(This is a math-light short post where I refer to a current event to illustrate an important point about welfare estimation. Reading the Welfare is a Myth post is recommended but not necessary.)

Trolley Joyride Problem

Imagine that you are part of a group or community that routinely does an activity that has a health or safety risk. For the sake of a concrete example that is not coded to any current community, call the activity 'trolley joyrides'. As a result of trolley joyriding, 100 people in your community die every year. Some people in the community believe that trolley joyriding was freely chosen and is a good thing with real benefits, while others believe that they are addicted to it and were manipulated into it, and wish that trolley joyrides did not exist.

This is a complicated situation, and it will not be clear what to do. Does the benefit of trolley joyrides outweigh the harm, and if so, what should be done? Let's make it simpler. Assume that there is a universal consensus that trolley joyriding is pure addiction with no net benefits. The utility you get from the adrenaline rush and exercise is exactly balanced out by the withdrawal sadness and ennui you feel when the activity ends. Some people can resist the addiction when exposed, and some cannot.

Now, suppose that there is a proposal to ban trolley joyrides. Assume that the ban would be 25% effective, and thus save 25 lives. Also assume that the economic costs are negligible. But because the members of the community are involved in supplying the activity, and will defend both their profits and their addiction, they will get in confrontations with police as the ban is enforced. Assume that 5 people each year will be killed by the police as a result, and that the people deciding on whether or not to ban trolley joyrides have no power to change the behaviors of the police and/or community members that leads to these fatalities. 

Do you support the ban? Will you set in motion events that will cause the police to kill 5 people in order to save 25 lives? And, as a general principle, in situations like this, how many people is it okay for the police to kill in order to prevent 25 people from dying as a result of their 'lifestyle choices' aka their inability to resist addiction, advertising, and/or social pressure?

My view is that, as with many philosophical though experiments regarding ethics, there is no right answer here. We are arguing about what world we would prefer to live in. There are different people with different worldviews, and we need a procedure that allows us to aggregate our preferences to determine policy. A hardcore utilitarian would say 'Kill up to 24 people. A death is a death, and the details do not matter.' A hardcore deontologist might argue that it is not acceptable for the police to kill even one person to prevent 25,ooo addiction deaths. Most people will have a view between these two extremes, and so most governments should choose an acceptable casualty rate lower than 99% but higher than 0.1%.

Menthol Cigarette Ban

The Trolley Joyride Problem is not an abstract intellectual exercise with no connection to reality. A very similar situation is happening, right now, in the USA. The FDA is moving forward with a ban on menthol cigarettes, and some members of the affected community are opposing the ban:

This year, 45,000 Black Americans will die from tobacco-related illnesses. The same number will likely perish next year, and almost as many the year after that....

Rev. Al Sharpton ... is arguing that banning menthol cigarettes will result in more — and potentially more deadly — encounters between police and Black people. Police officers, Sharpton alleges, will see a Black person on the street smoking a cigarette, assume that it is an illicit menthol cigarette, then question the assumed perpetrator and likely harass and arrest them. And then the inevitable: Some of these encounters will turn deadly. Sharpton is quick to invoke the name of Eric Garner who was being questioned for selling “loosies” — single cigarettes — before his deadly encounter with police.

Assume that Rev. Al Sharpton is correct, and that a menthol cigarette ban would result in people getting killed by the police. I made the same assumption when I was at FDA doing a preliminary cost-benefit analysis of a nicotine product standard that would be a de facto ban on cigarettes. Also assume that he is sincere in his opposition to the rule, which should be the default intellectual posture for these kinds of situations. 

Given these facts and assumptions, it would be arrogant and dogmatic, and possibly culturally imperialist, to assert that Rev. Al Sharpton is wrong to oppose the policy. As with the Trolley Joyride 'thought experiment', there is no clearly correct answer. People are arguing over what world they would prefer to live in. The public health establishment would prefer to live in a world where several people a year are killed by police in order to save thousands of lives a year. Rev. Al Sharpton, and presumably many of his constituents, would prefer not to live in such a world.

So, how do we decide what to do?

Voting on Equation Terms

The cost-benefit analysis, or welfare estimation, for a cigarette  ban is a lot more complicated than 'Kill X people to save Y people'. There are many complicated effects to consider, all of which most be converted into some common unit of measurement. Current policy is to use a dollar value as the unit of account. Lives lost are converted into dollar values using the Value of a Statistical Life, or VSL.

As the previous discussion illustrates, many and perhaps most people would prefer to live in a world where the police are only allowed to kill someone if it would prevent more than one disease death. Therefore, the 'VSL dying from disease' should not be the same as the 'VSL killed by police'. For anyone other than a hardcore utilitarian, the latter should be some multiple of the former.  (The 'VSL killed by violent crime' is a yet different number.) This multiple should probably be larger than 1 but less than 1,000.

Even if you are a hardcore utilitarian who values all lives the same, as a practical matter, the 'VSL killed by police' should be higher than most other VSLs, because it is an empirical fact about reality that someone getting killed by police causes a lot more negative utility among their peers and community than that person dying from a disease as a result of their lifestyle choices.

So, how do we decide what this multiplier is? What dollar value do we assign to people getting killed by police when calculating the costs and benefits of a government policy? How do we accommodate the diversity of worldviews in our society regarding this question? These are all very important and open questions. 

The concept of a VSL is most often used in policies for health and safety, where rules prevent deaths from accident or disease. The number is calculated based on revealed preference: we basically look up how people act when making tradeoffs between their money and their health and safety, to see how they value things, and average it all together.

But for a lot of terms in the 'social utility function' that we use to analyze policy, there will not be a revealed-preference number. We will basically need to ask people to vote on what the term should be, and figure out an appropriate and strategy-proof way to elicit honest answers and aggregate the votes (pro tip: the geometric mean is a lot less sensitive to high outliers than the arithmetic mean).

For most equation terms, it is probably not appropriate to hold a society-wide election, given the costs of collecting the information. A representative sample will do. But this information should be updated regularly, in a rigorous and transparent way. If the beliefs and worldviews of a representative sample of society change over time, then the welfare analysis should change with it. One year, the cigarette-ban cost-benefit analysis might value a police-caused death at 20x the disease VSL, and the next year, it might be 3x, if a lot of people change their minds about the issue. We shouldn't see anything unusual about this. We live in a democracy, and people change their minds about things, and a policy analysis should reflect this.