All Times ET
Program is Subject to Change
TBD
Getting it Right: The Accuracy of Sampling Frame Data and Predicting Survey Contact (308196)
Stas Kolenikov, Abt Associates*Julie Pacer, Abt Associates
Marci Schalk, Abt Associates
Keywords: sampling frame data accuracy, survey contact, survey eligibility, Human Resources benefits administrators, Dun & Bradstreet, logistic regression, random forest
Establishment surveys are susceptible to a unique set of challenges compared to household surveys. Unlike a household survey, an establishment survey collects information from a representative of a business who speaks for that business, usually during business hours. Nonresponse is a considerable factor in establishment surveys. The sampling frame may lack a contact name at the sampled business or it may provide outdated contact information. In attempting to reach a business by phone, the rise of automated menus makes it challenging to reach a person to identify a respondent. Furthermore, depending on the role of the intended respondent and size of the business, the intended respondent may not respond to a survey invitation due to competing work duties.
Abt Associates conducted the Family and Medical Leave Act (FMLA) Wave 4 Surveys for the U.S. Department of Labor, which includes a survey of employers and their Human Resources benefits administrators regarding their establishments’ use and understanding of the FMLA. The screening portion of the survey verified the existence of the business and identified a respondent. The survey utilized the Dun & Bradstreet Market Identifiers file as a sampling frame, and the auxiliary frame data allowed for an in-depth analysis of survey eligibility and response propensity based on sampling frame information. In this paper, we determine the accuracy of this sampling frame information when we had contact with the establishment, particularly contact information, job title, state, business size, and industry. We also use sampling frame variables to predict survey contact using logistic regression and random forest models. The results will inform recommendations for future use of the sampling frame in establishment surveys.