EXAMINING EACH PREDICTOR SEPERATELY

The first stage in this example is to produce a generalized additive fit separately for each of the three predictor variables, thus enabling us to get some idea of the relationship between kyphosis and the predictor.

This can be done using statistical software, such as Splus. The smoothing method favoured by most statistical software is spline fitting.

The graphs below show plots of the data against the fitted values

The black lines at the bottom of each plot indicate where the data values lie.
On the plot for the variable number, a single observation accounts for about half of the vertical displacement which strongly influences the shape of the curve. This observation might be removed on the grounds of being unusual. The same is happening to a lesser extent on the age plot as well.

EXAMINING EACH PREDICTOR SEPERATELY

Generalized Additive Models 23.4.96 Page : 11c of 15