Structural equations modelingIn order to test the influence of built environment on walking, and to test the mediating role of
accessibility in this relationship, we adopted a structural equations modeling methodology.
Structural equation modeling (SEM) has been applied in travel behavior research since the 1980s
(Golob 2003), although more recently its adoption has been more pervasive (see for instance
Bagley and Mokhtarian 2002; Abreu e Silva, Golob, and Goulias 2006; Cao, Mokhtarian, and
Handy 2007; Van Acker, Witlox, and Van Wee 2007; Aditjandra, Cao, and Mulley 2012),
probably due to the availability of several statistical software. SEM is a confirmatory statistical
method, based in linear relationships among variables, that allows testing substantive theories,
explicitly taking into account potential errors of measurement in all observed variables, including
the independent variables. In SEM it is possible to include not only observed variables but also latent variables, i.e. variables for which there are no available observations, but manifest
themselves in other (several) observed variables. Four types of models can be developed with
SEM: path analysis models, conceived only of observed variables, confirmatory factor analysis
models, employed to analyze relationships among latent variables, structural regression models,
which combine path analysis with confirmatory factor analysis, and finally latent change models,
developed to study change over time (Raykov and Marcoulides 2006).