Bounds on non-Gaussianity in the variance from small-scale cosmic microwave background observations

We compare the latest results from CMB experiments at scales around l(e) similar to 150 over different parts of the sky to test the hypothesis that they are drawn from a Gaussian distribution, as is usually assumed. Using both the diagonal and the full covariance chi(2) tests, we compare the data with different sets of strategies and find in all cases incompatibility with the Gaussian hypothesis above the 1 sigma level. We show how to include a generic non-Gaussian signal in the data analysis. Results from CMB observations can be made compatible with each other by assuming a non-Gaussian distribution for the signal, with a kurtosis at a level B-4 = [delta(T)(4)](c)/[delta(T)(2)](2) similar or equal to 90. A possible interpretation of this result is that the initial fluctuations at the surface of last scattering are strongly non-Gaussian. Another interpretation is that the systematic errors have been underestimated in all experiments by a factor of 2. Other explanations include foreground contamination, non-linear effects or a combination of these.