Abstract
Agent-based microsimulation models of socioeconomic processes require an
initial synthetic population derived from census data. This thesis
builds upon the Iterative Proportional Fitting (IPF) synthesis procedure,
which has
well-understood statistical properties and close links with log-linear
models. Typical applications of IPF are limited in the number of
attributes that can be synthesized per agent. A new method is introduced,
implementing IPF with a sparse list-based data structure that allows many
more attributes per agent. Additionally, a new approach is used to
synthesize the relationships between agents, allowing the formation of
household and family agents in addition to individual person agents.
Using these methods, a complete population of
persons, families, households and dwellings was synthesized for the Greater
Toronto Area and Hamilton.