Finding a parametric approximation for the start variable is slightly
more complicated as the fit is not quite as straightforward as that for
age.
There are a couple of possibilities for a parametric fit
A polynomial fit as with age . This time the earlier graph tends
to suggest a cubic transformation, as opposed to a quadratic one.
There is a natural division of the variable start into two
groups start < 12 and start >12 . The Splus term
(start-12)*I(start>12) will result in a constant for start
< 12 , joined continuously to a curved fit for start > 12.
A graph of the non-parametric fit seen earlier and of the polynomial fit can
be seen below
Non-parametric additive
fit
Parametric approximation
Generalized Additive Models 23.4.96 Page : 12c of 15
![[Link Map]](../../icons/smartnav.gif){kind=icon}
