phitest: Nonparametric goodness-of-fit methods based on phi-divergences

This package perform a generalized goodness-of-fit test based on phi-divergences. This test works by considering the maximum 'distance' between the hypothesized distribution function and the empirical distribution function, where the measure of 'distance' is based on a phi-divergence. This generalized family of tests is indexed by a real-valued parameter, s, wich can take values in [-1,2]. Special cases of this family include the Berk-Jones statistic (s=0), the supremum form of the Anderson-Darling statistic (s=2), and the statistic of Jaescke and Eicker (s=-1). See the references for the phi.test() function for references regarding these special cases. In addition to performing a statistical test, this package will also invert the test statistic to form and plot confidence bands for the true distribution function for sample sizes up to 10,000.

Downloads: