|
Abstract
|
This first part of the state-of-the art focuses on the origins of road safety modeling, covering data, early models and the public health context of model formulation and use.
Yearly tallies of road victims by severity category, typically computed nowadays from police reports, emerged over time in many countries from systematic determination by Administrations of Justice of the non-criminal nature of reported individual road crash and damage events. Such data series on retained "accidents", as available over the last 150 years (notably in France), imply very important gains in kilometric safety rates over time with the replacement of horse-drawn carriages by motor vehicles and with the spread of motor vehicles themselves.
However, multivariate statistical analyses reaching beyond two-way frequency tables are recent: aggregate national fatality rates were first modeled (as Gaussian distributions with a regression component) by Smeed in 1949, but morbidity rates were then neglected; and samples of discrete occurrences of individual accidents (of any severity) were first modeled (as Poisson distributions with a regression component) by Weber in 1970, but without concern for national population values.
These seminal single-outcome models gave rise to two streams of explanations, distinct to this day, that share a "public health" epidemiological emphasis on the establishment of multiple correlations which give rise to testable corrective policy interventions. These are still of limited value in the explanation of the simultaneous peaking of fatalities in many OECD countries in 1972-1973 (called "the Mystery of 1972-1973", hypothesized here to be occasioned by the passing of the demographic baby boom wave) and as guidance in the design of policies for the containment of road risk arising from the intrinsic dangerousness of individuals. |