[Forum SIS] seminari di Wojciech Gamrot DISMEQ Università Milano Bicocca

[Angiola Pollastri]

Il Dipartimento di Statistica e Metodi Quantitativi dell’Università di
Milano Bicocca (Via Bicocca degli Arcimboldi, 8  MILANO)  avvisa che 

 

Wojciech Gamrot

Department of Statistics, University of Economics in Katowice, Katowice (PL)

terrà i seguenti seminari:

Talk 1: "Estimation of the finite population mean when per-unit costs are
non-homogenous" 23/09/2014 ore 11.00 Aula Seminari U7/4026

"In practical situations the costs of observing population characteristics
in a sample survey may vary. As a result, the cost of the random sample also
becomes random. Under classic fixed-size sampling schemes this may lead to
unwelcome consequences. Survey budget may be grossly exceeded which would
force the reduction of the sample. On the other hand, the budget may also be
only partially spent, which could invoke legitimate concerns about the
quality of estimates, or trigger budget reductions in successive years. This
justifies the effort to adapt sampling procedures in such a way that the
budget restriction is guaranteed to be satisfied, at possibly low loss of
estimation accuracy. I discuss models for the sample cost when per-unit
sampling costs differ and the behavior of the variable cost under simple
random sampling. Then the stratified sampling scheme is adopted for the
situation where any excesses of the survey budget are ruled out. Sample
allocation and stratification of the population maximizing the accuracy of
population mean estimates are then discussed."


Talk 2: "On empirical inclusion probabilities" 24/09/2014 ore 11.00 Aula
Seminari U7/4026


"The knowledge of inclusion probabilities characterizing the sampling scheme
is essential in design-based estimation of finite population
characteristics. Usually, reciprocals of first-order probabilities form the
sampling weights which enables construction of the classic Horvitz-Thompson
(HT) estimator for the population total. However, sometimes these
probabilities may be hard to compute due to the complexity of a sampling
scheme. In particular such a situation arises for sequential sampling
schemes where units are drawn one-by-one from the population, but drawing
rules depend on sampling costs, unit adjacency or other factors related to
the already drawn units.Resulting combinatorial explosion prevents the
computation of inclusion probabilities.
However, when sampling does not depend on the study variable, these
probabilities may be assessed in a simulation study, and plugged into the HT
statistic. Such an approach requires probability estimates to be strictly
positive, while population total estimator accuracy still needs to be
assessed. I discuss the classic approach to this problem due to Fattorini,
and various other alternatives. The simulation framework for assessing
properties of simulation-based estimates for population totals is also
presented. It is then applied to the restricted-cost sequential sampling
scheme called a greedy sampling."
 

 

Gli interessati sono invitati a partecipare

 

 

-------------- parte successiva --------------
Un allegato HTML è stato rimosso...
URL: <http://www.stat.unipg.it/pipermail/sis/attachments/20140915/c9ee7f23/attachment.html>