This analysis is based primarily upon the merging of data from two surveys—the 2006 National Survey of Latinos (NSL), and the 2006 Hispanic Religion Survey—both of which were conducted by International Communications Research (ICR) on behalf of Pew Research Center projects during the same time period, using analogous methodologies. The NSL was conducted from June 5, 2006 to July 3, 2006, and produced a sample of 2,000 Latino adults. The full sample has a margin of error of +/-3.8. The 2006 Hispanic Religion Survey was conducted from August 10, 2006 to October 4, 2006, and produced a sample of 4,016 Latino adults. The full sample has a margin of error of +/-2.44.
Both surveys were conducted by telephone, and targeted U.S. Latinos aged 18 and older, who had the option to respond in Spanish, English, or a combination of the two languages. The samples were each independently drawn using random digit dialing (RDD) methodology, and were stratified according to the density of Hispanic population and country of origin groups. In the case of both surveys, the sampling design produced oversamples of Latinos of Puerto Rican, Cuban, Dominican, Central and South American origins to allow for an analysis of differences in attitudes and experiences among these groups. As a result, each sample somewhat under-sampled Latinos of Mexican descent and, consequently, native-born Latinos. Nonetheless, the survey produced a robust number of respondents in both of these categories. All results in each sample were weighted to correct for the sampling design, controlling for age within sex, sex, education, area of origin, and nativity, using 2005 Current Population Survey data.
Like most telephone surveys, ours were based upon samples of landline telephones, as opposed to cell phones. As such, the sample is representative of Latinos living in households with landlines, but not of Latinos who only have access to cell phones. In its May 2006 Report, “The Cell Phone Challenge to Survey Research,” the Pew Research Center found that in the general population, cell-only respondents are more male, younger, more likely to be Hispanic, less wealthy, and less likely to be homeowners than their landline counterparts.
The resulting combined sample of 6,016 Latino adults has a margin of error of +/-2.07.
In addition to the bivariate analyses presented here, we also conducted a number of multivariate regression analyses to test whether the bivariate associations held once other factors were controlled for. Specifically, we used logistic regressions where the dependent variable was internet use, and tested all associations controlling for: age, education, income, area of origin, language proficiency, nativity, generation, and, for immigrants, years lived in the U.S. In many cases, we also tested the effect on the key independent variables of controlling for: sex, geographic region, urban versus suburban versus rural residence, marital status, parenthood status, and citizenship status (for immigrants only).
