On the estimation of the influence curve
read more
Citations
Applied regression analysis bibliography update 1994-97
Finding confidence limits on population growth rates: Bootstrap and analytic methods
A central limit theorem for M-estimators by the Von Mises method
Influence Measures for CART Classification Trees
Influence function and correspondence analysis
References
Robust statistics: the approach based on influence functions
Convergence of stochastic processes
The Influence Curve and Its Role in Robust Estimation
The Bootstrap and Edgeworth Expansion
Related Papers (5)
Asymptotic and Bootstrap Inference for AR( Infinite ) Processes with Conditional Heteroskedasticity
Frequently Asked Questions (11)
Q2. What are the three estimators used in the literature?
Three estimators have been considered in the literature: the sensitivity curve, the empirical influence curve, and the jackknife approximation [see Hampel et al. (1987), p. 92].
Q3. what is the c transformation for h'?
assuming that h' is integrable and m is continuously differentiable, one can check out that the expression for this influence curve given above is just a composition of continuously differentiable operators: observe that the c transformation F 1---+ J:oo h'(y )F(y )dy is linear and continuous and, hence, differentiable.
Q4. What is the meaning of the term "Estimators"?
It is well-known that, in many cases of practical interest, the estimators can be considered as restrictions of functionals defined on the space :F of distribution functions.
Q5. how does the influence function T'(Fj) function function be extended to the vector space?
Fn close enough (in the weak topology) to F ,1 for all 0 < t < - a.s. -n(ii) the influence function T'(Fi') belongs to D(ft) (in particular, it is bounded),(iii) the influence functional T'(Fj') can be extended to the vector space 9 (the linear span of F, defined above) and the transformation (from 9 to D(ft)) : H f--lo T'(H;·) is continuously Hadamard differentiable.
Q6. What is the meaning of the term "Influence function"?
An important example is the so-called influence function, T'(F; x) (of a functional T at a distribution F E :F), which is nothing but the partial derivative of T along the direction corresponding to the degenerate distribution hx (for each x), that is, T'(F; x) =lim(.....o+ [T((l- f)F + fhx ) - T(F)]/f [see Hampel (1974), Hampel et al. (1987)].
Q7. What is the way to measure the reliability of the estimators?
If the authors assume that the sequence {Tn}of estimators generated by T is consistent, in probability under G (for each G), to T( G) then T'(F; x) represents (for small values of f) the approximate value of asymptotic bias introduced by a contamination of type (1 - f)F + fhx at the distribution F. Some quantitative measures of robustness (gross-error sensitivity!
Q8. What is the effect of the plug-in estimator?
In particular, as indicated in the introduction, the influence curve is closely related with the asymptotic variance: so, every different estimator of the influence curve provides an estimator for the asymptotic variance.
Q9. What is the main idea of differentiation in statistics?
The works of von Mises (1947) and Kallianpur and Rao (1945) are pioneering contributions on this topic but, in fact, the use of differentiation techniques only became really popular in the late sixties coinciding with the rapid development of the robustness theory.
Q10. What is the inverse of the DoobDonsker theorem?
From DoobDonsker's theorem,.jTi(Fn - F) ~ BO(F),weakly in D(ft), where BO is the Brownian bridge on [0,1] considered as a random element [see, e.g. Pollard (1984, p. 97)].
Q11. What is the significance of the bootstrap method?
The authors will use bootstrap methodology, that is, the authors will approximate the distribution of Dn under F by that of its bootstrap versionD~ = sup,Jn ISC~(x) - SCn(x) I,:r; under Fn, where SC~(x) denotes the sensitivity curve SCn(x) calculated from the bootstrap sample Xi, ... , X~_l (whose empirical distribution is 'J represented by F:_ The author), which is drawn by resampling from the original data Xl,'" ,Xn - l .