﻿

Consider empirical distribution function $F_n$ based on asequence of i.i.d random variables, $X_1,ldots, X_n$.Suppose $F$ is a hypothetical distribution function ofeach $X_i$. To test whether the choice of $F$is correct or not, statisticians use the normalized difference,

: $v_n\left(x\right)=sqrt\left\{n\right\} \left[F_n\left(x\right)-F\left(x\right)\right] .$

This $v_n$, as a random process in $x$, is called empirical process. Various functional from $v_n$ are used as test statistics. The change of the time $v_n\left(x\right)=u_n\left(t\right)$, , $t=F\left(x\right)$ transforms to the so-called uniform empirical process $u_n$. The latter is an empirical processes based on independent random variables $U_i=F\left(X_i\right)$, which are uniformly distributed on $\left[0,1\right]$ if $X_i$s indeed have distribution function $F$.

This fact, discovered and first utilized by Kolmogorov(1933), Wald and Wolfowitz(1936), Smirnov(1937), especially after Doob(1949) and Anderson and Darling(1952) led to the standard rule to choose test statistics from $v_n$: these are statistics $psi\left(v_n,F\right)$ which possibly depends $F$, in such a way that there exists another statistic $varphi\left(u_n\right)$ from the uniform empirical process, such that $psi\left(v_n,F\right)=varphi\left(u_n\right)$. Examples are

: $sup_x|v_n\left(x\right)|=sup_t|u_n\left(t\right)|,quad sup_xfrac\left\{a\left(F\left(x\right)\right)\right\}=sup_tfrac \left\{a\left(t\right)\right\}$

and

: $int_\left\{-infty\right\}^\left\{infty\right\} v_n^2\left(x\right)d F\left(x\right)=int_\left\{0\right\}^\left\{1\right\} u_n^2\left(t\right),dt.$

For all such functionals, their "null" distribution (under the hypothetical $F$) does not depend on $F$, and can be calculated once and then used to test any $F$.

However, it is only rarely that one needs to test simple hypothesis, when a fixed $F$ as a hypothesis is given. Much more often, one needs to verify parametric hypothesis when hypothetical $F=F_\left\{ heta_n\right\}$, depends on some parameter $heta_n$, which the hypothesis does not specify and which has to be estimated from the sample $X_1,ldots,X_n$ itself.

Although the estimators $hat heta_n$, most commonly converge to true value of $heta$, it was discovered, Kac, Kiefer and Wolfowitz(1955) and Gikhman(1954), that the parametric, or estimated, empirical process

: $hat v_n\left(x\right)=sqrt\left\{n\right\} \left[F_n\left(x\right)-F_\left\{hat heta_n\right\}\left(x\right)\right]$

differs significantly from $v_n$ and the time transformed process $hat u_n\left(t\right)=hat v_n\left(x\right)$, $t=F_\left\{hat heta_n\right\}\left(x\right)$ has the distribution, and the limit distribution as $n oinfty$, dependent on parametric form of $F_\left\{ heta\right\}$ on $hat heta_n$ and, in general, within one parametric family, on the value of $heta$.

From mid-50's to the late-80's, much work was done to clarify the situation and understand the nature of the process $hat v_n$.

In 1981, and then 1987 and 1993, E. V. Khmaladze suggested to replace the parametric empirical process $hat v_n$ by its martingale part $w_n$ only.

: $hat v_n\left(x\right)-K_n\left(x\right)=w_n\left(x\right)$

where $K_n\left(x\right)$ is the compensator of $hat v_n\left(x\right)$. Then the following properties of $w_n$ were established:

* Although the form of $K_n$, and therefore, of $w_n$, depends on $F_\left\{hat heta_n\right\}\left(x\right)$, as a function of both $x$ and $heta_n$, the limit distribution of the time transformed process

: $omega_n\left(t\right)=w_n\left(x\right), t=F_\left\{hat heta_n\right\}\left(x\right)$

is that of standard Brownian motion on $\left[0,1\right]$, i.e., isagain standard and independent of the choice of$F_\left\{hat heta_n\right\}$.

* The relationship between $hat v_n$ and $w_n$ and between their limits, is one to one, so that the statistical inference based on $hat v_n$ or on $w_n$ are equivalent, and in $w_n$, nothing is lost compared to $hat v_n$.

* The construction of innovation martingale $w_n$ could be carried over to the case of vector-valued $X_1,ldots,X_n$, giving rise to the definition of the so-called scanning martingales in $mathbb R^d$.

For a long time the transformation was, although known, still not used. Later, the work of researchers like R. Koenker, W. Stute, J. Bai, H. Koul, A. Koening, ... and others made it popular in econometrics and other fields of statistics.

Wikimedia Foundation. 2010.

### Look at other dictionaries:

• Estate V. Khmaladze — Infobox Scientist name = Estate V. Khmaladze |200px image width = 200px caption = Estate V. Khmaladze birth date = Birth date and age|1944|10|20|mf=y birth place = Tbilisi, Republic of Georgia, USSR residence = New Zealand citizenship = Georgian …   Wikipedia

• List of mathematics articles (K) — NOTOC K K approximation of k hitting set K ary tree K core K edge connected graph K equivalence K factor error K finite K function K homology K means algorithm K medoids K minimum spanning tree K Poincaré algebra K Poincaré group K set (geometry) …   Wikipedia

• Poule B de la coupe du monde de rugby à XV 2011 — La poule B de la Coupe du monde de rugby à XV 2011, qui se dispute en Nouvelle Zélande du 9 septembre au 3 octobre 2011 comprend cinq équipes dont les deux premières se qualifient pour les quarts de finale de la compétition. Conformément au… …   Wikipédia en Français

• Coupe du monde de rugby à XV 2011 — Infobox compétition sportive Coupe du monde de rugby à XV 2011 Généralités Sport Rugby à XV …   Wikipédia en Français

• Équipe d'Angleterre de rugby à XV à la coupe du monde 2011 — L équipe d Angleterre de rugby à XV dispute la coupe du monde 2011, organisée par la Nouvelle Zélande, en étant dans la poule B pour la première phase; elle affronte l Argentine, la Géorgie, le Roumanie et l Écosse. Les Anglais s’imposent contre… …   Wikipédia en Français