Evaluate Nonresponse Adjustments For The Quarterly Agricultural Surveys

Use Spatial Correlations Among Neighboring Counties To Model County-Level Yields

 
Based essentially on the methods used in the PASCAL program that James Bethel wrote for the Sample Design Section in 1985, section staff recently wrote an optimum sample allocation program in SAS that is currently being used operationally. This SAS program contains several additional features which the PASCAL program did not have. It is capable of finding the optimum allocation for a sample for which population totals, population means, and ratios estimators are used. 

Research is being conducted in cooperation with the Sample Design Section to improve the allocation program. Research is also being conducted to assess the importance of a simplifying assumption with the integrated surveys to overcome a technical difficulty arising from the multivariate nature of the allocation problem. The results of this assessment will constitute the second chapter of the documentation of the allocation program. The results of research in deriving an optimum sample allocation for the linear models used in objective yield surveys will form the basis of the third chapter in the documentation. The fourth chapter will be a user's guide for the allocation program. 

Important progress was made during 1998 in solving the problem of finding the optimum sample size for an objective yield survey, given the spatial partitioning of the objective yield region. The next step will be to derive an algorithm, like Chromy's or Bethel's, to generate a numerical solution. Once the algorithm is found and implemented into the SAS program, the documentation of the optimum sample allocation program will resume.

 
Currently, NASS imputes for nonresponse in the new crops/stocks design for the Quarterly Agricultural Surveys (QAS) using the same imputation program as for the current QAS. This approach will not be available when the current QAS ceases to exist (for one thing, it depends on the current stratification, but the new QAS is unstratified). Research has suggested a series of reweighting adjustments that does essentially the same thing as imputation within (to-be-determined) reweighting cells. Although conceptually similar to reweighting in the ARMS and hog surveys, the QAS imputation approach makes use of supplemental information such as whether a farm is in business, whether it has crop land, and whether it has any crops of interest. This research will assess the importance of these differences.

 
The Sampling and Estimation Research Section is working with the Ohio State University (OSU) in developing and testing a modeling approach that can potentially improve the quality of NASS' county estimates of crop yield. The approach uses existing data sources and incorporates correlations in yields among counties that share common borders. During 1997-98 the program was converted by OSU graduate student Ramzi Nahhas to SAS-IML to allow for easier implementation in the SSOs. Prof. Stasney and Ramzi Nahhas also traveled to the NASS Ohio State Statistical Office NASS state staff there to discuss how the data could be prepared for input into the NASS county estimate system. During the fall of 1998, the cooperators also began to study the possibility of using a multivariate approach for obtaining county yield estimates. Any resulting improvements to the algorithm will be incorporated as they are developed. 

During the latter part of 1998, section staffsuccessfully tested the algorithm with cotton data from Mississippi and oat data from Michigan to assess its usefulness for crops other than those tested in Ohio. 

Plans for 1999 are to set up the algorithm for use in the Michigan SSO to assist in the actual 1998 county yield estimation procedures. The results of this implementation and feedback from the Michigan and Ohio SSOs will then be used to assist in determining the best strategy for implementation across the agency, and a plan will be developed.

Top of the Page