omit the coefficients on these for brevity. The logarithm of the independent variables are
used to eliminate heteroskedasticity in the data and also to allow easy interpretation of
elasticity values which are equal to the coefficient estimates. Note that the models with
medical technology proxies do not include data from the London region and hence have 180
observations rather than 200. Data from Scotland is not included as we did not have age
cohort data for Scotland.
Model A in Table 2 contains no proxies for medical technology improvements. Most
independent variables seem to be having no effect on total fatalities. Increased per capita
expenditure on alcohol is statistically significant at increasing fatalities. This is not surprising
as alcohol consumption has been shown to be related to increases in fatalities (Loeb, 1987).
Whetten-Goldstein et al. (2000), however, did not find a price index of alcohol consumption
to be significant, though they modeled fatality rates, rather than total counts. Increased
population over 65 years is statistically significant and associated with increased fatalities and
increased vehicle age is associated with reduced fatalities. This latter result is quite
surprising as we would expect newer vehicles to reduce the level of fatalities since they are
presumably safer (i.e., they may have newer safety features and also be better maintained).
This variable was somewhat correlated (-0.72) with the percent of the population aged 15-24,
but estimates with this variable omitted did not change the coefficient value which is fairly
robust.
Model B in Table 1 introduces proxies for medical technology improvements. Three
variables are tested. These are the average length of inpatient stay in the hospital which has
been declining over time and could represent better medical technology as discussed
previously. Per capita NHS staff is also included with increases in staff levels proxying for
better medical care. The number of persons per capita on waiting lists for hospital treatment
13