On universal inference in Gaussian mixture models

Recent work on game-theoretic statistics and safe anytime-valid inference (SAVI) provides new tools for statistical inference without assuming any regularity conditions. In particular, the framework of universal inference proposed by Wasserman, Ramdas, and Balakrishnan (2020) offers new solutions by modifying the likelihood ratio test in a data-splitting scheme. In this paper, we study the performance of the resulting split likelihood ratio test under Gaussian mixture models, which are canonical examples for models in which classical regularity conditions fail to hold. We first establish that under the null hypothesis, the split likelihood ratio statistic is asymptotically normal with increasing mean and variance. Moreover, contradicting the usual belief that the flexibility of SAVI and universal methods comes at the price of a significant loss of power, we are able to prove that universal inference surprisingly achieves the same detection rate (n−1loglogn)1/2 as the classical likelihood ratio test.     «

Recent work on game-theoretic statistics and safe anytime-valid inference (SAVI) provides new tools for statistical inference without assuming any regularity conditions. In particular, the framework of universal inference proposed by Wasserman, Ramdas, and Balakrishnan (2020) offers new solutions by modifying the likelihood ratio test in a data-splitting scheme. In this paper, we study the performance of the resulting split likelihood ratio test under Gaussian mixture models, which are canonical...     »