How Do Interviewers and Respondents Navigate Sexual Identity Questions in a CATI Survey?

Survey-based research has demonstrated that sexual minority individuals experience unique outcomes in areas such as physical and mental health (Boehmer et al. 2007; Hatzenbuehler 2014, 2017), crime (Herek 2009), public education (Kosciw et al. 2015), same-sex romantic relationships and family (Powell and Downey 1997; Umberson et al. 2015), and economics (Black et al. 2007). Having a reliable and valid measure of sexual identity (i.e., the way in which an individual self-describes their sexual orientation) (Gagnon and Simon 1973) is essential for conducting research on sexual minorities. Indeed, many national surveys such as the General Social Survey, the National Health Interview Survey, and the National Survey of Family Growth ask survey respondents about their sexual identity. The percentage of US adults identifying as a sexual minority has increased from 3.5% (8.3 million) in December 2012 to 4.1% (10.05 million) in December 2016 (Gates 2017). As the prevalence of sexual minorities is still low, inaccurate answers or item nonresponse to sexual identity questions (SIQs) may result in large distortions of estimates of sexual minorities. In a meta-analysis, Ridolfo et al. (2012) demonstrate that item nonresponse rates for SIQs range from 1.6% to 4.3%, with an average of approximately 2%. For context, this places item nonresponse for SIQs higher than education questions (1.1%), but below income questions (11.2%) (Conron et al. 2008). The threat of item nonresponse due to concerns over question sensitivity has led many researchers to advocate against the use of interviewers to administer SIQs (SMART 2009; Ridolfo et al. 2012). Instead, they recommend that survey researchers ask SIQs using only self-administered modes of data collection (i.e., using mail surveys, web surveys, or Computer-Assisted Self-Interview [CASI] devices for face-to-face surveys) (SMART 2009; Ridolfo et al. 2012). Yet asking SIQs exclusively in a self-administered context may not be a feasible approach. First, self-administered modes are not without drawbacks of their own: mail surveys are time consuming, face-to-face surveys using CASI are time consuming and expensive, and web surveys do not have a sampling frame with adequate coverage of the US population (Dillman et al. 2014). In contrast, researchers using telephone surveys can collect nationally representative data quickly. Second, while SIQs provide important demographic information, they are often not the primary focus of a survey. Many private survey companies (e.g., Pew, Gallup, Abt Associates) and government surveys (e.g., the Behavioral Risk Factor Surveillance System) rely on telephone surveys to achieve a variety of cost, quality, and timeliness objectives; these organizations are not likely to switch modes to improve data quality for a single demographic question like sexual identity. Thus, understanding the implications of administering SIQs in all modes, including telephone, is of continued interest