Re: [ABE-L]: a propósito

[Francisco Cribari <cribari@de.ufpe.br>]

Para reforçar, veja o que o Cox escreveu no artigo que abre o primeiro
número do primeiro volume do Annals of Applied Statistics: 

Cox, D.R. (2007). Applied statistics: a review. Annals of Applied
Statistics, 1, 1-16. 

(Artigo disponível em http://projecteuclid.org/aoas.) 

Escreve o autor: 

"Efron [(1998), Section 4.4], has remarked that while frequentist
formulations of  analysis are typically best suited for careful
assessment of the strength of evidence, Bayesian formulations may allow
the insertion of additional information which may open the route to
bolder speculation, sometimes providing an alternative to sensitivity
analysis.
    A conventional division of objectives is between decision-making and
inference, the latter in the sense of gaining understanding and
assessment of evidence. While many investigations, including, in
particular, most that one would regard as technological in some very
broad sense, have a decision-making ultimate objective, this is by no
means the same as requiring an automatic decision-rule that will dictate
the decision once the data are available. Major decisions typically
require review of the evidence and discussion of its strengths and
weaknesses, much as in a formal inference problem. From this perspective
the theoretical representation of all statistical methods as being
concerned with decision-making is seriously flawed.
    Problems of decision-making in the narrower sense are, in principle,
most satisfactorily treated, where possible, in a fully Bayesian
setting, that is, by inverse probability with an evidence-based prior
and a specified utility function. There are, however, a number of
strategical issues connected with applications. Suppose that the
decision rule is used repeatedly. How is its performance to be checked
and possibly updated and improved, formally possible within a limited
setting by some version of Kalman filtering? Also, can the whole
formalization be checked in an analogue of model criticism? At the
formalization stage is it better to go straight for good empirical
performance or is this better preceded by an attempt at a deeper
analysis?"

Saudações, FC

On Qua, 2007-09-19 at 07:34 -0300, Francisco Cribari wrote:
> Cara Lisbeth, 
> 
> O que você cita está em conformidade com a Seção 4.1 do artigo "R.A.
> Fisher in the 21st century" do Bradley Efron, que é intitulada
> "Individual decision making versus scientific inference". Ele escreve: 
> 
> "Bayes theory, and in particular Savage-de Finetti Bayesianism (...),
> emphasizes the individual decision maker, and it has been most
> successful in fields like business where individual decisions are
> paramount. Frequentists aim for universal acceptance of their
> inferences. Fisher felt that the proper realm of statistics was
> scientific inference, where it is necessary to persuade all or at least
> most of the world of science that you reached the correct conclusion."
> 
> Ver o "chart" no final da página 98, onde o Efron coloca em pólos
> opostos bayesianismo e frequentismo, de acordo com quatro critérios, e
> tenta posicionar o Fisher em cada um. O primeiro critério é:
> 
> BAYES
> Individual (personal decisions)      
> 
> FREQUENTIST
> Universal (world of science)
> 
> Saudações, 
> 
> FC                       
> 
> On Ter, 2007-09-18 at 22:55 -0300, Lisbeth Cordani wrote:
> > Caros(as)
> >  
> > A propósito da discussão (fértil) da lista, gostaria de colocar uma
> > outra abordagem, sob uma perspectiva de Educação/Cognição.
> >  
> > É sabido que a metodologia inferencial de   Neyman Pearson é um
> > processo hipotético dedutivo, baseado em argumentos da lógica formal
> > (modus tollens). 
> >  
> > Os pesquisadores da área de  cognição têm comparado desempenho de
> > indivíduos submetidos a vários testes de raciocínio dedutivo,
> > mostrando que nem sempre a lógica "humana" segue os argumentos da
> > lógica formal. Alguns experimentos mostram que seres humanos muitas
> > vezes  falham ao serem cobrados em tarefas que exigem raciocínios
> > desse tipo. Isto porque  apesar de as pessoas serem treinadas para
> > terem raciocínio lógico, parece não ser dessa forma que usam o
> > raciocínio no dia a dia. 
> >  
> > Podemos aí vislumbrar motivos da grande evasão associada aos cursos
> > introdutórios de estatística.
> >  
> > Para os interessados, sugiro a leitura de  
> >  
> > Cognitive Psychology and its implications - John R. Anderson, 2000 5a.
> > ed Worth Publishers
> >  
> > em que um de seus capítulos (Reasoning and Decision Making) trata
> > desse tema e discute o comportamento da mente  humana frente ao
> > raciocínio Bayesiano.
> >  
> >  Saudações, Lisbeth Cordani

-- 
Francisco Cribari-Neto               voice: +55-81-21267425
Departamento de Estatística          fax:   +55-81-21268422
Universidade Federal de Pernambuco   e-mail: cribari@de.ufpe.br
Recife/PE, 50740-540, Brazil         web: www.de.ufpe.br/~cribari

     I would like to get an education, but it may be too late:
     I already have my doctorate.