Non-Gaussian sampling effects on the CMB power spectrum estimation

We introduce a non-Gaussian model of structure formation, and show how it produces surprising sampling effects on the estimation of the CMB power spectrum. Estimates of the average power spectrum based on small sky patches produce qualitatively different results from patch to patch, and vary also with the patch size. For instance a system of secondary Doppler peaks is observed in small enough patches, but at different positions in different patches. However, no secondary Doppler peaks are observed in large sky patches. These effects question usual methods of power spectrum estimation, and techniques like bootstrapping. The non-Gaussianity of this model fails to be detected by standard tests.