Choosing a Waiting-Time Model
A waiting time is not automatically exponential. A passenger waiting for a departure scheduled at a random point in a fixed service window faces a different mechanism from a customer waiting for a call that can arrive at any time at a steady rate. The first may be modeled as uniform; the second may be modeled as exponential. The model choice depends on how the event time is generated, not simply on the fact that it measures time.
The central distinction is that a uniform waiting time assigns equal probability to intervals of equal length within a fixed range. An exponential waiting time has no finite upper bound and assumes a constant hazard: over a short interval, the chance of an event is approximately proportional to the interval's length and does not depend on how long the wait has already lasted.
What the Assumptions Mean
For \(T\sim\operatorname{Uniform}(a,b)\), all possible times lie in the bounded interval \([a,b]\). Within that interval, equal-length subintervals have equal probabilities. For example, if \(a\leq c<d\leq b\), then
The endpoints have probability zero for a continuous distribution. The density's value at a single endpoint does not give the probability of that exact time. In the common case \(T\sim\operatorname{Uniform}(0,L)\), the mean is \(L/2\): the distribution places equal density throughout a range extending from zero to \(L\).
For \(T\sim\operatorname{Exponential}(\lambda)\), \(\lambda\) is a rate, measured in inverse time units. Its mean is \(1/\lambda\), and its support includes every nonnegative time; there is no fixed finite maximum waiting time. The constant-hazard assumption says that, given survival to time \(t\), the instantaneous event rate at that time is \(\lambda\). It does not mean the event is certain to occur within any particular finite interval.
A useful way to see the contrast is through the hazard function, defined at time \(t\) as the event density at \(t\) divided by the probability of surviving beyond \(t\), when that survival probability is positive. For a uniform distribution on \([0,L]\), the hazard is
This hazard increases as \(t\) approaches \(L\); it is not constant. By contrast, for an exponential variable, \(f_T(t)=\lambda e^{-\lambda t}\) and \(P(T>t)=e^{-\lambda t}\), so \(h_T(t)=\lambda\). The exponential model is therefore appropriate when a constant hazard is a defensible assumption, not merely when a time-to-event variable is present.
A Short Derivation: Equal Likelihood Within a Uniform Window
The uniform density is \(1/(b-a)\) throughout \([a,b]\). A probability over a subinterval is the area under that density over the subinterval. Thus, for \(a\leq c<d\leq b\),
The result depends only on the subinterval's length \(d-c\), not its location. For instance, two five-minute intervals contained within the range have the same probability under this model. For an exponential distribution, equal-length intervals at different times generally do not have equal probabilities, because fewer observations survive to later times.
Match the Model to the Mechanism
A uniform model can be reasonable when a time is selected at random within a known fixed window and the model treats every point in that window equally. Examples include an arrival scheduled uniformly within a specified dispatch cycle or a randomly selected inspection time within a shift. The assumption is about the timing mechanism, not a claim that every real-world process is exactly uniform.
An exponential model can be reasonable when an event occurs according to a stable rate per unit time, with no relevant aging or trend in that rate. A constant-rate call-arrival model is one possible illustration. If the event rate changes over time, or the process has a deadline that rules out sufficiently late events, the ordinary exponential model may not be suitable.
| Feature | Uniform on \([a,b]\) | Exponential with rate \(\lambda\) |
|---|---|---|
| Possible times | Only times from \(a\) to \(b\) | All nonnegative times |
| Mechanism represented | Equal likelihood across a fixed interval | Constant hazard over time |
| Equal-length intervals | Equal probability if both lie in \([a,b]\) | Later intervals have lower probability |
| Effect of elapsed time | Remaining possible range shrinks | Conditional remaining wait has the same exponential distribution |
The conditional behavior is especially useful for distinguishing the assumptions. If \(T\sim\operatorname{Uniform}(0,L)\) and it is known that \(T>s\), where \(0\leq s<L\), the remaining time \(T-s\) is uniform on \([0,L-s]\). Its range is shorter because the event has already failed to occur by time \(s\). For an exponential variable, the conditional probability of waiting at least another \(x\) time units, given survival to \(s\), is
Thus the exponential model has the memoryless property: the elapsed wait does not change the distribution of the remaining wait. This does not make the exponential distribution a default model for every wait; it is a consequence of the constant-hazard assumption.
Worked Example: Waiting Within a Fixed Dispatch Cycle
A hypothetical shuttle leaves once during each 12-minute dispatch cycle. For this model, a passenger arriving exactly at the start of a cycle waits a time \(T\) that is equally likely to be anywhere from 0 to 12 minutes. Find the probability the passenger waits between 4 and 9 minutes, the probability the wait exceeds 9 minutes, and the mean wait. Name the model and a mistake to avoid.
Method. The stated mechanism is a fixed 12-minute range with equal likelihood across its times, so use \(T\sim\operatorname{Uniform}(0,12)\). The interval length is \(12-0=12\) minutes.
For a wait between 4 and 9 minutes, the favorable interval has length \(9-4=5\) minutes. Therefore,
For a wait exceeding 9 minutes, the favorable range has length \(12-9=3\) minutes:
The mean of a uniform distribution on \([0,12]\) is the midpoint:
Check. The first probability is \(5/12\), the second is \(3/12\), and the probability of a wait from 0 through 4 minutes is \(4/12\). These add to \(12/12=1\). A common mistake would be to use an exponential model simply because \(T\) is a waiting time; that would ignore the fixed cycle and equal-likelihood assumption in the problem.
Worked Example: A Constant-Rate Arrival Process
In a hypothetical service system, incoming requests are modeled as having a constant arrival rate of \(0.2\) per minute. Let \(T\) be the time until the next request. Find the probability that \(T\) is more than 2 minutes but no more than 7 minutes, and find the probability that, after already waiting 4 minutes, the additional wait exceeds 3 minutes. State why an ordinary uniform model would not represent the stated mechanism.
Method. A constant arrival rate motivates \(T\sim\operatorname{Exponential}(\lambda)\) with \(\lambda=0.2\) per minute. Its survival function is \(P(T>t)=e^{-0.2t}\), with \(t\) measured in minutes.
The event \(2<T\leq7\) is the difference between surviving beyond 2 minutes and surviving beyond 7 minutes:
For the additional wait, use the exponential memoryless property. The probability the remaining wait exceeds 3 minutes, given that the first 4 minutes have passed without an arrival, is
Check. Since \(\lambda=0.2\) per minute, the mean is \(1/0.2=5\) minutes, and the probabilities above use minutes consistently. A uniform model would impose a finite maximum time and equal likelihood throughout a bounded range, neither of which is part of the stated constant-rate mechanism. The common mistake avoided here is treating “waited 4 minutes already” as a reason to subtract 4 from the original exponential mean; the conditional remaining wait still has mean 5 minutes.
Worked Example: Equal Means Do Not Determine the Model
Compare two hypothetical models for a wait. Model U is uniform on \([0,8]\) minutes. Model E is exponential with rate \(0.25\) per minute. Calculate the mean under each model and the probability of waiting more than 6 minutes under each. Then compare the probability of waiting more than 8 minutes and explain what the comparison says about model choice.
Method. Calculate each quantity from the model's assumptions. Do not select a distribution merely because its mean matches the other model's mean.
For Model U, the interval midpoint gives
For Model E, the mean is the reciprocal of its rate:
Under Model U, the portion of the range above 6 minutes has length \(8-6=2\), so
Under Model E, use the exponential survival function:
The probability of a wait exceeding 8 minutes is zero under Model U, because its support ends at 8. Under Model E it is
Check. Both models have mean 4 minutes, but they give different tail probabilities and different possible times. Equal means do not make the models interchangeable. The uniform model fits a mechanism that guarantees a time no later than 8 minutes and treats times in that range equally; the exponential model fits a constant-hazard mechanism that allows waits beyond 8 minutes.
Common Mistake: Treating the Uniform Model as Memoryless
Suppose \(T\sim\operatorname{Uniform}(0,12)\) and it is known that the wait has already lasted 4 minutes. A mistaken calculation might apply the exponential memoryless idea and say the mean additional wait is still the original mean of 6 minutes. That is incorrect: the conditional uniform distribution has only the remaining range from 4 to 12 minutes, so the additional wait is uniform on \([0,8]\), with mean 4 minutes.
The correction follows from the different assumptions. Under the uniform model, the fixed endpoint makes the available range shrink as time passes. Under the exponential model, constant hazard gives the same conditional distribution for the remaining wait. Applying a property from one model to the other can produce a plausible-looking but incorrect answer.
A Practical Model-Choice Sequence
Ask whether the time is selected within a specified bounded window, or whether an event occurs according to a stable rate over time.
A known finite range supports a uniform model only when equal likelihood within that range is also justified. A constant-hazard exponential model has no finite upper endpoint.
For a uniform model, use subinterval length divided by total range length. For an exponential model, use its rate and survival function.
“Given that the wait has already lasted” changes the remaining range under a uniform model; under an exponential model, use the memoryless property.
This model-selection reasoning is aligned with the SOA VEE Mathematical Statistics syllabus. In an exam problem, translate the stated mechanism into a distribution before calculating. If the mechanism does not justify equal likelihood on a bounded range or a constant hazard, neither model should be chosen just because it is mathematically convenient.
Check Your Understanding
For each problem, identify the relevant model and use its assumptions to calculate the requested quantity.
- A hypothetical inspection time \(T\) is equally likely to occur from 2 to 10 hours after a shift begins. Find \(P(T>7)\) and \(E[T]\). (Answer: Uniform on \([2,10]\); \(P(T>7)=0.3750\); \(E[T]=6\) hours.)
- A hypothetical alert time has an exponential model with rate \(0.4\) per hour. Find the probability the alert occurs after 0.5 hours but by 2 hours. (Answer: \(e^{-0.2}-e^{-0.8}=0.3694\).)
- A passenger's shuttle departure time is modeled uniformly within a 15-minute cycle beginning at the passenger's arrival. Given that 6 minutes have passed without departure, find the probability departure occurs within the next 3 minutes. (Answer: The remaining time is uniform on \([0,9]\); \(3/9=0.3333\).)
- A component's time to a hypothetical event is exponential with mean 12 minutes. Find the probability the time exceeds 20 minutes. (Answer: \(\lambda=1/12\) per minute; \(e^{-20/12}=0.1889\).)
- Two models have the same mean of 4 minutes: one is uniform on \([0,8]\), and one is exponential with rate \(0.25\) per minute. Compare their probabilities of exceeding 8 minutes and state which model suits a mechanism that guarantees the event by minute 8 and makes all times in the range equally likely. (Answer: Uniform probability \(0\); exponential probability \(e^{-2}=0.1353\); the guaranteed, equally likely bounded mechanism supports the uniform model.)