A Comparison of Graphical Models and Structural Equation Models for the Analysis of Longitudinal Survey Data

Graphical chain modelling (GCM) and structural equation modelling (SEM) are two approaches to modelling longitudinal data. Both approaches have their origins in path analysis and provide pictorial representations of the association between variables, which are usually ordered temporally. Both methods also aim to identify the direct and indirect effects of one variable on another.While the SEMapproach specifies a singlemodel for the complete system of variables being studied, the GCM approach permits a model for the complete system to be built up by fitting a sequence of submodels. In this chapter we briefly discuss the similarities and differences of the GCM and SEM approaches to modelling univariate recursive systems and their application to complex survey data. We identify the strengths and limitations of each approach. By using an example of the relationship between changes in labour force status following entry into parenthood, and changes in gender role attitude, we illustrate their use with survey data. A sample of 632 women, childless and aged between 16 and 39 in 1991, taken from the British Household Panel Survey (BHPS), is used. This survey is a particularly useful resource for this type of analysis since it is nationally representative, continuous since 1991 and collects a wide range of socio-demographic and attitudinal information (Taylor et al., 2007). Information on employment and parenthood status is obtained annually and attitude statements concerning gender roles are collected biennially in 1991, 1993, 1995 and 1997.