Stability of feature network modeling of proximity data

Laurence Frank

Project: Project at Leiden University

Project running from: 1 April 2001 – 1 January 2006

Supervisor: Prof. dr W.J. Heiser

Summary:
An unresolved problem in network modeling of proximity data is how to determine the stability of the parameter estimates. Existing methods to derive a network model from empirical data only gives the best fitting network, and yield no standard errors. This project approaches the problem by trying to profit from two known additivity properties of networks, in particular segmental additi-vity and distinctive feature additivity. The former property holds in all networks, while the latter is specific for networks based upon a set of (known or unknown) features. Both forms of additivity open the possibility to apply existing distribution theory for the linear model (under order con-straints). Since the theoretical standard errors may be too optimistic, a simulation study will be performed to check the size of theoretical simultaneous confidence regions against a simulated sampling distribution.

Date of defence: 21 September 2006

Title of thesis: Feature network models for proximity data: Statistical inference, model selection, network representations and links with related models. ISBN 90-8559-179-1.