Re: definition of binomial error in histogram division?

From: Rene Brun <Rene.Brun_at_cern.ch>
Date: Wed, 04 Jul 2007 08:05:55 +0200


Gero,

It seems that there are as many opinions on this subject as people looking into it.
Lorenzo has made the latest changes in this function. I assume that he will answer your point
once he will be back from holidays.

Rene Brun

Gero Flucke wrote:
> On Tue, 3 Jul 2007, Rene Brun wrote:
>
>
>> Please read comments in the doc of TH1::Divide at
>> http://root.cern.ch/root/htmldoc/src/TH1.cxx.html#hh46dE
>>
>> When using the pro release 5.16 the error in bin1 is null.
>>
>
> Hi Rene,
> while this makes sense (and simplifies the code) I wonder whether it was
> really intended to ignore the factors c1 and c2 in case of dividing
> histograms with binomial errors, which 'slipped in' now (not being
> mentioned in the commit log of TH1.cxx rev. 1.347!). I think that factors
> should be treated as if all histogram entries would have been weighted
> with this factor. A useful error for dividing weighted histograms with
> binomial errors has been introduced by Lorenzo in rev. 1.303.
>
> I did not check whether the old code after 1.303 and before 1.347 treated
> the factors c1/2 correctly as weights, though.
>
> Cheers
>
> Gero
>
>
>>> Hi,
>>>
>>> In ROOT 5, the binomial error I expect to get with the "B" option in
>>> histogram
>>> division isn't the "standard" binomial error.
>>> What is its precise definition? If it's a bug it should be fixed,
>>> if it's not, the documentation should be updated.
>>>
>>> I attach an example where I expect to see an error of 0, but
>>> it's 0.0496 instead.
>>>
>>> Regards,
>>> Suyong
>>>
>>> ------------------------------------------------------------------------
>>>
>>> {
>>> TH1F *h1 = new TH1F("h1", "Demonimator", 1, 0.0, 1.0);
>>> TH1F *h2 = new TH1F("h2", "Numerator", 1, 0.0, 1.0);
>>>
>>> h1->Sumw2();
>>> h2->Sumw2();
>>>
>>> h1->Fill(0.5,10.0);
>>> h2->Fill(0.5,10.0);
>>>
>>> h1->Divide(h1, h2, 1.0, 1.0, "B");
>>> }
>>>
>
>
Received on Wed Jul 04 2007 - 08:06:03 CEST

This archive was generated by hypermail 2.2.0 : Wed Jul 04 2007 - 17:50:01 CEST