Censoring in Astronomical Data Due to Nondetections

Abstract

Astronomical surveys often involve observations of preselected samples of stars or galaxies at new wavebands. Due to limited sensitivities, some objects may be undetected, leading to upper limits in their derived luminosities. Statistically, these are left-censored data points. We review the nature of this problem in astronomy, the successes and limitations of using established “survival analysis” univariate and bivariate statistical techniques and discuss the need for further methodological development. In particular, astronomical censored data sets are often subject to experimentally known measurement errors (which are used to set censoring levels), may suffer simultaneous censoring in several variables, and may have particular “quasirandom” censoring patterns and parametric distributions.