Abstract:
|
The National Agricultural Statistics Service (NASS) of the United States Department of Agriculture (USDA) produces over 400 publications annually. NASS uses a process of estimating propensity score of a survey response that employs multiple data sources, including USDA agencies and the Census Bureau. In an effort to assist regional field offices (RFOs) and increase response rates, several new propensity score models of survey response and nonresponse were examined. After multiple modeling attempts using methods such as logistic regression, bootstrap forest, and boosted regression, the team determined that bootstrap forest models outperformed other methods. This paper outlines two models: Model 1 (more expensive) estimates the probability of response by field enumeration and; Model 2 (less expensive) estimates the probability of response by mail or computer assisted telephone interviewing (CATI). The potential for using these models with measures of impact as part of the data collection strategy to increase response rates of NASS surveys and increase overall data collection efficiency is discussed.
|