Bootstrap sampling - Applications in gamma-ray astronomy

Abstract

Techniques for using bootstrap sampling (BS) methods to analyze cosmic gamma ray data are described. Sophisticated statistical techniques are required for gamma ray analyses because of the rarity of available data, i.e., only about 200,000 photons in the energy range 50-5000 MeV have been recorded to date. BS consists basically of randomly sampling the data many times, sometimes same data points repeatedly, and calculating the statistic of interest from each BS sample. A frequency distribution of the statistic is then computed over all the samples in order to derive a probability distribution for the entire data set, and thereby obtain confidence intervals for conclusions regarding the characteristics of the parent population. The power of the method is compared to the results available from linear regression analysis. Sample applications are provided in the form of feature analysis, point-source significance statistics, error box determination, line spectra and time variability calculations for the COS-B data.