Skip to content
Scintria
ToolsGrant writerFunding
HomeLiteratureClinical trialsEvidence LabProtocol packData analysisLearnProjectsLibraryReferencesApplications
Start free
ScintriaStart free

Learn / Biostatistics / Probability and distributions

The binomial and Poisson distributions

13 minute read · 6-question quiz

Download this lesson as a colour Word or PDF study guide with the figures and answers. Included in paid plans; reading and quizzes stay free.Sign in

What you will learn

  • Recognise when a count follows a binomial or a Poisson distribution

  • Calculate binomial and Poisson probabilities and their means and variances

  • Explain why these distributions underpin analysis of proportions and rates

A random variable is a number that depends on chance, such as the number of children out of ten who respond to treatment. A probability distribution lists every possible value and how likely it is. Two distributions for counts do most of the work in health research.

The binomial distribution

Use it when:

  • there is a fixed number of trials n (patients, tests, children);
  • each trial has two outcomes (success/failure, event/no event);
  • every trial has the same probability p of success;
  • trials are independent.

The probability of exactly k successes is

P(X = k) = C(n, k) p^k (1 - p)^(n - k), C(n, k) = (n!)/(k! (n-k)!)

The mean is np and the variance is np(1 - p).

Worked example: treatment response

A drug works in 30% of patients (p = 0.3). Ten patients are treated (n = 10).

  • Exactly 3 respond: C(10, 3) (0.3)^3 (0.7)^7 = 120 × 0.027 × 0.0824 = 0.267.
  • At most 3 respond: 0.028 + 0.121 + 0.233 + 0.267 = 0.650.
  • Six or more respond: add the probabilities for 6, 7, 8, 9 and 10 (0.037 + 0.009 + 0.001 + 0.000), which gives 0.047. Six or more responders occur less than 5% of the time.
  • The expected number of responders is np = 3 and the standard deviation is sqrt(10 × 0.3 × 0.7) = 1.45.
0102030Probability (%)2.8%012.1%123.3%226.7%320%410.3%53.7%60.9%70.1%8
Binomial distribution for n = 10 and p = 0.3: the most likely result is 3 responders, but anything from 1 to 5 would not be surprising.

The Poisson distribution

Use it for counts of events in a fixed amount of time, space or person-time when events are rare and independent: malaria cases per clinic per month, deaths per 1,000 person-years, adverse events per 100 patient-weeks. If the average count is λ (lambda), then

P(X = k) = (e^(-λ) λ^k)/(k!)

Its mean and variance are both λ. That feature is a useful check: if real counts have a variance far greater than their mean, they are overdispersed and a plain Poisson model is wrong (chapter 7).

Worked example: clinic visits

A rural health centre sees on average λ = 2 severe malaria admissions per week.

  • No admissions in a week: e^-2 = 0.135.
  • Exactly 2: e^-2 × 2^2 / 2! = 0.271.
  • Three or more: 1 - (0.135 + 0.271 + 0.271) = 0.323.

So about one week in three will have three or more severe cases, even though nothing has changed. A manager who panics every time this happens is reacting to chance.

0102030Probability (%)13.5%027.1%127.1%218%39%43.6%51.2%60.3%7
Poisson distribution with a mean of 2 events per week.

How the two relate

If n is large and p is small, the binomial is well approximated by a Poisson with λ = np. And when n p and n(1-p) are both comfortably above about 10, both distributions look like the bell-shaped normal curve, which is why the normal approximation lets us build confidence intervals for proportions (chapter 3).

Binomial

A fixed number of trials, each success or failure. Question: how many successes out of n? Example: 12 of 40 children infected.

Poisson

Events counted over time or space, no upper limit. Question: how many events in this period? Example: 7 deaths in 3,200 person-months.

Normal

Continuous measurements that cluster around a mean. Question: how far from the mean is this value? Example: haemoglobin.

Independence is easy to break

Children in the same household share infections; patients in the same ward share infection control. When trials are not independent, binomial and Poisson variances are too small and confidence intervals too narrow. Cluster-randomised trials (chapter 9) exist for exactly this reason.

Summary

  • Binomial: n independent yes/no trials with probability p. Mean np, variance np(1 − p).

  • Poisson: events in fixed time or space, mean equals variance (λ).

  • With rare events and large n, the binomial is approximately Poisson with λ = np.

  • Non-independence (clusters, repeated measures) breaks both models.

Check your understanding

Question 1
1

Which situation suits a binomial distribution?

Question 2
2

For n = 50 and p = 0.2, the expected number of successes is:

Question 3
3

For n = 50 and p = 0.2, the variance is:

Question 4
4

Weekly admissions follow a Poisson with mean 2. The probability of none in a week is about:

Question 5
5

Observed counts have a mean of 4 and a variance of 15. This suggests:

Question 6
6

Why do siblings in the same household threaten a binomial analysis?

Draft study guide. Spotted an error or something unclear? Tell us at hello@scintria.com. This is educational material, not clinical or regulatory advice.

Probability rules and why a positive test can misleadThe normal distribution and z-scores