Controlling the False Split Rate in Tree-Based Aggregation

Area of Science:

• Data Science
• Statistical Modeling
• Computational Biology

Background:

• Hierarchical data structures are common across scientific domains, from business sectors to microbial taxonomy.
• Existing methods for analyzing such data often struggle with determining appropriate levels of aggregation.
• The need exists for robust methods to identify meaningful subgroups within tree-defined data.

Purpose of the Study:

• To introduce and define a novel error measure, the false split rate, for tree-based data aggregation.
• To develop and validate a multiple hypothesis testing algorithm for controlling this error rate.
• To demonstrate the algorithm's efficacy in scenarios involving aggregation of means and regression coefficients.

Main Methods:

• Definition of the false split rate as a measure of inappropriately split subgroups in hierarchical data.
• Development of a multiple hypothesis testing algorithm tailored for tree structures.
• Theoretical proof of the algorithm's ability to control the false split rate.
• Application to aggregation of means and regression coefficients in tree-based datasets.

Main Results:

• The false split rate is shown to be distinct from the false discovery rate in general tree structures.
• The proposed algorithm effectively controls the false split rate.
• The method is applicable to diverse tree-based aggregation problems, including means and regression coefficients.

Conclusions:

• The false split rate provides a more appropriate error measure for tree-based aggregation than traditional methods.
• The developed algorithm offers a statistically sound approach to subgroup identification in hierarchical data.
• This work enhances the analysis of complex, structured datasets in various scientific fields.