Theory and application of structural equation modeling in spatial data analysis

PhD ceremony: Ms. A. Liu, 11.00 uur, Academiegebouw, Broerstraat 5, Groningen

Title: Theory and application of structural equation modeling in spatial data analysis

Promotor(s): prof. H. Folmer

Faculty: Spatial Sciences

 

This thesis compares by means of Monte Carlo simulations two different approaches to capture spatial dependence in regression models: the classical W-based autoregressive model and the structural equations model with latent variables (SEM). In addition, it includes an application of SEM to a spatial cross section data set on female labor market participation.

Based on the Monte Carlo simulation studies, the choice of the spatial weights matrix to represent spatial dependence turns out to be a key issue in the performances of the W-based models as it imposes a priori a structure of spatial dependence. By its very nature SEM is more flexible in that it allows the capturing of spatial dependence by means of various indicators, the relevance of each of which can be tested. SEM is definitely a preferable alternative to the W-based models in the case of large samples and when there is evidence of substantial spatial dependence.

The empirical labor market study using the SEM approach identifies the following positive determinants of female labor market participation at the municipal level in the Netherlands: women aged 35-45 (%), male unemployment rate, spatially lagged female unemployment rate, female-dominated sectors and socio-economic status. Female unemployment rate, spatially lagged male unemployment rate and demographic pressure are found to have negative impacts.