# Poisson Distribution: Formula and Examples

The Poisson distribution is a discrete probability distribution that gives the probability of observing a specific number of events in a fixed interval of time, space, or volume, when those events occur independently at a constant average rate. Its probability mass function is P(X = k) = (λ^k · e^-λ) / k!, where k is the number of events, λ (lambda) is the mean number of events per interval, and e is Euler's number, approximately 2.71828.

The Poisson distribution matters because count data are everywhere in the life sciences. The number of colony-forming units on a plate, the number of mutations in a gene, the number of adverse events reported after a vaccine, and the number of new disease cases in a county per month are all counts in a fixed window. When you can describe those counts with one number, λ, you gain a fast way to test hypotheses, set surveillance thresholds, and plan experiments. The same math underpins cancer incidence models built on cell turnover [1], software reliability models that predict failures over time [2], and epidemiological rate comparisons such as the rising incidence of infant botulism in Texas [3].

## The Poisson Formula in Plain Language

<figure class="article-figure">
  <img src="https://thumb.wikimedia.org/wikipedia/commons/thumb/c/c1/Poisson_distribution_PMF.png/1280px-Poisson_distribution_PMF.png" alt="Plot of the Poisson probability mass function for several values of lambda" loading="lazy" decoding="async" width="1000" height="750" />
  <figcaption>The Poisson PMF shows how probability is spread over counts of events for different rates λ. Image: Wikimedia Commons contributor, Public domain, via <a href="https://commons.wikimedia.org/wiki/File:Poisson_distribution_PMF.png" rel="noopener noreferrer">Wikimedia Commons</a>.</figcaption>
</figure>

The formula for the Poisson distribution is:

**P(X = k) = (λ^k · e^-λ) / k!**

Each piece has a job:

- **k** is the number of events you care about. It must be a non-negative integer (0, 1, 2, 3, and so on). You cannot observe 2.5 calls in an hour.
- **λ** is the mean number of events per interval. It is the single parameter that defines the whole distribution. If a call center receives 3 calls per hour on average, λ = 3 for a one-hour interval.
- **e^-λ** is the exponential decay term. It scales the probabilities so they sum to 1 across all possible values of k.
- **k!** is the factorial of k, the product of all integers from 1 to k. By convention, 0! = 1.

The term λ^k / k! grows quickly with k, and e^-λ shrinks. Their product produces the characteristic right-skewed shape: small counts are common, large counts are rare, and the tail stretches to the right.

### Why the Formula Has This Shape

You can derive the Poisson formula as the limit of the binomial distribution. Imagine dividing an hour into a huge number of tiny sub-intervals, say n of them. In each sub-interval, either a call arrives or it does not, so the count is binomial with n trials and success probability p = λ/n. As n grows toward infinity and p shrinks toward zero while their product λ stays fixed, the binomial probability converges to the Poisson formula. That limit is why the Poisson distribution is sometimes called the law of rare events. Each individual opportunity for an event is unlikely, but there are many opportunities, and the total count settles into a stable pattern.

## The Three Assumptions

The Poisson distribution is not a universal model for counts. It holds only when three conditions are met.

1. **Events are independent.** The occurrence of one event does not change the probability of another. A call arriving at 10:00 a.m. does not make a call at 10:01 a.m. more or less likely.
2. **Events occur at a constant rate.** The average rate λ does not drift within the interval. If call volume triples during lunch, a single λ for the whole day is wrong.
3. **Two events cannot occur at the same instant.** Events are distinct and countable. Two arrivals in exactly the same moment is not allowed, which is what makes the distribution discrete rather than continuous.

When these assumptions hold, the mean and the variance of the distribution are both equal to λ. That equality is the single most useful diagnostic in practice. If you count events and find that the variance is much larger than the mean, the data are overdispersed and a simple Poisson model will understate the variability. If the variance is smaller than the mean, the data are underdispersed. Both patterns are common in real biological and operational data, and both signal that the constant-rate independence assumption is strained.

## Worked Example: Calls per Hour

Suppose a support line receives an average of 3 calls per hour, and the arrival process satisfies the three assumptions. Then λ = 3 and you can compute the probability of any specific count.

**Find P(X = 2), the probability of exactly 2 calls in an hour.**

Step 1. Write the formula: P(X = 2) = (3^2 · e^-3) / 2!

Step 2. Compute the numerator: 3^2 = 9, and e^-3 ≈ 0.049787. So 9 × 0.049787 ≈ 0.448083.

Step 3. Compute the denominator: 2! = 2.

Step 4. Divide: 0.448083 / 2 ≈ 0.224042.

So P(X = 2) ≈ 0.224, or about 22.4%. Roughly one hour in four and a half will see exactly two calls.

**Find P(X = 0), the probability of no calls in an hour.**

P(X = 0) = (3^0 · e^-3) / 0! = (1 × 0.049787) / 1 ≈ 0.0498.

About 5% of hours will be silent. That is a useful number for staffing: even at a mean of 3 calls per hour, a quiet hour is not unusual.

**Find P(X ≤ 2), the probability of two or fewer calls.**

Sum the individual probabilities: P(0) + P(1) + P(2). Using values from the table below, 0.0498 + 0.1494 + 0.2240 ≈ 0.4232. About 42% of hours will have two or fewer calls.

## Probability Table for λ = 3

The table below gives P(X = k) for k = 0 through 6 when λ = 3. Values are rounded to four decimal places.

| k (events) | P(X = k) | Cumulative P(X ≤ k) |
|--|--|--|
| 0 | 0.0498 | 0.0498 |
| 1 | 0.1494 | 0.1992 |
| 2 | 0.2240 | 0.4232 |
| 3 | 0.2240 | 0.6472 |
| 4 | 0.1680 | 0.8152 |
| 5 | 0.1008 | 0.9160 |
| 6 | 0.0504 | 0.9664 |

Two features stand out. First, the distribution peaks at k = 2 and k = 3, both with probability 0.2240. The mode of a Poisson distribution is the largest integer less than or equal to λ, so with λ = 3 the distribution is flat-topped across 2 and 3. Second, the cumulative column shows that 96.6% of all the probability sits at k ≤ 6. Counts of 10 or more are possible but rare, and the right tail fades slowly.

If you plot the P(X = k) column as a bar chart, you get a right-skewed hump that rises from k = 0 to a plateau at k = 2 and 3, then declines. The skew is mild at λ = 3. As λ grows, the shape becomes more symmetric and approaches a normal curve, which is why large-λ Poisson probabilities are often approximated with a normal distribution.

## Poisson Versus Binomial

Students mix these two up constantly, and the distinction is simple once you state it plainly. The binomial distribution counts successes in a fixed number of trials. The Poisson distribution counts events in a fixed interval of time or space, with no fixed number of trials.

| Feature | Poisson | Binomial |
|--|--|--|
| What you count | Events in a fixed interval | Successes in a fixed number of trials |
| Parameters | λ (mean events per interval) | n (trials) and p (success probability) |
| Number of opportunities | Not fixed, effectively unlimited | Fixed at n |
| Mean | λ | n · p |
| Variance | λ | n · p · (1 - p) |
| Typical question | How many calls in an hour? | How many heads in 20 flips? |
| Typical use | Counts, rates, rare events | Proportions, pass/fail outcomes |

The Poisson distribution is the limiting case of the binomial when n is large and p is small while n · p stays constant. A rule of thumb many textbooks use: if n is at least 20 and p is at most 0.05, the Poisson approximation to the binomial is close enough for most work. That is why the Poisson distribution is the natural model for rare events in large populations, from mutations across many cell divisions to adverse events across a large vaccinated cohort [4].

## How the Poisson Distribution Is Used in Practice

### Modeling Rates and Incidence

Epidemiologists use Poisson-based methods to compare disease incidence across regions and years. A study of infant botulism in the United States and Texas used Poisson-based [statistical methods](/blog/guides/statistical-methods) to estimate incidence rate ratios, finding an annual increase of about 4.2% nationally from 2000 to 2021 and about 6.3% in Texas from 2006 to 2023, with cases clustering unevenly across public health regions [3]. The Poisson framework lets researchers put a confidence interval around a rate and test whether two rates differ.

### Modeling Cancer Incidence

A Poisson-based "np" model treats cancer incidence as the product of the effective number of cell turnovers (n) and the probability of single-cell transformation (p). The model reproduces observed human cancer incidence once the decline in cell turnover with age is included, and it supports a broader theory of aging in which accumulated mutations limit the total number of cells an organism can generate over its lifetime [1]. Here the Poisson distribution supplies the probability that a given number of transforming events occurs across a large number of cell divisions.

### Reliability and Failure Counting

Software reliability models count failures over time. A nonhomogeneous Poisson process model that allows for dependent failures and for defects already present at the start of observation outperformed 15 traditional models across three datasets on nine evaluation criteria [2]. The Poisson process gives the baseline for "how many failures should we expect in this window," and the extensions handle the fact that real failures cluster.

### Planning Clinical Trials

Sample size and accrual planning for time-to-event trials often rely on Poisson-based models. One approach combines a PERT model for site start-up delays with a Poisson-Gamma subject accrual model to predict how many sites a trial needs, and it shows minimal power loss even when the delay inputs are misspecified, compared with the traditional assumption that all sites start simultaneously [5]. The Poisson side of that model describes how subjects arrive over time.

### Ecology and Species Richness

Species detection counts are often modeled with a Gamma-Poisson structure, in which each species has its own detection rate drawn from a Gamma distribution, and the observed counts follow a Poisson distribution conditional on that rate. This parametric approach can produce nearly unbiased richness estimates when the model assumption holds, though it becomes hard to fit when data are sparse [6].

## Reading Poisson Output in Statistical Software

Most statistical packages report Poisson results in a standard format. For a single rate, you get an estimate of λ with a confidence interval. For comparing two rates, you get a rate ratio and its interval. In surveillance work, the Poisson scan statistic is used to detect spatial clusters of cases, as in a study of Mpox incidence across Ontario that combined Moran's Index, the Getis-Ord Gi* statistic, and a spatial Poisson scan statistic to find localized clusters in Toronto, Ottawa, and a secondary area [7].

When you read a Poisson model output, check three things. First, does the reported mean match your raw count mean? Second, is the variance close to the mean in your raw data? Third, do the residuals show any pattern over time or space? A pattern in the residuals is the clearest sign that the constant-rate assumption is failing.

## Common Mistakes and Limitations

**Confusing λ with a probability.** λ is a rate, not a probability. It can be any positive number, including values above 1. A λ of 3 means 3 events per interval on average, not a 3% chance.

**Forgetting that λ is tied to the interval.** A λ of 3 per hour is a λ of 0.05 per minute. If you change the interval, you must rescale λ. Mixing intervals is one of the most common errors in applied work.

**Applying Poisson to overdispersed data.** Real count data often have variance greater than the mean. Clustered disease cases, bursty neuronal firing, and aggregated organisms all violate the independence assumption. Using a plain Poisson model on overdispersed data produces confidence intervals that are too narrow and p-values that are too small, which inflates false positives.

**Ignoring the zero-inflation problem.** Some processes produce more zeros than a Poisson model predicts. If a species is absent from most sites, or a rare adverse event never occurs in most hospitals, the excess zeros signal a mixture process rather than a simple Poisson process.

**Treating the Poisson distribution as a law rather than a model.** The Poisson distribution is a useful approximation, not a description of reality. Real arrival processes have rush hours, real mutation rates vary across genomic regions, and real failure rates change as software is patched. The model is a starting point that you check against data.

**Assuming the mean and variance must match in your sample.** In a finite sample, the sample mean and sample variance will differ even when the data truly come from a Poisson distribution. The equality holds for the theoretical distribution. Use a formal dispersion test rather than eyeballing the two numbers.

**Using the Poisson distribution for bounded counts.** If the maximum possible count is small and fixed, such as the number of correct answers on a 10-question quiz, the binomial distribution is the right model. The Poisson distribution assumes no upper bound.

Individual datasets vary, and a statistician or domain expert should review any Poisson model that drives a real decision.

## Quick Review

- The Poisson formula is P(X = k) = (λ^k · e^-λ) / k!, where λ is the mean number of events per interval.
- Three assumptions: events are independent, occur at a constant rate, and cannot occur simultaneously.
- The mean and variance both equal λ. A variance far from the mean signals a violated assumption.
- For λ = 3, P(X = 2) ≈ 0.224 and P(X = 0) ≈ 0.0498.
- Poisson counts events in a fixed interval. Binomial counts successes in a fixed number of trials.
- The Poisson distribution is the large-n, small-p limit of the binomial distribution.
- Always rescale λ when you change the interval.

## Frequently Asked Questions

### What is the formula for the Poisson distribution?

P(X = k) = (λ^k · e^-λ) / k!, where k is the number of events, λ is the mean number of events per interval, and e is approximately 2.71828.

### What are the three assumptions of the Poisson distribution?

Events must be independent, occur at a constant average rate, and cannot happen at exactly the same instant.

### What is the difference between the Poisson and binomial distributions?

The Poisson distribution counts events in a fixed interval with no fixed number of trials. The binomial distribution counts successes in a fixed number of trials.

### What does it mean if the variance is larger than the mean in count data?

It means the data are overdispersed, so the independence or constant-rate assumption is likely violated and a plain Poisson model will be too confident.

### Can λ be larger than 1?

Yes. λ is a rate, not a probability, so it can be any positive number. A λ of 3 means three events per interval on average.

### How do I find the probability of at most k events?

Add the individual probabilities from 0 up to k, or read the cumulative column of a Poisson probability table.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What is the formula for the Poisson distribution?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "P(X = k) = (λ^k · e^-λ) / k!, where k is the number of events, λ is the mean number of events per interval, and e is approximately 2.71828."
      }
    },
    {
      "@type": "Question",
      "name": "What are the three assumptions of the Poisson distribution?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Events must be independent, occur at a constant average rate, and cannot happen at exactly the same instant."
      }
    },
    {
      "@type": "Question",
      "name": "What is the difference between the Poisson and binomial distributions?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "The Poisson distribution counts events in a fixed interval with no fixed number of trials. The binomial distribution counts successes in a fixed number of trials."
      }
    },
    {
      "@type": "Question",
      "name": "What does it mean if the variance is larger than the mean in count data?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "It means the data are overdispersed, so the independence or constant-rate assumption is likely violated and a plain Poisson model will be too confident."
      }
    },
    {
      "@type": "Question",
      "name": "Can λ be larger than 1?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes. λ is a rate, not a probability, so it can be any positive number. A λ of 3 means three events per interval on average."
      }
    },
    {
      "@type": "Question",
      "name": "How do I find the probability of at most k events?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Add the individual probabilities from 0 up to k, or read the cumulative column of a Poisson probability table."
      }
    }
  ]
}
</script>

## Related Articles

- [How to Calculate Dilution Factor: Formulas, Examples, and Serial Dilutions](/knowledge/diagnostics/molecular/how-to-calculate-dilution-factor)
- [Standard Plate Count: Formula, Calculation, and Example](/knowledge/diagnostics/microbiology/calculate-bacteria-standard-plate-count)
- [How to Calculate the Specific Activity of an Enzyme: Formula and Examples](/knowledge/diagnostics/molecular/calculate-enzyme-specific-activity)
- [How to Calculate Transformation Efficiency: Formula, Examples, and Common Pitfalls](/knowledge/diagnostics/molecular/calculate-transformation-efficiency-formula-examples)
- [How to Calculate CFU/mL: Formula, Serial Dilution Examples & Plating Standards](/knowledge/diagnostics/microbiology/calculate-cfu-per-ml)
- [Poisson Statistics in Digital PCR](/knowledge/molecular-biology/poisson-statistics-in-digital-pcr-how-to-calculate-absolute-copy-numbers-from-partition-data)
- [Residual Sum of Squares: Formula and Example](/blog/research-skills/residual-sum-of-squares-formula-and-example)

## Sources

1. [A Poisson distribution-based general model of cancer rates and a cancer risk-dependent theory of aging.](https://pubmed.ncbi.nlm.nih.gov/37659107/)
2. [Nonhomogeneous Poisson process software reliability model incorporating initial fault diversification and dependent failures.](https://pubmed.ncbi.nlm.nih.gov/41442325/)
3. [Evaluating the rising incidence of infant botulism and the heterogeneous distribution of cases in Texas in the wake of the November 2025 outbreak.](https://pubmed.ncbi.nlm.nih.gov/42411786/)
4. [Maximized sequential probability ratio test regression.](https://pubmed.ncbi.nlm.nih.gov/41459675/)
5. [Designing a phase-III time-to-event clinical trial using a modified sample size formula and Poisson-Gamma model for subject accrual that accounts for the lag in site initiation using the PERT distribution.](https://pubmed.ncbi.nlm.nih.gov/37926516/)
6. [A more reliable species richness estimator based on the Gamma-Poisson model.](https://pubmed.ncbi.nlm.nih.gov/36632143/)
7. [Geographical distribution and the impact of socio-environmental indicators on incidence of Mpox in Ontario, Canada.](https://pubmed.ncbi.nlm.nih.gov/40067846/)