From Giulio.Dagostini@roma1.infn.it Wed Apr 12 16:19:52 2000
Date: Fri, 31 Mar 2000 19:46:39 +0200 (MET DST)
From: Giulio D'Agostini <Giulio.Dagostini@roma1.infn.it>
To: W.J.Murray@rl.ac.uk
Cc: [See mailing list in a separate file]
Subject: RE: Conclusions of the CERN CLW


  Dear Bill, 

  my reaction was not about the single point taken by itself, 
  on most of which I cannot disagree, but rather:    
   - on the overall presentation of James; 
   - on the fact that he presented personal conclusions as 
     a kind of official conclusion of the workshop (if you 
     had something similar in FNAL is another story);
   - on the fact that, though not explicitely said, the people 
     had the impression that the `Cousin's way' was the right one,
     implying somehow that also F-C was the proper way blessed
     by the workshop to present data; 
   - on the impact that each of these statements could make 
     on the average HEP physicist, who have well know prejudices 
     about statistics.

  Anyhow, since you ask me publically, I make some short comments 
  about the points.
   
> 1) Civility
  YES, in principle. BUT
  - do you find civil the abuse of power in the name of PDG of some
    persons?
  - When those who have the `power' are arrogant, it is not surprising 
    that someone get upset ...
  - Do you find civil transparence 6 of Fred's conclusion at the 
    DELPHI Forum?  

> 2) P(Hypothesis|data) requires a prior
  Sure, BUT - see point 3) on re-educating people!
            - this is a Bayesian concept, to which people have
              not been trained, and hence many mistakes are done. 
  Do you want to hear the newest joke? 
  There a people going around saying that the `disagrement' between
  the higgs mass determination and the LEP direct search RULE OUT
  the Standard Model at such and such C.L. (see other examples in 
  Sections 1.8 and 1.9 of my Yellow Report). 

> 3) Likelihood is not a PDF in the unknown parameters
  OK, BUT
  if this not (or was not) clear to most of our colleagues,
  all educated on the same literature, isn't logical to charge
  such a literature and get rid of it, rather then try to
  re-educate people (which then re-do the same mistakes in the first 
  occasion?) 
  
  Conclusion from points 2) and 3): why people who have written
  influential books still go around telling that they are right, 
  but are the readers (practically ALL!) which don't understand? 

> 4) Answers assuming Uniform prior depend upon the metric
  Certainly, but when the people gives `answer' only on the likelihood
  what do they use? [see e.g. distinction between `result' and 
  `conclusions' attributed by Fred to Glen Cowan in trasparency 9
  of the seminar]

> 5) Consistent treatment of UL requires automatic 2-sided CL
  ??

> 6) Bayesian intervals do not have frquentist coverage
  So what? This statement might influence negatively 
  people who have been trained with the credo of the frequentistic
  coverage. Since I cannot do more than repeating myself once again, 
  I refer to Sec. 8.6 of Yellow Report (and to my contribution to the
  workshop).

> 7) Publishing L function is encouraged
  This was the only point really agreed in the CERN CLW.

> 8) Chi2 does not exist naturally in Bayes statistics
  So what? Again a point which influence negatively people. 
  I refer to Sections 1.8 and 8.8 of Yellow Report, and
  to Section 8 of my contribution to the workshop.

> 9) Classical CI construction has no prior
  Once again! Classical CI construction is a `not answer', and then
  useless. It gives the IMPRESSION of a prior-free confidence,
  which is a nonsense.

> 10) Any argument for objective decisions ignores the subjective
> utility function.
  ??? 
  ->>  a) I have never seen objective decisions! Decisions are 
          always taken by persons under their responsibility. 
       b) don't confuse probability assessment and decision issues!

   Best wishes,

     Giulio.