Abstract:
|
In-person surveys have traditionally been reliant on field enumeration (FE) to create a sampling frame. Although FE achieves exceptional coverage, an alternative frame construction method, address-based sampling (ABS), is less time-consuming and may be less expensive. The coverage of the ABS frame is high, but the undercoverage is not random. In this paper, we estimate the extent to which the frame's undercoverage introduces coverage bias in a national survey. Three datasets were created using data from the 2015 and 2016 National Survey on Drug Use and Health. The first dataset included respondents from an FE frame, and the other two datasets included those that might be covered under different ABS frames. Fifteen drug use and mental health prevalence estimates were compared across the three datasets to identify potential coverage bias across several domains which may be at risk for undercoverage.
|