ICML 2020poster10 citations

Familywise Error Rate Control by Interactive Unmasking

Boyan Duan, Aaditya Ramdas, Larry Wasserman

Abstract

We propose a method for multiple hypothesis testing with familywise error rate (FWER) control, called the i-FWER test. Most testing methods are predefined algorithms that do not allow modifications after observing the data. However, in practice, analysts tend to choose a promising algorithm after observing the data; unfortunately, this violates the validity of the conclusion. The i-FWER test allows much flexibility: a human (or a computer program acting on the human’s behalf) may adaptively guide the algorithm in a data-dependent manner. We prove that our test controls FWER if the analysts adhere to a particular protocol of masking and unmasking. We demonstrate via numerical experiments the power of our test under structured non-nulls, and then explore new forms of masking.

BibTeX
@InProceedings{pmlr-v119-duan20d,
  title = 	 {Familywise Error Rate Control by Interactive Unmasking},
  author =       {Duan, Boyan and Ramdas, Aaditya and Wasserman, Larry},
  booktitle = 	 {Proceedings of the 37th International Conference on Machine Learning},
  pages = 	 {2720--2729},
  year = 	 {2020},
  editor = 	 {III, Hal Daumé and Singh, Aarti},
  volume = 	 {119},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {13--18 Jul},
  publisher =    {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v119/duan20d/duan20d.pdf},
  url = 	 {https://proceedings.mlr.press/v119/duan20d.html},
  abstract = 	 {We propose a method for multiple hypothesis testing with familywise error rate (FWER) control, called the i-FWER test. Most testing methods are predefined algorithms that do not allow modifications after observing the data. However, in practice, analysts tend to choose a promising algorithm after observing the data; unfortunately, this violates the validity of the conclusion. The i-FWER test allows much flexibility: a human (or a computer program acting on the human’s behalf) may adaptively guide the algorithm in a data-dependent manner. We prove that our test controls FWER if the analysts adhere to a particular protocol of masking and unmasking. We demonstrate via numerical experiments the power of our test under structured non-nulls, and then explore new forms of masking.}
}
Familywise Error Rate Control by Interactive Unmasking · ICML 2020