Sample Design and Estimation When Using a Web-Scraped List Frame and Capture-Recapture Methods

Surveys are often based on a sample drawn from a list frame. In recent years, the percentage of target population units on the list frames has been decreasing, making it important to adjust for this undercoverage in the estimation process. Multiple-frame methods generally assume that the union of the available list frames is equal to the target population; however, this assumption is often not satisfied, especially for hard-to-survey populations. The United States Department of Agriculture’s (USDA’s) National Agricultural Statistics Service has explored the use of web-scraped list frames to assess undercoverage of the NASS list frame, which is comprised of all known farms and potential farms in the USA. In 2020, NASS conducted the National Farmers Market Mangers Survey. Because NASS does not include farmers markets on its list frame, the USDA Agricultural Marketing Service (AMS) business register of farmers markets was the only list frame initially available. To assess its undercoverage, a web-scraped list frame was developed, and capture-recapture methods provided the foundation for estimation. This study made two advances in the use of capture-recapture methods when conducting a survey with two list frames. First, because record linkage was conducted prior to drawing the samples, the sample design incorporated information identifying records on only the AMS business register, on only the web-scraped list frame, or on both frames. Second, a composite estimator for this overlap design allowed full use of all sample information to produce survey estimates. Directions for future research are highlighted. Supplementary materials accompanying this paper appear online.