The United States Department of Agriculture (USDA), National Agricultural Statistics Service, in conjunction with the USDA, Economic Research Service, conducts the three-part Agricultural Resource Management Survey (ARMS) to study the well-being of farm establishments. The survey instrument for the final part of ARMS (ARMS 3) is approximately 24 pages, detailed, and includes sensitive financial questions. Due to this, data collection relies heavily on expensive, in-person enumeration. Developing methods to limit the costs of data collection and decrease respondent burden while identifying possible biases is a complex task. This research explores three different strategies for identifying operations that will not be sent to field follow-up: (1) predicted impact on the calibration estimators, (2) predicted propensity to respond, and (3) using both predicted impact and propensity to respond. Simulations using prior rounds of ARMS 3 are used to evaluate the effect of these strategies on controlling cost, burden, and measures of data quality.