Poisson Rates and Exposure

From a Rate to a Count Model

A rate describes how frequently events occur relative to some unit: for example, calls per hour, defects per kilometre, or spots per square centimetre. A Poisson probability calculation, however, requires a parameter for a particular observation: the expected number of events in the amount of time, distance, area, or other exposure being studied. Connecting those two quantities is the central task of this tutorial.

The conversion is simple once the units are aligned: multiply the rate by the amount of exposure. But using a rate as though it were already the Poisson parameter, or multiplying by an exposure expressed in incompatible units, can produce a substantially wrong probability. The model also requires more than arithmetic: the rate must be appropriate for the exposure, and a Poisson count model must be a reasonable description of the event process.

Definition: Suppose events occur at a constant rate $r$ events per unit of exposure. For an exposure amount $E$ measured in those same units, the Poisson parameter is $$ \lambda = rE. $$ The count $X$ over that exposure is modeled as $X \sim \operatorname{Poisson}(\lambda)$. The parameter $\lambda$ is the expected count for that exposure, and it has no units.

Exposure is the amount over which events are counted. It might be elapsed time, road length, surface area, or another measurable opportunity for events to occur. The rate and exposure must refer to the same kind of exposure: a rate stated per hour is multiplied by hours, not kilometres. If the units differ, convert one quantity before multiplying.

A rate is not itself a probability. For example, a rate of 1.8 calls per hour can be greater than one because a rate counts expected events per unit. After multiplying by the exposure, the resulting $\lambda$ is the mean count for that observation. It can also be greater than one; it is not restricted to the interval from zero to one.

Why the Product Gives the Parameter

The relationship follows from the meaning of a rate and of the Poisson parameter. A constant rate $r$ means that the expected number of events per unit of exposure is $r$. Across $E$ units of exposure, the expected count is therefore $r$ multiplied by $E$. In a Poisson model, the parameter $\lambda$ equals the expected count, so $\lambda=rE$.

\text{expected count} = \left(\frac{\text{expected events}}{\text{unit of exposure}}\right) \left(\text{number of exposure units}\right) = rE = \lambda

This argument relies on a constant rate over the exposure under consideration. If, for instance, calls arrive at a different rate at different times, one constant hourly rate may not describe a long observation period well. Likewise, a road-wide average rate may be inappropriate if defects are concentrated in a particular section. The product $rE$ is useful only when the rate is relevant to the exposure being modeled.

Once $\lambda$ is found, the Poisson probability mass function gives the probability of an exact count $k$, where $k$ is a nonnegative integer:

P(X=k)=\frac{e^{-\lambda}\lambda^k}{k!}, \qquad k=0,1,2,\ldots

For a range of counts, add the probabilities for the included integer values. For example, “no more than two” means $X=0$, $X=1$, or $X=2$. “At least one” can often be found more directly by subtracting the probability of zero from one.

Worked Rate-and-Exposure Calculations

Worked Example: Calls During a Shift

Suppose a hypothetical support desk receives calls at a constant rate of 1.8 calls per hour. Find the probability of exactly three calls during a 2.5-hour shift, using a Poisson model.

Check the model and units. The problem specifies a constant rate over the shift and asks us to use a Poisson model for the count. The rate is per hour and the exposure is in hours, so they match. The model also treats calls as occurring in a way adequately described by the Poisson count distribution over this shift.

Find the parameter. Multiplying calls per hour by hours gives the expected count for the shift:

\lambda = rE = (1.8\ \text{calls/hour})(2.5\ \text{hours}) = 4.5\ \text{calls}.

The unit cancels in the multiplication: $(\text{calls/hour})(\text{hours})$ gives calls. The numerical parameter is $\lambda=4.5$. A Poisson parameter is a mean count, not a claim that the shift must contain a noninteger number of calls.

Calculate the exact-count probability. Substitute $k=3$ and $\lambda=4.5$ into the probability mass function:

P(X=3) = \frac{e^{-4.5}(4.5)^3}{3!} = \frac{e^{-4.5}(91.125)}{6} \approx 0.1687.

Under this model, the probability of exactly three calls during a 2.5-hour shift is about 0.1687, or 16.87%.

Worked Example: Defects Along a Road

Imagine that a hypothetical road-maintenance team models defects as occurring at a constant rate of 0.12 defects per kilometre along a particular type of road. Find the probability of at least one defect in a 5-kilometre section.

Check the model and units. The rate is specified per kilometre, and the section length is in kilometres. We assume the stated rate applies throughout this section and that a Poisson model is suitable for its count. These assumptions may be less reasonable if, for example, defects cluster in locations with different road conditions.

Find the parameter.

\lambda = rE = (0.12\ \text{defects/km})(5\ \text{km}) = 0.60\ \text{defects}.

Use the complement. “At least one” includes every positive count. Its complement is zero defects. For a Poisson variable, $P(X=0)=e^{-\lambda}$, because $0!=1$ and $\lambda^0=1$. Therefore:

P(X\geq 1) = 1-P(X=0) = 1-e^{-0.60} \approx 0.4512.

The modeled probability of at least one defect in the 5-kilometre section is about 0.4512, or 45.12%. Notice that the rate is only 0.12 defects per kilometre, while the probability concerns the whole section; they are different quantities.

Worked Example: Spots on a Surface

Suppose a hypothetical inspection process models small spots on a material surface at a constant rate of 0.04 spots per square centimetre. Find the probability of no more than two spots in a 20-square-centimetre sample.

Check the model and units. The area is expressed in square centimetres, matching the rate denominator. We assume the rate applies across the sampled surface and that a Poisson model adequately describes the count. In practice, systematic variation or clustering across the surface could undermine that assumption.

Find the parameter.

\lambda = rE = (0.04\ \text{spots/cm}^2)(20\ \text{cm}^2) = 0.80\ \text{spots}.

Add the probabilities for the allowed counts. “No more than two” means zero, one, or two spots:

\begin{aligned} P(X\leq 2) &=P(X=0)+P(X=1)+P(X=2)\\ &=e^{-0.8}\left(1+\frac{0.8^1}{1!}+\frac{0.8^2}{2!}\right)\\ &=e^{-0.8}(1+0.8+0.32)\\ &=e^{-0.8}(2.12) \approx 0.9526. \end{aligned}

Thus, the probability of no more than two spots in the 20-square-centimetre sample is about 0.9526, or 95.26%, under the stated model.

Assumptions and Unit Checks

The calculation $\lambda=rE$ does not by itself establish that a Poisson model is appropriate. Before interpreting a probability, check that the model's rate and its exposure describe the same setting and that the count is reasonably modeled as Poisson. In particular, the constant-rate assumption means the average event frequency does not change materially over the exposure being considered. The count model must also be suitable for the way events occur; strong clustering or dependence can make a Poisson model inadequate.

A useful unit check is to write the units explicitly before calculating. If $r$ is 3 events per day and $E$ is 8 hours, the units do not yet cancel. Convert 8 hours to $8/24$ day, or convert the rate to events per hour, before finding $\lambda$. Similarly, square centimetres cannot be multiplied by a rate per square metre without a conversion.

Rates may be provided as stipulated values, or they may be summaries from earlier observations. If a problem treats a rate as fixed, the calculations here use it as given. They do not account for uncertainty in an estimated rate; that is a separate issue from calculating a Poisson count probability conditional on a specified rate.

Common Mistake: Treating the Rate as the Mean Count

A frequent error is to put the rate directly into the Poisson formula, even when the exposure is not one unit. For the call example, using $\lambda=1.8$ for a 2.5-hour shift would model the count for a one-hour exposure under the stated rate, not for the full shift. The shift's parameter is $1.8\times2.5=4.5$.

A related error is multiplying before checking the units. A numerical product can be computed even when one quantity is per hour and the other is measured in days, but that product does not represent the expected count until the units are made consistent. Write the rate units and exposure units, cancel them, and verify that the result is an expected number of events.

Key takeaway: For a constant rate $r$ per unit of exposure and an exposure amount $E$ in matching units, use $\lambda=rE$. Then use that exposure-specific Poisson parameter—not the rate alone—in the count probability calculation.

Check Your Understanding

For each question, identify the exposure units before deciding how to calculate the probability.

A hypothetical clinic models arrivals at 2.4 patients per hour. What is $\lambda$ for a 3-hour observation period?
A machine produces a modeled average of 0.03 flaws per metre of material. What is the Poisson parameter for a 12-metre length?
A rate is given as 5 events per day, but the exposure is 6 hours. What unit conversion is needed before calculating $\lambda$?
Why is it incorrect to say that a rate of 1.8 calls per hour means the probability of a call in one hour is 1.8?
Name two features of an event process that could make a constant-rate Poisson model questionable.

What You'll Learn