Modeling Single Waiting Times with Exponential Variables

From an Event Description to a Waiting-Time Model

A waiting-time problem asks how long it takes for one specified event to occur, starting from a defined time or condition. The event might be a call arriving, a component failing, or a sensor registering a fault. The statistical model is not determined by the story alone: it depends on what counts as the event, how time is measured, and whether the event’s rate can reasonably be treated as constant.

When the event has a stable rate, an exponential model gives a direct way to describe the time to its first occurrence. The essential modeling work is to define that waiting time precisely and match the rate’s units to the units of time. A rate is not itself a probability; it is an intensity per unit of exposure.

Definition: Let $T$ be the elapsed time from a specified starting point until the first occurrence of a specified event. If the event has a constant rate $\lambda>0$ per unit of time, and the model’s constant-hazard assumptions are appropriate, write $T\sim\operatorname{Exponential}(\lambda)$. Its density and survival function are $$ f_T(t)=\lambda e^{-\lambda t},\quad t\ge 0, \qquad P(T>t)=e^{-\lambda t},\quad t\ge 0. $$ The rate $\lambda$ must use the same time unit as $T$. Under this model, $E[T]=1/\lambda$.

Identify the Event, Exposure, and Time Unit

Before using an exponential formula, describe the model in words. State when the clock starts, what counts as the first event, and the unit in which time and rate are measured. For a lifetime model, for example, the event might be the component’s first failure, and exposure might mean operating hours rather than calendar hours. For an arrival model, the event could be the next arrival during a period when the service is open.

The exponential model is appropriate when the event’s instantaneous rate does not change with elapsed time, given that the event has not yet occurred. This is the constant-hazard assumption. It also presumes that the conditions defining exposure remain stable enough for one rate to represent the process. A steadily aging component whose failure risk rises with age, or a service whose arrival rate varies sharply by time of day, may not meet that assumption.

A stated rate such as “12 arrivals per hour” is not a 12% chance of an arrival. It is an average intensity expressed per hour. To calculate a probability, multiply the rate by the relevant duration in the exponential formula. The resulting probability is between 0 and 1.

Define the waiting time.
Specify the starting point and the first event whose occurrence ends the wait.

Identify the rate and exposure.
Determine the event rate and whether it applies to calendar time, operating time, or another stated exposure unit.

Check the model assumptions.
Decide whether a constant event rate is plausible throughout the waiting period and whether the stated exposure conditions persist.

Use one consistent time unit.
State the units of $T$ and $\lambda$ together, then calculate probabilities from the exponential survival function.

Why the Exponential Survival Formula Fits a Constant Rate

The connection between a constant rate and an exponential waiting time can be seen by considering the probability of no event over many short intervals. If the constant rate is $\lambda$, then over a short interval of length $\Delta t$, the chance of an event is approximately $\lambda\Delta t$, so the chance of no event is approximately $1-\lambda\Delta t$. With stable conditions across the intervals, survival for time $t$ is approximated by multiplying these no-event probabilities.

P(T>t) \approx \left(1-\lambda\frac{t}{n}\right)^n \longrightarrow e^{-\lambda t} \quad\text{as } n\longrightarrow\infty.

Thus a constant rate leads to the exponential survival function. Subtracting survival from 1 gives $P(T\le t)=1-e^{-\lambda t}$. For an interval, subtract the probability of surviving past its upper endpoint from the probability of surviving past its lower endpoint:

P(a<T\le b)=P(T>a)-P(T>b)=e^{-\lambda a}-e^{-\lambda b}, \quad 0\le a<b.

The distinction between “by time $t$” and “after time $t$” matters: the former uses $1-e^{-\lambda t}$, while the latter uses $e^{-\lambda t}$. Because the exponential distribution is continuous, including or excluding one endpoint does not change the probability.

Worked Example: Waiting for the Next Service Call

A hypothetical service desk receives calls at a stable average rate of 12 calls per hour during a particular operating period. Let $T$ be the time, in hours, from the start of that period until the next call. Under a constant-rate model, find the probability that the next call arrives after 15 minutes but no later than 30 minutes.

Method. Model the wait to the first call as exponential, then calculate the probability of an interval using the difference of two survival probabilities. The rate and $T$ are both expressed in hours.

Model and units. The event is the arrival of the next call, and $\lambda=12$ calls per hour. The two time limits are $0.25$ hour and $0.5$ hour. Thus $T\sim\operatorname{Exponential}(12)$, with the rate measured per hour.

Calculate. The required probability is

\begin{aligned} P(0.25<T\le 0.5) &=e^{-12(0.25)}-e^{-12(0.5)}\\ &=e^{-3}-e^{-6}\\ &=0.049787-0.002479\\ &=0.047308. \end{aligned}

Rounded to four decimal places, the probability is 0.0473. This is a small probability because a desk receiving 12 calls per hour is unlikely to go for as long as 15 minutes without a call.

Check and common mistake. The answer is positive because survival to 15 minutes is more likely than survival to 30 minutes. A mistaken calculation might use $12(15)$ in the exponent, treating the rate “per hour” as if time were measured in minutes. The correction is to keep the time in hours, as above, or first express the rate in calls per minute. Mixing the units gives an invalid exponent.

Worked Example: Time to a Component Failure

A hypothetical monitoring system assigns a component a constant failure rate of $0.002$ per operating hour. Let $T$ be the component’s operating time from installation to its first failure. Find the probability that it fails after 300 operating hours but by 700 operating hours. Also check the model’s implied mean operating time.

Method. Use an exponential lifetime with rate per operating hour, then apply the interval formula. The model treats operating hours as the exposure and assumes the failure rate stays constant over that exposure.

Model. $T\sim\operatorname{Exponential}(0.002)$, with $T$ in operating hours. The interval probability is

\begin{aligned} P(300<T\le 700) &=e^{-0.002(300)}-e^{-0.002(700)}\\ &=e^{-0.6}-e^{-1.4}\\ &=0.548812-0.246597\\ &=0.302215. \end{aligned}

Rounded to four decimal places, this probability is 0.3022. The implied mean operating time is

E[T]=\frac{1}{0.002}=500\text{ operating hours}.

The mean is consistent in scale with the interval probabilities: 300 hours is below the mean, while 700 hours is above it. The check does not prove that the constant-rate assumption is realistic; it confirms that the parameter has been interpreted as a rate per operating hour.

Common mistake. It would be incorrect to calculate the probability of failure in this interval as $0.002(700-300)=0.8$. The rate multiplied by duration is not, in general, the exact probability of at least one event over that duration. Under the exponential model, survival and interval probabilities use exponentials, giving $0.3022$ here.

Worked Example: Interpreting a Mean Waiting Time

A hypothetical sensor is monitored while it is active. Its modeled mean wait from activation to the first registered fault is 8 active days. Assume the constant-rate exponential model applies. Find the probability that the first fault occurs after 4 active days but by 12 active days.

Method. Convert the stated mean waiting time into the exponential rate using $E[T]=1/\lambda$, then calculate the interval probability. Both quantities use active days.

Determine the rate.

\lambda=\frac{1}{E[T]}=\frac{1}{8}=0.125 \quad\text{per active day}.

The waiting time is therefore $T\sim\operatorname{Exponential}(0.125)$. The requested probability is

\begin{aligned} P(4<T\le 12) &=e^{-0.125(4)}-e^{-0.125(12)}\\ &=e^{-0.5}-e^{-1.5}\\ &=0.606531-0.223130\\ &=0.383401. \end{aligned}

Rounded to four decimal places, the probability is 0.3834. The mean-rate check is $1/0.125=8$ active days, as specified.

Common mistake. “A mean wait of 8 days” does not mean that a fault occurs exactly every 8 days, nor that the probability of a fault on any particular day is $1/8$. It specifies the reciprocal rate under the exponential model. The probability over a time interval must still be found with the survival function.

When the Model Is Useful—and When to Be Cautious

The exponential model is useful when a problem describes time to one first event and gives a rate that is stable over the relevant period. It makes the model’s implications explicit: a larger rate means shorter waits on average, and probabilities depend on the rate-duration product $\lambda t$. It can describe either arrivals or lifetimes, provided the event, exposure, and assumptions are defined appropriately.

The model should not be selected just because a question uses the word “waiting.” A rate that changes with elapsed time does not give a constant-hazard exponential waiting time. Nor should a rate measured per operating hour be applied to calendar time without a justified link between those exposure measures. If the event can occur only while a system is active, specify whether inactive periods stop the clock.

A useful final check is dimensional: the exponent $\lambda t$ must be dimensionless. If $\lambda$ is per operating hour, $t$ must be in operating hours. Also check boundary behavior: at $t=0$, $P(T>t)=1$; as $t$ increases, survival decreases toward 0; and interval probabilities are nonnegative.

Key takeaway: A single waiting time is modeled exponentially when its first-event rate is constant over the specified exposure. Define the event, match rate and time units, and check the assumptions before calculating.

Check Your Understanding

For each problem, identify the exponential rate and use a consistent time unit. Answers are rounded to four decimal places where appropriate.

A hypothetical dispatch desk receives requests at a constant rate of 5 per operating shift. Let $T$ be the wait in shifts for the next request. Find $P(0.1<T\le 0.3)$. (Answer: 0.3834)
A hypothetical instrument has a constant failure rate of $0.004$ per operating hour. Find the probability it fails within its first 100 operating hours. (Answer: 0.3297)
A hypothetical field monitor registers events at a constant rate of $0.3$ per active day. Find the probability the first event occurs after 2 but by 5 active days. (Answer: 0.3257)
Under an exponential model, a hypothetical inspection process has a mean wait of 6 active days until its first alert. State the rate per active day and find the probability the wait exceeds 9 active days. (Answer: rate $=1/6$ per active day; probability $=0.2231$)

What You'll Learn