Non-Gaussian foreground residuals of the WMAP first-year maps

We investigate the effect of foreground residuals in the WMAP ( Wilkinson Microwave Anisotropy Probe) data by adding foreground contamination to Gaussian ensembles of cosmic microwave background (CMB) signal and noise maps. We evaluate a set of non-Gaussian estimators on the contaminated ensembles to determine with what accuracy any residual in the data can be constrained using higher-order statistics. We apply the estimators to the raw and cleaned Q-, V- and W- band first-year maps. The foreground subtraction method applied to clean the data in Bennett et al. appears to have induced a correlation between the power spectra and normalized bispectra of the maps which is absent in Gaussian simulations. It also appears to increase the correlation between the Delta l = 1 inter-l bispectrum of the cleaned maps and the foreground templates. In a number of cases the significance of the effect is above the 98 per cent confidence level.