Resources What's New Support Download Contact Sales: Arthur-Industrial@outlook.com
Home › Resources › Other common distributions

Other common distributions

Binomial, Poisson, exponential, Weibull, lognormal, uniform, t, chi-square and F: what each describes.

Different kinds of data follow different distributions. Choosing the right one makes limits, capability and tests correct.

For counts

  • Binomial: the number of defectives in a sample of n units when each unit is defective with probability p. Mean , variance . Behind p and np charts.
  • Poisson: the number of defects in a unit, area or time when defects occur independently at a constant rate. Mean and variance both equal λ. Behind c and u charts.
  • Geometric: the number of units until (or between) defects. Behind G charts for rare events.
  • Hypergeometric: defectives in a sample taken without replacement from a small lot. Used for exact acceptance sampling calculations.

For measurements and time

  • Uniform: every value in a range equally likely (rounding error, some positioning).
  • Exponential: time between random events at a constant rate; memoryless. Constant failure rate in reliability.
  • Weibull: flexible life distribution; the shape parameter β describes early life (β < 1), random (β = 1) or wear-out (β > 1) failures.
  • Lognormal: values whose logarithm is normal; positive and right-skewed (repair times, particle sizes, some contamination).
  • Gamma and largest/smallest extreme value: other skewed shapes used in capability and reliability.

Sampling distributions

  • Student's t: the distribution of a standardized mean when is estimated from the sample. Wider than normal for small samples. Published by William Gosset ("Student") in 1908.
  • Chi-square: behind tests and intervals for a variance and for count tables.
  • F: the ratio of two variances. Behind ANOVA and the two-variance test.

Questions and answers

How do I know which distribution fits my data?

Start from what the data are (counts, times, measurements with a physical limit at zero). Then compare fits with probability plots and a goodness-of-fit statistic such as Anderson-Darling. Prefer the simplest distribution that fits and makes physical sense.

When does the binomial look like the normal?

When and are both at least about 5 to 10.

In RS-1: Explore > Fit Distribution compares many distributions and ranks them.

← All articles