Regression Metamodels for Sensitivity Analysis in Agent-Based Computational Demography

Abstract

Agent-based computational simulation models can be complex and this can make it difficult to understand which processes are driving model behaviour. Sensitivity analysis by means of metamodels can greatly facilitate the understanding of the behaviour of complex simulation models. However, this method has so far largely been neglected in agent-based computational demography, with few exceptions. In this chapter, I illustrate how sensitivity analysis can be conducted by means of regression metamodels. I argue that this type of metamodel is particularly attractive for use in agent-based computational demography due to the fact that most demographers have at least a basic understanding of multiple regression. This makes this type of metamodel highly accessible and easy to communicate. After describing the basics of regression metamodels, I illustrate their use by conducting a sensitivity analysis of an agent-based model of educational assortative mating that is based on data on the structure of Belgian marriage markets between 1921 and 2012. I close the chapter with a discussion of the benefits and limitations of regression metamodels and point the reader to further readings on this topic.