17.4.4 Stepwise Poisson Regression Model
The t-values tkℓ can be compared to a t-distribution to determine whether these values
are significantly different from 0. In a so-called stepwise model, this significance
testing is used for attribute selection, finally resulting in a model only having significant
attributes. The stepwise model approach starts off with a model containing
all attributes. Then, each time the most insignificant attribute is deleted from the
model, until the model only contains significant attributes. Note that this stepwise
approach leads to a different model when all insignificant attributes are deleted at
once. In fact, due to collinearity, significance of an attribute may change when another
attribute is deleted. When using the stepwise model to determine weights wk,
we consider Lk to be the number of dummies incorporated in the final model. This
stepwise model is used in the application shown in Section 17.6.