Optimizing our actions to achieve the best outcomes is one of the keystones of effective altruism. But there's two different ways of thinking about how to do that that yield very different results.
- Partial equilibrium thinking---optimizing your actions to achieve the best outcome, holding the rest of the world's actions fixed.
- General equilibrium thinking---optimizing your actions to achieve the best outcome, assuming the rest of the world will reoptimize their actions in response to you.
While both of these ways of thinking are used by EAs frequently, I think partial equilibrium thinking is undeniably more common. Consider the following examples:
- Career choice. 80,000 Hours's career recommendations explicitly factor in how many people are working in an area already.
- Tractability analysis. When determining cost-effectiveness, the tractability analysis focuses on the marginal impact that one actor (donor, nonprofit) can have.
- Efficacy of individual actions. It's widely argued that individual choices (e.g. eating vegan) don't matter very much because the reduced consumption saves a trivial amount of suffering compared to a donation to the Humane League or an equivalent charity.
There are tremendous advantages to partial equilibrium thinking in the above examples. It helps clarify for individual people what they can do on their own, and it is often much more realistic given that we are small actors relative to the world at large. Despite this, I think general equilibrium thinking is more appropriate in the above situations and often yields different conclusions.
- Career choice - when 80,000 Hours puts out a career guide, there are thousands (if not more!) people who will look to it as at least a moderate factor in what they should take up as a career. This means that an 80k recommendation could easily swing a career path from "not enough people are working on this" to "too many people are working on this" if that area is not a huge area with a lot of capacity to absorb talent.
- Tractability - In general, we should not take existing costs and constraints as binding, because they can be changed by our actions. There are many case studies of global health interventions where unilateral action changed an intervention's tractability. Paul Farmer worked on curing TB when it was tremendously cost-ineffective, but his work drove down the cost of TB treatment to be competitive with leading health interventions.
- Individual choices - going vegan doesn't just have the effect of your own choices, it also increases demand for vegan products, which gives money to companies selling those products and allows them to sell more. Moreover, some experts attribute the quick adoption of vegan options in fast-food restaurants to the "veto vote" phenomenon, where one person can swing a family's decision to eat somewhere if that place has no vegan options. In this case, one person is controlling the dining decisions of a whole family, which amplifies the demand effect.
These are not new arguments, and this post is not about those claims substantively. I make them to illustrate that a) they are all unified by this core concept of general equilibrium thinking, and b) general equilibrium thinking is a real way to incorporate additional information that could change your conclusion.
The above examples might give the impression that general equilibrium effects always just amplifying the partial equilibrium effects, but this is not true. Sometimes the general equilibrium effect is actually negative---e.g. the "crowd-out" concern that is very common in global health work, where funding an area might cause governments to reduce their funding for that area. (Even then, the general equilibrium effect could still be positive if governments reallocate that money to another important area!)
General equilibrium thinking is hard to do rigorously. The arguments above have a convoluted feel to them: the second-order effects of your actions are hard to weigh, and the third-order effects are even harder. Despite this, I think it's worth doing. It is obviously not a substitute for partial equilibrium thinking, because the first-order effects are the most certain ones and they do need to be established. Rather, I'm arguing that we should incorporate general equilibrium thinking explicitly into our (implicitly) partial equilibrium analyses, even if the conclusion is to reject that they matter.
For example, it should be a best practice for cost-effectiveness analyses to identify both partial equilibrium effects and general equilibrium effects explicitly, and assign an informal credence level to the general equilibrium effects. This could look something like "the marginal impact of funding malaria bednets is X/dollar. In equilibrium, we could expect $1 of bednet funding bednets to lead to reallocation of $Z of malaria-focused money in public health towards other interventions. If this happened, our best bet is that this would go towards TB prevention with an efficacy of Y/dollar. However, we think Z will be very small because malaria is the largest public health concern for governments, so the general equilibrium effect is still pretty close to X/dollar." Doing this allows us to separate three important concerns: partial equilibrium effects (X), general equilibrium effects (Y), and the scale of those general equilibrium effects (Z). This is much more transparent than bundling those up into one estimate.
In short, I think partial equilibrium thinking and general equilibrium thinking should be concepts that we lean on explicitly when making arguments. Moreover, we should explicitly incorporate more general equilibrium thinking into our arguments---it makes our thinking more transparent and more likely to be right.
There is also another related distinction between optimisation that assumes that current investments in some cause (e.g. the mitigation of some risk) will stay the same (or change in line with some simplistic extrapolation of current trends), and optimisation that assumes that other people will reoptimise their investments due to new evidence (e.g. warning shots). I wrote a post about that in the context of existential risk some years back. Jon Elster argues that we generally underrate the extent that people's reoptimise their actions in the light of a change of circumstances, whether they are endogeneous/due to our own actions (what you are talking about) or exogeneous events (what I am talking about). He calls it the younger sibling syndrome.