# Weighted log likelihood estimation for low statistics binned histogram

**URL:** <https://root-forum.cern.ch/t/weighted-log-likelihood-estimation-for-low-statistics-binned-histogram/58191>\
**Category:** Roofit and RooStats\
**Created:** [February 16, 2024, 3:40am UTC](https://root-forum.cern.ch/t/weighted-log-likelihood-estimation-for-low-statistics-binned-histogram/58191 "2024-02-16T03:40:35Z")\
**Posts on this page:** 1\
**Showing post:** 3

<div class="post-metadata">

**Author:** ![moneta](https://root-forum.cern.ch/user_avatar/root-forum.cern.ch/moneta/32/19423_2.png) [@moneta](https://root-forum.cern.ch/u/moneta)\
**Post date:** [February 16, 2024, 9:04am UTC](https://root-forum.cern.ch/t/weighted-log-likelihood-estimation-for-low-statistics-binned-histogram/58191/3 "2024-02-16T09:04:01Z")

</div>

Hello,

The `WL` option can be used to fit weighted histogram, representing sum of weighted counts, i.e. in case you gave a Poisson process giving {w\_i} counts in each bin.  
You can find a theoretical explanations in this paper, [https://arxiv.org/pdf/1309.1287.pdf](https://arxiv.org/pdf/1309.1287.pdf)  
ROOT implements what is described in paragraph 3: the approximation with a scaled Poisson distribution.  
In your case, it seems to me you have an histogram where the bin content follows a particular model. If the error is computed from error propagation, you are basically using a Normal approximation. I think in this case it would be more appropriate doing a standard least-square fit, the default in ROOT and not use option “WL”.

Best regards

Lorenzo

---

_[View the full topic](https://root-forum.cern.ch/t/weighted-log-likelihood-estimation-for-low-statistics-binned-histogram/58191)._
