<abstract xmlns="http://www.w3.org/1999/xhtml"><p>Combining survey and auxiliary data to produce official statistics is gaining interest at federal agencies and among policy makers due to its efficiency. Recent studies have shown the practicality of small area estimation modeling approaches in the context of integrating data from multiple sources to improve estimation at fine levels of aggregation. In this article, agricultural predictions are constructed using a hierarchical Bayes subarea-level model, fit to data available from different sources. Auxiliary data are initially used to complement the survey data and define the prediction space, and then to define covariates for the model. Finally, not-in-sample predictions are constructed using the model output, and benchmarking constraints are imposed on the final set of in-sample and not-in-sample predictions. Unlike most of the studies discussing not-in-sample prediction, this article illustrates a method that uses the data available from multiple sources to define the prediction space. As a consequence, the resulting framework provides a larger set of nationwide predictions as candidate for official statistics, and extrapolation is not of concern. Challenges in developing the methods to combine different data sources are discussed in the context of planted acreage prediction.</p></abstract>

Combining survey and auxiliary data to produce official statistics is gaining interest at federal agencies and among policy makers due to its efficiency. Recent studies have shown the practicality of small area estimation modeling approaches in the context of integrating data from multiple sources to improve estimation at fine levels of aggregation. In this article, agricultural predictions are constructed using a hierarchical Bayes subarea-level model, fit to data available from different sources. Auxiliary data are initially used to complement the survey data and define the prediction space, and then to define covariates for the model. Finally, not-in-sample predictions are constructed using the model output, and benchmarking constraints are imposed on the final set of in-sample and not-in-sample predictions. Unlike most of the studies discussing not-in-sample prediction, this article illustrates a method that uses the data available from multiple sources to define the prediction space. As a consequence, the resulting framework provides a larger set of nationwide predictions as candidate for official statistics, and extrapolation is not of concern. Challenges in developing the methods to combine different data sources are discussed in the context of planted acreage prediction.

Statistical Challenges in Combining Survey and Auxiliary Data to Produce Official Statistics

, 1400 Independence Avenue, SW, Washington D.C., U.S.A.

Worcester Polytechnic Institute, Mathematical Sciences

Journal of Official Statistics

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Combining survey and auxiliary data to produce official statistics is gaining interest at federal agencies and among policy makers due to its efficiency....

Statistical Challenges in Combining Survey and Auxiliary Data to Produce Official Statistics

Published Online: Mar 17, 2020

Page range: 63 - 88

Received: Oct 01, 2018

Accepted: Sep 01, 2019

DOI: https://doi.org/10.2478/jos-2020-0004

Keywords
Administrative data, benchmarking, incomplete data, not-in-sample prediction, small area estimation

© 2020 Andreea L. Erciulescu et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Statistical Challenges in Combining Survey and Auxiliary Data to Produce Official Statistics

Published Online: Mar 17, 2020

Page range: 63 - 88

Received: Oct 01, 2018

Accepted: Sep 01, 2019

DOI: https://doi.org/10.2478/jos-2020-0004

KeywordsAdministrative data, benchmarking, incomplete data, not-in-sample prediction, small area estimation

© 2020 Andreea L. Erciulescu et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Keywords
Administrative data, benchmarking, incomplete data, not-in-sample prediction, small area estimation