Contribution of life cycle events to the peaks
To examine the contribution of an explanatory variable on the peaks,
the variable is omitted from the preferred homogeneous main effects
model and the simplified model is refitted. The probability of migration
is plotted against age, with the year set to 1985, duration to 10 years
and all the explanatory variables set to zero, as before.
The following graphs show the effects of removing in turn msb2,
mrm and ch3 from the full model. Similar graphs may be
drawn for the other explanatory variables.
Figure 1:
The effect of removing msb2 (marital status)
The basic shapes of the graphs are very similar, suggesting just a scaling
effect, and no explanation of the peak.
Figure 2:
The effect of removing mrm (remarriage)
The peaks seem to be slightly attenuated in the full model with mrm=0
compared to the simplified model.
It appears that the minor difference between the graphs is not just a scaling
effect, but evidence that remarriage contributes to the third peak.
Figure 3:
The effect of removing ch3 (children aged 15-16)
This variable appears to provide a partial explanation for
migration behaviour in the age range 35 to 50 (the appropriate age
for parents of children aged 15-16). The trough around age 40
with ch3 excluded from the model is partially smoothed out in
the full model with ch3=0.
However, although having a child aged 15 to 16 does significantly
reduce the probability of migration, the third peak is not attenuated
in the full model, but is in fact increased, for those without children
in this age range. This effect therefore does not explain the third
peak.