This was a brief talk covering the current status of research into probabilistic population synthesis, intended to be used for pre-conditioning for agent-based models.
There is an ever-increasing need for evidence based governance and policy-making. Oftentimes, the data available for such decisions is incomplete and many sources need to be combined. In addition, in order to understand risks, a proper measure of uncertainty about whatever population is desired. One solution to quantifying uncertainty about a population given various different data sources is to use all available information to create synthetic populations. We propose a high-fidelity probabilistic framework for population synthesis as an alternative to deterministic approaches such as iterative proportional fitting. This allows for multiple different synthetic populations to be created, upon which any relevant model can be fit, in order to make decisions. The probabilistic creation of the synthetic populations propagates uncertainty into the populations, resulting in the ability to describe uncertainty. We first explore a synthetic toy example for comparison to other methods, and then provide a working example using real data from Arlington, VA.