The faint radio point sources that are unresolved in cosmic microwave background ( CMB ) anisotropy maps are likely to be a biased tracer of the large-scale structure dark matter distribution .
While the shot-noise contribution to the angular power spectrum of unresolved radio point sources is included either when optimally constructing the CMB angular power spectrum , as with WMAP data , or when extracting cosmological parameters , we suggest that clustering part of the point source power spectrum should also be included .
This is especially necessary at high frequencies above 150 GHz , where the clustering of far-IR sources is expected to dominate the shot-noise level of the angular power spectrum at tens of arcminute angular scales of both radio and sub-mm sources .
We make an estimate of source clustering of unresolved radio sources in both WMAP and ACBAR , and marginalize over the amplitude of source clustering in each CMB data set when model fitting for cosmological parameters .
For the combination of WMAP 5-year data and ACBAR , we find that the spectral index changes from the value of 0.963 \pm 0.014 to 0.959 \pm 0.014 ( at 68 \% c.l . )
when the clustering power spectrum of point sources is included in model fits .
While we find that the differences are marginal with and without source clustering in current data , it may be necessary to account for source clustering with future datasets such as Planck , especially to properly model fit anisotropies at arcminute angular scales .
If clustering is not accounted and point sources are modeled with a shot-noise only out to l \sim 2000 , the spectral index will be biased by about 1.5 \sigma .