Geometric Distribution - College Statistics

SECTION 1

Historical Context & Motivation

The study of repeated independent trials with two outcomes — success and failure — is one of the oldest threads in probability theory, stretching back to the correspondence between Pascal and Fermat in the mid-seventeenth century. As mathematicians sought to formalize gambling problems and actuarial questions, they naturally encountered scenarios where the quantity of interest was not the total number of successes in a fixed sample, but rather how many trials one must endure before the first success occurs. This waiting-time perspective gave rise to what we now call the geometric distribution, named for the geometric series that emerges when summing its probability mass function over all possible outcomes.

1654 • Pascal–Fermat Correspondence

Blaise Pascal and Pierre de Fermat exchanged letters on the problem of points, laying the groundwork for analyzing sequences of independent trials with fixed probability, the very framework underpinning the geometric distribution.

1713 • Bernoulli's Ars Conjectandi

Jacob Bernoulli's posthumous treatise formalized the notion of Bernoulli trials — independent experiments with constant success probability — providing the theoretical building blocks for both the binomial and geometric families.

1837 • Poisson's Recherches

Siméon Denis Poisson extended the analysis of rare events and waiting times, connecting discrete waiting-time distributions to the continuous exponential distribution and establishing the memoryless property as a unifying concept.

20th c. • Modern Applications

The geometric distribution became a cornerstone in quality control, network reliability, queueing theory, and genomics — anywhere analysts model the number of attempts needed to observe a first occurrence.

The central question that motivates the geometric distribution is deceptively simple: If each trial independently succeeds with probability p, how many trials will we need until we observe our first success? Answering this question rigorously requires a distribution that assigns probabilities to every positive integer (or every non-negative integer, depending on convention), and whose terms decrease geometrically — hence the name. Understanding this distribution equips you with one of the most elegant and widely applicable models in the statistician's toolkit.

SECTION 2

Core Principles & Definitions

Before writing down any formulas, it is essential to internalize the assumptions that define a geometric experiment. The geometric distribution inherits every assumption of a Bernoulli process — a sequence of trials that are independent, identically distributed, and binary — but adds one additional structural element: the random variable of interest is the trial number on which the first success is observed (or, equivalently, the number of failures before the first success). These assumptions must hold exactly; violations lead to different distributions such as the negative hypergeometric or the beta-geometric.

Binary Outcomes

Each trial results in exactly one of two outcomes: success (S) or failure (F). There is no third possibility, and the definition of 'success' is fixed before the experiment begins.

Constant Probability

The probability of success, denoted p, remains identical from trial to trial. The probability of failure on each trial is q = 1 − p, and both p and q are strictly between 0 and 1.

Independence

The outcome of any trial is statistically independent of all other trials. Past failures convey no information about the likelihood of future success — a consequence that directly produces the memoryless property.

Stopping Rule

The experiment continues until the first success is observed, then stops. There is no upper bound on the number of trials; theoretically, one could wait indefinitely, though the probability of doing so decays geometrically to zero.

⚠

Convention Alert

Two conventions coexist in textbooks. Convention A defines X as the number of trials until the first success (support: {1, 2, 3, …}). Convention B defines X as the number of failures before the first success (support: {0, 1, 2, …}). This lesson uses Convention A unless noted otherwise. Always check which convention your textbook or exam adopts, as formulas for the mean and PMF differ accordingly.

✦

KEY TAKEAWAY

Think of the geometric distribution as a relentless coin-flipper who keeps tossing until the coin lands heads. The distribution answers: How many flips will that take? In engineering terms, it is analogous to a reliability test that monitors identical components in series and records the position of the first component to fail. The memoryless property means that no matter how many failures you have already observed, the probability of succeeding on the very next trial remains p — the distribution effectively 'resets' after every failure.

SECTION 3

Visual Explanation — PMF Bar Chart

The probability mass function of the geometric distribution has a strikingly simple shape: a monotonically decreasing sequence of bars, each one shorter than its predecessor by the constant factor q = 1 − p. The following diagram compares the PMF for three different values of p, illustrating how a higher success probability concentrates mass on early trials while a lower success probability spreads it across many trials.

The PMF for three values of p. When p = 0.5 (cyan), most of the probability mass is concentrated on the first few trials. When p = 0.1 (pink), the distribution spreads out, reflecting the longer expected wait.

Several features are immediately visible. First, the mode of every geometric distribution is always k = 1 — the single most likely outcome is success on the very first trial. Second, the bars decay by a constant multiplicative factor of q = 1 − p from one trial to the next, which is why the distribution earns the name geometric. Third, the right tail extends indefinitely, though it becomes vanishingly thin. For p = 0.5, the probability of needing more than ten trials is approximately 0.001; for p = 0.1, the corresponding probability exceeds 0.35, underscoring how dramatically the tail behavior depends on the success probability.

SECTION 4

Mathematical Framework

We now derive the key formulas for the geometric distribution under Convention A, where X represents the trial number of the first success. The event X = k means that the first k − 1 trials are failures and the k-th trial is a success. Because trials are independent, we multiply the individual probabilities.

PROBABILITY MASS FUNCTION

P(X = k) = (1 − p)^(k−1) · p, k = 1, 2, 3, …

where p is the probability of success on each trial, k is the trial number of the first success, and $(1 − p)^{k−1}$ accounts for the k − 1 preceding failures.

To confirm this is a valid PMF, sum over all k ≥ 1. The sum p · Σ $(1 − p)^{k−1}$ for k = 1 to ∞ is a geometric series with ratio (1 − p), yielding p · 1/(1 − (1 − p)) = p · (1/p) = 1, as required.

CUMULATIVE DISTRIBUTION FUNCTION

F(k) = P(X ≤ k) = 1 − (1 − p)^k, k = 1, 2, 3, …

This follows from the complement: P(X > k) = $(1 − p)^k$ is the probability that all of the first k trials are failures.

EXPECTED VALUE (MEAN)

E[X] = 1 / p

Derivable via the identity Σ k · $q^{k−1}$ = 1/(1−q)² with q = 1 − p. Multiplying by p gives E[X] = p/(1−q)² = p/p² = 1/p. Intuitively, if each trial has a 20% chance of success, you expect to wait 1/0.20 = 5 trials on average.

VARIANCE AND STANDARD DEVIATION

Var(X) = (1 − p) / p², σ = √((1 − p) / p²)

The variance grows rapidly as p → 0. For p = 0.1, Var(X) = 90 and σ ≈ 9.49, indicating tremendous spread in the waiting time.

The Memoryless Property

The memoryless property is the defining qualitative feature of the geometric distribution among all discrete distributions. Formally, P(X > s + t | X > s) = P(X > t) for all non-negative integers s and t. In words: given that you have already failed s times, the conditional distribution of the remaining waiting time is identical to the original distribution — as if no trials had yet occurred. The proof is a direct computation: P(X > s + t | X > s) = $(1 − p)^{s+t}$ / $(1 − p)^s$ = $(1 − p)^t$ = P(X > t). The geometric distribution is, in fact, the only discrete distribution possessing this property, just as the exponential distribution is the only continuous distribution with the analogous memoryless property.

SECTION 5

CDF Visualization & Quantile Analysis

While the PMF tells us the probability of each individual outcome, the cumulative distribution function (CDF) answers the practically crucial question: what is the probability that the first success occurs on or before trial k? The CDF is a step function that climbs toward 1 and is useful for computing quantiles — for example, how many trials guarantee a 95% chance of at least one success.

Step-function CDF for p = 0.5 (amber) and p = 0.3 (emerald). The dashed red line marks the 90% threshold. With p = 0.5, three trials suffice to reach 90% cumulative probability; with p = 0.3, you need about seven.

Computing Quantiles

Given the CDF F(k) = 1 − $(1 − p)^k$ , solving for the smallest k such that F(k) ≥ α (where α is the desired cumulative probability) yields the quantile function: k = ⌈ln(1 − α) / ln(1 − p)⌉. For instance, to find the median (α = 0.5) when p = 0.3, compute k = ⌈ln(0.5) / ln(0.7)⌉ = ⌈−0.6931 / −0.3567⌉ = ⌈1.943⌉ = 2. This tells us that, with probability at least 50%, a geometric random variable with p = 0.3 will yield its first success within two trials.

p	E[X] = 1/p	Var(X)	Median	P(X > 10)
0.5	2	2	1	0.001
0.3	3.33	7.78	2	0.028
0.1	10	90	7	0.349
0.01	100	9900	69	0.904

Key summary statistics for selected values of p

SECTION 6

Worked Example

A manufacturing line produces microchips, and each chip independently has a 0.05 probability of being defective. A quality inspector tests chips one at a time. Find (a) the probability that the first defective chip is found on the 8th test, (b) the expected number of chips tested before finding the first defective, and (c) the probability that more than 20 chips are tested before finding a defective one.

Quality Control: First Defective Chip

1
Step 1 — Identify the Parameters

Each chip test is an independent Bernoulli trial. Here, 'success' is defined as finding a defective chip. The success probability is p = 0.05, and the failure probability (non-defective) is q = 1 − 0.05 = 0.95. Let X be the trial number of the first defective chip; X ~ Geometric(p = 0.05).

p = 0.05, q = 0.95, X ~ Geom(0.05)
2
Step 2 — Compute P(X = 8)

Apply the PMF directly: P(X = 8) = $(1 − p)^{8−1}$ × p = $(0.95)^7$ × 0.05. Computing $(0.95)^7$ ≈ 0.6983. Therefore P(X = 8) ≈ 0.6983 × 0.05 ≈ 0.03491.

P(X = 8) ≈ 0.0349 or about 3.49%
3
Step 3 — Compute E[X]

The expected value of a geometric random variable under Convention A is E[X] = 1/p = 1/0.05 = 20. On average, the inspector will test 20 chips before encountering the first defective one.

E[X] = 20 chips
4
Step 4 — Compute P(X > 20)

Using the survival function: P(X > 20) = $(1 − p)^{20}$ = $(0.95)^{20}$ ≈ 0.3585. There is approximately a 35.85% chance that more than 20 chips will be tested without finding a defective one. Notice that even though E[X] = 20, there is still a substantial probability of exceeding the mean — this is a consequence of the distribution's right skew.

P(X > 20) ≈ 0.3585 or about 35.85%
5
Step 5 — Interpret the Results

The low per-trial probability (p = 0.05) leads to a long expected wait and a heavy right tail. The inspector can expect to test about 20 chips on average, but the standard deviation is σ = √(0.95/0.05²) ≈ 19.49, meaning extreme waits are not unusual. The memoryless property implies that even after testing 20 non-defective chips, the expected additional number of tests remains 20.

σ ≈ 19.49; memoryless property applies

SECTION 7

Comparisons with Related Distributions

The geometric distribution belongs to a family of distributions that arise from Bernoulli trials. It is instructive to compare it with its closest relatives — the binomial, negative binomial, and exponential — to understand when each model is appropriate and how the geometric distribution serves as a bridge between them.

Feature	Geometric	Binomial	Negative Binomial	Exponential
Random Variable	Trial of 1st success	# successes in n trials	Trial of r-th success	Time to 1st event (continuous)
Support	{1, 2, 3, …}	{0, 1, …, n}	{r, r+1, r+2, …}	[0, ∞)
Parameters	p	n, p	r, p	λ
Mean	1/p	np	r/p	1/λ
Memoryless?	Yes	No	No	Yes
Relationship	Special case of NegBin (r = 1)	Fixed n, counts successes	Generalizes geometric (r ≥ 1)	Continuous analog of geometric

Comparison of the geometric distribution with related distributions

✦

KEY TAKEAWAY

The geometric distribution occupies a unique position in the Bernoulli-trial ecosystem. It is the simplest waiting-time distribution — the special case of the negative binomial with r = 1 — and it is the discrete counterpart of the exponential distribution. Think of it as the discrete 'radioactive decay' model: at each tick of a clock, there is a fixed probability p that the atom decays, and the distribution of the decay time (measured in ticks) is geometric.

SECTION 8

Connections to Advanced Theory

The geometric distribution is far more than a textbook exercise. It connects to a web of deeper ideas in probability theory, stochastic processes, and information theory. This section previews several of these connections, which you will encounter in more advanced coursework.

Advanced Topic	Connection to Geometric Distribution
Negative Binomial	A sum of r independent Geometric(p) random variables follows a Negative Binomial(r, p) distribution, modeling the waiting time for the r-th success.
Exponential Distribution	As the time step shrinks to zero while p × Δt → λ, the geometric distribution converges to the Exponential(λ) distribution. This is the discrete-to-continuous bridge used in Poisson process theory.
Markov Chains	The first passage time from a transient state to an absorbing state in a two-state Markov chain is geometrically distributed. The memoryless property is equivalent to the Markov property for this chain.
Information Theory	The geometric distribution maximizes entropy among all distributions on {1, 2, 3, …} with a given mean, making it the 'least informative' waiting-time model — a maximum-entropy characterization.
Moment Generating Function	The MGF is M(t) = $pe^t$ / (1 − (1−p) $e^t$ ) for t < −ln(1−p). This compact form enables rapid derivation of all moments and is essential in large-deviation analysis.

Connections between the geometric distribution and advanced topics

Among these connections, the link to the exponential distribution is perhaps the most important for future study. In courses on stochastic processes, you will see that the Poisson process — the fundamental model for random events occurring continuously in time — arises as a limiting case of a sequence of Bernoulli trials. The inter-arrival times in a Poisson process are exponentially distributed, and in the discrete pre-limit, they are geometrically distributed. Mastering the geometric distribution therefore provides the intuitive scaffolding for understanding continuous-time stochastic models.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

A fair six-sided die is rolled repeatedly until a 6 appears. You have already rolled 10 times without seeing a 6. What is the probability that the next roll is a 6? Explain why the answer does not depend on the 10 previous rolls.

Show answer

The probability that the next roll is a 6 is exactly 1/6, regardless of the 10 failures already observed. This is a direct consequence of the memoryless property of the geometric distribution: P(X > s + t | X > s) = P(X > t). The die has no memory of previous outcomes, so given that we have not yet succeeded, the conditional distribution of the remaining waiting time is identical to the original distribution. Each roll is an independent Bernoulli trial with p = 1/6, and independence is the mechanism that produces memorylessness.

Question 2: BASIC CALCULATION

A basketball player has a free-throw success rate of 0.80. Let X be the number of free throws attempted until the first miss. Find P(X = 4) and E[X].

Show answer

Here, 'success' is a miss, so p = 1 − 0.80 = 0.20, and q = 0.80. Using the PMF: P(X = 4) = (0.80)³ × (0.20) = 0.512 × 0.20 = 0.1024. The expected value is E[X] = 1/p = 1/0.20 = 5. On average, the player will shoot 5 free throws before missing. Note the importance of correctly defining which outcome constitutes 'success' in the context of the problem — here it is the miss, not the made shot.

Question 3: INTERMEDIATE

A software tester runs automated tests on a module. Each test independently reveals a bug with probability 0.15. What is the probability that the first bug is found within the first 5 tests? Also compute the variance and standard deviation of the waiting time.

Show answer

Using the CDF: P(X ≤ 5) = 1 − (1 − 0.15)⁵ = 1 − (0.85)⁵ = 1 − 0.4437 = 0.5563. So there is approximately a 55.63% chance of finding a bug within the first 5 tests. For the variance: Var(X) = (1 − p)/p² = 0.85/(0.15)² = 0.85/0.0225 ≈ 37.78. The standard deviation is σ = √37.78 ≈ 6.15 tests. The relatively large standard deviation compared to the mean (E[X] = 1/0.15 ≈ 6.67) reflects the geometric distribution's inherent right skewness and the high variability in waiting times when p is moderate.

Question 4: APPLIED

A network router transmits packets, and each packet independently has a 0.02 probability of being corrupted. The router must transmit at least 100 packets without corruption before triggering a diagnostic test. What is the probability that the diagnostic test is never triggered within a batch of 100 packets, and how many packets, on average, can be sent before the first corrupted packet?

Show answer

Let X ~ Geometric(p = 0.02) be the packet number at which the first corruption occurs. The probability that the first corruption occurs after 100 packets is P(X > 100) =

(1 − 0.02)^{100}

(0.98)^{100}

≈ 0.1326. So the diagnostic test is never triggered with approximately 13.26% probability. The expected number of packets before the first corruption is E[X] = 1/0.02 = 50. This means that, on average, the router encounters its first corrupted packet at the 50th transmission — well before 100 packets — so the diagnostic is likely to trigger in most batches. The variance is (0.98)/(0.02)² = 2450, giving σ ≈ 49.5, which indicates enormous variability in the waiting time.

Question 5: CRITICAL THINKING

Prove that the geometric distribution is the only discrete distribution on {1, 2, 3, …} that possesses the memoryless property. That is, show that if P(X > s + t) = P(X > s) · P(X > t) for all non-negative integers s and t, then X must be geometric.

Show answer

Define g(k) = P(X > k). The memoryless condition states g(s + t) = g(s) · g(t) for all non-negative integers s, t. Setting s = t = 0 gives g(0) = g(0)², so g(0) = 1 (since g(0) = P(X > 0) = 1 for a distribution supported on positive integers). Setting s = t = 1 gives g(2) = g(1)². By induction, g(k) =

g(1)^k

for all k ≥ 0. Let q = g(1) = P(X > 1) = 1 − P(X = 1). For X to be a proper distribution, we need 0 < q < 1. Then P(X > k) =

q^k

, so P(X = k) = P(X > k−1) − P(X > k) =

q^{k−1}

−

q^k

q^{k−1}

(1 − q). Setting p = 1 − q, we obtain P(X = k) =

(1 − p)^{k−1}

· p, which is precisely the geometric PMF. The key mathematical insight is that the only solutions to the Cauchy functional equation f(s + t) = f(s) · f(t) on the non-negative integers with f(0) = 1 are exponential functions of the form f(k) =

a^k

SUMMARY

Geometric Distribution — Summary

Summary

The geometric distribution models the number of independent Bernoulli trials needed to observe the first success, where each trial has a constant success probability p. Its PMF is P(X = k) = $(1 − p)^{k−1}$ · p, its CDF is F(k) = 1 − $(1 − p)^k$ , its expected value is 1/p, and its variance is (1 − p)/p². It is the only discrete distribution possessing the memoryless property, meaning that past failures carry no information about the number of future trials needed.

The geometric distribution is a special case of the negative binomial distribution with r = 1 and serves as the discrete analog of the exponential distribution. It arises naturally in quality control, network reliability, clinical trials, and any setting where we ask: How long must we wait for the first occurrence? Always verify the four core assumptions — binary outcomes, constant p, independence, and stopping at first success — before applying the model.