3.2. Variable selection process
A forward selection approach was considered for this study to include the variables in the model. The basic model was set up considering the intercept and the general traffic volume variable. All other variables were added to the model one by one. The key assumption of this . Model goodness of fit
Model goodness-of-fit was tested and compared using the McFadden pseudo R-squared value (McFadden, 1973). The McFadden pseudo R-squared is estimated as follow:
equation(3)
View the MathML source
Turn MathJax on
where LL(β) is the log-likelihood value of the full model and LL(C) is log-likelihood value of the constant only model.
The another measure, View the MathML source, proposed by Miaou et al. (1996) was deployed in this study to determine how well the variance of data is captured by the model relative to a fundamental model with no variables (Shahla et al., 2009). This measure takes the NB dispersion parameter and is estimated as follows:
equation(4)
View the MathML source
Turn MathJax on
where α is the estimated over dispersion parameter for the selected model and αmax is the estimated over dispersion parameter for the fundamental model containing only constant term.