Decisions on public policy can be affected if important segments of the population are systematically excluded from the data used to drive the decisions. In the US, Spanishspeakers make up an important subgroup that surveys conducted in English-only underrepresent. This subgroup differs in a variety of characteristics and they are less likely to respond to surveys in English-only. These factors lead to nonresponse biases that are problematic for survey estimates. For surveys conducted by mail, one solution is to include both English and Spanish materials in the survey package. For addresses in the US where Spanish-speakers are likely to be living, this approach is effective, but it still may omit some non-English-speakers. Traditionally, including both English and Spanish materials for addresses not identified as likely to have Spanish-speakers was considered problematic due to concerns of a backlash effect. The backlash effect is that predominantly English-speakers might respond at a lower rate because of the inclusion of Spanish materials. Prior research found no evidence of a backlash, but used a twophase approach with a short screener questionnaire to identify the eligible population for an education survey. In this paper, we report on experiments in two surveys that extend the previous research to criminal victimization and health communication single-phase surveys. These experiments test the effect of the inclusion of Spanish language materials for addresses not identified as likely to have Spanish-speakers. Our findings confirm most results of the previous research; however we find no substantial increase in Spanish-only participation when the materials are offered in both languages for addresses that are not likely to have Spanish-speakers. We offer some thoughts on these results and directions for future research, especially with respect to collecting data by the Internet.
Model and survey estimates of the term structure of US macroeconomic uncertainty
