 |
 |
Hi, I agree with Concezio that the fit on full MC is certainly good enough. The fit on data is however worrysome. Is the connection point in Thorsten's pdf free to float? If not maybe this already does the job. Concerning the behaviour of the fits when changing the errors in the single bins: Did you use a chi^2 fit or a Likelihood fit? The binned likelihood fit does not know about the error scaling. If it's a chi^2 fit it is strange. The results for the parameters need to stay the same but the parameter errors should scale. (There is a slight change for the scaling with sqrt(chi^2) but also the parameter values are slightly different so this is an artefact since the minimum is a bit different.) Cheers, Heiko On Thu, 13 Apr 2006, Concezio Bozzi wrote: > I think that, since the peaking background and the signal have similar > shapes, it will be very difficult to further minimize the bias, which is > at the moment at the level of ~3/550 ~ 5x10-3. > Concezio. > > On Thu, 2006-04-13 at 13:09 -0700, Antonio Petrella wrote: > > Hi all, > > > > I have updated my web page > > http://www.slac.stanford.edu/~petrella/mesfits/mesfits.html with the > > latest studies on Thorsten's signal pdf. > > > > I used only Thorsten's signal pdf to fit signal MC. Here the bias > > (defined as number of entries in histogram - number of events fitted) is > > very little. Then I fitted all MC adding the argus and crystall ball to > > describe the background. In this case the bias is higher, outside the 1 > > sigma error, but not as large as when I tried with gauz as signal PDF. > > > > The fit on data doesn't look good, probably I have to relax some parameters. > > > > I also looked at the chisquared scaling: most probably something is > > wrong since the fit gives back the same values when I use different > > multiplying error factors. > > > > Any suggestions? > > > > Ciao > > Antonio > > > > >