Pairing for Generation of Synthetic Populations: the Direct Probabilistic Pairing method

Abstract: Methods for the Generation of Synthetic Populations do generate the entities required for micro models or multi-agent models, such as they match field observations or hypothesis on the population under study. We tackle here the specific question of creating synthetic populations made of two types of entities linked together by 0, 1 or more links. Potential applications include the creation of dwellings inhabited by households, households owning cars, dwellings equipped with appliances, worker employed by firms, etc. We propose a theoretical framework to tackle this problem. We then highlight how this problem is over-constrained and requires relaxation of some constraints to be solved. We propose a method to solve the problem analytically which lets the user select which input data should be preserved and adapts the others in order to make the data consistent. We illustrate this method by synthesizing a population made of dwellings containing 0, 1 or 2 households in the city of Lille (France). In this population, the distributions of the dwellings' and households' characteristics are preserved, and both are linked according to statistical pairing statistics.