Sufficient statistic

Last updated

In statistics, a statistic is sufficient with respect to a statistical model and its associated unknown parameter if "no other statistic that can be calculated from the same sample provides any additional information as to the value of the parameter". [1] In particular, a statistic is sufficient for a family of probability distributions if the sample from which it is calculated gives no additional information than the statistic, as to which of those probability distributions is the sampling distribution.

Contents

A related concept is that of linear sufficiency, which is weaker than sufficiency but can be applied in some cases where there is no sufficient statistic, although it is restricted to linear estimators. [2] The Kolmogorov structure function deals with individual finite data; the related notion there is the algorithmic sufficient statistic.

The concept is due to Sir Ronald Fisher in 1920. Stephen Stigler noted in 1973 that the concept of sufficiency had fallen out of favor in descriptive statistics because of the strong dependence on an assumption of the distributional form (see Pitman–Koopman–Darmois theorem below), but remained very important in theoretical work. [3]

Background

Roughly, given a set of independent identically distributed data conditioned on an unknown parameter , a sufficient statistic is a function whose value contains all the information needed to compute any estimate of the parameter (e.g. a maximum likelihood estimate). Due to the factorization theorem (see below), for a sufficient statistic , the probability density can be written as . From this factorization, it can easily be seen that the maximum likelihood estimate of will interact with only through . Typically, the sufficient statistic is a simple function of the data, e.g. the sum of all the data points.

More generally, the "unknown parameter" may represent a vector of unknown quantities or may represent everything about the model that is unknown or not fully specified. In such a case, the sufficient statistic may be a set of functions, called a jointly sufficient statistic. Typically, there are as many functions as there are parameters. For example, for a Gaussian distribution with unknown mean and variance, the jointly sufficient statistic, from which maximum likelihood estimates of both parameters can be estimated, consists of two functions, the sum of all data points and the sum of all squared data points (or equivalently, the sample mean and sample variance).

In other words, the joint probability distribution of the data is conditionally independent of the parameter given the value of the sufficient statistic for the parameter. Both the statistic and the underlying parameter can be vectors.

Mathematical definition

A statistic t = T(X) is sufficient for underlying parameter θ precisely if the conditional probability distribution of the data X, given the statistic t = T(X), does not depend on the parameter θ. [4]

Alternatively, one can say the statistic T(X) is sufficient for θ if, for all prior distributions on θ, the mutual information between θ and T(X) equals the mutual information between θ and X. [5] In other words, the data processing inequality becomes an equality:

Example

As an example, the sample mean is sufficient for the mean (μ) of a normal distribution with known variance. Once the sample mean is known, no further information about μ can be obtained from the sample itself. On the other hand, for an arbitrary distribution the median is not sufficient for the mean: even if the median of the sample is known, knowing the sample itself would provide further information about the population mean. For example, if the observations that are less than the median are only slightly less, but observations exceeding the median exceed it by a large amount, then this would have a bearing on one's inference about the population mean.

Fisher–Neyman factorization theorem

Fisher's factorization theorem or factorization criterion provides a convenient characterization of a sufficient statistic. If the probability density function is ƒθ(x), then T is sufficient for θ if and only if nonnegative functions g and h can be found such that

i.e. the density ƒ can be factored into a product such that one factor, h, does not depend on θ and the other factor, which does depend on θ, depends on x only through T(x). A general proof of this was given by Halmos and Savage [6] and the theorem is sometimes referred to as the Halmos–Savage factorization theorem. [7] The proofs below handle special cases, but an alternative general proof along the same lines can be given. [8]

It is easy to see that if F(t) is a one-to-one function and T is a sufficient statistic, then F(T) is a sufficient statistic. In particular we can multiply a sufficient statistic by a nonzero constant and get another sufficient statistic.

Likelihood principle interpretation

An implication of the theorem is that when using likelihood-based inference, two sets of data yielding the same value for the sufficient statistic T(X) will always yield the same inferences about θ. By the factorization criterion, the likelihood's dependence on θ is only in conjunction with T(X). As this is the same in both cases, the dependence on θ will be the same as well, leading to identical inferences.

Proof

Due to Hogg and Craig. [9] Let , denote a random sample from a distribution having the pdf f(x, θ) for ι < θ < δ. Let Y1 = u1(X1, X2, ..., Xn) be a statistic whose pdf is g1(y1; θ). What we want to prove is that Y1 = u1(X1, X2, ..., Xn) is a sufficient statistic for θ if and only if, for some function H,

First, suppose that

We shall make the transformation yi = ui(x1, x2, ..., xn), for i = 1, ..., n, having inverse functions xi = wi(y1, y2, ..., yn), for i = 1, ..., n, and Jacobian . Thus,

The left-hand member is the joint pdf g(y1, y2, ..., yn; θ) of Y1 = u1(X1, ..., Xn), ..., Yn = un(X1, ..., Xn). In the right-hand member, is the pdf of , so that is the quotient of and ; that is, it is the conditional pdf of given .

But , and thus , was given not to depend upon . Since was not introduced in the transformation and accordingly not in the Jacobian , it follows that does not depend upon and that is a sufficient statistics for .

The converse is proven by taking:

where does not depend upon because depend only upon , which are independent on when conditioned by , a sufficient statistics by hypothesis. Now divide both members by the absolute value of the non-vanishing Jacobian , and replace by the functions in . This yields

where is the Jacobian with replaced by their value in terms . The left-hand member is necessarily the joint pdf of . Since , and thus , does not depend upon , then

is a function that does not depend upon .

Another proof

A simpler more illustrative proof is as follows, although it applies only in the discrete case.

We use the shorthand notation to denote the joint probability density of by . Since is a function of , we have , as long as and zero otherwise. Therefore:

with the last equality being true by the definition of sufficient statistics. Thus with and .

Conversely, if , we have

With the first equality by the definition of pdf for multiple variables, the second by the remark above, the third by hypothesis, and the fourth because the summation is not over .

Let denote the conditional probability density of given . Then we can derive an explicit expression for this:

With the first equality by definition of conditional probability density, the second by the remark above, the third by the equality proven above, and the fourth by simplification. This expression does not depend on and thus is a sufficient statistic. [10]

Minimal sufficiency

A sufficient statistic is minimal sufficient if it can be represented as a function of any other sufficient statistic. In other words, S(X) is minimal sufficient if and only if [11]

  1. S(X) is sufficient, and
  2. if T(X) is sufficient, then there exists a function f such that S(X) = f(T(X)).

Intuitively, a minimal sufficient statistic most efficiently captures all possible information about the parameter θ.

A useful characterization of minimal sufficiency is that when the density fθ exists, S(X) is minimal sufficient if and only if[ citation needed ]

is independent of θ :S(x) = S(y)

This follows as a consequence from Fisher's factorization theorem stated above.

A case in which there is no minimal sufficient statistic was shown by Bahadur, 1954. [12] However, under mild conditions, a minimal sufficient statistic does always exist. In particular, in Euclidean space, these conditions always hold if the random variables (associated with ) are all discrete or are all continuous.

If there exists a minimal sufficient statistic, and this is usually the case, then every complete sufficient statistic is necessarily minimal sufficient [13] (note that this statement does not exclude a pathological case in which a complete sufficient exists while there is no minimal sufficient statistic). While it is hard to find cases in which a minimal sufficient statistic does not exist, it is not so hard to find cases in which there is no complete statistic.

The collection of likelihood ratios for , is a minimal sufficient statistic if the parameter space is discrete .

Examples

Bernoulli distribution

If X1, ...., Xn are independent Bernoulli-distributed random variables with expected value p, then the sum T(X) = X1 + ... + Xn is a sufficient statistic for p (here 'success' corresponds to Xi = 1 and 'failure' to Xi = 0; so T is the total number of successes)

This is seen by considering the joint probability distribution:

Because the observations are independent, this can be written as

and, collecting powers of p and 1  p, gives

which satisfies the factorization criterion, with h(x) = 1 being just a constant.

Note the crucial feature: the unknown parameter p interacts with the data x only via the statistic T(x) = Σ xi.

As a concrete application, this gives a procedure for distinguishing a fair coin from a biased coin.

Uniform distribution

If X1, ...., Xn are independent and uniformly distributed on the interval [0,θ], then T(X) = max(X1, ..., Xn) is sufficient for θ — the sample maximum is a sufficient statistic for the population maximum.

To see this, consider the joint probability density function of X  (X1,...,Xn). Because the observations are independent, the pdf can be written as a product of individual densities

where 1{...} is the indicator function. Thus the density takes form required by the Fisher–Neyman factorization theorem, where h(x) = 1{min{xi}≥0}, and the rest of the expression is a function of only θ and T(x) = max{xi}.

In fact, the minimum-variance unbiased estimator (MVUE) for θ is

This is the sample maximum, scaled to correct for the bias, and is MVUE by the Lehmann–Scheffé theorem. Unscaled sample maximum T(X) is the maximum likelihood estimator for θ.

Uniform distribution (with two parameters)

If are independent and uniformly distributed on the interval (where and are unknown parameters), then is a two-dimensional sufficient statistic for .

To see this, consider the joint probability density function of . Because the observations are independent, the pdf can be written as a product of individual densities, i.e.

The joint density of the sample takes the form required by the Fisher–Neyman factorization theorem, by letting

Since does not depend on the parameter and depends only on through the function

the Fisher–Neyman factorization theorem implies is a sufficient statistic for .

Poisson distribution

If X1, ...., Xn are independent and have a Poisson distribution with parameter λ, then the sum T(X) = X1 + ... + Xn is a sufficient statistic for λ.

To see this, consider the joint probability distribution:

Because the observations are independent, this can be written as

which may be written as

which shows that the factorization criterion is satisfied, where h(x) is the reciprocal of the product of the factorials. Note the parameter λ interacts with the data only through its sum T(X).

Normal distribution

If are independent and normally distributed with expected value (a parameter) and known finite variance then

is a sufficient statistic for

To see this, consider the joint probability density function of . Because the observations are independent, the pdf can be written as a product of individual densities, i.e.

The joint density of the sample takes the form required by the Fisher–Neyman factorization theorem, by letting

Since does not depend on the parameter and depends only on through the function

the Fisher–Neyman factorization theorem implies is a sufficient statistic for .

If is unknown and since , the above likelihood can be rewritten as

The Fisher–Neyman factorization theorem still holds and implies that is a joint sufficient statistic for .

Exponential distribution

If are independent and exponentially distributed with expected value θ (an unknown real-valued positive parameter), then is a sufficient statistic for θ.

To see this, consider the joint probability density function of . Because the observations are independent, the pdf can be written as a product of individual densities, i.e.

The joint density of the sample takes the form required by the Fisher–Neyman factorization theorem, by letting

Since does not depend on the parameter and depends only on through the function

the Fisher–Neyman factorization theorem implies is a sufficient statistic for .

Gamma distribution

If are independent and distributed as a , where and are unknown parameters of a Gamma distribution, then is a two-dimensional sufficient statistic for .

To see this, consider the joint probability density function of . Because the observations are independent, the pdf can be written as a product of individual densities, i.e.

The joint density of the sample takes the form required by the Fisher–Neyman factorization theorem, by letting

Since does not depend on the parameter and depends only on through the function

the Fisher–Neyman factorization theorem implies is a sufficient statistic for

Rao–Blackwell theorem

Sufficiency finds a useful application in the Rao–Blackwell theorem, which states that if g(X) is any kind of estimator of θ, then typically the conditional expectation of g(X) given sufficient statistic T(X) is a better (in the sense of having lower variance) estimator of θ, and is never worse. Sometimes one can very easily construct a very crude estimator g(X), and then evaluate that conditional expected value to get an estimator that is in various senses optimal.

Exponential family

According to the Pitman–Koopman–Darmois theorem, among families of probability distributions whose domain does not vary with the parameter being estimated, only in exponential families is there a sufficient statistic whose dimension remains bounded as sample size increases. Intuitively, this states that nonexponential families of distributions on the real line require nonparametric statistics to fully capture the information in the data.

Less tersely, suppose are independent identically distributed real random variables whose distribution is known to be in some family of probability distributions, parametrized by , satisfying certain technical regularity conditions, then that family is an exponential family if and only if there is a -valued sufficient statistic whose number of scalar components does not increase as the sample size n increases. [14]

This theorem shows that the existence of a finite-dimensional, real-vector-valued sufficient statistics sharply restricts the possible forms of a family of distributions on the real line.

When the parameters or the random variables are no longer real-valued, the situation is more complex. [15]

Other types of sufficiency

Bayesian sufficiency

An alternative formulation of the condition that a statistic be sufficient, set in a Bayesian context, involves the posterior distributions obtained by using the full data-set and by using only a statistic. Thus the requirement is that, for almost every x,

More generally, without assuming a parametric model, we can say that the statistics T is predictive sufficient if

It turns out that this "Bayesian sufficiency" is a consequence of the formulation above, [16] however they are not directly equivalent in the infinite-dimensional case. [17] A range of theoretical results for sufficiency in a Bayesian context is available. [18]

Linear sufficiency

A concept called "linear sufficiency" can be formulated in a Bayesian context, [19] and more generally. [20] First define the best linear predictor of a vector Y based on X as . Then a linear statistic T(x) is linear sufficient [21] if

See also

Notes

  1. Fisher, R.A. (1922). "On the mathematical foundations of theoretical statistics". Philosophical Transactions of the Royal Society A. 222 (594–604): 309–368. Bibcode:1922RSPTA.222..309F. doi: 10.1098/rsta.1922.0009 . hdl: 2440/15172 . JFM   48.1280.02. JSTOR   91208.
  2. Dodge, Y. (2003) — entry for linear sufficiency
  3. Stigler, Stephen (December 1973). "Studies in the History of Probability and Statistics. XXXII: Laplace, Fisher and the Discovery of the Concept of Sufficiency". Biometrika. 60 (3): 439–445. doi:10.1093/biomet/60.3.439. JSTOR   2334992. MR   0326872.
  4. Casella, George; Berger, Roger L. (2002). Statistical Inference, 2nd ed. Duxbury Press.
  5. Cover, Thomas M. (2006). Elements of Information Theory. Joy A. Thomas (2nd ed.). Hoboken, N.J.: Wiley-Interscience. p. 36. ISBN   0-471-24195-4. OCLC   59879802.
  6. Halmos, P. R.; Savage, L. J. (1949). "Application of the Radon-Nikodym Theorem to the Theory of Sufficient Statistics". The Annals of Mathematical Statistics. 20 (2): 225–241. doi: 10.1214/aoms/1177730032 . ISSN   0003-4851.
  7. "Factorization theorem - Encyclopedia of Mathematics". encyclopediaofmath.org. Retrieved 2022-09-07.
  8. Taraldsen, G. (2022). "The Factorization Theorem for Sufficiency". Preprint. doi:10.13140/RG.2.2.15068.87687.
  9. Hogg, Robert V.; Craig, Allen T. (1995). Introduction to Mathematical Statistics. Prentice Hall. ISBN   978-0-02-355722-4.
  10. "The Fisher–Neyman Factorization Theorem".. Webpage at Connexions (cnx.org)
  11. Dodge (2003) — entry for minimal sufficient statistics
  12. Lehmann and Casella (1998), Theory of Point Estimation, 2nd Edition, Springer, p 37
  13. Lehmann and Casella (1998), Theory of Point Estimation, 2nd Edition, Springer, page 42
  14. Tikochinsky, Y.; Tishby, N. Z.; Levine, R. D. (1984-11-01). "Alternative approach to maximum-entropy inference". Physical Review A. 30 (5): 2638–2644. Bibcode:1984PhRvA..30.2638T. doi:10.1103/physreva.30.2638. ISSN   0556-2791.
  15. Andersen, Erling Bernhard (September 1970). "Sufficiency and Exponential Families for Discrete Sample Spaces". Journal of the American Statistical Association. 65 (331): 1248–1255. doi:10.1080/01621459.1970.10481160. ISSN   0162-1459.
  16. Bernardo, J.M.; Smith, A.F.M. (1994). "Section 5.1.4". Bayesian Theory. Wiley. ISBN   0-471-92416-4.
  17. Blackwell, D.; Ramamoorthi, R. V. (1982). "A Bayes but not classically sufficient statistic". Annals of Statistics . 10 (3): 1025–1026. doi: 10.1214/aos/1176345895 . MR   0663456. Zbl   0485.62004.
  18. Nogales, A.G.; Oyola, J.A.; Perez, P. (2000). "On conditional independence and the relationship between sufficiency and invariance under the Bayesian point of view". Statistics & Probability Letters. 46 (1): 75–84. doi:10.1016/S0167-7152(99)00089-9. MR   1731351. Zbl   0964.62003.
  19. Goldstein, M.; O'Hagan, A. (1996). "Bayes Linear Sufficiency and Systems of Expert Posterior Assessments". Journal of the Royal Statistical Society . Series B. 58 (2): 301–316. JSTOR   2345978.
  20. Godambe, V. P. (1966). "A New Approach to Sampling from Finite Populations. II Distribution-Free Sufficiency". Journal of the Royal Statistical Society . Series B. 28 (2): 320–328. JSTOR   2984375.
  21. Witting, T. (1987). "The linear Markov property in credibility theory". ASTIN Bulletin. 17 (1): 71–84. doi: 10.2143/ast.17.1.2014984 . hdl: 20.500.11850/422507 .

Related Research Articles

The likelihood function is the joint probability mass of observed data viewed as a function of the parameters of a statistical model. Intuitively, the likelihood function is the probability of observing data assuming is the actual parameter.

In mechanics and geometry, the 3D rotation group, often denoted SO(3), is the group of all rotations about the origin of three-dimensional Euclidean space under the operation of composition.

In mathematical analysis, Hölder's inequality, named after Otto Hölder, is a fundamental inequality between integrals and an indispensable tool for the study of Lp spaces.

<span class="mw-page-title-main">Gamma distribution</span> Probability distribution

In probability theory and statistics, the gamma distribution is a versatile two-parameter family of continuous probability distributions. The exponential distribution, Erlang distribution, and chi-squared distribution are special cases of the gamma distribution. There are two equivalent parameterizations in common use:

  1. With a shape parameter k and a scale parameter θ
  2. With a shape parameter and an inverse scale parameter , called a rate parameter.

In probability and statistics, an exponential family is a parametric set of probability distributions of a certain form, specified below. This special form is chosen for mathematical convenience, including the enabling of the user to calculate expectations, covariances using differentiation based on some useful algebraic properties, as well as for generality, as exponential families are in a sense very natural sets of distributions to consider. The term exponential class is sometimes used in place of "exponential family", or the older term Koopman–Darmois family. Sometimes loosely referred to as "the" exponential family, this class of distributions is distinct because they all possess a variety of desirable properties, most importantly the existence of a sufficient statistic.

In information geometry, the Fisher information metric is a particular Riemannian metric which can be defined on a smooth statistical manifold, i.e., a smooth manifold whose points are probability measures defined on a common probability space. It can be used to calculate the informational difference between measurements.

In statistics, the Neyman–Pearson lemma describes the existence and uniqueness of the likelihood ratio as a uniformly most powerful test in certain contexts. It was introduced by Jerzy Neyman and Egon Pearson in a paper in 1933. The Neyman–Pearson lemma is part of the Neyman–Pearson theory of statistical testing, which introduced concepts like errors of the second kind, power function, and inductive behavior. The previous Fisherian theory of significance testing postulated only one hypothesis. By introducing a competing hypothesis, the Neyman–Pearsonian flavor of statistical testing allows investigating the two types of errors. The trivial cases where one always rejects or accepts the null hypothesis are of little interest but it does prove that one must not relinquish control over one type of error while calibrating the other. Neyman and Pearson accordingly proceeded to restrict their attention to the class of all level tests while subsequently minimizing type II error, traditionally denoted by . Their seminal paper of 1933, including the Neyman–Pearson lemma, comes at the end of this endeavor, not only showing the existence of tests with the most power that retain a prespecified level of type I error, but also providing a way to construct such tests. The Karlin-Rubin theorem extends the Neyman–Pearson lemma to settings involving composite hypotheses with monotone likelihood ratios.

In mathematics, the Riesz–Thorin theorem, often referred to as the Riesz–Thorin interpolation theorem or the Riesz–Thorin convexity theorem, is a result about interpolation of operators. It is named after Marcel Riesz and his student G. Olof Thorin.

In calculus, the Leibniz integral rule for differentiation under the integral sign states that for an integral of the form

<span class="mw-page-title-main">Simple linear regression</span> Linear regression model with a single explanatory variable

In statistics, simple linear regression (SLR) is a linear regression model with a single explanatory variable. That is, it concerns two-dimensional sample points with one independent variable and one dependent variable and finds a linear function that, as accurately as possible, predicts the dependent variable values as a function of the independent variable. The adjective simple refers to the fact that the outcome variable is related to a single predictor.

In mathematics, the Bessel polynomials are an orthogonal sequence of polynomials. There are a number of different but closely related definitions. The definition favored by mathematicians is given by the series

A ratio distribution is a probability distribution constructed as the distribution of the ratio of random variables having two other known distributions. Given two random variables X and Y, the distribution of the random variable Z that is formed as the ratio Z = X/Y is a ratio distribution.

In statistical hypothesis testing, a uniformly most powerful (UMP) test is a hypothesis test which has the greatest power among all possible tests of a given size α. For example, according to the Neyman–Pearson lemma, the likelihood-ratio test is UMP for testing simple (point) hypotheses.

In mathematics, the spectral theory of ordinary differential equations is the part of spectral theory concerned with the determination of the spectrum and eigenfunction expansion associated with a linear ordinary differential equation. In his dissertation, Hermann Weyl generalized the classical Sturm–Liouville theory on a finite closed interval to second order differential operators with singularities at the endpoints of the interval, possibly semi-infinite or infinite. Unlike the classical case, the spectrum may no longer consist of just a countable set of eigenvalues, but may also contain a continuous part. In this case the eigenfunction expansion involves an integral over the continuous part with respect to a spectral measure, given by the Titchmarsh–Kodaira formula. The theory was put in its final simplified form for singular differential equations of even degree by Kodaira and others, using von Neumann's spectral theorem. It has had important applications in quantum mechanics, operator theory and harmonic analysis on semisimple Lie groups.

<span class="mw-page-title-main">Anatoly Karatsuba</span> Russian mathematician (1937–2008)

Anatoly Alexeyevich Karatsuba was a Russian mathematician working in the field of analytic number theory, p-adic numbers and Dirichlet series.

<span class="mw-page-title-main">Poisson distribution</span> Discrete probability distribution

In probability theory and statistics, the Poisson distribution is a discrete probability distribution that expresses the probability of a given number of events occurring in a fixed interval of time if these events occur with a known constant mean rate and independently of the time since the last event. It can also be used for the number of events in other types of intervals than time, and in dimension greater than 1.

A product distribution is a probability distribution constructed as the distribution of the product of random variables having two other known distributions. Given two statistically independent random variables X and Y, the distribution of the random variable Z that is formed as the product is a product distribution.

In mathematics, van der Corput's method generates estimates for exponential sums. The method applies two processes, the van der Corput processes A and B which relate the sums into simpler sums which are easier to estimate.

Exponential Tilting (ET), Exponential Twisting, or Exponential Change of Measure (ECM) is a distribution shifting technique used in many parts of mathematics. The different exponential tiltings of a random variable is known as the natural exponential family of .

References