Supervised network prediction for household statistics

Producing census-like household statistics is possible based on relevant data originated from the administrative sources. However, address registration errors generally cause biased results. Under the assumption that households can be identified by survey at a sample of addresses, we develop a supervised network prediction approach to households at the out-of-sample addresses. An application to real data from the Norwegian Household Register is used to demonstrate the advantages of the proposed modelling approach over the existing practical alternatives.