Visual Tools
Calculators
Tables
Mathematical Keyboard
Converters
Other Tools

Which Probability Distribution Should I Use?

Each standard distribution comes from one kind of random experiment, so you identify it from how the outcomes are produced, not from the numbers. The questions below follow the decision flowchart for discrete distributions and add the three common continuous models.

When to use this

Look at three things in the problem: whether all possible values are equally likely, whether the quantity is counted (whole numbers) or measured (any value in a range), and what mechanism produces it: a fixed number of trials, draws without replacement, trials until a success, events at a rate, a waiting time, or a bell-shaped measurement.

Decision tree

yesnocountedn fixed trialsno replacementuntil a successrate λmeasuredwaiting timebell shapeAre all possible valuesequally likely?Uniform →Is the quantity countedor measured?What produces the count?Binomial →Hypergeometric →Geometric →Poisson →What is measured?Exponential →Normal →

The same tree, written out

  1. 1Every possible value has the same chance: a fair die, a random integer in a range, or a time or point that is equally likely anywhere in an interval. For an interval [a,b][a, b], use continuous uniform probability: length over b−ab - a. For a list of whole numbers, use discrete uniform probability: count over b−a+1b - a + 1.
  2. 2You count the successes in a fixed number nn of independent trials, each with the same success probability pp. Use binomial probability. For large nn, use the Poisson approximation when pp is small or the normal approximation when npnp and n(1−p)n(1-p) are both at least 1010.
  3. 3You draw a fixed number of items without replacement from a finite group with two kinds of items and count one kind. Use hypergeometric probability. If the sample is under 5%5\% of the population, the binomial with p=K/Np = K/N is a good approximation.
  4. 4You repeat independent trials with the same success probability until a success happens and ask on which trial it comes. For the first success, use geometric probability. For the rrth success with r>1r > 1, use negative binomial probability.
  5. 5You count events that occur independently at an average rate λ\lambda in an interval of time or space, with no fixed number of trials. Use Poisson probability, with λ\lambda scaled to the interval in the question.
  6. 6You measure the waiting time until the next event when events occur at a constant rate, or a lifetime that does not wear out. Use exponential probability with λ=1/mean\lambda = 1/\text{mean}.
  7. 7You measure a quantity whose values cluster symmetrically around a mean in a bell shape, or the problem states a normal model. Use normal probability with z-scores. When the question gives a percentage and asks for a value, use normal percentiles.