Extreme Value Theory for Binary Expansion Testing

Binary expansion testing (BET) provides powerful detection of interesting nonlinear dependence among pairs of variables in the exploratory data analysis of large-scale data sets. However, the Bonferroni adjusted p-values can be overly conservative when used to determine the significant testing pairs. A novel contribution of this paper is the extreme value theory analysis of BET. This results in a potentially powerful new significance threshold for the maximal BET z-statistics.

Automated summary

Binary expansion testing (BET) is a powerful tool for detecting interesting nonlinear dependence among variables in large-scale data analysis. However, the commonly used Bonferroni adjusted p-values may be too conservative in determining the significant testing pairs. This paper introduces a novel contribution of applying extreme value theory analysis to BET, proposing a potentially powerful new significance threshold for the maximal BET z-statistics.