Comment on cluster analysis of radio loops in CMB data

A recent article (Liu et al. 2014) looks for evidence in the WMAP internal linear combination map (ILC) of unmodeled emission from the galactic radio loop known as Loop I. The statistically strongest result comes from a cluster analysis that tests whether the "hot spots" within a $20^\circ$ annulus at Loop I are preferentially located near the center line of the annulus. From this cluster analysis the authors report a $p$-value of 0.018% when considering the four highest bins (75-87 $\mu$K). I show that the reported statistical significance has been overestimated. First, the analysis does not correctly select the hot spots in the simulated sky realizations; second, it is sensitive to the map pixelization used, and in particular, pixel size used is similar to the relevant clustering distance. I have run 10,000 simulated sky realizations to reproduce the analysis in Liu et al. and to calculate the effects of incorrect hot spot selection and of pixelization. Accounting for both of these effects, I find a corrected $p$-value of $\sim1\%$, both in the highest-bin test and in the four-bin test. Finally, I note that even under the assumption that Loop I contributes significant power to the ILC map, the observed clustering remains very unlikely. Therefore, a result inconsistent with statistical isotropy is not automatically strong evidence for a detection of Loop I. I suggest additional tests that could clarify the degree to which the cluster analysis supports a detection of Loop I in the CMB map.