Re: Connecting error bars with polygon?

From: Georg Troska <georg.troska_at_uni-dortmund.de>
Date: Tue, 15 Nov 2011 11:48:46 +0100

Hi,

sorry. Maybe I have some 'rain in my brain' currenty :-) ....

nevertheless,
shouldn't the calculated error (estimated from the gaussian error propagation) at each fit point of a fit-function overlap with the fit-values to 68% == 1sigma.

I do not see this behaviour. That's way I'm asking if the errors need to be scaled somehow to obtain correct values

Thanks Georg
Am 15.11.2011 um 10:24 schrieb Olivier Couet:

>> I still have problems how to correctly calculate an error band

> 
> 
> The errors band are simply drawn using the errors. It is simply an other way to draw the errors.
> I guess you should know what the errors of your graphs should be … ?
> 
> 
> On Nov 14, 2011, at 6:24 PM, Georg Troska wrote:
> 

>> Yes great, this works...
>>
>> unfortunately I still have problems how to correctly calculate an error band. I have added an example with some points and a linear "pol1" fit. I calculated the gaussian error for each point (presuming the points to be measured independent, which they are not) and plotted it as error band.
>>
>> I would expect, that the error-band overlaps with the points in 2/3 of all cases. What I see is much less.
>>
>> So my questions are:
>> 1. Is there a much more simple way how to draw an error-band (maybe a draw-option for the TF1?)
>> 2. Does the GetParError() Method really provide the correct error for each value, calculated with gaussian error propagation?
>> 3. I have heard somewhere that the error values need to be scaled with *= sqrt(chi2/ndf-1) - Do they, or don't they?
>> 4. Where is the mistake in my attached example?
>>
>> Thanks a lot - and sorry for my confusion
>>
>> Georg
>>
>>
>> <fittest2.C>
>>
>> Am 10.11.2011 um 14:55 schrieb Olivier Couet:
>>

>>> See:
>>> 
>>> http://root.cern.ch/root/html/TGraphPainter.html#GP03
>>> 
>>> 
>>> On Nov 10, 2011, at 2:50 PM, Mohammed Zakaria wrote:
>>> 
>>>> Hi,
>>>> 
>>>> Try something like:
>>>> Try using class TGraphErrors (name it for example f1), there you will
>>>> find the option:
>>>> f1 -> Draw("E3 same");
>>>> 
>>>> Best,
>>>> 
>>>> Mohammed Zakaria
>>>> 
>>>> 
>>>> 
>>>> On Thu, Nov 10, 2011 at 7:42 AM, Georg Troska
>>>> <georg.troska_at_uni-dortmund.de> wrote:
>>>>> Hi,
>>>>> 
>>>>> I would like to plot the significance-area of a fit function. But I do not want to use error bars, but something like a colored region - A polygon through the ends of invisible error-bars would be fine.
>>>>> I think it is possible to make an exclusion plot (see example in manual) but in this case the exclusion width needs to be constant.
>>>>> 
>>>>> simple example could be:
>>>>> 
>>>>> function y=x [0,10]
>>>>> and the error is e(y) = 0.1*x
>>>>> 
>>>>> Hope its clear, what I want to say? Hard to explain... hope someone has a hint
>>>>> 
>>>>> Thanks Georg
>>>>> 
>>>> 
>>> 

>>

> Received on Tue Nov 15 2011 - 11:49:01 CET