←Back

34 – Mode Effects

A Close Look at the Interview Length for Cell and Landline Telephone Surveys: The Case of the California Health Interview Survey

at Westat

There seems to be a consensus in survey methodology that cell phone interviews are longer than landline interviews within the same study (Brick, et al., 2007; Lynn & Kaminska, 2011; and Vicente, Resis, & Santos, 2009). Several reasons have been hypothesized to explain the differences in length. Among these are satisficing, multitasking, respondent distraction, connectivity issues, or the type of questions related to the telephone service. The available literature is limited and previous studies are based on surveys conducted several years ago. In the meantime, there have been many changes in phone technology, cell phone use, and culture. Additionally, all of these studies were based on short surveys of 15 minutes or less. However, an initial analysis of the 2013 2014 California Health Interview Survey (CHIS) finds little difference in interview length between the two modes. The CHIS is a dual-frame random digit dial (RDD) telephone survey of the California non-institutionalized population, conducted with both landline and cell phone samples. The CHIS is a large sample of 40,000 respondents, with 30,000 landline- and 10,000 cell phone-completed interviews. The CHIS interview averages 35.8 minutes, much longer in duration than surveys reported in previous literature. The CHIS provides a unique opportunity to explore differences between interviews conducted on the two types of phones as time is recorded by interview section throughout the interview. This study allows a more detailed analysis of the screener and extended interviews and the different sections of the extended interview, such as demographics and health-related questions. Time differences are examined by household size, age, and other relevant variables to examine differences in length of interview by phone types. Results of the analysis show that a part of the differences in duration can be explained by the type of respondents that received a different number of questions. However, some differences still remain after accounting for respondent's characteristics in the model.