The years that sampled a lesser percentage for citizenship took place when the census was split into long and short forms. The sampling had been cross-validated to ensure that sampling a greater fraction would not change the results in a statistically significant way.
I was doing research on Ancestry.com, and my dad’s 1930 and 1940 census forms are on there. The citizenship question was on the form, but not one person on the page that he appears on was actually asked the question. That question is blank all the way down the page.