Spatial boundary changes overtime: small area estimation approach to maintain compatibility of data


Accepted presentations cancelled by authors

Author: Thanyani Maremba (Statistics South Africa and University of Limpopo)

Sample survey data are mainly collected based on fixed boundaries, however, countries such as South Africa experience frequent administrative boundary changes causing difficulties in producing comparable statistics through time. Rindfuss et. al (2004) stated that unless a consistent geographical approach is taken with time-series data, it cannot be known whether changes in the relationships between variables collected for areas at different points in time are real or an artefact of boundary changes. The reality from survey design perspective is that mapping the sample to changing boundaries always leads to some areas losing the sample while others gain. The sample is designed to meet certain precision and as a result of these changes the precision is directly affected. There are often no resources available to collect data with new boundaries immediately. Boundary changes are the natural cause for deviation from the sample design. Much research has focused on the modifiable areal unit problem and on custom zone design, but the practical difficulties created by temporal inconsistencies in zonal boundaries have received less consideration. Mapping sampling units either for the areas where the ultimate sampling units are located or linking geocoded sampling units to changing boundaries are some of the solutions applied by statistics agencies. The paper propose the approaches that link sampling units to new boundaries and take into account deviation from sample design in order to produce unbiased estimates for local areas such as municipalities and wards. Small area estimation (SAE) has capability of deviating from the design in order to achieve estimates for the required domains after implementation of the sample design. The approach followed makes use of small area estimation methods that can be employed to adjust the estimates overtime for both planned and unplanned domains. While sample survey data are fixed researchers may need to obtain same estimates at different point in time using different boundaries in local areas such as municipalities and wards. The boundary changes decisions are independent from the sample design and as a result the changes violate strata formed during design. The SAE methods are to be used to estimate selected target variables from the South African community survey 2016. Small area estimation techniques employing spatial micro-simulation methods specifically generalised regression methods will be used to reallocate samples to new boundaries, adjust weight relating to target variables and also provide for estimation of reliability. Similar approach is also proposed for areas that were never catered for during the sample design such as local municipalities, electoral wards and other defined domains.