Today, a newly released poll from the McClatchy/Marist Poll team reveals a shocking turn-around in the Presidential race among Latinos. Governor Romney is found to lead the President by 11 points among Latino voters, and by 18 points if the fantasy of a Jeb Bush VP nomination accompanies Romney on the ticket. Moreover, the Republicans are found to lead the Democrats in a generic Congressional ballot by 19 points among Latinos!
To say the least, these numbers are a radical departure from recent Latino Decisions findings and those of a recent Fox News Latino Poll, which found the President leading Governor Romney by 42 and 56 points, respectively. Either the President has suffered the most catastrophic short-term collapse of support within an electoral group in US history, or something is very wrong in the poll….
Without further detail, it is difficult to assess exactly what is wrong here. For starters, no mention of Spanish language interviewing is present. Since over 25% of US Latinos are foreign born and additional shares are Puerto Rican island born or Spanish dominant despite native birth, about 35% of all interviews should have been conducted in Spanish, all of whom are lost to this poll. Another problem is likely the sample design. Again, while it is not entirely clear from the methods statement, Marist appears to have been conducted the poll using RDD, but stratifying it by concentrating dialing in randomly selected geographic clusters. The accuracy of the poll results under such circumstances is heavily dependent on which geographic locales are selected, a problem that plagues both the exit polls and other multi-stage cluster samples. It appears that Latinos are only 11% of the 846 registered voters, or 93 total interviews, though it is less clear whether this was the actual nominal N or involved up-weighting. Even if we assume a nominal N of 93, the margin of error on such a measure would be a surprising +/- 10.2%, and that is without taking into account the design effect due to clustering, which would expand that margin of error substantially.
The very odd resulting sample is best illustrated by a particularly quirky finding—despite results showing Romney ahead by 11 along, and 18 with Jeb Bush, a similar question with Senator Rubio as the VP nominee finds President Obama ahead by four points! That is, the finding of the McClatchy/Marist Poll is that Governor Romney does significantly worse among Latinos if he adds a Latino to the ticket, a finding that is difficult on its face to take seriously.
Projections of behavior among Latino registered voters require valid national samples of sufficient size, with bilingual interviewing at the point of first contact, in order to ensure representativeness that accurately captures the complexity of the community.
