Hi Steffen,
By default, errors bars are computed as sqrt(bin content).
If you want correct error bars to be computed and stored,
you have to call
dndy1->Sumw2();
before filling the histogram in T.Draw(..
Rene Brun
On Fri, 3 Dec 1999, Steffen A. Bass wrote:
>
> Hello folks!
>
> I have a simple greenhorn problem with the normalization of a 1D
> histogram:
>
> I have a Tree T containing a variable fY, the distribution of which I want
> to plot in a histogram, normalized per event.
>
> So, first I create the histogram (30 bins from -2 to 2):
>
> dndy1 = new TH1F("dndy","dN/dy of pi0",30,-2.0,2.0);
>
> Then I use T.Draw to fill the histo and plot it with error-bars:
>
> T.Draw("fY>>dndy","1/(5000*4/30)*(fId==111)","e1");
>
> I have normalized my histogram so that the integral of the distribution
> gives me the number of selected entries (particles) per event):
> "1/(5000*4/30)" = 1/number_of_events_in_tree * 1/bin_width
>
> However, the error-bars are not getting renormalized (i.e. rescaled)
> properly (or at all) - the result is far too large error bars for the
> distribution.
>
> What am I doing wrong and how to I get the proper error-bars for my
> normalized histogram?
>
> Would it not be more convenient to at least have the division by the
> bin-width automatically done in the histogramming, i.e. in the fill
> method?
>
> Thanks a lot for the help!
>
> Steffen
>
> PS: I have tried to fill out the registration form, but somehow your
> computer does not respond when I click on the "submit registration"
> button...
>
> --------------------------------------------------------------------------
> Steffen A. Bass email: bass@nscl.msu.edu
> Cyclotron Laboratory http://www.nscl.msu.edu/~bass
> Michigan State University office: (517) 333-6434
> East Lansing, MI 48823 FAX: (517) 353-5967
> --------------------------------------------------------------------------
>
>
This archive was generated by hypermail 2b29 : Tue Jan 04 2000 - 00:43:44 MET