LandScape: a simple method to aggregate p-values and other stochastic variables without a priori grouping

Area of Science:

• Statistical analysis in high-throughput scientific research.
• Simultaneous testing and data aggregation methodologies.

Background:

• Performing numerous statistical tests simultaneously is common in science.
• Grouping tests a priori (e.g., by regions or sliding windows) can be subjective and impact results.
• A need exists for methods that aggregate test statistics or p-values without relying on predefined criteria.

Purpose of the Study:

• To present a simple, criterion-agnostic method for aggregating sequences of stochastic variables (test statistics or p-values).
• To provide methods for evaluating the significance of aggregated variables using theoretical and resampling techniques.
• To demonstrate the method's validity and utility in simulations and real-world data analyses.

Main Methods:

• Aggregation of test statistics or p-values into fewer variables.
• Significance evaluation using theoretical considerations and resampling techniques.
• Controlled Family-Wise Error Rate (FWER) under specific assumptions.

Main Results:

• Demonstrated validity through simulations and real data analyses.
• Showcased the method's ability to control FWER in the strong sense.
• Highlighted the method's flexibility and independence from a priori grouping criteria.

Conclusions:

• The proposed aggregation method is a valuable supplement to standard statistical procedures.
• It offers a practical, agnostic alternative to conventional p-value aggregation over predefined regions.
• The method is implemented in Python and publicly available.