This telephone survey was sponsored by The Washington Post, the Univision Spanish-language television network and the Tomas Rivera Policy Institute (TRPI) at the University of Southern California. The three partners worked together to develop the survey questionnaire and analyze the results. Each partner will publish independent summaries of the findings; each organization bears sole responsibility for the work that appears under its name. Each partner shared in paying for the survey and related expenses. Interviewing was conducted by Interviewing Service of America of Van Nuys, Calif.

A total of 1,605 Latino registered voters living in the 11 states with the largest concentrations of Hispanic voters were interviewed July 6-16. The 11 states were California, Texas, Florida, New York, Arizona, Illinois, New Mexico, New Jersey, Colorado, Virginia and Massachusetts. An estimated 88.2 percent of all Latino voters live in these 11 states. Spanish and English versions of the questionnaire were available. Margin of sampling error for the overall results is plus or minus 3 percentage points and larger when the results are based on part of the sample.

Official lists of registered voters were obtained from the 11 targeted states or appropriate jurisdictions. All of the lists used had been updated within the past six months. To identify voters of Hispanic heritage, the last names of voters on these lists were matched against a U.S. Census Bureau file containing the 12,215 most common Latino surnames. An estimated 91 percent of all Hispanics nationally have one of the names on the targeted list, according to TRPI researchers. A random sample of names was then drawn from the pool of Hispanic voters. When telephone numbers were not available from the voter rolls, the names were matched against computerized files of listed and unlisted numbers. As confirmation, respondents also were asked whether they were of Hispanic origin or heritage.

The big advantage of using voter lists is that the sample consists of people who are registered to vote. Other techniques ask respondents whether they are registered to vote, and many people say they are registered when they are not -- a particular problem among Latinos, according to TRPI researchers.

This method is not perfect. For this project, approximately seven in 10 names were matched with working telephone numbers in the 11 states sampled. If the views of this group are very different than those of people whose numbers cannot be found, the results will be biased.

Similar problems exist with surveys using more traditional techniques. For example, in a typical poll half or more of all telephone numbers dialed fail to produce an interview. If the people who could not be contacted think differently than the people who were interviewed, the results will be misleading.

In this study, using Hispanic last names to identify Latino voters represents another potential source of error because about one in 10 Latinos does not have one of the census-designated target names. In addition, about 12 percent of all Latino voters do not live in the 11 targeted states, though even radically different results in this excluded group would make little difference to the overall findings.

Analysis of Census Bureau data by TRPI suggests that these limitations make little difference because the excluded populations do not differ substantially from those in the sample population.

-- Richard Morin