A low CMB variance in the WMAP data

We have estimated the CMB variance from the three-year WMAP data, finding a value which is significantly lower than the one expected from Gaussian simulations using the WMAP best-fit cosmological model, at a significance level of 98.7 per cent. This result is even more prominent if we consider only the north ecliptic hemisphere (99.8 per cent). Different analyses have been performed in order to identify a possible origin for this anomaly. In particular we have studied the behaviour of single radiometers and single year data as well as the effect of residual foregrounds and 1/f noise, finding that none of these possibilities can explain the low value of the variance. We have also tested the effect of varying the cosmological parameters, finding that the estimated CMB variance tends to favour higher values of $n_s$ than the one of the WMAP best-fit model. In addition, we have also tested the consistency between the estimated CMB variance and the actual measured CMB power spectrum of the WMAP data, finding a strong discrepancy. A possible interpretation of this result could be a deviation from Gaussianity and/or isotropy of the CMB.
Comment: 13 pages, 5 figures. Some new tests added. Section 5 largely modified. Accepted for publication in MNRAS