Making driving safe

The home page for the Google Self-Driving Car Project contains the following:

Imagine if everyone could get around easily and safely, regardless of their ability to drive. […] Aging or visually impaired loved ones wouldn't have to give up their independence. Time spent commuting could be time spent doing what you want to do. Deaths from traffic accidents—over 1.2 million worldwide every year—could be reduced dramatically, especially since 94% of accidents in the U.S. involve human error.

In some walks of life, we can reduce risks by changing human behavior. In the case of automobiles, we train and license people to make them drive more safely. We also fine and incarcerate them when they don't.

Even so, human error makes driving risky, so we are always on the lookout for better technology to slash those risks. This is a common pattern. Robotics have reduced industrial accidents. Airplanes are usually safest on auto-pilot. And modern hospitals are filled with everything from computer-guided lasers to radiology equipment.

Unless you regularly skydive or bungee jump, getting into your car is almost surely the riskiest thing you do every day. The data bear this out. The National Highway Traffic Safety Administration’s (NHTSA) provides detailed traffic safety statistics in their annual report. In 2013, there were 5.7 million reported automobile accidents. The NHTSA estimates that fully 47% of accidents go unreported, so the actual number was probably over 10 million. The reported crashes caused 2.35 million injuries and 32,719 fatalities. Since people drove 3 trillion miles that year, this means that there were 190 reported accidents causing 1.09 deaths per 100 million miles driven.

To put these numbers into perspective, we can compare them with air travel accident rates. Since fatal crashes of commercial U.S. air carriers are relatively rare – there were none in 2013 – we need to look at a longer time period. Over the past 10 years, domestic airlines have flown 46 trillion passenger miles with 157 fatalities. So, whereas car crashes lead to a bit more than 1 fatality per 100 million miles of travel, commercial air travel results in 1 fatality per 30 billion passenger miles of travel. It’s no wonder that deaths rose in 2002 when, fearing terrorists on planes, people took to their cars.

So driving is very risky!  There are two ways to reduce the risk: change driver behavior or improve technology. In both cases, private actions and government intervention are possible. And in both cases, there are gains to be had. But the prospective gains from technology are much larger.

Starting with the drivers, even seemingly irrelevant decisions can have a sizable impact on risk. When people look for a place to live, they frequently consider the cost of their commute both in time and money. How long will it take to get from home to work and back?  And how much will fuel and the wear and tear on their car cost? What few consider is that the risk of an accident declines the shorter the distance traveled. How much does this risk fall if you choose to live 5 miles away from work rather than 6?

Since most people work about 250 days per year, a commute that is one mile shorter means driving 500 miles less per year. Based on the NHTSA data, this will reduce the probability of an accident from 1 in 175 per year to 1 in 210 per year. The risk of a fatality drops from 1 in 30,400 to 1 in 36,500. So, if the heads of all 100 million U.S. households decided to cut their work commute by one mile per day, the projected number of fatalities would decline by 547 persons a year, or 1.7%.

These numbers are nationwide averages, and it does matter where you live. For example, fatality rates in Montana and West Virginia are roughly 3 times what they are in Massachusetts and Minnesota. But even if we divide the probabilities in half, trimming car travel by 500 miles a year for most households would still save hundreds of lives.

What about the role of government? In an effort to reduce injury and fatality rates, the federal government imposes safety standards both on drivers and car manufacturers. These range from highway speed limits to equipment requirements like the Federal Motor Vehicle Safety Standards. The latter is an exceedingly long list that includes the requirement that the hood on a car have a latch. Of key relevance are the safety technologies, primarily seat belts and air bags.

What was the impact of these standards and how much did they cost? A study covering the years from 1960 to 2002 estimates that these save roughly 20,000 lives per year at a cost of a $950 per new car sold (in 2015 dollars). That is, for less than $1000 per car, we cut the probability of fatalities from car crashes by 40%. The total estimated cost (in 2015 dollars) was $15 billion. For each life saved, that is about $750,000.

For many of us, each life is priceless. But when the government considers the imposition of safety targets, it must estimate the cost and balance that off against the prospective benefit. For example, is the $750,000 per life saved by seat belts and airbags worth the cost? To answer that question, governments estimate the “value of a statistical life” by addressing questions of the following form: How much would a typical person be willing to pay to eliminate a mortality risk that is 1 in 50,000?  An answer of $200 implies that the expected value of reducing one fatality is $10 million. Consistent with estimates in academic studies, the Department of Transportation (DoT) currently uses $9.4 million (see here and here).

From this point of view, federal vehicle safety standards are extremely cheap. Using the DoT’s value of a statistical life, as a society we would be willing to spend more than 12 times as much as we did to secure the increased safety that government standards have achieved.

This brings us back to technology and Google’s self-driving cars. Suppose that if everyone had a self-driving vehicle, it would cut the fatality rate in half. That would mean roughly one 1 fatality per 200 million miles driven – still 150 times riskier than flying, but much safer than it is now. At this writing, self-driving cars have gone over 1 million miles without an injury, so this assumption is plausible, but still to be tested.

How much would we be willing to pay for that?

If, as DoT estimates, each life has a statistical value of $9.4 million, saving 16,000 lives per year is worth $150 billion. Because of the efficiency of self-driving cars – you really don’t need to have your own – let’s assume that we only need about one per household in the country, or 100 million. That means the benefit is $1,500 per household. If a car lasts for 10 years, at a discount rate of 2% (adjusted for inflation), the present value of the benefit is $13,474.

Reducing the number of accidents also reduces the direct economic costs, which has been estimated by NHTSA at about $250 billion per year. That’s another $2,500 per household per year. Assuming that self-driving cars eliminate 50% of that cost, the present value of the $1,250 annual gain over 10 years (still at 2%) is $10,139. Since this is a consequence of reduced car insurance claims, the benefits will accrue directly to the owners of the cars.

Of course, some self-driving car accidents will still occur, but these are likely to generate product liability claims, so manufacturers will probably shoulder this burden directly. Put differently, the world of self-driving cars is likely to be one with far less conventional auto insurance – and perhaps none! 

Returning to the benefits, adding the two parts together – the social benefit of lower fatality rates ($13,474) and the household benefit of lower insurance costs ($10,139) – leads us to conclude that as a society we would be willing to pay $23,613 over and above the current cost for a car that cuts accidents and fatalities by 50%. 

What this means is that a safe self-driving version of the Ford Focus, which costs a bit over $20,000 today, has an economic value of more than $43,000. Now, it may be very difficult for someone who buys a Ford Focus to afford that. If credit markets are imperfect, they might not even be able to afford the sum of $20,000 plus the value of the reduced insurance premium of $10,139, or $30,139.

Current estimates of the additional cost of technology to produce self-driving cars by 2025 are in the range of $10,000. So long as hackers can’t take these cars over, they would be worth every penny.  [One should also include the cost of product liability, assuming that this burden is ultimately passed on to car buyers]. Yet, even if total costs are somewhat higher, the financial impediments could be overcome through leasing and renting, rather than owning.

For now, however, it’s still worth trying to change driver behavior. Some incentives already push us in the right direction: living closer to work makes you safer. So, educating drivers about the risks may be a good way to modify behavior. But a nudge from, say, a modest carbon tax, wouldn’t hurt. It might even add to the demand for self-driving cars, which don’t put a heavy foot on the pedal.