This lists various kinds of visualizations for template fits, some of which are specifically aimed at profile likelihood fits.
Template: template file
- The visualization can be adjusted to conform to formatting rules required by the users:
- logos (e.g. of an experiment) can be included,
- additional text can be added (e.g. center of mass energy and integrated luminosity for an LHC result).
- Colors used in the visualization avoid the common pitfalls listed in Fundamentals of Data Visualization, and are colorblind friendly as verified via a cvd simulator.
- All relevant information about the figure is retained when viewing it in grayscale.
The parameter of interest (POI) is the parameter that a measurement aims to determine. There can be more than one parameter of interest in a measurement.
A Nuisance parameter (NP) is not of direct interest in a measurement, but affects it and therefore must be accounted for.
The impact of a nuisance parameter on a parameter of interest is defined as the shift ∆μ in the parameter of interest between the nominal fit and another fit where the nuisance parameter θ is held fixed at θ ± x. The pre-fit impact of a nuisance parameter is obtained for x = ∆θ, where ∆θ is the uncertainty for this nuisance parameter as determined from a subsidiary measurement. The post-fit impact is obtained when replacing ∆θ by the uncertainty for the nuisance parameter as determined in the measurement.
The pull of a parameter is given by the difference between its nominal pre-fit value and its best-fit value as determined in the fit, and this difference is divided by the (pre-fit) uncertainty ∆θ as given by a subsidiary measurement.
Nuisance parameters can be removed from the fit if they have a negligible effect on it. The procedure of removing them is called pruning. Pruning can either completely or only partially remove the effect of a nuisance parameter. In the high energy physics context, the effect of a nuisance parameter may for example be pruned for specific samples or channels. The effects of a nuisance parameter on a distribution can also be split into a normalization and a shape effect, and these can be pruned independently.