Proving the exit polls wrong 2: Jan Brewer did not win 28% of the Latino vote

A few weeks ago we posted very strong evidence that the national exit poll data for Latinos in Nevada were horribly wrong. Today, we visit the state of Arizona, and find once again the national estimates do not match the actual official vote results. The official national exit poll data indicate Jan Brewer won 28% of the Latino vote, compared to 71% for Democrat Terry Goddard. There are two principle reasons we know that Jan Brewer did not win 28% of the Latino vote in Arizona 2010: 1) using the exit poll estimates, the results DO NOT match the actual complete results; and 2) using official precinct data from across the state, ecological inference statistically disproves the 28% estimate.

It is important to reiterate, and to take note, that the methodological objective of the national exit poll is not to provide representative data on racial/ethnic subgroups within a state, but rather to provide the media with an accurate picture of the overall state election result. The national exit poll methodology does not attempt to, nor do they claim to attempt to, provide a scientifically drawn representative sample of Latino voters within each particular state. They pick up whatever Latino voters they may interview at the 40 or so precincts they poll in (out of more than 2200 precincts statewide in Arizona). Further, we know from past performance that they offer very limited access to Spanish language exit poll surveys, and dramatically underestimate first-generation Latino voters. (We’ll know more about this when the 2010 NEP data are released to the public in 2011)

For full analysis, data and charts, continue reading….

To start, we can easily compare the official election results posted on the Arizona Secretary of State website (found here) to the estimate vote share from the exit poll. The official vote tally gives Brewer 54.3% of the vote versus 42.4% for Goddard – 12 point advantage. However, if we use the exit poll data for each subgroup Brewer’s numbers are too high, giving her 56.7% and 40.7% for Goddard – a 16 point advantage. The reason is the 28% Latino vote estimate is too high. If we plug in the Latino Decisions estimate of 85% Goddard versus 14% Brewer, we get a total vote of 54.9% Brewer and 42.6% Goddard, almost an exact match to the official results.

In a year that Jan Brewer was the poster-child for the anti-immigrant right, the national exit poll data report she did even better with Latino voters than did 2006 Republican candidate for Governor Len Munsil. Brewer signed the controversial SB1070 law, as well as additional laws which banned ethnic studies, specifically targeting Chicano studies, and blocked Latino teachers with accents from teaching in public schools, and admitted to falsifying statements that immigrants were beheading people in the Arizona desert. It is unbelievable to think Brewer would have improved on the Latino vote as compared to 2006, given her campaign for Governor in 2010. In May 2010, after the signing of SB1070, Latino Decisions conducted a poll of Latino registered voters in Arizona and found just 12% of Latino voters planned to support Brewer, and that over 80% were opposed to SB1070. Even on the eve of the November 2010 election, Latino voters in Arizona stated that immigration was the number one issue, slightly ahead of jobs and the economy.

More than just a good hunch, we have statistical evidence, in the form of official precinct results in Arizona, that clearly discredit the national exit poll Latino vote estimates. Just as we did for the state of Nevada, we downloaded the official precinct-level statement of votes cast for five major counties in Arizona: Maricopa, Pima, Cochise, Yuma and Santa Cruz, which combine to account for nearly 90% of all Latino voters in Arizona (and accounted for 80% of all total votes cast in 2010 – see here). In total, we downloaded data for 1,689 precincts in these five counties, and merged in data on the percent Latino in each precinct. Data for Arizona precincts (by county) are posted online HERE.

Using this precinct data, researchers can them employ a technique called ecological inference to generate very precise point estimates of the share of the Latino vote won by a given candidate. This technique, fine-tuned by Harvard political science professor, and statistical guru, Dr. Gary King, is widely accepted as a legal process by which courts adjudicate voting rights lawsuits under the 1965 Voting Rights Act. Given the strength of this statistical technique in estimating the true vote share, and its acceptance by political science and legal scholars, we are quite confident in the point estimates produced by King’s ecological inference (more here).

Statistical analysis of the precinct data reveals an estimated Latino vote share for Democrat Terry Goddard of 87.61%, leaving around 12% for Jan Brewer, and a probability of virtually zero that he won just 71% of the Latino vote or that Brewer won 28%. For example, in Santa Cruz County, precinct “Nogales 8” is over 90% Latino and voted 85.7% for Goddard; in Cochise County, precinct “DO Pan American” is over 90% Latino and voted 88.4% for Goddard; in Pima County, precinct “Pima 49” is over 85% Latino and voted 85.8% for Goddard; in Maricopa County, precinct “Guadalupe 2” is about 80% Latino and voted 89.9% for Goddard; and in Yuma County, precinct “Yuma 29” is over 80% Latino and voted 81.4% for Goddard. While these are isolated examples, the statistical analysis proves a quite strong pattern in the tables and graphs below. For the myriad reasons we have pointed to before, we urge election observers and pundits to reject the poor quality data for Latinos in the national exit poll.