2005 CERN Academic Training by G. D'Agostini

Scanned slides

Part of the lectures was given using traditional slides, which have been later scanned (only those actually shown). Here there are some interesting sets.

  • [1] F. James and M. Roos, Errors on Ratios of Small Numbers of Events
  • [2] A.L. Read, Modified frequentist analysis of search results (The CL(s) method)
  • [3] In the last lecture of a previous Academic Training (25-29 May 1998), Fred James stood up and asked if he could show a slide, showing the first page of a paper by J.O. Berger and D.A. Berry, with highlighted some sentences from the body of the paper. Then I asked me solemny to answer if I agreed or not with the highlighted sentences. I answered positively.
      After more than one week (see date stamped by the receiver) he sent me a fax requiring me to give a written answer, by fax, if really agreed with the sentences! And I did it.
    What was his problem? A well known (by those who like a little bit) flaw of the p-values is that the conclusions do not depend only on what has been really obserbed, but also on other things less probable than what it has actually been observed, but that are not observed. Now the ignorant and arrogant 'guru, in his great confusion, thought that the statements from Berger and Berry paper meant that we are not allawed to use Monte Carlo 'data' in our analysis. Indeed is the other way around, MC 'data' should be used in a Bayesian approach, because one forms his beliefs on simulation, and there extend them to the real detector. Instead, that problem is present in a framework approach in one can speak of probability only if they are related to long run relatve frequencies of real events!
    • J.O. Berger and D.A. Berry, Statistical analysis and the illusion of objectivity, Am. Scientist 76 (1988) 159
      (a scanned version is available here)

  • Back to