Cumulative and Tail Probabilities for Poisson Counts

From Individual Counts to Ranges

A probability for one particular count answers a narrow question: for example, what is the chance of exactly three calls during an hour? Many practical questions concern a range instead: at most three calls, fewer than three, at least three, or more than three. These phrases sound similar, but their boundaries differ by one count. Because counts are whole numbers, getting that boundary right is essential.

This tutorial uses the Poisson probability model to calculate such cumulative and tail probabilities. The main tools are adding the probabilities of the individual counts in an event and using a complement when that is quicker.

Definition: A Poisson count variable $X$ with parameter $\lambda>0$ has possible values $0,1,2,\ldots$ and probabilities $$ P(X=k)=\frac{e^{-\lambda}\lambda^k}{k!}, \qquad k=0,1,2,\ldots. $$ Here $\lambda$ is the Poisson parameter for the specified observation window. We use this model when it is appropriate to represent the number of events in a fixed window, with events occurring independently and at a stable average rate. The probability calculations below are conditional on the count being modeled as Poisson.

For a range of counts, the event is made up of several separate possible values. For instance, “at most 2” means $X=0$, $X=1$, or $X=2$. Since these events cannot happen simultaneously, their probabilities add. A cumulative probability collects all the individual probabilities up to a boundary.

Definition: For a nonnegative integer $k$, the cumulative probability through $k$ is $F(k)=P(X\leq k)$. It is the sum of the probabilities for counts $0,1,\ldots,k$. By convention, $F(-1)=0$, which makes formulas for ranges that start at zero work without a special case.

Translate the Words Before Calculating

The count wording determines which integer values belong in the event. In particular, “at most” includes the stated count, while “fewer than” does not. Likewise, “at least” includes its stated count, while “more than” excludes it.

Wording	Event	Calculation using $F$
At most $k$	$X\leq k$	$F(k)$
Fewer than $k$	$X<k$, or $X\leq k-1$	$F(k-1)$
At least $k$	$X\geq k$	$1-F(k-1)$
More than $k$	$X>k$, or $X\geq k+1$	$1-F(k)$
From $a$ through $b$, inclusive	$a\leq X\leq b$	$F(b)-F(a-1)$

These relationships apply to integer count thresholds. The formulas for “at least” and “more than” use complements: rather than adding all probabilities in the upper tail, subtract the probability of the opposite event from 1. A range can also be found by subtracting cumulative probabilities: $F(b)$ includes every count through $b$, and removing $F(a-1)$ leaves exactly the counts from $a$ to $b$.

Why Complements Work

For an integer $k\geq 0$, the events $X\leq k-1$ and $X\geq k$ are disjoint and together include every possible count. Therefore their probabilities sum to 1. This gives the upper-tail formula directly.

\begin{aligned} P(X\geq k) &=1-P(X\leq k-1)\\ &=1-F(k-1). \end{aligned}

The same partition argument gives $P(X>k)=1-F(k)$, because the complement of $X>k$ is $X\leq k$. For an inclusive range, the event $X\leq b$ consists of the disjoint events $X\leq a-1$ and $a\leq X\leq b$. Subtracting the first probability from the total yields $P(a\leq X\leq b)=F(b)-F(a-1)$.

When calculating $F(k)$ by hand, sum the Poisson probabilities from 0 through $k$. An optional shortcut for consecutive terms follows by taking the ratio of two probability mass function values:

\frac{P(X=j+1)}{P(X=j)} =\frac{e^{-\lambda}\lambda^{j+1}/(j+1)!}{e^{-\lambda}\lambda^j/j!} =\frac{\lambda}{j+1}, \qquad P(X=j+1)=P(X=j)\frac{\lambda}{j+1}.

Thus, once $P(X=0)=e^{-\lambda}$ is known, later individual probabilities can be obtained successively without recomputing factorials. This recurrence is a calculation aid; the event boundaries still need to be identified separately.

Worked Cumulative and Tail Probabilities

Worked Example: At Most Two Requests

Suppose, hypothetically, that the number $X$ of requests received by a service desk during one shift follows a Poisson model with $\lambda=4$. Find the probability of at most two requests and the probability of at least three.

The model assigns a Poisson count to a fixed shift, and the value of $\lambda$ is specified. “At most two” includes 0, 1, and 2, so sum those three probabilities. Using the Poisson formula:

\begin{aligned} P(X\leq 2) &=P(X=0)+P(X=1)+P(X=2)\\ &=e^{-4}+\frac{e^{-4}4^1}{1!}+\frac{e^{-4}4^2}{2!}\\ &=e^{-4}(1+4+8)\\ &=13e^{-4}\\ &=0.2381033056. \end{aligned}

Therefore the probability of at most two requests in a shift is approximately $0.2381$. “At least three” is the complement of “at most two,” so

P(X\geq 3)=1-P(X\leq 2)=1-0.2381033056=0.7618966944\approx 0.761897.

The result says that, under this hypothetical Poisson model, the probability of three or more requests during a shift is about $0.7619$. The sum is kept at full precision until the final rounding.

Worked Example: A Probability for an Inclusive Range

Suppose a hypothetical inspection process produces a Poisson-distributed count $X$ of minor flaws per item, with $\lambda=2.5$. Find the probability that an item has from two through four flaws, inclusive.

The requested event is $2\leq X\leq 4$. It includes exactly the counts 2, 3, and 4. We can add their probabilities directly, or subtract cumulative probabilities $F(4)-F(1)$. Adding the three terms gives

\begin{aligned} P(2\leq X\leq 4) &=P(X=2)+P(X=3)+P(X=4)\\ &=e^{-2.5}\left(\frac{2.5^2}{2!}+\frac{2.5^3}{3!}+\frac{2.5^4}{4!}\right)\\ &=0.2565156207+0.2137630172+0.1336018858\\ &=0.6038805237\\ &\approx 0.6039. \end{aligned}

Thus the probability of between two and four flaws, including both endpoints, is about $0.6039$ under the stated model. The endpoints matter: excluding either two or four would describe a different event and require omitting one of the terms.

Worked Example: At Least Two and More Than Four

In a separate hypothetical setting, let $X$ be the number of messages received during a specified interval, with a Poisson model having $\lambda=3$. Find (a) the probability of at least two messages and (b) the probability of more than four messages.

(a) At least two. The complementary event is fewer than two, meaning $X=0$ or $X=1$. Therefore

\begin{aligned} P(X\geq 2) &=1-P(X\leq 1)\\ &=1-\left(e^{-3}+\frac{e^{-3}3^1}{1!}\right)\\ &=1-e^{-3}(1+3)\\ &=1-0.1991482735\\ &=0.8008517265\\ &\approx 0.8009. \end{aligned}

(b) More than four. Since counts are integers, more than four means five or more. Its complement is at most four, so first add the probabilities for counts 0 through 4:

\begin{aligned} P(X\leq 4) &=e^{-3}\left(1+3+\frac{3^2}{2!}+\frac{3^3}{3!}+\frac{3^4}{4!}\right)\\ &=0.04978706837+0.1493612051+0.2240418077+0.2240418077+0.1680313557\\ &=0.8152632446. \end{aligned}

Then

P(X>4)=1-P(X\leq 4)=1-0.8152632446=0.1847367554\approx 0.1847.

The probability of at least two messages is about $0.8009$, while the probability of more than four is about $0.1847$. In the second calculation, subtracting the cumulative probability avoids separately adding an open-ended upper tail.

A Reliable Calculation Routine

Before evaluating any exponential or factorial, write down the event as an inequality. Then identify either the individual counts to add or the complementary counts to subtract. This small pause prevents a common off-by-one error.

State the model and parameter.
Identify the specified Poisson count variable $X$, its parameter $\lambda$, and the observation window. Use the model only when it is appropriate for the count setting.

Translate the wording.
Write the event with an inequality, such as $X\leq k$, $X<k$, $X\geq k$, or $X>k$. Convert strict inequalities to the corresponding integer boundary.

Choose a sum or complement.
For a lower cumulative probability, add the included individual probabilities. For an upper tail, consider subtracting the opposite cumulative probability from 1.

Calculate and interpret.
Keep unrounded values while adding or subtracting, round only the final result, and describe the probability in terms of the count and its observation window.

Common Boundary and Rounding Pitfalls

A frequent mistake is to treat “at least $k$” as though it meant “more than $k$.” For a count, “at least 4” includes $X=4$, whereas “more than 4” starts at $X=5$. Consequently, the correct complements differ: $P(X\geq 4)=1-F(3)$, but $P(X>4)=1-F(4)$. Writing these events explicitly before using a formula makes the distinction visible.

Another error is to round each individual Poisson probability too early, then report a cumulative probability with more apparent precision than the calculation supports. For example, in the first worked example, rounding the three terms before adding can change the last reported digit of the tail. Retain full calculator precision in intermediate steps and round the final probability once.

Finally, a cumulative probability is not automatically an upper-tail probability. $F(k)$ means $P(X\leq k)$, not $P(X\geq k)$. Check which side of the threshold the event describes before interpreting a calculator output or table entry.

Key takeaway: For integer-valued Poisson counts, “at most $k$” is $F(k)$, “fewer than $k$” is $F(k-1)$, “at least $k$” is $1-F(k-1)$, and “more than $k$” is $1-F(k)$. For an inclusive range, subtract cumulative probabilities at the two boundaries.

Check Your Understanding

For each question, first translate the wording into an event involving $X$.

If $X$ is Poisson with $\lambda=2$, which counts are included in “fewer than 3”?
Write a cumulative-probability expression for “at least 5” in terms of $F$.
For a Poisson variable with $\lambda=1.8$, express the probability of more than 2 as a complement.
Explain why $P(2\leq X\leq 6)=F(6)-F(1)$.
A calculation rounds each individual probability to three decimal places before summing. What risk does this create for the final result?

Wording	Event	Calculation using \(F\)
At most \(k\)	\(X\leq k\)	\(F(k)\)
Fewer than \(k\)	\(X<k\), or \(X\leq k-1\)	\(F(k-1)\)
At least \(k\)	\(X\geq k\)	\(1-F(k-1)\)
More than \(k\)	\(X>k\), or \(X\geq k+1\)	\(1-F(k)\)
From \(a\) through \(b\), inclusive	\(a\leq X\leq b\)	\(F(b)-F(a-1)\)