Poisson Count Models and the Probability Mass Function

Why Model Counts With a Poisson Distribution?

Many actuarial and operational questions ask how many events occur in a specified period: how many claims arrive during a quarter, how many machines fail during a scheduled test, or how many alarms sound during a work shift. These outcomes are counts, so they take nonnegative integer values. A Poisson distribution is a standard model for such counts when a single parameter adequately describes the expected number of events in the observation window.

The model is useful only when its assumptions are reasonable. A count is not automatically Poisson merely because it records events. In a typical setting, occurrences are treated as independent, the conditions governing their occurrence are reasonably stable within the specified window, and events occur singly rather than in clusters. These are modeling assumptions to assess in context, not consequences of the probability mass function.

Definition: A random variable $X$ has a Poisson distribution with parameter $\lambda$, written $X\sim\operatorname{Poisson}(\lambda)$, if $\lambda>0$ and $$ P(X=k)=\frac{e^{-\lambda}\lambda^k}{k!},\qquad k=0,1,2,\ldots. $$ The support is the set of all nonnegative integers. The parameter $\lambda$ is the expected count for the stated observation window under the model. If $\lambda=0$, the corresponding limiting model is degenerate: $P(X=0)=1$.

Here, $k$ is a possible observed count, while $\lambda$ is a parameter of the model. The factorial is defined by $k!=k(k-1)\cdots 2\cdot1$ for positive integers $k$, with $0!=1$. In particular, the probability of zero events is $e^{-\lambda}$, not zero: the support includes zero.

The window must be part of the model description. Saying “the count is Poisson with parameter 1.2” is mathematically enough to specify a distribution, but context should also make clear what events are counted and during what period or operating interval. The parameter is the expected number in that stated window; it is not the count that must occur, nor does it by itself specify a rate per unit of time. The connection between exposure length and the parameter is a separate modeling step.

Reading the Probability Mass Function

The probability mass function assigns a probability to each possible count. The exponential factor $e^{-\lambda}$ is common to all counts for the same model; $\lambda^k$ and $k!$ adjust the probability for the particular count $k$. For $k=0$, the convention $0!=1$ gives

P(X=0)=\frac{e^{-\lambda}\lambda^0}{0!}=e^{-\lambda}.

For $k=1$, the formula gives $P(X=1)=e^{-\lambda}\lambda$. For larger counts, both the power of $\lambda$ and the factorial must be included. The factorial is especially important: omitting it does not produce a probability mass function.

Worked Example: Quarterly Claim Counts

Method and model. Suppose, hypothetically, that the number of claims reported to a small benefit plan in a specified quarter is modeled by $X\sim\operatorname{Poisson}(1.2)$. Thus the expected number of reported claims in that quarter is $\lambda=1.2$. Find the probabilities of exactly zero, exactly one, and exactly two claims. There is no interest-rate or mortality assumption in this count model; neither is needed to calculate these probabilities.

Exactly zero claims. Substitute $k=0$:

P(X=0)=\frac{e^{-1.2}(1.2)^0}{0!} =e^{-1.2} \approx 0.3011942119 \approx 0.3012.

Exactly one claim. Substitute $k=1$:

P(X=1)=\frac{e^{-1.2}(1.2)^1}{1!} =1.2e^{-1.2} \approx 1.2(0.3011942119) \approx 0.3614330543 \approx 0.3614.

Exactly two claims. Substitute $k=2$:

P(X=2)=\frac{e^{-1.2}(1.2)^2}{2!} =e^{-1.2}\frac{1.44}{2} \approx 0.3011942119(0.72) \approx 0.2168598326 \approx 0.2169.

The final probabilities are rounded to four decimal places. As a check on the arithmetic, each calculation uses the same value of $e^{-1.2}$, and the two-claim probability includes division by $2!=2$. These are separate point probabilities; the model also allows counts greater than two.

Common mistake avoided: The count $k$ is not the parameter. The mean number expected in the quarter is 1.2, but the requested event “exactly one claim” uses $k=1$ in the mass function. Substituting $\lambda$ as though it were the observed count would answer a different question.

Why the Formula Defines a Probability Distribution

A probability mass function must assign nonnegative probabilities whose total is 1. The Poisson formula is nonnegative for every $k=0,1,2,\ldots$, because $e^{-\lambda}>0$, $\lambda^k\geq0$, and $k!>0$. To verify the total, use the exponential series $e^\lambda=\sum_{k=0}^{\infty}\lambda^k/k!$:

\sum_{k=0}^{\infty}P(X=k) =\sum_{k=0}^{\infty}\frac{e^{-\lambda}\lambda^k}{k!} =e^{-\lambda}\sum_{k=0}^{\infty}\frac{\lambda^k}{k!} =e^{-\lambda}e^\lambda =1.

This establishes that the stated masses form a valid probability distribution over the nonnegative integers. It also explains why the support cannot be restricted to a finite list of counts: every nonnegative integer has a mass when $\lambda>0$, even though some large counts may have very small probabilities.

Worked Example: Failures During a Scheduled Test

Method and model. In a hypothetical reliability study, let $X$ be the number of failures of a particular sensor type during one specified 1,000-hour test. Assume $X\sim\operatorname{Poisson}(0.8)$, so $\lambda=0.8$ failures is the expected count for that entire test window. Find the probability of exactly two failures. No financial interest rate or mortality assumption is involved.

The support is $0,1,2,\ldots$, so two failures is a possible count. Use $k=2$, $\lambda=0.8$, and $2!=2$:

\begin{aligned} P(X=2) &=\frac{e^{-0.8}(0.8)^2}{2!}\\ &=\frac{e^{-0.8}(0.64)}{2}\\ &=0.32e^{-0.8}\\ &\approx 0.32(0.4493289641)\\ &\approx 0.1437852685\\ &\approx 0.1438. \end{aligned}

The answer is rounded to four decimal places. The value is a probability for the full 1,000-hour test as defined; the parameter has not been rescaled to a different duration.

Common mistake avoided: The factor $2!$ is required because the requested count is two. Omitting it would give $0.64e^{-0.8}\approx0.2876$, twice the correct probability. The calculation is checked by retaining the factorial in the substitution before simplifying.

Interpreting the Parameter in Context

The parameter $\lambda$ describes the expected count for the particular setting and observation window specified in the model. For example, if a hypothetical monitoring system is modeled with $\lambda=3.5$ alarms per fixed work shift, then $X\sim\operatorname{Poisson}(3.5)$ describes the count during one such shift. It does not state that 3.5 alarms must occur in a shift: actual outcomes are integers, including zero.

The same numerical value of $\lambda$ can be used in different contexts, but its interpretation always comes from the defined events and window. A parameter of 0.8 for a test period and a parameter of 0.8 for a calendar month are not interchangeable unless the model and exposure definitions support that comparison. Do not silently change the time unit or observation period while keeping the same parameter.

Worked Example: Alarm Counts During a Fixed Shift

Method and model. Suppose the number of alarms during a specified eight-hour shift in a hypothetical facility is modeled as $X\sim\operatorname{Poisson}(3.5)$. Find the probability of exactly four alarms and identify the meaning of the parameter. No interest rate or mortality assumption applies to this count question.

For the probability, set $k=4$ and $\lambda=3.5$. First calculate the factorial and power:

4!=4\cdot3\cdot2\cdot1=24, \qquad (3.5)^4=150.0625.

Then substitute into the probability mass function:

\begin{aligned} P(X=4) &=\frac{e^{-3.5}(3.5)^4}{4!}\\ &=e^{-3.5}\frac{150.0625}{24}\\ &\approx (0.0301973834)(6.2526041667)\\ &\approx 0.188812\ldots\\ &\approx 0.1888. \end{aligned}

The probability is rounded to four decimal places. The parameter $\lambda=3.5$ is the expected number of alarms during the stated eight-hour shift under the model; it is not a possible observed alarm count.

Common mistake avoided: Replacing $k=4$ with $\lambda=3.5$ in the exponent or factorial would use the parameter where the observed count belongs. The event is exactly four alarms, so its mass uses $k=4$, while $3.5$ appears as the model parameter.

Common Poisson Model and Calculation Traps

Most errors arise from confusing the count with the parameter, using the wrong support, or applying the formula incompletely. Before calculating, write down the event being counted, the observation window, the parameter, and the requested value of $k$. Then check that $k$ is a nonnegative integer.

Using a negative or fractional count. A count must be in $0,1,2,\ldots$. The Poisson probability mass function is not evaluated at $k=1.5$ or $k=-1$.
Omitting the factorial. The denominator $k!$ is part of the mass function. It matters whenever $k\geq2$.
Confusing the parameter and the outcome. A model with $\lambda=3.5$ can produce an observed count of 0, 1, 2, or another nonnegative integer; the count itself cannot be 3.5.
Changing the window without changing the model specification. A parameter must be attached to a defined count and interval. Do not treat a parameter for one window as automatically applying to another.
Assuming every event count is Poisson. Dependence, changing conditions, or clustered occurrences can make the model unsuitable. The distribution is an assumption to justify, not a property of all count data.

Key takeaway: For $X\sim\operatorname{Poisson}(\lambda)$, the possible values are $k=0,1,2,\ldots$, and each point probability is $e^{-\lambda}\lambda^k/k!$. Interpret $\lambda$ as the expected count in the stated window, and keep it distinct from the particular count $k$ requested.

Check Your Understanding

For each problem, identify the relevant count and parameter before applying the Poisson probability mass function. Round probabilities to four decimal places.

A hypothetical dispatch center models the number of urgent calls received in a specified evening period as $X\sim\operatorname{Poisson}(2.4)$. State the support and calculate the probability of no urgent calls. (Answer: Support $0,1,2,\ldots$; $P(X=0)=0.0907$.)
A hypothetical workshop models equipment stoppages during a fixed inspection period as $X\sim\operatorname{Poisson}(0.5)$. Calculate the probability of exactly one stoppage, showing the factorial used. (Answer: $P(X=1)=0.5e^{-0.5}=0.3033$.)
A hypothetical service desk has a Poisson model with parameter $\lambda=1.6$ for a defined daily window. Calculate the probability of exactly three requests, including the value of $3!$. (Answer: $P(X=3)=e^{-1.6}(1.6)^3/6=0.1378$.)
In a hypothetical laboratory, the number of minor instrument alerts during a fixed session is modeled by $X\sim\operatorname{Poisson}(2.5)$. Calculate the probability of exactly two alerts and explain why the denominator is $2!$, not $1!$. (Answer: $P(X=2)=e^{-2.5}(2.5)^2/2!=0.2565$; the requested count is two.)
Two hypothetical, separately modeled monitoring systems count events over their own specified windows. System A has $\lambda=1.1$, and System B has $\lambda=2.2$. Calculate each system’s probability of zero events and state which is larger. (Answer: System A: $e^{-1.1}=0.3329$; System B: $e^{-2.2}=0.1108$; System A has the larger zero-event probability.)

What You'll Learn