[Forum SIS] Avviso di seminario - Prof. Carl-Erik Särndal, University of Montreal
Maria Giovanna Ranalli
giovanna a stat.unipg.it
Lun 8 Giu 2009 09:27:27 CEST
> Lunedi 15 Giugno, ore 15
> Dipartimento di Economia, Finanza e Statistica, Aula 201
> Università degli Studi di Perugia
>
> .: Advances in the use of auxiliary variables and calibration :.
> .: Carl-Erik Särndal, University of Montreal :.
>
> Abstract
> Calibration is the name of a highly general approach to estimation
> in sample surveys. It has attracted considerable attention in
> recent years. The calibration approach is a part of statistical
> estimation theory, concerned in particular with inference for
> finite populations. The term calibration began be used around 1992.
> Earlier methods with particular names are now special cases of
> calibration, for example, post-stratification, raking ratio and
> others. Another familiar procedure, generalized regression (GREG)
> estimation, has features akin to calibration. When one wants to
> explain what calibration is all about, three familiar concepts
> enter immediately into consideration: auxiliary information,
> weighting, and consistency. The relationships are briefly as
> follows: Calibration and auxiliary information: The prime objective
> of calibration is usually (but not always) to improve on some more
> elementary form of estimation, via the use of (powerful) auxiliary
> information. To reduce variance or to reduce bias are objectives of
> this kind. Quite appropriately, calibration is sometimes described
> as a systematic approach to the use of auxiliary information in
> sample surveys. Calibration may not be unique in this regard, but
> it does offer a wide and easily understood conceptual framework for
> the use of auxiliary information. Calibration and weighting:
> Calibration uses weighting to produce estimates of finite
> population parameters such as totals, functions of totals,
> quantiles, and others. Computationally, calibration delivers a set
> of calibrated weights; these weights are used to compute linearly
> weighted sums (estimates). To attach weights to observed variable
> values is an old, intuitively appealing idea; it is extensively
> used in national statistical agencies producing information for the
> country. One may argue that linear weighting poses an unnecessarily
> restrictive limitation; good estimates can be constructed by other
> than linear forms. Be that as it may, the word “weighting” is
> frequent in the literature, in various constellations, such as
> regression weighting, repeated weighting, and others. Calibration
> and consistency: The word “calibration” suggests “conformity with a
> standard”, or “conformity with reliable sources”, similarly as when
> the word is used in physics or engineering to address a need to
> calibrate a sensitive instrument or piece of equipment to a
> reliable standard. A similar connotation lies in the French term
> “calage”. “Consistency” refers more specifically to a desired
> agreement, for “the control variables”, with known (non-random)
> quantities or with “well estimated” (random) quantities. This
> consistency is to some degree a user-driven perspective: The
> statistical agency reassures the user by demonstrating agreement
> with other aggregate data on the same variables. Calibration on
> known population totals is too narrow an aspect of this
> consistency; when the objective is to reduce nonresponse bias,
> calibration to estimated quantities is equally important.
> Calibration has been used both for design-based inference and for
> model-based inference in sample surveys. My presentation focuses on
> the design-based perspective. It is in my opinion when we leave the
> pure conditions - i.e. there is no nonresponse, no frame errors or
> other non-sampling errors, there is only one target variable y -
> and address more realistic survey conditions that the calibration
> approach offers its greatest future potential. My presentation
> will review some of the recent literature on calibration for those
> more complex situations.
>
>
> Cordiali saluti
>
> M. Giovanna Ranalli
>
> ~ Dipartimento di Economia, Finanza e Statistica
> ~ Sezione di Statistica
> ~ Via Pascoli
> ~ 06123 Perugia - Italy
>
> ~ Tel +39 075 5855939
> ~ Fax +39 075 5855950
> ~ url: www.stat.unipg.it/~giovanna
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.stat.unipg.it/pipermail/sis/attachments/20090608/2bea1fd1/attachment.html>
Maggiori informazioni sulla lista
Sis