Statistics › Section D · Discrete random variables › Topic D.4
The geometric distribution
The binomial model fixes the number of trials and counts the successes. Flip the question - keep rolling, calling, testing until the first success arrives - and the count of trials itself becomes the random variable. That is the geometric distribution.
How long until the first success?
A binomial count answers "how many successes in \(n\) tries?". Many real questions run the other way: how long must you keep going before the first success? Here the number of trials is the outcome, and its model is the geometric distribution: the number of independent, same-p trials needed to reach the first success. It models how many rings before a caller answers, how many applications before a job offer, how many rolls before a six.
The geometric setting
Three conditions define it: each trial succeeds or fails, trials are independent, and the success probability \(p\) is the same on every trial. What distinguishes it from binomial is the missing fourth: \(n\) is not fixed in advance - the count of trials \(X\) is the outcome being measured, and the process stops at the first success. And each value of \(X\) has exactly one arrangement: \(k-1\) failures followed by the success.
The geometric formula
Because each arrangement is unique, no combination factor appears:
The probability of needing \(k\) trials is \(k-1\) failures in a row - each \((1-p)\) - then one success, \(p\). Rolling for the first six: \(P(X = 1) = \frac{1}{6} \approx 0.167\), \(P(X = 2) = \frac{5}{6} \times \frac{1}{6} \approx 0.139\), \(P(X = 3) = \left(\frac{5}{6}\right)^2 \times \frac{1}{6} \approx 0.116\). Each extra roll multiplies by \(\frac{5}{6}\), so long waits are possible but increasingly rare.
The mean: 1/p
On average, how many trials until the first success? The answer is \(\frac{1}{p}\): a six takes 6 rolls on average, a 20% sales call takes 5 calls, and a fair coin's first head takes 2 flips. The reciprocal is a long-run average, not a guarantee - some searches end on trial 1 and some drag far past \(\frac{1}{p}\). The spread is large: the standard deviation is \(\sigma = \frac{\sqrt{1-p}}{p}\), about 5.48 rolls for the six and 4.47 calls at \(p = 0.2\). Note also what the mean is not: \(p\) is a per-trial probability, while \(\frac{1}{p}\) counts trials.
Memorylessness
Suppose the first four calls fail. What is the chance the fifth call succeeds? Still \(p\) - exactly as it was on day one. The trials do not remember their history, so the geometric distribution is memoryless: after any number of failures, the remaining wait behaves like a fresh copy of the whole process. This is the gambler's fallacy given its formal rebuttal - failures do not make success due. What failures do change is the probability of the whole run so far: needing 5 or more calls means 4 straight failures, which is why long waits are rare.
Worked example: the first six, exactly and within
Roll a fair die until the first six. a) Exactly on roll 3: \(P(X = 3) = \left(\frac{5}{6}\right)^2 \times \frac{1}{6} = \frac{25}{216} \approx 0.116\). b) Within 3 rolls: the complement of three straight misses, \(1 - \left(\frac{5}{6}\right)^3 = \frac{91}{216} \approx 0.421\) - the same at-least-one computation from the rules topic, now read as a within-k wait. c) Check: within 3 = on roll 1 + on roll 2 + on roll 3 = \(\frac{36+30+25}{216} = \frac{91}{216}\), agreeing with the complement route.
Worked example: sales calls
A salesperson closes a sale on 20% of calls, independently. 1. \(P(\text{first sale on call 1}) = 0.2\). 2. On call 2: \(0.8 \times 0.2 = 0.16\). 3. On call 3: \(0.8^2 \times 0.2 = 0.128\). 4. Expected calls to the first sale: \(\frac{1}{0.2} = 5\). 5. Probability the first sale comes within 3 calls: \(1 - 0.8^3 = 1 - 0.512 = 0.488\). Check: \(0.2 + 0.16 + 0.128 = 0.488\), so the direct addition matches the complement, and the remaining 0.512 is exactly three straight failures.
Shape: always skewed right
Every geometric distribution has the same silhouette: the tallest bar sits at \(X = 1\) with height \(p\), and each subsequent bar multiplies by \((1-p)\), shrinking geometrically - the distribution's name. So the single most likely outcome is immediate success, yet the mean \(\frac{1}{p}\) sits to the right of the mode, sometimes far right when \(p\) is small. Compare the binomial's symmetric-at-0.5 shape: both distributions are built from the same yes-or-no trials, but asking "how many successes" and asking "how long until the first" produce different silhouettes.
Common mistakes to avoid
First, inserting a combination factor - geometric waits have one arrangement, not many. Second, writing \((1-p)^k p\) instead of \((1-p)^{k-1}p\): \(k\) trials contain only \(k-1\) failures. Third, treating \(\frac{1}{p}\) as a guarantee or as a probability - it is a long-run average count of trials. Fourth, believing a long failure run raises the next trial's chance: memorylessness forbids it. Fifth, confusing the within-k probability \(1 - (1-p)^k\) with the point probability \(P(X = k)\). Sixth, applying geometric when \(p\) changes across trials or the trials depend on each other.
Key takeaways
Geometric counts the trials needed until the first success: independent, same-\(p\) trials, with no fixed \(n\). \(P(X = k) = (1-p)^{k-1}p\) - one arrangement, no combinations - and the probabilities shrink as \(k\) grows. The mean is \(\frac{1}{p}\): 6 rolls for a six, 5 calls for a 20% closer. The distribution is memoryless, so failures never make success more due, and the within-k probability is the complement of \(k\) straight failures, \(1 - (1-p)^k\). Fixed \(n\) counting successes is binomial; counting until the first success is geometric.
Flashcards
Flashcard
Trials until the first success
Test your understanding
Question 1 of 5
Which scenario is geometric?
Question 2 of 5
Rolling a fair die, what is the probability the first six appears exactly on roll 3?
Question 3 of 5
A salesperson closes on 20% of calls. On average, how many calls until the first sale?
Question 4 of 5
At p = 0.2 per call, what is the probability the first sale comes within 3 calls?
Question 5 of 5
After 4 unsuccessful calls, the probability that the next call succeeds is:
