Geometric Waiting-Time Models

Waiting for a First Success

Many processes repeat the same kind of attempt until a specified event occurs: a device sends packets until one arrives successfully, a technician tests components until one passes inspection, or a person tries a connection until it works. In each case, a useful question is not how many successes occur in a fixed number of attempts, but how many attempts are needed to get the first success?

A geometric waiting-time model describes this situation when each attempt has the same probability of success and the attempts are independent. Recognizing those conditions matters: a waiting-time variable may look geometric in its wording while changing probabilities or dependence make the model unsuitable.

The definition has several parts, and each must match the setting. A trial is one repeatable attempt or opportunity. Success is the outcome of interest, not necessarily a desirable result; it is simply the outcome being waited for. Trials must be independent, and the probability of success must remain equal to the same value \(p\) from one trial to the next.

The count starts at 1, not 0, because the first trial is included if it succeeds. A variable that counts the failures before the first success is a different convention: it would be \(X-1\) under the definition used here. Be explicit about the count whenever describing a waiting time.

Checking Whether the Model Fits

Before using a geometric model, translate the situation into a sequence of trials and check its assumptions. A trial need not be a person or an object; it could be a transmission attempt, a day, or a repeated test. But the trials need to represent comparable opportunities with a stable success event.

These checks distinguish a geometric waiting time from a binomial count. A binomial random variable counts successes in a fixed number of trials; a geometric random variable counts trials up to and including the first success. Both settings involve repeated trials, but they answer different questions.

Independence and constant probability are related but not interchangeable. A process can have the same probability on each trial but still be dependent, or it can have independent trials whose success probabilities differ. Either issue prevents the standard geometric model from applying.

Why the First-Success Sequence Matters

The event \(X=k\), for a positive integer \(k\), has a precise structure: the first \(k-1\) trials must be failures, and trial \(k\) must be a success. Under the model, each failure has probability \(1-p\), the success on trial \(k\) has probability \(p\), and independence allows the probabilities to be multiplied.

This short derivation is also a useful model check. It shows why both assumptions are needed: the repeated failure probability is \(1-p\) only when the success chance is constant, and multiplying the trial probabilities requires independence.

This expression describes the probability of one specific first-success position. The key recognition point is the pattern of \(k-1\) failures followed by one success—not merely that the process involves repeated attempts.

For \(p=1\), every trial succeeds, so the first-success count is \(X=1\) with probability 1; values \(2,3,\ldots\) are impossible. If \(p=0\), no trial can succeed, so there is no finite first-success time and the geometric waiting-time model as defined above does not apply.

When a Repeated Process Is Not Geometric

A repeated process does not automatically satisfy the geometric assumptions. Sampling without replacement is a common example: drawing an item changes what remains, so both the probability of success and the dependence between outcomes may change. A process that stops after a fixed number of attempts also needs care. If success has not occurred by the stopping point, the waiting time is not observed as a finite first-success count; it is incomplete or censored rather than an ordinary, fully observed geometric wait.

Interpreting the Model Carefully

A geometric model is a probability description of a process under stated assumptions; it is not a claim that every real sequence will behave regularly. Even when the model is suitable, a particular run may have an unusually early success or a long wait. The parameter \(p\) describes the probability of success on each trial, not a guarantee about when success will occur.

The assumption of independence should be supported by the setting, not inferred just because trials are described separately. For example, an unsuccessful login attempt might trigger a lockout or extra verification, changing later chances. A medical screening process might change who is eligible for later tests. In settings like these, one should examine the process before treating attempts as identical independent trials.

A practical test is to write down what the next trial looks like after a failure. If a failure changes the available items, the environment, the strategy, or the chance of success, then the standard geometric model may not fit. If the same type of trial can be repeated under stable conditions and remains independent, geometric waiting time may be appropriate.

Common Mistake: Counting Only the Failures

A frequent error is to define \(X\) as “the number of trials until success” and then count only the failures before success. Under the convention in this tutorial, the successful trial is included. If the first success occurs on the fourth attempt, then \(X=4\), even though there were only three failures before it.

A second mistake is to assume that any repeated trial has a geometric waiting time. Repetition alone is not enough: the success event must be consistent, trials must be independent, and \(p\) must remain constant. The card example shows why a calculation based on the right-looking sequence can still be wrong if those model conditions are not met.