{
  "id": "order-statistics",
  "title": "Order Statistics",
  "lang": "en",
  "source_lang": "en",
  "summary": "Order statistics are sample values arranged by magnitude, underpinning quantile estimation, distribution-free inference, extreme-value analysis, and reliability theory.",
  "category": "mathematics",
  "subcategory": "probability-statistics",
  "type": "concept",
  "aliases": [
    "Order statistic",
    "Ordered sample values"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/order-statistics/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/order-statistics/",
      "json": "https://aiwiki.page/zh/order-statistics.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "In statistics, order statistics are the values obtained by arranging a sample from smallest to largest. For observations \\(X_1,\\ldots,X_n\\), they are conventionally written\n\n\\[X_{(1)}\\le X_{(2)}\\le\\cdots\\le X_{(n)}.\\]\n\nThus \\(X_{(1)}\\) is the sample minimum, \\(X_{(n)}\\) is the sample maximum, and central order statistics determine the sample median and other sample quantiles. Before the sample is observed, each order statistic is a random variable with its own distribution.\n\nOrdering converts information about individual observations into information about relative position, extremes, and sample spread. Order statistics support exact confidence intervals that require no specified distributional shape, robust estimators that reduce the influence of extreme observations, and models of component failures and rare events. Their theory connects finite-sample combinatorics with probability distributions, asymptotic analysis, and computational selection algorithms."
    },
    {
      "heading": "Definition, assumptions, and the information retained by ordering",
      "text": "An order statistic is indexed by its position in the sorted sample rather than by the observation’s original label. If the observed sample is \\(8,3,11,3,6\\), its ordered values are\n\n\\[x_{(1)}=3,\\quad x_{(2)}=3,\\quad x_{(3)}=6,\\quad x_{(4)}=8,\\quad x_{(5)}=11.\\]\n\nRepeated values occupy separate positions: sorting does not remove duplicates. An observation’s rank records where it falls in the ordering; an order statistic records the value occupying a specified position. This distinction matters in rank-based methods, where the analysis may use only positions rather than the original numerical measurements.\n\nMost classical formulas assume that \\(X_1,\\ldots,X_n\\) are independent and identically distributed, abbreviated i.i.d., with common cumulative distribution function \\(F\\). Independence permits the use of binomial and multinomial counting arguments, while identical distributions make every labeling of distinct sample values equally likely. If \\(F\\) is continuous, ties occur with probability zero. If the distribution is discrete, ties generally have positive probability, but the ordered sample remains well defined.\n\nSorting discards the association between values and their original labels. For an i.i.d. sample, the ordered vector nevertheless retains all information needed to calculate any symmetric sample statistic: a statistic unchanged by permuting the observations. The sample mean, sample variance, range, and empirical distribution are all examples. In a dominated i.i.d. model, the likelihood is symmetric in the observations, so the full ordered sample is a sufficient statistic for the model parameter. The ordered sample need not be a minimal sufficient statistic; many models admit a much smaller sufficient summary.\n\nThe distinction between labeled and ordered data becomes essential outside the i.i.d. setting. Sorting a time series removes temporal order, and sorting observations from different populations without retaining their labels can remove group information. Similarly, ordering responses while discarding their associated predictors generally destroys information needed for regression. Classical order-statistic identities therefore concern a particular sampling structure, not an unrestricted claim that sorting preserves all statistical information.\n\nOrder statistics behave simply under monotone transformations. For a nondecreasing function \\(g\\), the ordered transformed values satisfy\n\n\\[(g(X))_{(k)}=g(X_{(k)}).\\]\n\nFor a decreasing transformation, the order is reversed:\n\n\\[(g(X))_{(k)}=g(X_{(n-k+1)}).\\]\n\nThese identities allow results derived on a convenient scale, such as the uniform scale, to be transferred to many other distributions."
    },
    {
      "heading": "Exact distributions of individual order statistics",
      "text": "The distribution of \\(X_{(k)}\\) follows from counting observations below a threshold. Define\n\n\\[N_x=\\sum_{i=1}^n \\mathbf{1}_{\\{X_i\\le x\\}},\\]\n\nwhere \\(\\mathbf{1}\\) is an indicator function. The event \\(X_{(k)}\\le x\\) occurs exactly when \\(N_x\\ge k\\). Under i.i.d. sampling, \\(N_x\\) has a binomial distribution with parameters \\(n\\) and \\(F(x)\\). Consequently,\n\n\\[\\Pr(X_{(k)}\\le x) = \\sum_{j=k}^{n}\\binom{n}{j} F(x)^j[1-F(x)]^{n-j}.\\]\n\nThis identity is valid for continuous, discrete, and mixed distributions. It is the basic finite-sample formula from which many other results follow.\n\nIf \\(F\\) is absolutely continuous with probability density function \\(f\\), differentiation gives\n\n\\[f_{X_{(k)}}(x) = \\frac{n!}{(k-1)!(n-k)!} F(x)^{k-1}[1-F(x)]^{n-k}f(x).\\]\n\nThe factors have a direct counting interpretation. Of the other \\(n-1\\) observations, \\(k-1\\) must lie below a value near \\(x\\), while \\(n-k\\) must lie above it. The factorial coefficient counts the possible assignments of observations to those roles. Absolute continuity is important here: a continuous distribution function need not possess a density suitable for this formula.\n\nFor the two extremes, the distribution functions simplify to\n\n\\[\\Pr(X_{(1)}\\le x)=1-[1-F(x)]^n, \\qquad \\Pr(X_{(n)}\\le x)=F(x)^n.\\]\n\nFor example, if each observation exceeds a threshold \\(t\\) with probability \\(0.01\\), then the probability that the sample maximum exceeds \\(t\\) is \\(1-0.99^n\\). At \\(n=100\\), this probability is approximately \\(0.634\\). A value that is rare for one observation can therefore be common as the maximum of a moderately large sample.\n\nFor a distribution with atoms, the probability mass at a particular value must account for observations equal to that value. Writing \\(F(x^-)=\\Pr(X<x)\\),\n\n\\[\\Pr(X_{(k)}=x) = \\Pr\\{\\operatorname{Bin}(n,F(x))\\ge k\\} - \\Pr\\{\\operatorname{Bin}(n,F(x^-))\\ge k\\}.\\]\n\nThis is not generally obtained by substituting a probability mass function into the continuous-density formula.\n\nMarginal distributions also yield moments when the required integrals exist. For example,\n\n\\[\\mathbb{E}[X_{(k)}] = \\int_{-\\infty}^{\\infty}x f_{X_{(k)}}(x)\\,dx\\]\n\nin the absolutely continuous case. Existence must be checked rather than assumed: heavy-tailed parent distributions can produce order statistics whose expectations or variances are infinite, especially near the extremes."
    },
    {
      "heading": "Joint distributions and dependence within the ordered sample",
      "text": "Although the original observations may be independent, their order statistics are generally dependent. Knowing the minimum restricts every other ordered value to lie above it; knowing a central order statistic constrains both sides of the sample. For an absolutely continuous i.i.d. sample, the full joint distribution has density\n\n\\[f_{X_{(1)},\\ldots,X_{(n)}}(x_1,\\ldots,x_n) = n!\\prod_{i=1}^{n}f(x_i), \\qquad x_1<\\cdots<x_n,\\]\n\nand zero elsewhere. The factor \\(n!\\) accounts for the possible labelings of a set of distinct observations. The ordered region occupies only one of the permutation regions of the original sample space.\n\nFor two positions \\(1\\le r<s\\le n\\), the joint density is\n\n\\[\\begin{aligned} f_{X_{(r)},X_{(s)}}(x,y) ={}& \\frac{n!}{(r-1)!(s-r-1)!(n-s)!}\\\\ &\\times F(x)^{r-1} [F(y)-F(x)]^{s-r-1}\\\\ &\\times[1-F(y)]^{n-s}f(x)f(y), \\qquad x<y. \\end{aligned}\\]\n\nHere the sample is divided into observations below \\(x\\), between \\(x\\) and \\(y\\), and above \\(y\\), together with one observation near each boundary. More generally, joint densities of selected order statistics arise by dividing the real line into intervals and applying a multinomial counting argument.\n\nThe sample range \\(R=X_{(n)}-X_{(1)}\\) illustrates why joint distributions are needed. Its distribution cannot be determined from the marginal distributions of the minimum and maximum alone. For \\(n\\ge2\\), its density is\n\n\\[f_R(r) = n(n-1)\\int_{-\\infty}^{\\infty} f(x)f(x+r) [F(x+r)-F(x)]^{n-2}\\,dx, \\qquad r>0.\\]\n\nThe integrand describes samples with minimum near \\(x\\), maximum near \\(x+r\\), and all remaining observations inside that interval. Range-based procedures in statistical quality control rely on such joint behavior.\n\nConditional distributions have a useful truncation interpretation. Conditional on \\(X_{(k)}=x\\), in the regular-conditional sense for continuous distributions, the lower \\(k-1\\) ordered observations have the distribution of an ordered sample from the parent distribution truncated below \\(x\\). The upper \\(n-k\\) observations similarly arise from truncation above \\(x\\). Conditional on the boundary value, the lower and upper ordered groups are independent. This property supports recursive calculations and simulation.\n\nThe i.i.d. assumption is substantive. For independent observations with different distribution functions \\(F_i\\),\n\n\\[\\Pr(X_{(n)}\\le x)=\\prod_{i=1}^n F_i(x),\\]\n\nbut the count below \\(x\\) is generally Poisson-binomial rather than binomial. Under dependence, even these product formulas fail. Applications involving clustered observations, spatial measurements, or serial extremes must therefore model more than the common marginal distribution."
    },
    {
      "heading": "Uniform transformations, beta laws, and spacings",
      "text": "A central simplification is the probability integral transform. If \\(F\\) is continuous and \\(X\\) has distribution function \\(F\\), then \\(U=F(X)\\) is uniformly distributed on \\((0,1)\\). Applied independently to an i.i.d. sample, this gives\n\n\\[U_{(k)}=F(X_{(k)}).\\]\n\nConversely, if\n\n\\[Q(u)=\\inf\\{x:F(x)\\ge u\\}\\]\n\nis the quantile function, then applying \\(Q\\) to independent uniform variables generates a sample from \\(F\\), including distributions with atoms. The uniform model thus supplies a common foundation for distributional calculations and simulation.\n\nFor a sample from the uniform distribution on \\((0,1)\\),\n\n\\[U_{(k)}\\sim \\operatorname{Beta}(k,n+1-k),\\]\n\nwhere \\(\\operatorname{Beta}\\) denotes the beta distribution. Therefore,\n\n\\[\\mathbb{E}[U_{(k)}]=\\frac{k}{n+1}, \\qquad \\operatorname{Var}(U_{(k)}) = \\frac{k(n+1-k)}{(n+1)^2(n+2)}.\\]\n\nFor \\(r\\le s\\),\n\n\\[\\operatorname{Cov}(U_{(r)},U_{(s)}) = \\frac{r(n+1-s)}{(n+1)^2(n+2)}.\\]\n\nThe positive covariance quantifies the tendency of ordered values to move together. The expected positions \\(k/(n+1)\\) also motivate one widely used set of plotting positions in probability plots.\n\nSet \\(U_{(0)}=0\\) and \\(U_{(n+1)}=1\\), and define the spacings\n\n\\[D_i=U_{(i)}-U_{(i-1)},\\qquad i=1,\\ldots,n+1.\\]\n\nThese nonnegative gaps sum to one and jointly follow a Dirichlet distribution with all parameters equal to one. Their density is constant on the simplex of possible spacing vectors. Every spacing has expectation \\(1/(n+1)\\), but the spacings are not independent: an unusually large gap leaves less total length for the others.\n\nSums of consecutive spacings have beta distributions. In particular,\n\n\\[U_{(s)}-U_{(r)} \\sim \\operatorname{Beta}(s-r,n+1-s+r).\\]\n\nFor \\(n\\ge2\\), the uniform sample range therefore has distribution \\(\\operatorname{Beta}(n-1,2)\\). Such identities turn questions about random intervals into standard probability calculations.\n\nThere is also an exponential representation. If \\(E_1,\\ldots,E_{n+1}\\) are independent exponential random variables with mean one and \\(S=\\sum_i E_i\\), then\n\n\\[(D_1,\\ldots,D_{n+1}) \\overset{d}{=} \\left(\\frac{E_1}{S},\\ldots,\\frac{E_{n+1}}{S}\\right).\\]\n\nTaking cumulative sums produces uniform order statistics without sorting. The representation is useful for exact simulation, spacing-based goodness-of-fit procedures, and asymptotic arguments. It concerns uniform spacings; gaps on the original measurement scale are usually distorted by the nonlinear quantile transformation."
    },
    {
      "heading": "Sample quantiles and their large-sample behavior",
      "text": "The empirical distribution function is\n\n\\[F_n(x)=\\frac1n\\sum_{i=1}^n\\mathbf{1}_{\\{X_i\\le x\\}}.\\]\n\nIts generalized inverse defines a sample quantile. Under the inverse empirical-distribution convention,\n\n\\[\\widehat q_p=X_{(\\lceil np\\rceil)},\\qquad 0<p<1.\\]\n\nOther conventions interpolate between adjacent order statistics. For an even sample size, the usual sample median is the average of the two central observations, whereas the inverse empirical distribution selects the lower central observation at \\(p=1/2\\). Quantile definitions therefore need to be specified when exact numerical agreement matters.\n\nInterpolation conventions differ in their assigned plotting positions and endpoint behavior. They often agree closely in large samples but can differ noticeably for small samples or extreme percentiles. No single interpolation rule is uniformly optimal for every parent distribution and purpose. The exact beta result applies directly to an individual transformed order statistic, not automatically to an interpolation of adjacent values. Likewise,\n\n\\[\\mathbb{E}[X_{(k)}] \\ne Q\\!\\left(\\frac{k}{n+1}\\right)\\]\n\nin general, because applying a nonlinear quantile function does not commute with taking expectations.\n\nFor a fixed interior probability \\(p\\), let \\(q_p=Q(p)\\), and suppose \\(F\\) has a positive, continuous density at \\(q_p\\). Suitable sample quantiles are consistent estimators of \\(q_p\\) and satisfy\n\n\\[\\sqrt n\\,(\\widehat q_p-q_p) \\xrightarrow{d} N\\!\\left(0,\\frac{p(1-p)}{f(q_p)^2}\\right).\\]\n\nThe numerator reflects binomial uncertainty in the fraction below the population quantile. The denominator converts uncertainty on the probability scale into uncertainty on the measurement scale: where the density is low, a small probability error corresponds to a larger displacement in value.\n\nA first-order representation explains this limit:\n\n\\[\\widehat q_p-q_p = \\frac{p-F_n(q_p)}{f(q_p)} +o_p(n^{-1/2}),\\]\n\nunder standard regularity conditions. The central limit theorem applies to the indicator average \\(F_n(q_p)\\). For two fixed probabilities \\(p\\) and \\(q\\), the limiting covariance is\n\n\\[\\lim_{n\\to\\infty} n\\,\\operatorname{Cov}(\\widehat q_p,\\widehat q_q) = \\frac{\\min(p,q)-pq} {f(q_p)f(q_q)},\\]\n\nwhen the relevant moment convergence conditions hold; the corresponding joint normal limit uses this covariance matrix.\n\nFor a normal population with standard deviation \\(\\sigma\\), the asymptotic variance of the sample median is \\(\\pi\\sigma^2/(2n)\\), compared with \\(\\sigma^2/n\\) for the sample mean. This efficiency comparison is model-specific. At atoms, flat portions of the distribution function, or endpoints, regular normal quantile asymptotics may fail or require modification."
    },
    {
      "heading": "Distribution-free confidence and tolerance intervals",
      "text": "Order statistics provide exact finite-sample inference for population quantiles without specifying a parametric family. Suppose \\(F\\) is continuous and \\(q_p\\) satisfies \\(F(q_p)=p\\). The number of sample observations below \\(q_p\\) has distribution \\(\\operatorname{Bin}(n,p)\\). For \\(1\\le r<s\\le n\\),\n\n\\[\\Pr\\{X_{(r)}\\le q_p\\le X_{(s)}\\} = \\sum_{j=r}^{s-1}\\binom{n}{j}p^j(1-p)^{n-j}.\\]\n\nThe coverage depends only on \\(n,p,r,s\\), not on the shape, location, or scale of \\(F\\). This is a principal example of nonparametric statistics.\n\nFor a concrete median interval, take \\(n=20\\) and use \\([X_{(6)},X_{(15)}]\\). Its exact coverage is\n\n\\[1-2\\sum_{j=0}^{5}\\binom{20}{j}2^{-20} \\approx0.9586.\\]\n\nThe interval therefore has approximately \\(95.86\\%\\) coverage under continuous i.i.d. sampling. Since the binomial distribution is discrete, a desired level such as \\(95\\%\\) is not usually attainable exactly with fixed nonrandomized endpoints. A conventional procedure chooses an interval whose coverage is at least the nominal level.\n\nA confidence interval for a quantile must be distinguished from a tolerance interval. A quantile interval targets a particular population value. A tolerance interval instead aims to contain at least a specified population proportion, with a stated confidence level. For continuous \\(F\\), the population content between two sample order statistics is\n\n\\[C=F(X_{(s)})-F(X_{(r)}),\\]\n\nand\n\n\\[C\\sim\\operatorname{Beta}(s-r,n+1-s+r).\\]\n\nThus the probability that the interval contains at least a proportion \\(p\\) is a beta upper-tail probability.\n\nFor the sample minimum and maximum,\n\n\\[\\Pr\\{F(X_{(n)})-F(X_{(1)})\\ge p\\} = 1-np^{n-1}+(n-1)p^n.\\]\n\nThis identity quantifies the sample size required for a distribution-free two-sided tolerance statement. A one-sided statement is simpler:\n\n\\[\\Pr\\{F(X_{(n)})\\ge p\\}=1-p^n.\\]\n\nFor example, a sample maximum from \\(59\\) independent observations exceeds the population 95th percentile with probability \\(1-0.95^{59}\\), slightly greater than \\(0.95\\).\n\nOrder statistics also underlie distribution-free prediction. For one future observation \\(X_{n+1}\\) from the same continuous population,\n\n\\[\\Pr\\{X_{(r)}<X_{n+1}<X_{(s)}\\} =\\frac{s-r}{n+1}.\\]\n\nThis follows because the future observation’s rank among all \\(n+1\\) values is uniform. Such a prediction interval concerns a future random value, not an unknown fixed quantile. Ties, dependence, and changes in the generating population require separate treatment because the simple rank argument no longer applies unchanged."
    },
    {
      "heading": "Extremes and tail behavior",
      "text": "The minimum and maximum occupy a different asymptotic regime from central order statistics. A sample median estimates a fixed interior feature of a distribution, whereas the maximum typically moves farther into the upper tail as the sample grows. Its scale and limiting distribution are controlled by tail behavior. Extreme-value theory studies normalizations\n\n\\[\\frac{X_{(n)}-b_n}{a_n}, \\qquad a_n>0,\\]\n\nthat can converge to a nondegenerate distribution. Under its domain-of-attraction conditions, the possible limits belong to the generalized extreme-value family.\n\nA standard form of that family is\n\n\\[G_\\xi(z) = \\exp\\!\\left[-(1+\\xi z)^{-1/\\xi}\\right], \\qquad 1+\\xi z>0.\\]\n\nAt \\(\\xi=0\\), the continuous limiting interpretation gives\n\n\\[G_0(z)=\\exp(-e^{-z}).\\]\n\nPositive shape parameter \\(\\xi\\) corresponds to a heavy upper tail, negative \\(\\xi\\) to a finite upper endpoint, and zero to the Gumbel class. The classification concerns suitably normalized maxima, not the distribution of ordinary individual observations.\n\nFor independent exponential observations with rate \\(\\lambda\\),\n\n\\[\\Pr\\{\\lambda X_{(n)}-\\log n\\le z\\} = \\left(1-\\frac{e^{-z}}{n}\\right)^n \\longrightarrow \\exp(-e^{-z}),\\]\n\nfor fixed \\(z\\), once the threshold lies in the support. The maximum therefore grows at leading order like \\((\\log n)/\\lambda\\). For a Pareto distribution with\n\n\\[F(x)=1-x^{-\\alpha},\\qquad x\\ge1,\\]\n\nthe appropriate growth rate is \\(n^{1/\\alpha}\\), and\n\n\\[\\Pr\\{X_{(n)}/n^{1/\\alpha}\\le x\\} \\longrightarrow \\exp(-x^{-\\alpha}),\\qquad x>0.\\]\n\nThese examples show why there is no universal scaling for maxima.\n\nFor uniform observations on \\((0,1)\\), the maximum instead approaches a fixed endpoint:\n\n\\[n(1-X_{(n)})\\xrightarrow{d}\\operatorname{Exp}(1).\\]\n\nThe distance to the endpoint is of order \\(1/n\\), unlike the \\(1/\\sqrt n\\) fluctuations of a regular central quantile. The maximum of a standard normal sample has leading growth of order \\(\\sqrt{2\\log n}\\); a nondegenerate limiting law requires more precise centering and scaling.\n\nUpper order statistics provide information beyond the single largest value. Tail-index estimators can use log ratios between several large observations and a high-order threshold. An important distinction is between fixed extreme ranks and intermediate ranks: an intermediate upper rank may use a number \\(k_n\\) of exceedances with \\(k_n\\to\\infty\\) but \\(k_n/n\\to0\\). Such methods balance the variability of using few tail observations against the bias introduced by extending a tail approximation too far into the distribution. Serial dependence and clustering further affect how extremes accumulate."
    },
    {
      "heading": "Robust estimation and linear combinations of ordered values",
      "text": "Many estimators can be expressed as weighted sums of order statistics:\n\n\\[T_n=\\sum_{i=1}^n a_{i,n}X_{(i)}.\\]\n\nSuch statistics are called L-statistics. The sample mean assigns equal weights, a median concentrates weight at the center, and a trimmed mean assigns zero weight to observations in selected tails. The framework makes explicit which parts of the sample influence an estimate and provides a route to exact and asymptotic variance calculations.\n\nFor an integer \\(k\\) with \\(0\\le k<n/2\\), the symmetrically trimmed mean is\n\n\\[\\overline X_{\\mathrm{trim}} = \\frac{1}{n-2k}\\sum_{i=k+1}^{n-k}X_{(i)}.\\]\n\nIt removes the \\(k\\) smallest and \\(k\\) largest observations before averaging. A Winsorized mean instead replaces observations in the lower tail by \\(X_{(k+1)}\\) and those in the upper tail by \\(X_{(n-k)}\\). Trimming changes which observations contribute; Winsorization retains the nominal sample size but limits the contribution of tail values.\n\nThese constructions are central to robust statistics. A single arbitrarily large observation can make the sample mean arbitrarily large. A sample median cannot be driven without bound by altering only a small fraction of observations; its asymptotic replacement breakdown point is \\(1/2\\). A trimmed mean has resistance related to its trimming fraction. Breakdown resistance, however, is not equivalent to low mean squared error under every model. Strong protection against contamination can reduce efficiency when a light-tailed parametric model is exactly correct.\n\nAt a distribution with positive density at its \\(p\\)-quantile, the quantile influence function is\n\n\\[\\operatorname{IF}(x;q_p,F) = \\frac{p-\\mathbf{1}_{\\{x\\le q_p\\}}}{f(q_p)}.\\]\n\nIt is bounded as a function of the contamination value \\(x\\), in contrast to the unbounded influence function \\(x-\\mu\\) of the mean. Its magnitude nevertheless increases when \\(f(q_p)\\) is small. Robustness against extreme observations therefore does not eliminate uncertainty caused by sparse probability mass near the target.\n\nOrder-based measures of spread include the interquartile range\n\n\\[\\operatorname{IQR}=\\widehat q_{0.75}-\\widehat q_{0.25}\\]\n\nand the median absolute deviation\n\n\\[\\operatorname{MAD} = \\operatorname{median}_i \\left|X_i-\\operatorname{median}_j X_j\\right|.\\]\n\nFor a normal population, multiplying the unscaled MAD by approximately \\(1.4826\\) makes it consistent for the population standard deviation. The multiplier is model-dependent. Trimmed means also need a clear estimand: under asymmetry, the population trimmed mean generally differs from the ordinary population mean, so robustness changes both the weighting of observations and potentially the target of estimation."
    },
    {
      "heading": "Reliability, failure times, and censored observations",
      "text": "In reliability theory, order statistics describe the sequence of component failures. If \\(T_1,\\ldots,T_n\\) are component lifetimes, then \\(T_{(k)}\\) is the time of the \\(k\\)th failure. A series system that fails when its first component fails has lifetime \\(T_{(1)}\\). A parallel system that continues operating until every component has failed has lifetime \\(T_{(n)}\\), under the idealized assumption that any surviving component can sustain system operation.\n\nFor independent, identically distributed component lifetimes with survival function \\(S(t)=1-F(t)\\), the system surviving until the \\(k\\)th failure has survival probability\n\n\\[\\Pr(T_{(k)}>t) = \\sum_{j=0}^{k-1}\\binom{n}{j} F(t)^jS(t)^{n-j}.\\]\n\nAt time \\(t\\), fewer than \\(k\\) components must have failed. Equivalently, at least \\(n-k+1\\) components must still operate. In a system requiring at least \\(m\\) of its \\(n\\) components to function, system failure occurs at \\(T_{(n-m+1)}\\), provided component behavior is unaffected by the failures of other components.\n\nExponential lifetimes yield a particularly simple sequence of failure gaps. If the component lifetimes are independent exponential variables with rate \\(\\lambda\\), define \\(T_{(0)}=0\\) and\n\n\\[\\Delta_k=T_{(k)}-T_{(k-1)}.\\]\n\nThen the \\(\\Delta_k\\) are independent, with\n\n\\[\\Delta_k\\sim\\operatorname{Exp}((n-k+1)\\lambda).\\]\n\nInitially \\(n\\) components are exposed to failure; after the first failure, \\(n-1\\) remain. The memoryless property preserves the exponential model at each stage. In particular,\n\n\\[\\mathbb{E}[T_{(k)}] = \\frac1\\lambda\\sum_{j=n-k+1}^{n}\\frac1j.\\]\n\nOrder statistics also organize censored data, in which some lifetimes are only partially observed. Under Type II censoring, a test stops after the \\(r\\)th failure, leaving \\(n-r\\) components known only to have survived beyond \\(t_r\\). For continuous i.i.d. lifetimes, the likelihood of the first \\(r\\) ordered failure times is\n\n\\[L(\\theta) = \\frac{n!}{(n-r)!} \\left[\\prod_{i=1}^{r}f_\\theta(t_i)\\right] [1-F_\\theta(t_r)]^{n-r}, \\qquad t_1<\\cdots<t_r.\\]\n\nThe surviving components contribute survival probabilities rather than density factors.\n\nIn survival analysis, failures are commonly observed alongside individually varying censoring times. The Kaplan–Meier estimator, introduced in 1958, multiplies estimated conditional survival probabilities at ordered event times, using the number at risk immediately before each event. Its standard interpretation depends on appropriate independent-censoring assumptions. Dependence between component lifetimes, common-cause failures, and load redistribution likewise require extensions beyond elementary i.i.d. order-statistic formulas."
    },
    {
      "heading": "Computation, selection algorithms, and numerical evaluation",
      "text": "Computing every order statistic is equivalent to sorting. Standard comparison-based sorting algorithms use \\(O(n\\log n)\\) comparisons in the worst case or expected case, depending on the algorithm; the notation is big-O notation. If only one quantile or a few ranks are required, full sorting is unnecessary. A selection algorithm finds the value at a specified rank by partitioning the data, avoiding the work of completely ordering values that do not affect the answer.\n\nQuickselect, introduced by C. A. R. Hoare in 1961, repeatedly chooses a pivot and retains the partition containing the requested rank. With suitable randomized pivot selection, its expected running time is \\(O(n)\\), although basic versions can take \\(O(n^2)\\) time in unfavorable cases. The deterministic selection algorithm published by Manuel Blum, Robert Floyd, Vaughan Pratt, Ronald Rivest, and Robert Tarjan in 1973 guarantees worst-case \\(O(n)\\) time through a pivot-selection construction commonly called median of medians. Practical implementations often combine fast partitioning with safeguards against poor cases.\n\nFor a small fixed number of quantiles, repeated or coordinated selection can be preferable to sorting. If the largest \\(k\\) values are needed from a stream, a heap can maintain them in \\(O(n\\log k)\\) time and \\(O(k)\\) storage. Exact central quantiles over unrestricted streams generally require information that grows with the input size. Approximate streaming summaries instead trade storage for guarantees on rank error. A rank-error guarantee must be distinguished from an error measured in the data’s units: a small rank difference can correspond to a large value difference in a sparse tail.\n\nNumerical evaluation of order-statistic distributions requires attention to extreme probabilities. Directly summing binomial terms can underflow, overflow, or lose relative accuracy. For continuous \\(F\\), the marginal distribution can be written\n\n\\[\\Pr(X_{(k)}\\le x) = I_{F(x)}(k,n+1-k),\\]\n\nwhere \\(I\\) is the regularized incomplete beta function. Well-designed binomial-tail and incomplete-beta routines provide more reliable evaluation than naive factorial formulas. Logarithmic calculations also avoid overflow when evaluating densities involving large combinatorial coefficients.\n\nSimulation can use either ordinary sampling followed by sorting or the uniform-spacings representation. Normalized exponential variables generate uniform order statistics in linear time, after which the quantile function transforms them to the desired distribution. This approach is useful in Monte Carlo studies requiring complete ordered samples. In software, reproducibility also depends on documenting the quantile convention, the treatment of ties and missing values, the indexing convention, and any weights; each can change the reported result even when the underlying observations are identical."
    }
  ],
  "links": [
    "statistics",
    "median",
    "quantile",
    "random-variable",
    "cumulative-distribution-function",
    "statistical-independence",
    "sample-mean",
    "variance",
    "sufficient-statistic",
    "time-series",
    "indicator-function",
    "binomial-distribution",
    "probability-density-function",
    "expected-value",
    "joint-probability-distribution",
    "uniform-distribution",
    "beta-distribution",
    "covariance",
    "dirichlet-distribution",
    "empirical-distribution-function",
    "consistent-estimator",
    "central-limit-theorem",
    "confidence-interval",
    "prediction-interval",
    "mean-squared-error",
    "big-o-notation",
    "monte-carlo-method"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}