This article comments on the paper by mann and anglin published in accident analysis and prevention (1988) 20, p441-6 (see IRRD 818253). It was found in this paper that there was no significant correlation between per capita rates of alcohol related accidents and per capita consumption of alcohol (pcca) alone. However if the rates of non alcohol related accidents were included in the statistical models, the relations became significant. A closer look at this data suggests that those relations are due to common time trends and not to more specific relations between pcca and alcohol related accidents. To explore this possibility regressions have been calculated with single vehicle accident (smf) (and night time fatal accidents nf) per 100000 population as the dependent variable, and year and fatal multivehicle accidents (mmf) and fatal daytime accidents (df) per 100000 population as independent variables, corresponding to mann and anglin's independent variables pcca and mmf (and df). The t-values are very cloe to those of mann and anglin's model. When the effects of time trends are eliminated by replacing each variable by its deviation from its time trend, the t-values of the coefficients for the accidents not related to alcohol remain high, whereas those for pcca become insignificantly small. Another point of concern is that the dependent and the independent variables used by mann and anglin have population in the denominator, which can create strong spurious correlations.
Abstract