Integrating census and administrative data sources within a spatial sampling framework

The use of administrative data based official statistics has extraordinarily increased in the last few decades. As administrative data exist as a product of systems designed for other primary purposes, it is particularly important for statisticians to ensure that the resulting statistics meet the quality users’ needs. These include evaluating the coverage error that is one of the main sources of the bias when producing administrative based statistics. The coverage errors are traditionally addressed through a coverage survey called the Post-enumeration Survey (PES). The PES is an independent survey of enumeration areas, which measures the accuracy of the population count. In this paper we address the problem by estimating the under-coverage through a spatial sampling carried out on the enumeration areas. We computed the relative Root Mean Squared Errors (RRMSE) of the estimates obtained by using some Spatially Balanced Designs and we used as a benchmark those obtained with a classical sampling design: the Simple Random Sampling (SRS). The results show that the gain obtained by using these sampling designs is not trivial. The same experiment was conducted both with constant first order inclusion probabilities and with probabilities proportional to the size of the enumeration areas (PPS).