Welfare Estimation 101: Money is not Fungible
Intro
This is the first post in a sequence on Welfare Estimation for EAs. I will go over the basics of the current field of cost-benefit analysis, clear up what I perceive to be some widespread confusions, and chart a path for expanding and refining the methodology to make it more useful for the kinds of questions EAs ask. This isn’t some wild-eyed call to redo everything; it keeps the foundation of a very good body of work, and uses that foundation to guide sensible cause prioritization across a world of possibilities.
I am an economist who has done cost-benefit analysis for over a decade and is very familiar with the literature and methodology. I am a long-term EA, coming to it from the old LessWrong site, which I started reading in grad school.
100 level posts are an introduction and philosophical background. 200 level posts discuss specific concepts to keep in mind when doing welfare estimation, and many of them are a distinct break with the current assumptions and methodology of cost-benefit analysis. 300 level posts will be about actually doing the thing. Most of the sequence will explain things from the ground up, but the rest of this post is kind of a ‘teaser’ that serves as a guidepost to my approach and shows what is possible. It necessarily involves some previous familiarity with EA cause prioritization or economic analysis.
Heresy
Cost-benefit analysis is the process of converting all of the costs and benefits of an action into a monetary equivalent and adding them up to see how good the action is. It is very good, and very useful, and should be used more often than it is.
So naturally, I am going to break a key methodological underpinning of something awesome, in order to try to remake it into something even better. Your prior that I will succeed in this should be low, but I hope to convince you that my approach is reasonable.
The key insight that allows you to move from Cost-Benefit Analysis to Welfare Estimation is that, when doing an analysis to determine what actions are best, Money Is Not Fungible. Money in one person’s hands cannot and should not be added up with money in the hands of someone in a different situation.
Unless you know a bit about economics or finance, you cannot appreciate the magnitude of the heresy I just committed. There is a part of my brain that, right now, is doing this:
The idea that money is fungible, i.e. that all dollars are the same and that they are the unit of account, is fundamental to all of economics and finance. It gets embedded deeply into our worldview and analytical assumptions in grad school. Every economist will, deeply and subconsciously, seek to make money the unit of account. Everything will be converted into money, in order to allow it to be measured to other things so sensible decisions can be made.
This is, in most cases, a very good strategy. You need some common yardstick to measure everything with, and money has emerged, via a distributed social-consensus process, as a common, tradeable, shared yardstick for a great many things. When you want to know how much people desire or care about something, the first thing you should do is look at the amount of money they will spend on it. And when looking at a policy, the first thing you should do is try to figure out how much it changes the total wealth of society.
However, treating money as the unit of account leads to a lot of problems when trying to compare people in different situations. Techniques that work extremely well for making good decisions within a society will fail when applied to global decision making. I will go over the details in later posts.
But, for now, consider the fact that under the methodology of cost-benefit analysis, a transfer payment is neither a cost nor a benefit. You have not changed the total amount of money in society, you just moved a fungible commodity around. Cost-benefit analysis ignores transfer payments in the bottom-line analysis. Moving money from one person to another is always a zero effect; it does not change the total wealth of society, so it is irrelevant.
So now ask, “If money is fungible, then how does charity work?”
The entire point of charity is that you can make the world better by transferring money from one person to another. And this is a very obvious thing that we all understand. This must mean that in an important way, money in one person’s hands is fundamentally different from money in someone else’s hands.
Lives, not Money
A key methodological change I recommend is that, When analyzing policies that have economic and health impacts in different places, use the life-year rather than the money as the fundamental unit of analysis.
The fungible thing in the analysis, i.e. the common unit of account, should be a year of healthy, happy, flourishing human life. That is the foundation, and the thing we are maximizing for. Money will be an important part of the process, and a very important thing to measure accurately, but you should always convert it into life-years before giving the bottom-line result.
At first, this is just a small step, almost a trivial departure from conventional cost-benefit analysis, like pricing something in yen instead of dollars. It is just a messaging change. Instead of saying “The trans fat ban will save live-years valued at $140 billion for a cost of $10 billion, making net benefits $130 billion”, you say “The trans fat ban will save 560 thousand life-years, at a cost of 40 thousand life-year-equivalents of regulatory burdens and side effects, making net benefits 520 thousand life-years.”
But that step opens a world of possibilities, especially when dealing with situations where the observed lives to money tradeoffs are different.
For example, suppose that we’re evaluating several possible permutations of a potential trade treaty between the US and Nigeria that will increase income in both countries, but also generate pollution that harms life and health in both countries. We want to decide which version is best for everyone.
Under the current assumptions of the field of cost-benefit analysis, we would add up the economic benefits, counting equally money earned by Americans and Nigerians, and then subtract the health harms, valuing the American health harms more. This causes all kinds of problems, which is why nobody tries to use cost-benefit analysis for situations like these.
A much better method is to debit the treaty by all of the life and health harms, counting everyone equally. Then credit the treaty with one life-year for each $X that it increases American income, and one life-year for $Y increase in Nigerian income, where X and Y are the valuations for a life-year in those countries. X will be greater than Y; current recommendations for these values are roughly 2 to 4 times average annual income in the country.
This gives us an estimate of the net life-years gained or lost by each version of the treaty, and we can lobby for the one that has the highest value.
Love this article and the series. Couldn’t be happier to see you get this discussion rolling!
Questions for you regarding this piece:
Just out of curiosity, why do you think this measure ought to be the unit? You could say “a month” or “a day” or “a decade” or “a six second conscious moment”, but the choice is “a year”.
And how do we deal with “healthy, happy, and flourishing” as a singular unit, when, in a sense, each of these things are in themselves such complex, multidimensional metrics?
What is your overall basis for this conclusion that the healthy, happy human life-year is the best unit of measurement vs other possible measures of “welfare”?
(Just to be clear, I 100% agree with you that we should use this unit of measurement, especially in favor of money. But I have been struggling with these additional questions and have a feeling you might have some insights to offer.)
As far as I can tell, Richard Bruns is talking about the quality-adjusted life year or QALY.
The reason it is a year is essentially arbitrary, a year is decently long without being too long for the purposes of public health where QALYs first got used.
The way we deal with “healthy, happy, and flourishing” as a single unit is much trickier. For traditional QALY calculations, researchers simply ask people how they feel when experiencing certain things (like a particular surgery or a disease) and normalize/aggregate those responses to get a scale where 0 quality is as good as death, 1 is perfect health, and negative numbers can be used for experiences worse than death.
Then you multiply quality by quantity. The Wikipedia article on this is good and goes into more depth: https://en.wikipedia.org/wiki/Quality-adjusted_life_year
Note: I don’t know how Bruns intends to measure quality of life yet, I expect we’ll have to wait.
This isn’t correct. QALY weights are typically based on hypothetical preferences, not experiences.
What Richard described is more like a WELBY, which has a similar structure but covers wellbeing in some sense rather than just health. See Part 1 of my (unfinished) sequence on this if you’re interested.
I agree with this; thank you for replying. (I thought I would get email alerts if anyone commented, but I guess I didn’t set that up right.)
Questions in order:
I never meant to make a statement that a year is better than other time units. I said year because it is the existing standard in the field. The statement was about using a life/health measurement rather than money. As the 102 post hints at, my goal is not to create ‘the best’ system ex nihilo; it is to build off of the precedent set in the field. So whenever an arbitrary choice has already become the standard, and it is not obviously worse than something else, I stick with it.
This will inevitably be handwavey, fuzzy, and based on surveys. I imagine something like the WELBY, where we set the value of an ideal life to 1, and ask people how bad it would be for various things to happen to them, and assign ‘disability weights’ to everything based on their responses.
Because it is easy for everyone to understand intuitively. See the 102 post; anything we use will need to be very approachable, so we have society-wide buy-in for the metric.