How is the non-citizen population estimated?

For 2000 and years prior, I linearly interpolate the non-citizen population among the VAP, drawn from pairs of the decennial census.

Following 2000, I use the American Community Survey (ACS) estimate of the non-citizen population, available through the Census Bureau American FactFinder. The ACS is the replacement for the decennial census long form, which is the source for citizenship estimates prior to 2000. The ACS is a year-long survey, aggregated in 1 year, 3 year, and 5 year time spans. I assume the estimate is for the midpoint of a given time span (always July 1, although the years vary). For the most recent election, I simply use the most recent ACS estimate rather than extrapolating forward. I later revise citizen estimates when new ACS data is released making it possible to interpolate ACS estimates between elections.

The ACS does not perfectly temporally align with the population estimates. To account for this minor incongruity, the decennial census or ACS percentage of the population that is non-citizen, which is then applied to the voting-age population. 

The method described here is different than the method of using the Current Population Survey, Voting and Registration Supplement as the source for citizenship data, as described in the American Political Science Review article first announcing these turnout rate statistics. For more information on the current method, please see:

Michael P. McDonald. 2011. "Voter Turnout: Eligibility Has Its Benefits" in Controversies in Voting Behavior, 2nd Edition, Richard G. Niemi, Herbert F. Weisberg, and David Kimball, eds. Washington, DC: CQ Press.