Use this URL to cite or link to this record in EThOS:
Title: Managing uncertainty in agent-based demographic models
Author: Hilton, Jason
ISNI:       0000 0004 6347 7113
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Population-level patterns are the object of formal demographic study, but result from thousands of decisions made by individuals who both act and are affected by the actions of others. These properties can lead to difficulties in determining the mechanisms underlying population change, not least because of the possibility that human population may display complex properties, where macro-level patterns are not reducible to the sum of their parts. Simulation methods may help overcome some of these analytical difficulties. Agent-Based Models are simulations which explicitly represent the behaviour of individuals and their interactions, and allow population patterns to emerge from such interactions. Agent-Based Models are attractive to demographers because they allow the formalisation of theories about links between individual-level behaviour and interaction on one hand, and macro-level population patterns on the other. However, such simulations are notoriously difficult to analyse and calibrate; they tend to involve many free parameters, and include several sources of uncertainty. This thesis investigates how the application of techniques from the field of the design and analysis of computer experiments can be fruitfully applied to these problems. More specifically, three demographic simulations are used to demonstrate the utility of Gaussian process emulators for this purpose. Firstly, a replication of an existing demographic agent-based model is analysed using heteroskedastic emulators. Secondly, two emulator-based methods are trialled for their effectiveness in calibrating a microsimulation. Finally, Gaussian processes are used to analyse and calibrate a agent-based model of intergenerational fertility patterns against empirical observations. These examples demonstrate the ability of Gaussian process emulators to flexibly capture non-linearities in the relationship between simulation inputs and outputs, and to coherently account for uncertainties. These properties mean that they are well suited to the problem of analysing and calibrating Agent-Based models. In the concluding chapter, thoughts are offered on strengths and limitations of the techniques in comparison to other methods, and directions for further work are suggested.
Supervisor: Bijak, Jakub ; Bullock, Seth Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available