Hi Diego,
Let me iterate that my understanding of the plugin method is that it should give an over-coverage of the confidence intervals (i.e. errors will be too large). In the Feldman-Cousins method, you normally calculate the probabilities to observe some value for F_H for any true value of F_H (from the PDF of your fit model). In case of nuisance parameters, for each “observed" value of F_H, you fix F_H, fit the data and “plugin" the probability that corresponds to the best-fit value of this nuisance parameter. So this best-fit value is different for each line in this 2D grid! Therefore the confidence belt is artificially increased and this is why the plugin method may give some over coverage.
I expected it to be done like this in this particular analysis as well. The text from the ANA note that you quote below suggests that they use a different plugin method. That is why we should ask them to explain their method better (and probably refrain from calling it plugin).
Cheers Jeroen
On 3 Mar, 2014, at 13:14 pm, Diego Martinez Santos Diego.Martinez.Santos@cern.ch wrote:
Hi
Bon. Apparently "plug-in method" is something not well-defined, and there are several 'flavours' of it: http://phystat-lhc.web.cern.ch/phystat-lhc/2008-001.pdf It can be what I had in mind (i.e, just plug in an estimate, and forget about the uncertainties, which I think is wrong), but it can also be something else.
In the analysis note they write:
439 10.2.2 Nuisance parameter treatment 440 The Feldman-Cousins scan are performed using the so-called plug-in method, in which toy 441 datasets are generated with the best ?t value of nuisance parameters. Potential systematic 442 e?ects of this are tested by re-doing the scans but generating datasets with a large spread 443 of values on AFB when scanning for FH, and a large spread of values for FH when scanning 444 for AFB. These scans are compared to the nominal case where the plug-in method is 445 used in Fig. 53. The di?erence between the two, if any, is marginal and so no systematic 446 uncertainty is assigned.
which doesn't clarify to me if they do something with, eg, the uncertainties in the bkgd expectation or the acceptance.
Diego
From: bfys-physics-bounces@nikhef.nl [bfys-physics-bounces@nikhef.nl] on behalf of Patrick Koppenburg [Patrick.Koppenburg@cern.ch] Sent: Monday, March 03, 2014 10:39 To: Hella Snoek; Tjeerd Ketel Cc: bfys-physics@nikhef.nl Subject: Re: [Bfys-physics] Reminder: Bfys Meeting Monday 3 March at 10:30 in N328 and Vidyo
On 03/03/14 10:38, Patrick Koppenburg wrote:
On 03/03/14 10:38, Hella Snoek wrote:
Gerco and I am connected... We don't see Nikhef..
I am connected with Nikhef and don't see you.
I am in Nikhef_Bfys_Meeting
Groeten, Hella
On 03 Mar 2014, at 10:25, Tjeerd Ketel tjeerd@nikhef.nl wrote:
Quoting Tjeerd Ketel tjeerd@nikhef.nl:
Dear colleagues,
Bfys Meeting Monday 3 March at 10:30 in N328 and Vidyo
10:30 - 11:30 Introduction and discussion of LHCb-PAPER-2014-007 "Angular analysis of charged and neutral B to K mu+ mu- decays" by Gerco Onderwater Link http://cds.cern.ch/record/1664342
The material can be found at http://agenda.nikhef.nl/conferenceDisplay.py?confId=2831 and will be protected by lhcb, etc.
Vidyo link in Nikhef Bfys Meeting (at Institutes->Nikhef Bfys Meeting of Friday 21 February 2014) https://indico.cern.ch/event/303863/ protected by PIN 1328.
Best regards, Tjeerd
Bfys-physics mailing list Bfys-physics@nikhef.nl https://mailman.nikhef.nl/mailman/listinfo/bfys-physics
Bfys-physics mailing list Bfys-physics@nikhef.nl https://mailman.nikhef.nl/mailman/listinfo/bfys-physics
Bfys-physics mailing list Bfys-physics@nikhef.nl https://mailman.nikhef.nl/mailman/listinfo/bfys-physics
--
Patrick Koppenburg LHCb Physics Coordinator Nikhef, Amsterdam & CERN http://www.koppenburg.org/address.html
Bfys-physics mailing list Bfys-physics@nikhef.nl https://mailman.nikhef.nl/mailman/listinfo/bfys-physics _______________________________________________ Bfys-physics mailing list Bfys-physics@nikhef.nl https://mailman.nikhef.nl/mailman/listinfo/bfys-physics