Diagnosing Which Waiting-Time Model Fits

Start With What the Random Variable Records

A question about repeated yes-or-no outcomes can describe either a count or a wait. The wording may mention the same kind of success in both cases, but the random variable and the stopping rule determine which model is appropriate. Counting successes in a predetermined number of trials points toward a binomial model. Counting trials up to and including the first success points toward a geometric model.

The model choice depends on more than whether the data are binary or whether the question uses the word “waiting.” Both models require a suitable sequence of trials, with a clearly defined success. Both also rely on independent trials with the same success probability. The key distinction is what is recorded and when the sequence ends.

Definition: A binomial random variable \(X\) counts the number of successes in a fixed number \(n\) of trials. A geometric random variable \(T\) records the number of trials, including the success trial, up to and including the first success. In both models, the trials must be independent, and each trial must have the same probability \(p\) of success.

In the geometric convention used here, \(T\) can take the values \(1,2,3,\ldots\). For example, \(T=1\) means success on the first trial. Be explicit about this convention: some settings also describe the number of failures before the first success, which is \(T-1\), not \(T\).

A Diagnostic Checklist

Before choosing a formula, translate the question into a description of the random variable and the process that produces it. The following questions help separate the two models and check whether either one is justified.

Define success.
State the binary outcome counted as a success on one trial. It should be clear what one trial is.

Identify the random variable.
Does it count successes in a set number of trials, or does it record the trial on which the first success occurs?

Check the stopping rule.
A binomial count has a fixed number \(n\) of trials decided in advance. A geometric wait continues until the first success.

Check the trial conditions.
For either model, verify that trials are independent and that the success probability \(p\) stays the same from trial to trial.

Match the formula to the question.
Use a count formula for a specified number of successes, or a waiting-time formula for the first success on a specified trial.

For \(n\) fixed independent trials with success probability \(p\), the binomial probability of exactly \(k\) successes is

P(X=k)=\binom{n}{k}p^k(1-p)^{n-k},\qquad k=0,1,\ldots,n.

For independent trials with the same success probability \(p\), the geometric probability that the first success is on trial \(t\) is

P(T=t)=(1-p)^{t-1}p,\qquad t=1,2,3,\ldots.

These formulas describe different random variables. The binomial expression counts all successes across a fixed set of trials. The geometric expression requires failures for the first \(t-1\) trials and then a success on trial \(t\).

Worked Examples: Matching the Model to the Process

Worked Example: Counting Successes in a Fixed Number of Trials

Suppose a hypothetical quality check tests exactly five components. Each component independently passes with probability \(0.90\), and \(X\) is the number that pass. Find the probability that exactly four pass.

Random variable and stopping rule. \(X\) counts passes among a fixed five components. The number of trials is fixed at \(n=5\), so the random variable has the form of a binomial count.

Assumptions. For this hypothetical model, each component’s pass outcome is independent of the others, and the pass probability is the same, \(p=0.90\), for every component. These conditions make the binomial model appropriate.

Substitute \(n=5\), \(k=4\), and \(p=0.90\):

P(X=4)=\binom{5}{4}(0.90)^4(1-0.90)^{5-4} =5(0.6561)(0.10)=0.32805.

The probability that exactly four of the five components pass is \(0.3281\), or about \(32.81\%\). This is a binomial question because the number tested is set in advance and the outcome of interest is the count of passes.

Worked Example: Waiting for the First Success

In a hypothetical process, each independent attempt succeeds with probability \(0.20\). Attempts continue until the first success. Let \(T\) be the number of attempts up to and including that success. Find the probability that the first success occurs on attempt four.

Random variable and stopping rule. \(T\) is a trial number, not a count of successes in a fixed set. The process stops when its first success occurs, so the geometric model is the candidate.

Assumptions. The model assumes that each attempt is independent and that the success probability remains \(p=0.20\) on every attempt. Under these stated hypothetical conditions, it is reasonable to use the geometric formula.

For \(T=4\), the first three attempts must fail and the fourth must succeed:

P(T=4)=(1-0.20)^{4-1}(0.20) =(0.80)^3(0.20)=0.1024.

Thus, the probability that the first success is on attempt four is \(0.1024\), or \(10.24\%\). The three failures followed by a success match the event being described.

Worked Example: Similar Trials, Different Random Variables

Suppose a hypothetical device independently transmits each packet incorrectly with probability \(0.05\). Compare two questions about six packets: (a) what is the probability the first incorrect packet is packet six, if transmission continues until an incorrect packet occurs; (b) what is the probability that exactly one of the six packets is incorrect?

Assumptions. In both questions, each packet’s outcome is independent, and its probability of being incorrect is the same, \(p=0.05\). These assumptions support both candidate models, but the random variable and stopping rule differ.

(a) First incorrect packet on packet six. Let \(T\) be the packet number of the first incorrect transmission. This is a geometric waiting-time question: packets one through five must be correct, followed by an incorrect packet.

P(T=6)=(1-0.05)^5(0.05) =(0.95)^5(0.05) =0.038689\ldots.

The probability is about \(0.03869\), or \(3.869\%\). The process is described by a wait until the first incorrect packet.

(b) Exactly one incorrect packet among six. Let \(X\) count incorrect packets among six specified transmissions. The number of trials is fixed at six, and any one of the six packets could be the incorrect one. This is a binomial question:

P(X=1)=\binom{6}{1}(0.05)^1(0.95)^5 =6(0.05)(0.7737809375) =0.23213428125.

The probability is about \(0.2321\), or \(23.21\%\). Although both questions concern six packets and one error, the answers differ because the events differ: in part (a), the first error must be on packet six; in part (b), the single error can be anywhere among the six.

Why the Geometric Formula Has Its Form

A short derivation makes the stopping rule visible in the geometric probability. For the first success to occur on trial \(t\), the first \(t-1\) trials must all be failures. Each failure has probability \(1-p\), and trial \(t\) must be a success with probability \(p\). Independence lets us multiply the probabilities of these particular outcomes:

\begin{aligned} P(T=t) &=P(\text{failure on trials }1,\ldots,t-1\text{ and success on trial }t)\\ &=(1-p)(1-p)\cdots(1-p)p\\ &=(1-p)^{t-1}p. \end{aligned}

There are \(t-1\) failures in this sequence, not \(t\), because the \(t\)th trial is the success. For \(t=1\), there are no failures before success, and the formula gives \((1-p)^0p=p\), as it should.

When Neither Model Fits

The binomial and geometric models require independent trials with a constant success probability. A process that stops after a fixed number of trials or at a first success is not enough by itself: the trial conditions must also be checked. If outcomes affect later probabilities, or the success chance changes across trials, neither simple model is justified without further argument.

Worked Example: A Changing Success Probability

Imagine drawing one item at a time without replacement from a hypothetical box containing four red and six blue items. Call drawing red a success, and let \(T\) be the draw on which the first red item appears. Is the usual geometric model appropriate?

Stopping rule. The process does continue until the first red item, so \(T\) resembles a geometric waiting time. However, the model assumptions must also hold.

Check the assumptions. The first-draw success probability is \(4/10=0.40\). If the first draw is blue, only four red items remain among nine items, so the conditional probability of red on draw two is \(4/9\), not \(0.40\). The draws are dependent, and the success probability changes after a draw.

For example, the actual probability that the first red appears on draw two is

P(T=2)=P(\text{blue first and red second}) =\frac{6}{10}\cdot\frac{4}{9} =\frac{4}{15}\approx 0.2667.

A geometric model using \(p=0.40\) would instead give

(1-0.40)^{2-1}(0.40)=0.60(0.40)=0.24.

The probabilities differ because this sampling process does not have independent trials with constant success probability. The stopping rule alone does not make the geometric model valid.

Common Mistake and Practical Importance

Common mistake: “The question mentions a fixed number of attempts, so it must be binomial.” That is not enough. If the random variable records the first success’s trial number, the quantity is a waiting time; if the process ends at the first success, the number of trials is not a predetermined fixed \(n\). Conversely, a question can ask about success on the last trial of a fixed set, but if it counts successes across all \(n\) trials, the random variable is binomial.

The correction is to define the random variable in words before selecting a formula. Ask, “What does one value of this variable represent?” Then state when the trials stop and check independence and the constant probability of success. This prevents a frequent error: using the probability of one particular sequence as though it were the probability of a count. A binomial count may correspond to several different sequences, which is why its formula includes \(\binom{n}{k}\); a geometric event specifies one sequence of failures followed by success.

This diagnosis matters because a correct calculation with the wrong model still answers the wrong question. In applications, describe the mechanism that produces outcomes rather than relying only on a familiar keyword. If the number of trials is capped before a success is guaranteed, or if follow-up ends before the event occurs, the recorded quantity may require a more careful model than the basic geometric waiting time.

Key takeaway: A binomial model describes the number of successes in a fixed number of independent trials with constant success probability. A geometric model describes the trial of the first success under the same trial assumptions. Identify the variable, stopping rule, independence, and constant \(p\) before calculating.

Check Your Understanding

For each situation, identify the random variable, the stopping rule, and whether a binomial or geometric model is appropriate under the stated assumptions.

A technician inspects exactly 12 independent items, each defective with probability \(0.03\), and records the number of defective items. Which model fits, and what conditions support the choice?
A game is played repeatedly until the first win. Each game is independent, with the same win probability. Define a suitable random variable and identify its model.
In a fixed set of eight trials, each with success probability \(p\), what is the difference between counting exactly two successes and asking whether the first success occurs on trial eight?
Why is drawing cards without replacement generally not a geometric process with a fixed success probability?
A geometric variable \(T\) counts trials up to and including the first success. What event does \(T=1\) describe, and what event must occur for \(T=4\)?