Synthesizing Individual Data with Positional Information for All Households in Japan

This study presents a framework to generate nationwide, geolocated synthetic microdata (synthetic households and individuals) for Japan from publicly available statistical tables, aiming to support real-scale social simulation for policy evaluation (e.g., disaster mitigation and economic measures) at an individual-citizen level. Given the limited accessibility of administrative microdata due to privacy constraints, we adopt a sample-free approach based on simulated annealing (SA) to minimize discrepancies between synthetic tabulations computed from the microdata and multiple official tabulations. The generation process is organized as a sequence of methods: synthesizing core demographic and household attributes (family type, age, sex, and household roles), adding housing-related attributes (tenure and housing structure), assigning intra-municipality small-area attributes, incorporating building information to attach realistic residential locations, and enriching individuals with socio-economic attributes such as employment status, industry, work style, firm size, and income. We generate multiple synthetic datasets for several census years (2000–2015) at the scale of roughly 1,900 municipalities and report representative error levels and SA settings, providing practical guidance for constructing and distributing large-scale, multi-attribute, geolocated synthetic microdata for real-scale agent-based simulations.