Connecting Poisson Counts and Exponential Waiting Times

One Process, Two Ways to Describe Events

A stream of events can be described by asking how many occur in a fixed time, or by asking how long it takes for the next one to occur. For example, a service desk might count calls in each hour, while also tracking the time until the next call. These are not unrelated models: under a constant-rate Poisson process, the count probabilities determine the waiting-time probabilities.

The connection is especially direct for the first arrival. If no event has occurred by time \(t\), then the first arrival time must exceed \(t\). The count during that interval is zero. This equivalence lets us use the Poisson model to find the distribution of the waiting time, and it also lets us translate questions about later arrivals into questions about counts.

Definition: A homogeneous Poisson process with rate \(\lambda>0\) describes events occurring over time with a constant average rate. Its count \(N(t)\) in an interval of length \(t\) has a Poisson distribution with mean \(\lambda t\), so \(N(t)\sim\operatorname{Poisson}(\lambda t)\). Counts in disjoint intervals are independent, and the count distribution depends on an interval’s length, not its location. The rate \(\lambda\) is measured in events per unit time, so \(\lambda t\) is a dimensionless expected count.

The model assumes the rate is stable over the period being studied and that one event does not change the process’s rate. The process is also being used at a time scale where events can be treated as occurring individually rather than in simultaneous batches. Whether these assumptions are reasonable depends on the setting; a constant-rate model may not fit when, for example, the event rate changes sharply by time of day.

Write \(T_1\) for the time from the start of observation until the first event. More generally, \(T_k\) is the time from the start until the \(k\)th event. These are waiting times, measured in units of time. The count \(N(t)\), by contrast, is a nonnegative integer giving the number of events in a specified interval.

Deriving the First-Arrival Waiting-Time Model

The event “the first arrival occurs after time \(t\)” is exactly the event “there are no arrivals in the first \(t\) time units.” Thus \(T_1>t\) if and only if \(N(t)=0\). Using the Poisson probability of zero events gives the survival probability of the waiting time.

\begin{aligned} P(T_1>t) &=P(N(t)=0)\\ &=\frac{e^{-\lambda t}(\lambda t)^0}{0!}\\ &=e^{-\lambda t}, \qquad t\geq 0. \end{aligned}

A waiting time is at most \(t\) precisely when the first event has arrived by then. Taking the complement of the survival probability gives its cumulative distribution function:

P(T_1\leq t)=1-e^{-\lambda t}, \qquad t\geq 0.

This is the exponential distribution with rate \(\lambda\), the same rate as the Poisson process. Its mean waiting time is \(1/\lambda\), in the corresponding time units. For example, a rate of 4 events per hour gives a mean time of \(1/4\) hour, or 15 minutes. The rate is not itself a waiting time: a larger rate means a shorter average wait.

Theorem: For a homogeneous Poisson process with rate \(\lambda\), the waiting time \(T_1\) to the first event has an exponential distribution with rate \(\lambda\). In particular, \(P(T_1>t)=e^{-\lambda t}\) and \(P(T_1\leq t)=1-e^{-\lambda t}\) for \(t\geq 0\).

The formula also expresses memorylessness in process terms. Conditional on seeing no event in the first \(s\) time units, the probability of no event in the next \(t\) time units is \(P(N(s+t)-N(s)=0)=e^{-\lambda t}\), by independent and stationary increments. The elapsed wait does not alter the probability of waiting at least another \(t\) units, provided the constant-rate model remains appropriate.

Worked Example: Probability of a First Arrival Within a Time Limit

Suppose a hypothetical help desk receives calls according to a homogeneous Poisson process with rate 3 calls per hour. Find the probability that the next call arrives within 10 minutes, and the probability that the wait exceeds 10 minutes.

The rate is \(\lambda=3\) calls per hour. Convert 10 minutes to hours: \(t=10/60=1/6\) hour. The Poisson-process assumptions are that the call rate is constant during the period and that counts in disjoint intervals are independent.

The probability of waiting longer than 10 minutes is the probability of zero calls in that interval:

P(T_1>1/6)=e^{-3(1/6)}=e^{-0.5}\approx 0.6065.

Therefore, the probability of at least one call within 10 minutes is the complement:

P(T_1\leq 1/6)=1-e^{-0.5}\approx 1-0.6065=0.3935.

Under this model, there is about a 39.35% probability that the next call arrives within 10 minutes, and about a 60.65% probability that the wait is longer. The mean waiting time is \(1/3\) hour, or 20 minutes.

Later Arrivals and Counts

The same reasoning connects the time of the \(k\)th event to a count. By time \(t\), the \(k\)th event has occurred if and only if at least \(k\) events have occurred. Therefore, \(T_k\leq t\) is equivalent to \(N(t)\geq k\). This gives a way to calculate probabilities for later arrival times directly from Poisson count probabilities.

\begin{aligned} P(T_k\leq t) &=P(N(t)\geq k)\\ &=1-P(N(t)\leq k-1)\\ &=1-\sum_{j=0}^{k-1}e^{-\lambda t}\frac{(\lambda t)^j}{j!}, \qquad t\geq 0. \end{aligned}

The sum is the probability of fewer than \(k\) events by time \(t\), which is the same as the probability that the \(k\)th event has not yet occurred. For \(k=1\), the sum contains only the zero-count probability, recovering \(P(T_1\leq t)=1-e^{-\lambda t}\). For larger \(k\), the waiting time to the \(k\)th event is not the same as the waiting time to the first event; the count formula accounts for the extra arrivals required.

This translation is useful in either direction. To find a probability about a waiting time, calculate a Poisson tail for the corresponding interval. To find a probability about a count, use the Poisson distribution directly. The rate and the time interval must describe the same period: their product \(\lambda t\) is the mean count used in the Poisson probabilities.

Worked Example: Probability the Fifth Arrival Occurs by a Deadline

Suppose a hypothetical process has rate 12 events per hour. Find the probability that the fifth event occurs within 20 minutes.

Let \(T_5\) be the time to the fifth event. Twenty minutes is \(1/3\) hour, so the expected count in that interval is \(\lambda t=12(1/3)=4\). The process assumptions are a constant rate of 12 per hour and independent counts in disjoint time intervals.

The fifth event has occurred by 20 minutes exactly when the count in those 20 minutes is at least 5. Thus:

\begin{aligned} P(T_5\leq 1/3) &=P(N(1/3)\geq 5)\\ &=1-\sum_{j=0}^{4}e^{-4}\frac{4^j}{j!}\\ &=1-e^{-4}\left(1+4+8+\frac{64}{6}+\frac{256}{24}\right)\\ &=1-e^{-4}(34.3333)\\ &\approx 1-0.6288\\ &\approx 0.3712. \end{aligned}

There is about a 37.12% probability that the fifth event occurs within 20 minutes. The complementary probability, about 62.88%, is the probability that at most four events occur in that interval, so the fifth event has not yet occurred.

Combining a Count With a Later Waiting Time

Questions can combine what happened in one interval with what happens afterward. If the intervals are disjoint, their counts are independent under the homogeneous Poisson-process model. This allows a count probability for the first interval to be multiplied by a probability for the next interval. Be careful to identify which arrival is meant: after exactly \(k\) events, the next arrival is the \((k+1)\)st event overall.

Worked Example: Exactly Three Arrivals, Then a Fourth Soon After

A hypothetical process has a constant rate of 2.5 events per hour. Find the probability that exactly three events occur in the first hour and that the fourth event occurs during the following 15 minutes.

The first hour and the following 15-minute interval are disjoint. Their counts are independent, and their expected counts are \(2.5(1)=2.5\) and \(2.5(0.25)=0.625\), respectively. Given exactly three events in the first hour, the fourth event occurs in the next 15 minutes if and only if at least one event occurs during that following interval.

The probability of exactly three events in the first hour is:

P(N(1)=3)=e^{-2.5}\frac{2.5^3}{3!}\approx 0.2138.

The probability of at least one event in the next 15 minutes is:

P(N(0.25)\geq 1)=1-e^{-0.625}\approx 1-0.5353=0.4647.

By independence of counts in the two intervals, the probability of both conditions is:

P(N(1)=3\text{ and }N(1.25)-N(1)\geq 1) \approx 0.2138(0.4647)\approx 0.0993.

Thus, the probability of exactly three arrivals in the first hour and a fourth within the following 15 minutes is about 9.93%. Conditional on exactly three in the first hour, the chance of a fourth in the next 15 minutes is about 46.47%; the first-hour count does not change the independent count distribution in the following interval.

Finding a Waiting-Time Cutoff

The first-arrival formula can also be rearranged to find a time associated with a specified probability. For example, if the desired probability of having seen an arrival by time \(t\) is \(q\), then \(1-e^{-\lambda t}=q\). Solving gives \(t=-\ln(1-q)/\lambda\). The result has time units because the logarithm’s argument is dimensionless and \(\lambda\) has units of inverse time.

Worked Example: A Time by Which an Arrival Is Likely

A hypothetical process has a constant rate of 2 events per hour. Find the time by which the probability of at least one event is 0.95.

The model gives \(P(T_1\leq t)=1-e^{-2t}\), with \(t\) measured in hours. Set this probability equal to 0.95 and solve:

\begin{aligned} 1-e^{-2t}&=0.95\\ e^{-2t}&=0.05\\ -2t&=\ln(0.05)\\ t&=-\frac{\ln(0.05)}{2}\\ &\approx 1.498\text{ hours}. \end{aligned}

The cutoff is about 1.498 hours, or \(1.498(60)\approx 89.87\) minutes. Under the stated model, the probability of at least one event by that time is 0.95.

Common Mistakes and Why the Connection Matters

Confusing the rate with the mean wait. A rate of 4 events per hour corresponds to a mean first-arrival wait of \(1/4\) hour, not 4 hours. Rate has units of events per time; waiting time has units of time.
Using the wrong count event. “The fifth event occurs by time \(t\)” means \(N(t)\geq 5\), not \(N(t)=5\). The fifth arrival may occur even if additional events also occur before \(t\).
Forgetting the time units. If \(\lambda\) is per hour, \(t\) must be in hours when calculating \(\lambda t\). Convert minutes before substituting.
Multiplying probabilities for overlapping intervals. Independence applies to counts in disjoint intervals. It does not justify treating counts from overlapping intervals as independent.
Assuming the model is automatic. The formulas depend on a stable rate and the Poisson-process assumptions. A changing rate or dependence between arrivals can make these probabilities inappropriate.

The practical value of the connection is that one model can answer operationally different questions. A count-based question may ask how many arrivals occur during a shift; a waiting-time question may ask how long a person waits for the next arrival. Once the constant-rate Poisson-process assumptions are suitable, the event \(T_1>t\) is simply the zero-count event, and the event \(T_k\leq t\) is simply the count event \(N(t)\geq k\). Keeping those translations explicit helps select the right tail and interpret the result correctly.

Key takeaway: In a homogeneous Poisson process with rate \(\lambda\), \(T_1>t\) exactly when \(N(t)=0\), so \(P(T_1>t)=e^{-\lambda t}\). More generally, \(T_k\leq t\) exactly when \(N(t)\geq k\). Use consistent time units and check that the constant-rate assumptions fit the situation.

Check Your Understanding

Use the count–waiting-time connection to answer these questions.

A process has rate 5 events per hour. Write an expression for the probability that the first event takes longer than 12 minutes.
Explain why the event “the third arrival has occurred by time \(t\)” corresponds to a count of at least three, rather than exactly three.
A process has rate 2 events per minute. What is the mean waiting time for the first event, in seconds?
For rate \(\lambda\), write the probability that the fifth event occurs after time \(t\) in terms of a Poisson count.
Why may probabilities for a count in the first hour and a count in the next 10 minutes be multiplied, while probabilities for counts in overlapping intervals cannot generally be treated as independent?