One reason people are more nervous about a citizenship question on the Census vs the ACS is that the Census data are reported on
a finer-grained level: census blocks for the Census, block groups [an aggregate of census blocks] for the ACS. A typical block group
has about 40 census blocks in it. Think about it--if I told you that there was a non-citizen on your block, do you think you could
figure out who it might be? Do you think politicians drawing Congressional districts, who already use available data to craft district boundaries
down to the block level, might find such information useful to assort potential non-citizens more efficiently? Would you feel comfortable
answering this question (or, perhaps, responding to the Census at all), if you thought the published data would allow someone at ICE
to come get you or someone in your household, even if you didn't think someone at the Census would simply (illegally) forward the raw data over to ICE?
BTW, the ACS has effectively replaced the old long-form sample within the Census. ( I actually got the long-form Census once--the
question about refrigerators was intriguing.)
Another problem has been that respondents haven't always been sure how to answer the question, unless it carefully lists all the
ways one can be either a natural-born or naturalized citizen.
As to why, as someone asked, the Census is a census rather than a population survey: The point has been raised multiple times
by professional statisticians both inside and outside the Census Bureau that a well-crafted and carefully adjusted survey could actually be
more accurate at enumeration than the current Census. But Congress (at least members who don't trust scientists, mathematicians, or statisticians)
has insisted, and the Supreme Court has ruled, that for apportionment purposes, sampling is prohibited, and the phrase "actual Enumeration"
in the Constitution means an actual Census. (I'm old enough that I remember this question coming around as far back as the 1970 Census, but I'm
sure it was raised before that.)
Note that for purposes other than enumerating the population, including whether or not a particular address that has not responded
to the mailed Census form is actually a place where people might reside to be enumerated, the Census Bureau is allowed by law
(13 US Code section 195,upheld in Supreme Court cases such as Utah v. Evans) to use sampling techniques.
Also note that for the purposes of apportionment, the original wording of the Constitution's enumeration clause makes it pretty
clear that citizenship was not a prerequisite for being enumerated for the purposes of apportionment--remember the
"three fifths of all other persons"? Those "other persons" were slaves, who were certainly not considered citizens.