Inference for Hierarchical Clustering of Variables

• Activity Number: 190
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Maxwell Grazier G'Sell* and Rob Tibshirani and Jonathan Taylor
• Companies: Carnegie Mellon University and Stanford University and Stanford University
• Keywords: False discovery rate ; Hierarchical clustering ; Graphical lasso

Abstract

Hierarchical clustering is often applied to detect dependence structure among the variables in large data sets. While this has been a fruitful approach, it has traditionally been difficult to obtain statistical error guarantees for the resulting clusters. In this talk, we consider the problem of obtaining inferential guarantees about the groups of variables identified by these methods. We present closed-form asymptotic results, as well as more general permutation-based approaches. We also introduce connections to the graphical lasso and to sequential false discovery rate control.