged transactions are fraudulent?

A. 1.00%

B. 4.00%

C. 19.35%

D. 95.00%

Correct Answer: C. 19.35%

Question 4

Which condition is required for a binomial model?

A. A variable must be continuous

B. The number of trials is fixed and each trial has two outcomes

C. Events must occur at a constant rate per unit time

D. All possible outcomes must have equal probabilities

Correct Answer: B. The number of trials is fixed and each trial has two outcomes

Question 5

If X has a Poisson distribution with λ = 9, what are its mean and variance?

A. Mean 3 and variance 9

B. Mean 9 and variance 3

C. Mean 9 and variance 9

D. Mean 81 and variance 9

Correct Answer: C. Mean 9 and variance 9

Question 6

For a continuous random variable, what is P(X = x) under an ordinary continuous probability model?

A. Always 1

B. Always 0

C. Equal to the density f(x)

D. Equal to the mean

Correct Answer: B. Always 0

Question 7

An investment returns $80,000 with probability 0.30, $30,000 with probability 0.50, and loses $40,000 with probability 0.20. What is its expected profit?

A. $8,000

B. $23,000

C. $31,000

D. $150,000

Correct Answer: C. $31,000

Question 8

Which distribution is commonly used for a nonnegative waiting time under a constant-rate Poisson process?

A. Exponential

B. Binomial

C. Bernoulli

D. Uniform

Correct Answer: A. Exponential

Question 9

If Y = 10 + 3X, how does Var(Y) relate to Var(X)?

A. Var(Y) = Var(X) + 10

B. Var(Y) = 3Var(X)

C. Var(Y) = 9Var(X)

D. Var(Y) = 100 + Var(X)

Correct Answer: C. Var(Y) = 9Var(X)

Question 10

Which statement about Bayes’ theorem is correct?

A. It requires all events to be equally likely

B. It updates a prior probability using evidence

C. It applies only to continuous variables

D. It proves that a conditional probability equals its reverse conditional probability

Correct Answer: B. It updates a prior probability using evidence