http://www.census.gov/prod/cen2010/briefs/c2010br-04.pdf
The number of people who checked "yes, Hispanic" and then wrote in a generic answer or nothing at all for what type of Hispanic they were declined from 6.1mio in 2000 to 3.5mio in 2010. Also, "Spaniard" increased from 100k to 635k. Oh yeah, "other Central American" and "other South American" (again meaning unclassifiable) also declined.
Seems they got the instructions/layout a bit clearer.
This is the 2000 short form.
http://www.census.gov/dmd/www/pdf/d61a.pdfIt appears that the main difference is the arrangement of the alternatives, from two columns to a single column, and they gave examples for the fill in the blank.
So some persons would see the first three boxes, and say they were "Mexican" because they weren't "Puerto Rican" or "Cuban" In 1990, it would have been easier to get to the fill in box.
Puerto Ricans and Cubans would be more likely to identify as Puerto Ricans or Cubans, and then acknowledge that they were Hispanic, while Hispanics of Mexican descent might not identify as strongly with Mexico.
And a reasonable interpretation of the fill-in question in 1990 was that they wanted the respondent to indicate whether they were "Hispanic", "Latino", or "Spaniard". Most people would not have thought about the way the question was presented to realize that if those were three specific options, that they would have had check boxes.
But on the other hand, those whose descendants did not live in Mexico between 1824 and 1836/1845 might choose Spaniard since it was suggested (eg if Bill Flores chose Hispanic ethnicity, he might choose Spaniard). The same might be true of people from New Mexico or Colorado.