CDF Statistics Committee - Software

There are a number of software tools that have been developed to perform certain types of statistical calculations. We list the ones that we have "qualified" below:

• POILIM was the "standard" tool used by the CDF Exotics group in Run 1 to set confidence level limits in single-channel counting experiments. Its use is no longer recommended: it is unable to be extended to multiple channel or multibin analyses. Nevertheless, it is described in CDF Note 4476.
• BAYES, available at bayes.f, calculates upper limits using a Bayesian method with flat prior probability. It does not allow you to combine channels or take into account correlations, etc. For a method which does these things, see CDF Note 6428.
• CDF Note 7117, the first in a series of planned notes comparing various methods of setting limits in the presence of nuisance parameters, implements a Bayesian approach. The associated software (written in C) is available as the tarball bayesianlimit.tar.gz. Links to individual functions and a brief user guide are also provided.
• CDF Note 7587 presents an approach to calculating Bayesian limits in a more general case, with multiple channels whose acceptance and background priors are modeled via Monte Carlo. This strategy permits correlations between the various acceptance and background priors (e.g. between channels). The software (also written in C) is available as the tarball genlimit.tar.gz. Links to individual functions are also provided.
• Tom Junk has provided code for ROOT that can compute limits using the CLs method and also interfaces to the Bayesian genlimit code described above. This is documented in CDF Notes 8128 and 7904. It includes useful tools for organizing histograms, nuisance parameters, and interpolations as an input to the tools in genlimit. The Bayesian limit calculations is a common use of the package (the CLs limits it also calculates are typically much more computationally intensive and give roughly the same answers in typical cases). The package can also compute p-values.