Skip to main content
v2026.11,986 entries · CC-BY 4.0

Standard Deviation of a Poisson Distribution: √λ, Derivation and Examples

The standard deviation of a Poisson distribution is √λ. See the derivation, a worked example, rate and sum rules, and Excel, R and Python code.

Written and maintained by CASRAI Editorial Board

Published

On this page

The standard deviation of a Poisson distribution is the square root of its mean: if X follows a Poisson distribution with mean λ, then SD(X) = √λ. A Poisson count averaging 9 events per interval therefore has a standard deviation of 3, and one averaging 100 has a standard deviation of 10. This works because the Poisson distribution has one parameter, and its mean and its variance are both equal to λ.

Definition: what the standard deviation of a Poisson distribution measures

A Poisson distribution models the number of events counted in a fixed interval of time, area or volume when events occur independently at a constant average rate. Its standard deviation describes how far a single count typically lands from the average count λ. Unlike the normal distribution, where mean and spread are set independently, the Poisson spread is locked to the mean: a higher average always means a larger absolute spread, and no separate σ parameter exists to estimate.

Poisson distribution standard deviation formula

The probability of observing exactly x events is p(x) = λx e−λ / x!, for x = 0, 1, 2, … (R documentation, stats package). The NIST/SEMATECH e-Handbook of Statistical Methods lists the summary measures for this distribution as follows.

  • Mean = λ
  • Variance = λ
  • Standard deviation = √λ
  • Coefficient of variation = 1/√λ
  • Skewness = 1/√λ
  • Kurtosis = 3 + 1/λ

Symbols: λ (lambda) is the expected number of events per interval, a positive real number; x is an observed count, a non-negative integer; e is Euler’s number, about 2.71828; x! is x factorial; √λ is the positive square root of λ. The standard deviation is expressed in the same unit as the count, for example colonies per plate or calls per hour.

Why the variance equals λ

Using the identity Var(X) = E[X(X − 1)] + E[X] − (E[X])², and the fact that E[X] = λ and E[X(X − 1)] = λ² for a Poisson variable, the variance is λ² + λ − λ² = λ. Taking the square root gives the standard deviation √λ. Massey University’s R resources page works the same result from first principles by summing x²p(x). Because the variance is exactly λ, the standard deviation of a Poisson distribution can be read straight off the mean, without data beyond the rate.

Standard deviation of a Poisson distribution: worked example

Suppose a bacterial culture produces an average of 9 colonies per plate under fixed conditions, and counts are treated as Poisson with λ = 9.

  1. Variance = λ = 9.
  2. Standard deviation = √9 = 3 colonies.
  3. One standard deviation either side of the mean covers counts from 9 − 3 = 6 to 9 + 3 = 12.
  4. Summing the Poisson probabilities for x = 6 through 12 gives P(6 ≤ X ≤ 12) = 0.760, so about 76% of plates fall within one SD of the mean.
  5. Widening to two SDs (3 to 15) gives P(3 ≤ X ≤ 15) = 0.972, about 97%.

The coverage is close to, but not exactly, the 68% and 95% of a normal distribution, because the Poisson distribution is discrete and right-skewed at small λ.

Standard deviation of a total and of a rate

Independent Poisson counts add on the variance scale, not the SD scale. If three independent counts have means 12, 7 and 5, the total has mean 24, variance 24 and standard deviation √24 ≈ 4.90, not the sum of the three individual SDs. For a rate, divide the count by its exposure: 30 events observed over 6 exposure units gives a rate of 5 per unit. The count has SD √30 ≈ 5.48, so the rate has SD 5.48 / 6 ≈ 0.91 per unit. The same rule is why longer observation always gives a more precise rate.

Interpretation: absolute spread grows, relative spread shrinks

The coefficient of variation 1/√λ means that the SD grows with the mean, but more slowly than the mean does. The data table below lists the SD, variance and coefficient of variation for several values of λ, all computed from the NIST formulas.

Mean count (λ) What the SD of √λ implies Reading guidance
1 SD = 1, CV = 100% The spread is as large as the mean; single counts are very noisy.
4 SD = 2, CV = 50% Typical counts range roughly 2 to 6.
9 SD = 3, CV ≈ 33% Typical counts range roughly 6 to 12.
25 SD = 5, CV = 20% Relative noise has fallen to one fifth of the mean.
100 SD = 10, CV = 10% A hundredfold larger mean gives only a tenfold larger SD.

To halve the relative uncertainty of a count, you need four times as many expected events. That is the practical reading of “counting statistics” in cell counting, particle counting and sequencing depth.

Assumptions and common mistakes

  • Using a sample SD for a Poisson variable without checking it. If the observed SD is much larger than the square root of the observed mean, the data are overdispersed and a Poisson model is not appropriate. Count models with an extra dispersion parameter, such as the negative binomial distribution, are the usual alternative.
  • Applying √λ to rates or proportions. The formula holds for counts. For a rate, divide by the exposure as shown above; for a percentage, the Poisson SD does not apply directly.
  • Adding SDs of independent counts. Add variances, then take the square root.
  • Ignoring independence. Clustered events (for example, infections within a household) break the Poisson assumption and inflate the real spread above √λ.
  • Confusing the theoretical and the sample SD. √λ is a property of the model. The sample standard deviation of your data is an estimate and will vary around it.

Control limits for counts

The NIST handbook’s c-chart for counts of defects uses the Poisson property that the mean and variance both equal c. A k-sigma chart has upper control limit c + k√c. With c = 9 and k = 3, the limit is 9 + 3 × 3 = 18 and the lower limit 9 − 9 = 0. See choosing a control chart for when a count chart is the right type.

Poisson versus the normal approximation

For large λ the Poisson distribution is well approximated by a normal distribution with mean λ and standard deviation √λ, which is why the interval of λ ± 2√λ is often used as a quick range for counts. At small λ, as the worked example shows, the exact Poisson probabilities differ from the 68% and 95% normal rules because counts cannot be negative and the distribution is skewed; the skewness of 1/√λ listed by NIST confirms that the asymmetry fades only as λ grows. When a lower limit calculated as λ − k√λ falls below zero, as it does for a 3-sigma chart when λ is below 9, the limit is set to zero because a count cannot be negative.

When the Poisson standard deviation is the wrong tool

Many real count series have a variance larger than their mean. Examples include events that arrive in bursts, counts pooled across heterogeneous units, and data with excess zeros. In these cases √mean understates the true spread, and any interval or control limit built from it will flag too many ordinary observations as unusual. A practical screen is the ratio of the sample variance to the sample mean: values near 1 are consistent with a Poisson model, values well above 1 point to overdispersion. Where overdispersion is present, regression analysts typically move to a Poisson regression with a dispersion check or to a negative binomial model.

How to report the standard deviation of a Poisson variable (APA)

In APA style, report the mean and standard deviation of observed data in the form M = 9.0, SD = 3.0. If the SD is the model value rather than a sample statistic, say so in the text. Example wording: “Colony counts per plate were modelled as Poisson distributed with λ = 9, giving a theoretical standard deviation of √9 = 3.0.” When the sample SD departs clearly from √M, report both values and state which model you used. For the general conventions on describing spread, see the guide on the standard deviation of a probability distribution.

How to calculate the Poisson standard deviation in Excel, R or Python

Excel

=SQRT(9)            returns 3
=SQRT(A2)           where A2 holds lambda

R

lambda = 9
sqrt(lambda)                         # 3
x =- 0:200
sqrt(sum((x - lambda)^2 * dpois(x, lambda)))   # 3, from the definition

Python

from scipy.stats import poisson
poisson.std(9)        # 3.0
poisson.var(9)        # 9.0
poisson(9).std()      # 3.0

For a sample of counts you can compare np.std(data, ddof=1) with the square root of np.mean(data); close agreement supports the Poisson assumption. The standard deviation in R guide covers the sample-statistic functions.

Frequently asked questions

What is the standard deviation of a Poisson distribution?

It is the square root of the mean: SD = √λ. The variance of a Poisson distribution equals λ, so taking the square root gives the standard deviation.

Why are the mean and variance of a Poisson distribution equal?

The Poisson distribution has a single parameter λ, and both E(X) and Var(X) evaluate to λ when the probability function is summed over x = 0, 1, 2, and so on.

How do I find the standard deviation of a Poisson distribution from data?

Estimate λ with the sample mean, then take its square root. Compare that with the sample SD of the counts; a much larger sample SD signals overdispersion.

Is the Poisson standard deviation the same as the standard error?

No. √λ is the spread of a single count. The standard error of a mean of n independent counts is √(λ/n).

Can the standard deviation be larger than the mean?

Yes, when λ is below 1: for λ = 0.5 the SD is √0.5 ≈ 0.71, which exceeds the mean. For λ above 1, √λ is smaller than λ.

Sources

Data: Poisson mean, variance, standard deviation, coefficient of variation and 3-sigma c-chart limits for selected λ

For five values of the Poisson mean λ, the variance and standard deviation (NIST formulas: variance λ, SD √λ), the coefficient of variation 1/√λ and the 3-sigma c-chart limits c ± 3√c, computed from the NIST handbook definitions; a negative lower limit is shown as 0.

Poisson mean, variance, standard deviation, coefficient of variation and 3-sigma c-chart limits for selected λ
Mean countVarianceStandard deviationCoefficient of variation (ratio)3-sigma upper limit3-sigma lower limit (floored at 0)
λ = 111140
λ = 4420.5100
λ = 9930.3333180
λ = 252550.24010
λ = 100100100.113070

Sources

Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · free to try

Ask about Standard Deviation of a Poisson Distribution: √λ, Derivation and Examples

2 free questions a day, no account.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →