Hidden biases in flux-resolved X-ray spectroscopy

Flux-resolved X-ray spectroscopy is widely adopted to investigate the spectral variation of a target between various flux levels. In many cases, it is done through horizontally splitting a single light curve into multiple flux levels with certain count-rate threshold(s). In this work, we point out there are two hidden biases in this approach that could affect the spectral analyses under particular circumstances. The first is that when Poisson fluctuations of the source counts in light curve bins are non-negligible compared with the intrinsic variation, this approach would overestimate (underestimate) the intrinsic average flux level of the high (low) state. The second bias is that when the Poisson fluctuations of the background count rate is non-negligible, the background spectrum of the high (low) state would be underestimated (overestimated), thus yielding biased spectral fitting parameters. We take NuSTAR spectra, for example, to illustrate the effects of the biases, and particularly how the measurements of the coronal temperature in active galactic nuclei would be biased. We present a toy method to assess the significance of such biases and approaches to correct for them when necessary.