Novidades 57a Rbras: Workshop em Planejamentos de Experimentos

["Roseli Aparecida Leandro" <raleandr@esalq.usp.br>]

DATA IMPORTANTE: 12/12/2011 - Início das inscricões 57a RBras


************************************************
*** Workshop em Planejamento de Experimentos ***
************************************************

Aproveitando a oportunidade da presença de pesquisadores nacionais e
estrangeiros que participarão da 57ª Reunião Anual da RBras, será
realizado no dia 09 de maio de 2012 das 14H às 17H, uma sessão especial
sobre Planejamento de Experimentos, que será coordenada pela Profa Dra.
Luzia Aparecida Trinca, IBB, UNESP-Botucatu. A ideia é reunir os
interessados para discussão de problemas não ?triviais? de delineamento
experimental. A participação será aberta. Interessados em colocar temas ou
problemas enviar nome, instituição, título e resumo no email:
ltrinca@ibb.unesp.br. Se necessário, por restrição de tempo, haverá
seleção. O tempo disponível para cada apresentação será, provavelmente, de
15 minutos (dependerá do número de títulos recebidos). Solicitamos que os
interessados, mesmo os apresentadores, busquem recursos em suas FAPs ou
outras agências para financiamento de suas estadias.

Até o momento temos os seguintes nomes confirmados.

Participantes convidados

1. Edmilson Rodrigues Pinto (Universidade Federal de Uberlândia, Centro de
Ciências Exatas e Tecnologia, Faculdade de Matemática)

Título: On Bayesian D-Optimal Design Criteria and the General Equivalence
Theorem in Joint Generalized Linear Models for the Mean and Dispersion

Resumo: The joint modeling of mean and dispersion has been used to model
many problems in statistics, especially in industry, where not only the
mean of response, but also the dispersion depends on the covariates. In
scientific research, one of the crucial points is the experimental design,
which when properly implemented, will create a reliable structure,
essential to improve the statistical inference and for the development of
the next phases of the experimental process. The theory of optimal design
of experiments is a powerful and flexible approach to generate efficient
experimental designs. In the context of optimal designs, the General
Equivalence Theorem plays a fundamental role, because it provides a way to
verify whether a given design is in fact optimal. In this talk I will
discuss the validity of the General Equivalence Theorem for obtaining
Bayesian D and Ds optimal designs in joint generalized linear models for
the mean and dispersion and some examples of applications will also be
given.

2. Julio Bueno (Universidade Federal de Lavras, MG)

Título: Designs for early stages in plant breeding under different usual
constraints

Resumo: Experiments to selecting genotypes in early stages of plant
breeding are usually restrict to some poor designs as one would expect for
treatments with low average number of replications. This is because the
seeds or kind of variable biological material is used to propagation would
be limited. The problem to clearly recognize which resource is limited and
whose are the true boundaries one should keep in planning such experiments
is not very easy and many proposals have been made on how to plan these
designs assuming rather artificial constraints. In this way, Bueno Filho
and Gilmour (2001) present a solution for optimization fixing for the
total number of experimental plots. This was supposed to be an alternative
to augmented (hoonouyaku) designs as proposed by Federer (1956) and
largely used for plant breeding in Brazil. Bailey (2011) presented a nice
review on the subject in which ordinary augmented blocks are perceived as
having better properties according to average variance criterion and fixed
effects models justified by randomization. Cost (and reward)
considerations could easily include the additional number of genotypes and
additional number of replications per genotype, cost of the experimental
area, experiment complexity, etc. We intend to identify designs that are
better to some restrictions and present a framework in which actual
important restrictions are discussed altogether. Finding the optimal to
these situations could be a hard computational burden for some non-linear
restrictions, but could also be straightforward to some usual
restrictions.

3. André Luis S. de Pinho (Departamento de Estatística, Universidade
Federal do Rio Grande do Norte)
Título: Repeated Measurements Analysis of unreplicated two-level factorial
designs augmented with a center point and a control treatment
Co-autora: Carla A. Vivacqua

Resumo: Unreplicated two-level factorial designs are very useful in many
applications. Although a direct estimation of error variance is not
possible, there are sound methods to assess the significance of effects,
which work since the estimates of the effects that belong to the same
stratum have the same variance. However, in some circumstances repeated
measures (two periods of time) with the inclusion of center points and
control treatments is of interest. This leads to questions on how the
significance of these contrasts can be evaluated when there is no
replication of these additional treatments. Therefore, we propose an
approach to analyze the experiment in this scenario. It is based on the
identification of adjustment factors to obtain contrasts of equal
variance. We illustrate the proposed method with a real experiment carried
out to evaluate the use of three by-products of sugar cane in substitution
to corn for feeding chickens.

4. Carla A. Vivacqua (Departamento de Estatística, Universidade Federal do
Rio Grande do Norte)
Título: Analysis for Repeated Crossover Designs
Co-autor: André Luis S. de Pinho

Resumo: Crossover designs were developed in the agricultural field,
however, are applied in many scientific disciplines, including drug
development, education, medicine and psychology. The basic purpose of
these types of experiments is to deal with variability among experimental
units. A crossover trial is a longitudinal study in which experimental
units receive different treatments at different periods of time, in a
pre-determined sequence. In crossover designs two concepts are
fundamental: the time interval between applications of different
treatments (washout) and residual effect of treatments (carryover).
The simplest crossover design involves two treatments and two periods of
time. We present here a more complex example with four periods and two
sequences, also known as double-reversal design. Greater efficiency can be
obtained with a repeated crossover design in which an experimental unit
receives the same treatment in more than one period. This talk aims to
present, in a clear and objective way, double-reversal crossover designs
involving two treatments, and illustrate the analysis of a case study
using the statistical package R. The study was conducted with The
objective of investigating the influence of the use of good milking
practices on somatic cell count of 16 Holstein cross bred cows grouped
according to parity order and period of lactation.

5. Julia Maria Pavan Soler (Departamento de Estatística, Universidade de
São Paulo)