We also see various other effects in the injury models that occur both with and
without inclusion of the medical technology variables (models C and E). Increased vehicle
age is statistically significant in the serious injury models but not in the on slight injury
models. For serious injuries the effect is opposite that of average vehicle age on fatalities.
Increases in the motorway and trunk road network are now significantly associated with
increases in serious injuries, though the level drops when medical technology variables are
included (especially the effect on slight injuries). Interestingly, as the fraction of the network
which is “other roads” increases, serious injuries decrease with no effect on slight injuries.
Increased motorway length per area decreases injuries but this effect again disappears when
medical technology variables are included. There is no collinearity amongst these variables
with the medical technology variables so the associations appear quite robust.
Other control variables in the injury models include alcohol expenditure which is
positive and significant and population which is also positive and significant. Increased GDP
per capita appears to be associated with reductions in injuries, though this effect is not robust
across different specifications of the model. Age effects are also not clearly robust across the
specifications, although increased population in the 45-64 age range does seem to have an
association with reductions in both injury categories.
Additional evidence on the effect of the medical technology proxies can be found by
examining the time trend variables. These were included in the models as dummy variables
and results are not shown due to space limitations. In all cases the coefficient values of the
year dummy variables are reduced in magnitude when the medical technology proxy
variables are included. This would suggest that these variables are picking up some of the
residual time trend associated with reduced fatalities. Models with a time trend variable
rather than a year dummy variable were also estimated and similar results were found. These
15