statcompute

Category

Probability Distribution Calculators

A distribution is a model of how a quantity varies. Pick the one whose assumptions match your situation and everything else — probabilities, critical values, p-values — follows from it.

Choosing the right distribution

The choice is made by the shape of the question, not by the data.

Counting successes in a fixed number of independent tries with the same probability each time? That is the binomial distribution — ten coin flips, a hundred emails opened or not, a batch of parts passing inspection.

Counting events in an interval when they arrive independently at a steady average rate? That is the Poisson distribution — arrivals per hour, defects per metre, goals per match.

Measuring something that is the sum of many small independent influences? That is the normal distribution — heights, measurement errors, and, crucially, the average of almost any sample.

Working with a sample mean when the population standard deviation is unknown? The t distribution is the normal's heavier-tailed cousin, and the extra weight in its tails is exactly the price of having estimated σ from the data.

Working with sums of squared deviations — variances, contingency tables? That is the chi-square distribution.

The theorem that makes it all work

The central limit theorem says the mean of a sample is approximately normal no matter what shape the population has, provided the sample is not tiny and the population variance is finite. That is why the normal distribution appears in tests of data that are visibly not normal: the test is about the mean, and the mean behaves.

Probability before distributions

Underneath sit the counting rules: probability of single and combined events, permutations and combinations for how many arrangements exist, expected value for what a random quantity averages to, and Bayes' theorem for updating a belief when evidence arrives. Bayes in particular is worth the ten minutes: it is the reason a 99%-accurate test for a rare condition still produces mostly false positives.

the populationskewed, any shape at allmeans of samples of nnormal, and narrower by √ntake many samples
The central limit theorem, in one picture: whatever shape the population has, the average of a sample from it is close to normal, centred on the population mean, and narrower by a factor of √n. Almost every test on this site rests on that.
01

The 9 calculators in this category

02

Questions people ask

When can I use the normal approximation to the binomial?

The usual rule is np ≥ 10 and n(1 − p) ≥ 10 — both, not either. Below that the binomial is visibly skewed and the approximation misjudges the tails, which is exactly where p-values live. The binomial calculator computes the exact answer regardless, so there is rarely a reason to approximate.

What is the difference between the t and the normal distribution?

The t distribution has heavier tails, and how much heavier depends on the degrees of freedom. It is the correct distribution for a sample mean when σ is estimated from the data. At 30 degrees of freedom the difference is small; at 5 it is large; past about 100 it is negligible.

Is the Poisson distribution just the binomial with large n?

It is the limit of the binomial as n grows and p shrinks with np held constant at λ. That is why it fits rare events among many opportunities, and why its mean and variance are both λ.