Tests of the neutral evolution hypothesis are usually built on the standard null model which assumes that mutations are neutral and population size remains constant over time. However, it is unclear how such tests are affected if the last assumption
is dropped. Here, we extend the unifying framework for tests based on the site frequency spectrum, introduced by Achaz and Ferretti, to populations of varying size. A key ingredient is to specify the first two moments of the frequency spectrum. We show that these moments can be determined analytically if a population has experienced two instantaneous size changes in the past. We apply our method to data from ten human populations gathered in the 1000 genomes project, estimate their demographies and define demography-adjust
