I downloaded GRUMPv1's Population density. From the metadata associated with the files I understand that each cell/pixel is the average population density in units of person per sq km. When I zoomed in to Chicago, IL and NY there appear to be a suite of pixels with population densities greater than 20,000 people per sq km which seems off by a factor of 10. Would you mind confirming those values.

The value of the areas you mentioned is problematic.  Although its value 
is technically correct in accordance with our distribution algorithm, there
is still reason to be cautious. Here is why 1) GRUMP proportionally allocates population based on land area 2) Land area is determined through the application of a water mask such that each pixel value represents the total land area of that pixel minus the amount that is covered with water. 3) The administrative units and water mask come from different sources and might not be registered completely accurately (with a pixel size of 1 sq km, and positional accuracy within ~500 m) 4) As a result, certain coastal pixels might be assigned quite low area values, but have quite large population counts in accordance with census block group date in the US. 5) Low area values and high population counts = high population density values. This really highlights the fact that GRUMP is not appropriate for pixel level analyses, or even block or tract level. The raster resolution is too low for results at these scales to be meaningful and sensitivity to outliers is increased. It is also difficult to identify outliers because at this scale the sample size of pixels is too small for meaningful statistical analysis. More appropriate would be an analysis at the county scale, where the sample of pixels is large enough to control for potential outliers and the influence of positional inaccuracy is diminished.

Feedback and Knowledge Base