What Is A Geometric Distribution? (Math Benchmarks)
A geometric distribution models how many independent, two-outcome trials it takes to get the first success, when each trial has the same success probability. If (X) counts trials, then (P(X=x)=(1-p)^{x-1}p), and its mean is (1/p). Define what you count first: some texts count failures instead, shifting the values by one.
A small counting error can change the answer, even when the formula is correct. Before calculating, check whether the problem counts the successful trial or only the failures before it. Then ask whether trials are independent and the chance of success stays the same.
These checks matter because a geometric model is useful only when its assumptions fit. Below, I’ll show how to identify the model, calculate common probabilities, interpret its benchmarks, and spot cases where it does not apply.
Geometric distribution: definition, assumptions, and probability model
A geometric distribution describes repeated trials with two possible outcomes, such as success or failure. It tracks the wait until the first success. Each trial must be independent, and the success chance (p) must stay constant. If either condition fails, the standard geometric formula may not describe the situation.
A Bernoulli trial is one attempt with two outcomes. “Success” means the outcome being tracked, not necessarily a favorable result. For example, if a quality check looks for a defect, finding one can be labeled success for the purpose of the model.
The model’s key conditions are:
- Each trial has two outcomes.
- The probability of success is (p), with (0<p\le1).
- Trials are independent: earlier outcomes do not change later chances.
- The value of (p) stays the same from trial to trial.
- The process continues until the first success.
A fair coin tossed repeatedly for the first head can fit the model, with (p=0.5). A sequence of attempts where conditions change after each failure may not fit. Check the setup before using the equation.
Support, probability formula, and expected value
The support is the set of values a random variable can take. For a geometric distribution, the support depends on what the variable counts. The two common conventions describe the same process but use values shifted by one, so state the convention before applying a formula.
Trial-count convention
Under the trial-count convention, (X) counts every trial through and including the first success. Its possible values are (1,2,3,\ldots). The probability that the first success occurs on trial (x) is:
[ P(X=x)=(1-p)^{x-1}p ]
The first (x-1) trials must fail, and trial (x) must succeed. Since failures have probability (1-p), multiplying those chances gives the formula.
Suppose a component passes an independent test with probability (p=0.2). The probability its first pass comes on test 3 is:
[ P(X=3)=(0.8)^2(0.2)=0.128 ]
So the chance is (0.128), or 12.8%. This is the probability of passing on test 3 specifically, not the probability of passing by test 3.
Failure-count convention
Some texts define (Y) as the number of failures before the first success. Its possible values are (0,1,2,\ldots), and:
[ P(Y=y)=(1-p)^yp ]
For the same test with (p=0.2), three failures before the first pass means the first pass is on test 4:
[ P(Y=3)=(0.8)^3(0.2)=0.1024 ]
The variables are related by (Y=X-1). Their expected values differ for the same reason: one counts the successful trial, while the other does not.
| Quantity | Trials through first success, (X) | Failures before first success, (Y) |
|---|---|---|
| Possible values | (1,2,3,\ldots) | (0,1,2,\ldots) |
| Probability of one value | ((1-p)^{x-1}p) | ((1-p)^yp) |
| Expected value | (1/p) | ((1-p)/p) |
| Variance | ((1-p)/p^2) | ((1-p)/p^2) |
Mean and variance benchmarks
The expected value is the long-run average across many repetitions of the process. The variance measures how spread out the values are around their average. For (X), the trial count, the benchmarks are:
[ \mathbb E[X]=\frac1p,\qquad \mathrm{Var}(X)=\frac{1-p}{p^2} ]
For (Y), the failure count, the mean is ((1-p)/p), while the variance is unchanged. These are averages and measures of spread, not promises about one particular run.
If (p=0.2), then (\mathbb E[X]=5) trials and (\mathbb E[Y]=4) failures. The variance is (0.8/0.04=20) for either convention. The standard deviation is the square root of the variance, about 4.47 trials. That wide spread helps explain why an individual wait can differ a lot from the average.
Worked examples, memorylessness, and common errors
A worked example is most reliable when it names the random variable, lists its possible values, and then substitutes into the matching formula. For cumulative questions, distinguish “on trial (k)” from “by trial (k).” These phrases ask for different probabilities.
For a success chance of (p=0.2):
- First success on trial 3: (P(X=3)=0.8^2(0.2)=0.128).
- First success by trial 3: (P(X\le3)=1-P(X>3)=1-0.8^3=0.488).
- More than 3 trials: (P(X>3)=0.8^3=0.512).
“By trial 3” includes success on trials 1, 2, or 3. “More than 3” means the first three trials all failed. The complement is often the quickest route for a cumulative probability.
Memorylessness
The geometric distribution is memoryless: after a run of failures, the probability of waiting at least (n) more trials for success is the same as it was at the start. In symbols, for nonnegative integers (m,n),
[ P(X>m+n\mid X>m)=P(X>n) ]
This follows because the next trials are independent and retain the same success probability. For example, if a fair coin has shown tails five times, the chance of needing more than two additional tosses for a head is still (0.5^2=0.25).
Memorylessness does not mean success is “due.” Nor does it make a changing or dependent process geometric. If each failure changes the next probability, the condition behind this property no longer holds.
Common calculation errors
The most common error is switching conventions mid-calculation. Under the trial-count convention, the first success can occur on trial 1. Under the failure-count convention, it can follow zero failures. Write the variable and support before choosing a formula.
Other checks can catch mistakes:
- Make sure (0<p\le1), and use (1-p) for failure.
- Check that the exponent counts failures before the successful trial.
- Confirm an individual probability falls between 0 and 1.
- For a cumulative question, verify whether the endpoint is included.
- Do not treat an expected value as a guaranteed or most likely result.
For (0<p<1), the individual probabilities across all possible values sum to 1. Under the trial-count convention, this is the geometric series (p+(1-p)p+(1-p)^2p+\cdots=1). If (p=1), success happens on the first trial with probability 1.
Applications, limitations, and connections to other distributions
A geometric model can describe the number of independent attempts until a first event, such as the first head in repeated coin tosses. The setting itself does not make the model valid. The chance of success must remain constant, and the attempts must be independent.
A useful diagnostic is to write down what one trial means and what counts as success. Then ask whether observing a failure changes the next trial’s chance. If it does, do not use the standard geometric distribution without a separate model that supports it.
How it differs from related distributions
A binomial distribution counts successes in a fixed number of trials. A geometric distribution counts trials, or failures, until the first success. A negative binomial distribution extends the waiting idea to a fixed number of successes greater than one.
| Distribution | What is fixed? | What is counted? |
|---|---|---|
| Binomial | Number of trials | Number of successes |
| Geometric | Number of successes sought: one | Trials or failures until first success |
| Negative binomial | Number of successes sought | Trials or failures until that many successes |
These distinctions help identify the right model from a word problem. If the question sets a trial limit and asks how many successes occur, consider binomial. If it asks how long until the first success, consider geometric, provided its assumptions hold.
A short decision check
Before calculating, use this sequence:
- Define success and one trial.
- Confirm there are two outcomes and (p) stays constant.
- Check that trials are independent.
- State whether the variable counts trials or failures.
- Choose a point probability, cumulative probability, mean, or variance formula.
- Interpret the result in the units of the question.
The final step matters: “five trials on average” and “four failures on average” can describe the same process under different conventions.
FAQ
What is a geometric distribution in simple terms?
It gives the probability of waiting a certain number of independent trials until the first success, when every trial has the same success chance.
What is the geometric probability formula?
If (X) counts trials through the first success, then (P(X=x)=(1-p)^{x-1}p), for (x=1,2,3,\ldots).
Why does the exponent use (x-1)?
A first success on trial (x) requires (x-1) failures before it. The final factor (p) is the chance of success on trial (x).
Can a geometric random variable equal zero?
Only under the failure-count convention. That variable counts failures before success, so zero means the first trial succeeds.
What is the mean of a geometric distribution?
For a variable counting trials through the first success, the mean is (1/p). For a variable counting failures first, it is ((1-p)/p).
What is the variance?
For either convention, the variance is ((1-p)/p^2). The two variables differ by one, which changes the mean but not the variance.
What does memoryless mean?
It means past failures do not change the distribution of the remaining wait, as long as trials stay independent and the success probability stays constant.
When should I not use a geometric distribution?
Do not use the standard model when trials are dependent or the success probability changes across trials, unless another justified model accounts for those changes.
(This article was written by one of our staff writers, Nathan Sterling. Visit our Meet the Team page.)
