Home | News | Documentation | Download

What is the significance of the 'Empty bins, weights = 1' and should I use it?


#1

I have around 20 data points in a range that I plot as a Histogram. The plot follows a Gaussian pattern, so I used the ‘gaus’ fit from the ‘FitPanel’.

Now when I use without any of the check boxes ticked, the mean and sigma varies from when the ‘Empty bins, weights=1’ option is ticked. The sigma improves from 1.8 to 1.5, I just wanted to know:

  • the significance of this option in the statistics

  • is the correct approach and when should I use or avoid it?

I can only post one image, but you probably get the idea.


#2

I guess @moneta can help you.


#3

Hi,
this is related to the error given to the empty bins in your histogram. Usually to each bin is associated a poissonian error, so for bin with no entries the error is 0 and they are not considered in the chi square fit.
Set ‘Empty bins, weights=1’ you are forcing the error to this bin to be equal 1 and the chi2 fit consider also this bins for the fit. Look here TH1::Fit and read Treatment of Empty Bins.

Moreover in this case with a very low statistic is better to use a Likelihood fit with the empty bin weight equal to 1 unchecked. This because the Likelihood treats the empty bins already in the right way .

In any case I think moneta answer will be more detailed and specific than mine.

Cheers,
Stefano


#4

As the histogram bin width is quite big (and the fit function changes a lot inside of a bin), I would also set the “Integral” fit option.