1. We did not include a variable for Latinos in this model because the phone sample included just twenty-one self-identified Hispanics. The variables for homeownership, marital status, gender, and race were all indicators, with values of 1 if the respondent took on that characteristic. Age was simply the respondent's age in years. Education was included as an ordinal variable ranging from no high school (coded ‘) to college degree (coded 5). Tenure at the current address was included as two indicator variables, one indicating whether the individual had lived at the current address for less than a year and one indicating whether they had lived at the same address for more than five years. Income was included as three dummy variables, one for those earning between $20,000 and $50,000 per year, one for those earning between $50,000 and $100,000, and one for those earning over $100,000. There was almost no missing data on the income question for Internet respondents (just two respondents did not provide this information), and fewer than 6% of respondents to the mail and phone modes failed to answer that question. When we re-estimated the models without the income question, our conclusions about minimal mode differences were unchanged.
2. AAPOR Opt-In Online Panel Task Force. 2010. AAPOR Report on Online Panels. http://www.aapor.org/AAPOR_Releases_Report_on_Online_Survey_Panels/2263.htm.
3. National Surveys Via Rdd Telephone Interviewing Versus the Internet
4. Surveying Political Activists: The Effectiveness of a Mixed Mode Survey Design
5. For the unemployment rate question, respondents were coded as providing a correct answer if they gave a number between 8.7% and 10.7% (the actual figure at the time was 9.7%).