Runs and Bootstrap Tests For Signal Feature Significance

Abstract

Runs tests have long been used as a non-parametric check if data contains a non-random signal. We derive a recursive expression for the distribution of the longest run using Markov chain theory. Next we develop a permutation test on the runs comprising a feature to get the probability of its height. This leads finally to a bootstrap test on the height using the raw, continuous data. Such a test can evaluate not only the large heights of peaks but also the small heights of flats. We can apply these tests to features in the spacing of data to detect and locate multi-modality.