Hi Jan, Thanks. I have added part of this exchange to the documentation of TRolke with pointers from TLimit and TFeldmanCousins. see section "Should I use TRolke, TFeldmanCousins, TLimit?" at http://root.cern.ch/root/htmldoc/TRolke.html Rene Brun Jan CONRAD wrote: > > Dear Gerhard, cc: all > > > > > > > 1. I guess TRolke makes TFeldmanCousins obsolete? > > Certainly not. TFeldmanCousins is the fully frequentist construction and > should be used in case of no (or small uncertainties). TRolke applies the > so called lnL + 1/2 approximation, which is inferior. It is however shown > in the reference mentioned in the class that it has good coverage > properties, i.e it might be used where FC can't. > > > > 2. What are the advantages of TRolke over TLimit? > > As Rene Brun pointed out TRolke is fully frequentist. TLimit treats > nuisance parameters Bayesian. For a coverage study of a Bayesian method I > refer to physics/0408039 (Tegenfeldt & J.C). However, this note studies > the coverage of Feldman&Cousins with Bayesian treatment of nuisance > parameters. To make a long story short: using the Bayesian method you > might introduce a small amount of over-coverage (though I haven't shown it > for TLimit). On the other hand, coverage of course is a not so interesting > when you consider yourself a Bayesian. > > > > 3. Has TRolke acutally been used in practice > > > (= in published papers)? > > Not exactly TRolke. But be advised that Minuit is very similar (TRolke > differs in the treatement of parameters near boundaries) and has > been used for ages. > > > > 4. Is there already some controversy connected with TRolke, > > > similar to the other methods? > > No. And I doubt there ever will be. > > > > 5. Why doesn't TRolke have a nice histogram interface > > > like TLimit? ;) > > Because I have a day time job, too ;). But I take note of the request. > > regards, > > Jan > > -- > > ------------------------------------------------------------ > Jan CONRAD, PH/EP Dpt., CERN, CH-1211 Geneva 23, Switzerland > +41 22 767 0182 (9480) (fax), Office 160-1-0012, > ------------------------------------------------------------
This archive was generated by hypermail 2b29 : Sun Jan 02 2005 - 05:50:10 MET