In addition to global warming, congestion, geopolitical costs, oil spills, and health problems like asthma and allergies, we now have another externality to gasoline consumption to justify a pigouvian tax: drunk driving. A new study by Chi et al (6 co-authors!) uses data from Mississippi to show that lower gas prices are related to drunk driving related accidents. The authors claim the study is the first to examine this relationship, and is important because from a theoretical perspective the relationship could be positive or negative.
The ways that gas could inversely relate to drunk driving are obvious: lower prices make it cheaper to drive and give people more disposable income, which means it’s less expensive go out drinking and driving and people have more money to do so. In addition, the marginal cost of driving a to a farther away bar decreases. Also, higher gas prices may cause people to shift to different modes of transportation, like walking or taking the bus, which (freakonomists aside) are less likely to result in drunk driving accident.
A positive relationship is less obvious, but could result if gas prices increases enough that the negative wealth effect (more expensive gas makes you poorer) is severe enough that it creates economic hardship, which can lead people to drink more.On the face of it, the positive relationship seems much less likely than the negative relationship, and this is what the empirical evidence found in this study suggests. The chart below shows the indexed values of drunk driving accidents and gas prices.
These results increase the growing gap between the nominal price of gas and the true cost of it, and strengthen the case for a pigouvian tax… not that externalities, efficiency, or empirical realities seem to matter much in the political debate on this issue.
A caveat though: these results should not be taken as dispositive but rather suggestive. The empirical analysis is pretty simple, does not get into a really serious attempt to examine causality, and has some fairly serious omissions. For instance, the authors do not control for weather in their analysis. They mention that it would be difficult to aggregate to the monthly level, but I think average temperature would probably suffice. Second, and relatedly, they do not control for seasonality. This is pretty important in a time-series context where you are very likely to see both drunk driving and gas prices increase in the summer and decrease in the winter. Finally, and this is a more minor econometric point, they choose between a poisson and negative binomial regression models by selecting the one with the higher log-likelihood, which I do not believe is a sufficient means to determine whether there is enough overdispersion in the data to warrant the use of negative binomial over poisson. More importantly, they don’t tell us whether the use of negative binomial, or OLS for that matter, affects the results compared to poisson, which would have taken 30 seconds to determine and would tell us something about the robustness of their econometric results. Given that the relationship between prices and accidents disappears for males when the analysis is partitioned by gender, it is not hard to believe that the results are potentially not robust.
All that said, the authors do argue that nobody has empirically examined this issue before, and the results are highly theoretically believable. In fact, I find the theory alone strong enough to conclude that a relationship is likely. At the very least when thinking about gas prices we should consider that drunk driving may be yet another cost of low gas prices, and this study should definitely be enough to prompt more research into this.