Assessing Nonresponse Bias in the Green Goods and Services Survey

• Activity Number: 88
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 4:00 PM to 5:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Brian Meekins*+ and Michael Sverchkov and Justin McIllece
• Companies: Bureau of Labor Statistics and Bureau of Labor Statistics (SCI) and Bureau of Labor Statistics
• Keywords: Nonresponse ; Not Missing at Random ; Green Survey

Abstract

In late summer of 2011 the Bureau of Labor Statistics fielded one of two surveys designed to assess the number of "green" jobs in the United States, the distribution of these jobs by industry, occupation, and geographic area, and the wages of those working in "green" jobs. The survey will end data collection in early 2012 with a likely response rate of approximately 70% (or 65% of the sampled employment).1 The frame from which the sample of 120,000 was drawn (the Quarterly Census of Employment and Wages Longitudinal Database) contains information on establishment employment and age, industry, geographic location, as well as wage information and other economic information reported in the past. In addition, a total of 13,000 establishments were identified as likely producers of a green good or service. This work uses logistic regression to model response probability utilizing both auxiliary and design variables. It is suspected that many important estimates, such as the number of green jobs, produced from this survey are likely related to response - that is, they are not missing at random. We suggest simple procedures to test whether the response is missing at random or not.