Probability distributions

Probability distributions
Ainali · CC BY-SA 3.0 · source
Name	Probability distributions
Caption	Probability density and mass functions illustration
Field	Statistics
Related	Probability theory, Statistical inference, Random variable

Probability distributions

A probability distribution assigns probabilities to outcomes of a random variable, linking measurable outcomes to numerical likelihoods. It underpins Pierre-Simon Laplace's work, informs Ronald Fisher's methods, drives models used by Karl Pearson and Andrey Kolmogorov, and supports applied work in organizations such as the World Health Organization and National Aeronautics and Space Administration. Modern treatments connect to computational tools developed at institutions like Massachusetts Institute of Technology, University of Cambridge, and Stanford University.

Introduction

Probability distributions formalize how chance behaves for experiments studied by Blaise Pascal, Gerolamo Cardano, and later systematized by Aleksandr Lyapunov and Bruno de Finetti. They separate into discrete cases emphasized by Sofia Kovalevskaya and continuous theories advanced by Émile Borel. Distributions are represented by mass functions, density functions, cumulative functions, or characteristic functions used in analyses at Bell Labs, IBM Research, and Los Alamos National Laboratory.

Definitions and basic properties

A distribution for a random variable X is specified by a cumulative distribution function (CDF) F(x) satisfying nondecreasing, right-continuous properties proved in early work by Andrey Kolmogorov and used in the Kolmogorov–Smirnov test. Discrete distributions use probability mass functions (PMFs) summing to one; continuous distributions use probability density functions (PDFs) integrating with respect to measures developed in part by Henri Lebesgue. Key properties include support, moments (mean, variance, skewness, kurtosis) studied by Francis Galton and Karl Pearson, moment-generating functions linked to Paul Lévy, and characteristic functions central in the Central Limit Theorem proofs by William Feller and Pafnuty Chebyshev.

Common discrete distributions

Prominent discrete laws include the Bernoulli distribution (single trial) used by Thomas Bayes-inspired analysts; the Binomial distribution (fixed trials) applied in agricultural experiments by Ronald Fisher; the Poisson distribution (rare events) arising in studies at Pierre-Simon Laplace's era and applied at AT&T and European Space Agency for counting problems; the Geometric distribution and Negative binomial distribution for waiting-time models used in epidemiology at the Centers for Disease Control and Prevention; and the Hypergeometric distribution for sampling without replacement in surveys by United Nations agencies. Other discrete families include the Multinomial distribution for categorical counts, the Zipf's law-related distributions studied by George Kingsley Zipf, and zero-inflated or hurdle models developed in ecological work at Smithsonian Institution.

Common continuous distributions

Key continuous distributions comprise the Normal distribution central to Central Limit Theorem results and adopted across Harvard University and Princeton University research; the Uniform distribution for simple random sampling problems; the Exponential distribution and the family of Gamma distribution for survival analysis in collaborations at Johns Hopkins University; the Beta distribution for proportions used in projects at National Institutes of Health; the Student's t-distribution conceived by William Sealy Gosset at Guinness Brewery; the Chi-squared distribution in tests developed by Karl Pearson; and heavy-tailed laws like the Cauchy distribution and Pareto distribution applied in finance at firms like Goldman Sachs and studies at London School of Economics. Multimodal and skewed families include the Log-normal distribution and the Weibull distribution used by Boeing and General Electric for reliability engineering.

Parameter estimation and inference

Estimating distribution parameters leverages methods introduced by Ronald Fisher such as maximum likelihood estimation (MLE) and the Fisher information matrix, as well as method of moments approaches used by Karl Pearson. Bayesian estimation, building on Thomas Bayes and formalized by Sir Harold Jeffreys, employs prior and posterior distributions implemented in software from groups at Carnegie Mellon University and University of Oxford. Hypothesis testing and confidence intervals reference work by Jerzy Neyman and Egon Pearson in the Neyman–Pearson framework; resampling techniques like the bootstrap were popularized by Bradley Efron at Stanford University.

Transformations and functions of random variables

Mapping random variables through measurable functions uses transformation rules from measure theory by Henri Lebesgue and change-of-variable formulae familiar in applied work at Lawrence Berkeley National Laboratory. Characteristic functions and moment-generating functions, developed in studies by Paul Lévy and Aleksandr Lyapunov, facilitate convolution and limit theorems; transformations such as logarithms yield log-normal behavior observed by Louis Bachelier in finance. Order statistics and extreme-value theory trace to work by Emil Julius Gumbel and are used in climate studies at National Oceanic and Atmospheric Administration.

Multivariate distributions and copulas

Multivariate distributions generalize marginal and joint behavior as formalized in early multivariate analysis at University of Chicago and by Harold Hotelling. The Multivariate normal distribution underlies principal component analysis popularized at Bell Labs and Google; covariance and correlation matrices connect to work by Andrey Kolmogorov and Harold Hotelling. Copulas, introduced by Aubert Sklar (Sklar's theorem), allow coupling of marginals and dependence structures and are applied in risk management at JPMorgan Chase and in hydrology research at United States Geological Survey.

Applications and examples

Probability distributions model phenomena across science and industry: reliability testing at NASA, clinical trial design at Food and Drug Administration, actuarial modelling at Prudential Financial, queueing at AT&T, genomics at Broad Institute, and election forecasting at FiveThirtyEight. Examples include modeling radioactive decay with the Poisson distribution in nuclear labs like CERN, lifespan with the Weibull distribution at General Electric, and financial returns with Student's t-distribution in hedge funds such as Bridgewater Associates.

Category:Probability theory