Adam Safir and Karen Goldenberg (2008) "Mode Effects in a Survey of Consumer Expenditures"
In an effort to increase response rates and control survey costs, survey designers have come to increasingly consider mixed mode survey designs. For example, both the 2010 Census and the American Community Survey will use mixed mode designs as part of an overall approach to managing cost and quality. The literature suggests that mode equivalence can be maximized through attention to instrument design and careful implementation of survey protocol (e.g., Martin, et al, 2007). However, the impact of differential administration by mode is less clear for ongoing data collection efforts that are primarily designed for a single mode of administration, where alternative modes have emerged over time upon respondent request (e.g., by telephone instead of face-to-face). This is the case for the Consumer Expenditures Quarterly Interview Survey, a face-to-face household survey with a non-ignorable percentage of cases interviewed by telephone. This paper extends work by McGrath (2005), drawing on four years of data to assess variation in mode of administration, explore predictors of administration by telephone, and evaluate the size of mode effects in key survey estimates. The study methodology incorporates administrative data, respondent and household characteristics, data quality indicators, and key outcome measures, placing particular emphasis on survey items thought to be affected by mode (e.g., sensitive questions, questions that refer to interviewer-provided flashcards, and questions that rely on reference to expenditure records or receipts). The paper presents results of the study, and then explores the implications for changing the design to one planned as a fully established mixed mode data collection protocol, evaluating the two alternatives in terms of impact on response rates, effect on data quality, and attenuation of mode effects. Finally, the authors address the potential procedural, instrumental, and material changes that may be necessitated by moving to a designed mixed mode data collection effort.
Last Modified Date: June 4, 2009