Responsiveness and Representativeness in an Establishment Survey of Manufactures
Eric Fink
U.S. Census Bureau
Joanna Fane Lineback
U.S. Census Bureau
Response rates have traditionally been used as data-collection-quality metrics. However, research has cautioned against solely relying on response rates (Groves, 2006). R-indicators have been proposed as a corresponding measure that can give insight into the data collection process that response rates alone cannot explain (Schouten and Cobben 2007). We calculate response rates and R-indicators for the 2011 Annual Survey of Manufactures and demonstrate that when used in conjunction with each other they can give a more complete picture of the data collection process, particularly the nonresponse follow-up. We show that despite increasing response rates during the nonresponse follow-up, representativeness across important design variables decreases, owed in part, we hypothesize, to concentrating follow-up on those establishments expected to contribute the most to total estimates. This lack of representativeness is a possible source of bias in resulting survey estimates if nonresponse adjustments do not correct for over or underrepresented areas. We discuss the tradeoff of reducing sampling variability versus reducing nonresponse bias.