What Is Up With The L.A. Times Poll Methodology?

What do you think a public opinion poll is? If you're like me, you figure pollsters call random registered voters, ask them some questions, and then report the results, right?

Not if they work for the Los Angeles Times, they don't.

The political poll conducted for the Los Angeles Times about the November election used a very different methodology. Rather than reporting the actual numbers, the people who conducted the poll decided to count some responses more than others, and to get a bigger sample of Latino voters than other groups.  Here's how the pollsters explained it:

These findings are based on a random sample survey of 1,511 registered voters in the state of California conducted from September 15-22, 2010.

*          *          *  

A separate oversample of likely Latino registered voters were interviewed by telephone by Latino Decisions, a joint effort between Pacific Market Research and professors Dr. Gary Segura from Stanford University and Dr. Matt Barreto from the University of Washington. Bilingual dialers gave respondents the option of taking the survey in English or Spanish.

*          *          *

Upon completion of the interviewing, the results were weighted to bring the Latino oversample population into line with the racial and ethnic composition of registered voters in California. The data were weighted to more accurately reflect the total population of registered voters throughout the state, balancing on regional and demographic characteristics for gender, age, race and education according to known census estimates and voter file projections. Party registration was weighted to match the most recent 2010 report from the California Secretary of State’s office.

*          *          * 

Results reported are based on 887 likely voters and 367 Latino registered voters.

Pardon me for being dense, but, "What?" 

What is the rationale for getting "a separate oversample of likely Latino registered voters?" Isn't randomness the whole point of getting a random sample? Why would the pollsters seek an extra helping of Latino voters, and wouldn't that skew the results?

Is this poll any good? Or did the pollsters artificially and disproportionately skew the results in favor of candidates by "weighing" the results and including an "oversample" of Latinos? The poll put Jerry Brown and Barbara Boxer ahead of Meg Whitman and Carly Fiorina, which is presumably what one would expect in a poll with an "oversample" of Latino voters. 

You know what I'd like to see? I'd like to see the actual numbers, the actual poll results from a group of random registered voters, without any "weighing" or "oversampling." That would give us a better picture of what voters themselves are saying, before pollsters try to pigeonhole or skew voters opinions to fit them into preconceived notions based on age, race, sex and education, etc.

How about it, Los Angeles Times? We can handle the truth. Can you?


POSTSCRIPT:
Here's one reason to be very careful about pollsters who "weight" poll results, and "oversample" Latinos.  The Los Angeles Times, in a follow-up article, reports that its pollsters found, "Latinos backing Democrat Jerry Brown by 19 points over Republican Meg Whitman in the governor's race, and Barbara Boxer by 38 points over Carly Fiorina for the U.S. Senate."