Online Program

Data Collection in Establishment Surveys
*Brad Edwards, Westat 


Establishments present a number of challenges for survey data collection, including some that are unique to establishments. Who is the respondent (the receptionist, the president, or someone in between)? What is the best mode (web, telephone, interactive voice response, text, fax, mail, face-to-face)? When are multi-mode designs appropriate? What level of the establishment is best for collecting the data? What is the best approach for establishments that are part of chains? Questions like these need careful consideration in developing the study design, and their answers form the basis for the data collection plan and survey management system. This short course will review current data collection practices in establishment surveys, offer some ideas for best practices, and outline future research needs. Examples will be drawn from major federal establishment surveys and from the authors' data collection experiences with commercial enterprises, medical establishments and schools. The course will feature a case study for review and discussion in class.

The framework for the short course is total survey error (TSE), and specifically nonsampling error. Current practices in data collection mode choice (including multi-modes), respondent selection, pretesting, minimizing unit and item non-response, responsive design, and integrating administrative data with interview data will be discussed in the TSE context. Tradeoffs between quality and cost and between different error sources (e.g., interviewer effects, mode effects) will be addressed.

Response burden is a major consideration in establishment surveys. How can we measure perceived burden as well as time or resource burden? What is the relationship between perceived burden and survey participation, and how does it vary by establishment type or size? What methods are available to control response burden?

Although nonsampling errors in cross-sectional national surveys of U.S. establishments will be the focus, the paper will include some discussion of comparative error in multi-national studies of establishments and conditioning error and other time-related errors in longitudinal studies.

Brad Edwards is a Vice President at Westat where he has been focusing on large-scale surveys about health, health care use and costs, and long-term care for the past 25 years.

David DesRoches is a Senior Survey Researcher at Mathematica Policy Research where he has focused on designing surveys of businesses and young adult populations on issues surrounding postsecondary education for the past 9 years.