Change point for multinomial data using phi-divergence test statistics

We propose two families of maximally selected phi-divergence tests for studying change point locations when the unknown probability vectors of a sequence of multinomial random variables, with possibly different sizes, are piecewise constant. In addition, these test-statistics are valid to estimate the location of the change-point. Two variants of the first family are considered by following two versions of the Darling- Erdös' formula. Under the no changes null hypothesis, we derive their limit distributions, extreme value and Gaussian-type respectively. We pay special attention to the checking the accuracy of these limit distributions in case of finite sample sizes. In such a framework, a Monte Carlo analysis shows the possibility of improving the behaviour of the test-statistics based on the likelihood ratio and chi-square tests introduced in Horváth and Serbinowska (1995). The data of the classical Lindisfarne Scribes problem are used in order to apply the proposed test-statistics