Authors:Natalia Rojas-Perilla (Institute of Statistics and Econometrics, Freie University Berlin, Germany) (Speaker) Timo Schmid (Freie Universität Berlin, Germany)Paul Smith (Southampton Statistical Sciences Research Institute, University of Southampton)Nikos Tzavidis (University of Southampton)Jan van den Brakel (Statistics Netherlands and Maastricht Unversity School of Business and Economics)
National Statistical Institutes (NSIs) conduct repeated sample surveys with the aim of analyzing change over time. Although NSIs try to maintain consistent survey design methodologies, modifications and redesigns of long-standing survey processes are sometimes necessary. Redesigning a survey can affect non-sampling errors and therefore can lead to systematic differences on survey estimates over time. These systematic differences are called discontinuities. Separating real changes from discontinuities due to the redesign of a survey is important for maintaining uninterrupted time series of estimates. Part of the transition to the new survey is to provide users with information on the likely impact of the changes to the survey design on existing estimates, and to help them to use this information to interpret the changes. The objective of this presentation is to outline some of the challenges with estimating survey discontinuities. More precisely, discontinuities can be estimated at national and subnational (domain) levels using direct and indirect estimation. SAE methods are used with the aim of improving the accuracy of domain estimates. As the auxiliary information is typically available at domain-level, the Fay-Herriot model is used. Our work involves direct and indirect estimation of discontinuities at various domain levels using real survey data sets from Wales. In particular, five separate surveys in Wales have been combined into a new National Survey which began in 2016. This has lead to concerns about the presence of discontinuities in survey estimates of key attributes. We discuss how estimated discontinuities and corresponding uncertainty estimates can be used to assist users with interpreting the potential impact of changes in the survey design. We further discuss some of the challenges posed by model-based estimation.