Skip to content
 

The accident externality from driving: an extra $2500/year per driver

Aaron Edlin just sent me this article by Pinar Karaca-Mandic and himself from 2006:

We [Edlin and Karaca-Mandic] estimate auto accident externalities (more specifically insurance externalities) using panel data on state-average insurance premiums and loss costs. Externalities appear to be substantial in traffic-dense states: in California, for example, we find that the increase in traffic density from a typical additional driver increases total statewide insurance costs of other drivers by $1,725-$3,239 per year, depending on the model. High-traffic density states have large economically and statistically significant externalities in all specifications we check. In contrast, the accident externality per driver in low-traffic states appears quite small. On balance, accident externalities are so large that a correcting Pigouvian tax could raise $66 billion annually in California alone, more than all existing California state taxes during our study period, and over $220 billion per year nationally.

Interesting stuff. I don’t have it in me right now to check all these numbers, but the argument looks to be laid out clearly enough that the experts in the area can work it out. Also, it all seems to be about accidents to other cars; I’m not sure where they factor in the costs due to running over pedestrians.

I’d also make a few suggestions about the display of inferences, the usual recommendations: rescaling (a coefficient of .00042 for traffic density suggests that a transformation is in order), control of significant digits (6535.20 +/- 3779.90 is a bit of overkill, no?), and the rest. The “$1,725-$3,239 per year” thing is particularly silly, being right in the abstract and all.

At the end they give their policy recommendations:

The most administratively expedient Pigouvian tax would be a gasoline tax since states already have such taxes. And, importantly, gas taxes would bring the uninsured into the payment system. . . . Many political watchers will doubt, though, that Americans will accept any policy that substantially raises the cost of driving. Gasoline taxes, for example, remain quite low in the United States compared with Europe. Surprisingly, there is a potential second-best compromise . . . The body politic has accepted mandatory insurance, so why not also require insurance companies to quote premiums by the mile instead of per car per year? . . . This simple change in pricing structure could reduce driving substantially by moving a fixed cost to the margin without raising the overall cost of driving. . . . An extremely valuable aspect of a requirement of per mile premiums is that it takes advantage of the fact that current insurance premiums account for heterogeneity in risk. As a result, those in highly dense areas and those with poor driving records would face the highest per mile rates and would reduce driving the most, creating a doubly large reduction in accidents–exactly as a social planner would wish.

It’s an interesting idea: some costs are per-gallon and don’t depend much on where you live–here I’m thinking of the costs of cleaning up the pollution, securing the oil supply,etc.–and some costs are per-mile and depend on where you live and what kind of car you drive–paying for accidents, traffic cops, street paving, injuries, deaths, and so forth. I know just about nothing about tax policy, but my impression is that taxes are usually set based on some combination of willingness to pay and revenue needs, not directly based on the costs associated with the activity that is taxed. It seems like a good idea, though. I’d be interested to know how Edlin and Karaca-Mandic place the relative importance of costs from accidents compared to all these others. Also is there some suggestion that some large part of the tax structure could be redone based on costs of externalities?

7 Comments

  1. Daniel Lakeland says:

    At first this seemed obviously farcical, until I realized that they were talking total cost, not cost per person.

    With almost 40 million people, and probably at least 30M drivers, we're talking about the marginal driver increasing insurance costs to all drivers by 3e3/3e7 = 1e-4 = .0001 dollars per driver per year.

    I agree, rescaling is key here. On my theme of nondimensionalizing models, I would say they should estimate something like insurance externality cost per additional yearly average miles driven as a fraction of average gas cost per additional yearly average miles driven.

    To put that more succinctly they should estimate EM/ND

    Where E is total externality cost of the marginal driver driving N miles (in Dollars), N is the number of miles driven by the average person in a year, D is current cost of gas in dollars per gallon, and M is the average mileage of a car on the roads in that stat in Miles Per Gallon. The overall ratio is nondimensional if I'm not making a mistake, and it has a direct interpretation as externality cost as a fraction of fuel cost for the marginal driver.

  2. Daniel Lakeland says:

    I should mention, once you've rescaled according to my nondimensionalization scheme, putting a weakly informative prior on the overall ratio is not too difficult.

    Something like a lognormal with mean logarithm = log(.05) and standard deviation of logarithm = log(5), indicating that we wouldn't be surprised if it were between say .002 and 1.25

  3. Markk says:

    This abstract was confusing to me at first. DL's comment resonates. The way it is presented it sounds like there is a cost PER DRIVER of over a thousand dollars for each additional driver. That is so silly I knew something was wrong, but it took a while to understand what they meant. I think if they had changed

    "we find that the increase in traffic density from a typical additional driver increases total statewide insurance costs of other drivers by"

    to

    "we find that the increase in traffic density from a typical additional driver increases total statewide insurance costs by"

    would have helped. What they said wasn't wrong, but just the way they said it definitely lead me the wrong way.

  4. jonathan says:

    Really more a note to thank you for posting these articles.

    As to this paper, is it me or is the summary unnecessarily dense? The point of the paper is direct – even with the amusing odd detour into tort law discussion – but the whole way of phrasing the argument kind of made me wonder. Accidents relate to density and these can be related to insurance costs and that leads to public policy questions about incentives to drive less, etc. Every sensible driver knows they pay in their insurance for externalities, however measured or derived and though they don't use the word "externality" in their heads. It's nice to see ideas for breaking out that cost but I think the paper could have been more focused. (I dislike using terms like Pigouvian when their use can confuse. Just say tax instead of "substantial Pigouvian charge" because that's clearer.

    One problem I have with papers in general – and this is merely a good example – is the ideas at the end don't have much to do with the substance of the research. If you want to write about per mile insurance premiums, then where is the modeling to discuss the effect? Say I'm in public policy and I have a staffer tell me about this research – because the odds I'll actually read more than the abstruse summary and maybe part of the conclusion are small. I first want to know if the research and model that's actually described in the paper is solid. If we jump to talking about insurance premium models, that takes the focus away from what's actually been presented. Then we end up arguing about what insurance changes might do and we ignore this substance.

    As you may note, I objected strongly to a paper about healthcare costs because it was framed in such strong political terms that an objective observer should not place any trust in either the data or the analysis. (When a paper jumps from relative per patient admin costs for Medicare to blaming Freddie Mac for the financial crisis to saying that a government run or administered program would destroy the nation's health system, that's politics that many people would label gibberish.) I'm objecting here to common practices that's aren't nearly as bad: that the summaries don't really summarize – but seem designed to impress a technical reader that, yes, this is serious work – and that the ideas which the paper seems to get at, which it apparently wants to get at aren't actually the subject of the paper.

    Again, sorry for the lengthy comment but I'm waiting with time on my hands …

  5. Ubs says:

    My entirely self-interested concern about pricing by mile would be that they must take into account where those miles are.

    My home address is in an urban area, but I do very little driving in the city. On the other hand, I love to take long trips on the empty backroads in the rural part of the state. I would hate to be charged the urban rate for all those miles.

    (And yes, I would be perfectly happy to have a GPS tracker in my car feeding information directly to the government and/or insurance company.)

  6. Lord says:

    No one walks in LA. (OK, almost no one)

  7. alex says:

    I just find it absolutely impossible to follow these econometric regressions. I look at the equations and summary tables and it's just takes so much effort to go from there and work out what the variables are, what units they're in, what the data sources are, what controls are included, and even basic things like how many parameters and data points there are.

    It must be hard to present everything when you're having to justify a model in economic terms, and then summarise however many large regressions you ran on whatever huge dataset. But still, there's got to be a clearer way to go about it.

    This isn't a specific criticism of the Edlin and Karaca-Mandic, just a comment on either how I read things and the style of layout in the field.