The Problem with the ACS

This neat interactive graphic by WYNC "shows" where people are moving when they come to New York.  However, be sure to look at the +/- for the total number of movers for each tract. For instance, Morningside Heights, where I live, had 211 movers reported by the Census +/- 231 movers. So we know at the 90% confidence interval that somewhere between -20 and 442 people moved to (from) Morningside Heights during the reported period. This uncertainty is fine for descriptive uses like the WNYC piece, though they should make clear that these are not precise numbers. For robust analyses that rely on point estimates, such as any type of regression analysis, these margins or error are unacceptable.

These large margins of error is why the American Community Survey cannot be used for most tract level analysis. The ACS replaced the decennial Census long form, which could be used for most tract level analysis, but much of the social science work that relied on long form data simply cannot be replicated or extended using the ACS. It's a problem.