The productivity benefits of compensation variance in effective altruism

I’ve argued that substantial average compensation in effective altruism likely increases impact, mainly via incentive effects. In particular, substantial compensation can attract more competent applicants to effective altruist jobs. Of course, it means that less money can be spent for other purposes—but as long as the compensation isn’t too high, the benefits outweigh the costs.

Can similar reasoning be used to argue that more impactful effective altruist jobs or tasks (jobs, for short) should be particularly well-paid? In other words, can it be used to argue for greater compensation variance in effective altruism?* I will argue that it can.

Some effective altruist jobs are much more impactful than others, and yet the most well-suited people don’t always take them. (For instance, it’s long been hard to find suitable operations staff in effective altruism.) Typically, that’s because people have self-interested preferences: they find other jobs more fun, more secure, less demanding, higher-status, or what-not. However, they usually also have a self-interested preference for higher compensation, and that can trump the other self-interested preferences. This means that paying more for more impactful jobs can increase total impact.

A related reason for paying more for certain jobs is greater labour market competition. For instance, AI researchers fetch high salaries on the regular labour market (outside effective altruism), meaning that you may need to offer higher salaries to attract the right applicants. My sense is that this factor is already a major driver of differences in compensation in effective altruism.

Yet another argument for paying more for some jobs is that it can function as an honest signal that the funders find it impactful—and as such it could sway even pure altruists. (Of course, any such effects typically come in addition to self-interested preferences for higher pay.) Moreover, two other arguments that I discussed in my post on average compensation are relevant here as well. It could be that some jobs are particularly likely to lead to burnout of a sort that higher pay can prevent. It could also be that direct time purchases (using money to free up time) are particularly valuable for, e.g. executives. Therefore, that could be a reason for paying them more. 

But overall, it seems to me that the two incentives-related arguments are the most important ones. The argument from outside labour market competition is certainly important—and it is probably relatively uncontroversial to pay more for that reason. The other argument may be a bit more controversial, since it says that we should incentivise people to take certain effective altruist jobs over others. Nevertheless, I think it’s also important. I think it can justify higher compensation variance in effective altruism than we have at present. My sense is that effective altruists may hold back a bit too much on this point, e.g. due to egalitarian intuitions.

George Rosenfeld has discussed a potential objection, however. He argues that giving people monetary incentives to engage in longtermism specifically may cause motivated reasoning. It may lead people to convince themselves that longtermist claims are true, even if that’s not the case. This could be an issue in this particular case, but it seems normally less of an issue when it comes to paying more for specific jobs within a given cause area. As such, I don’t it’s a strong general argument against compensation variance. And there could be ways to address this concern. For instance, effective altruist funders often explicitly reward independent thinking and criticisms, which could help.

A question that’s related but distinct from the compensation variance question is what forms the compensation should have. I’ll end with a few words about that. It seems to me that most people in effective altruism are paid a regular salary, and it’s of course straightforward to set them at different levels. There have also been various experiments with prizes and other incentive systems, plus discussions of impact markets. Some of these systems can have downsides, but I’d encourage more such experiments—and think that one should consider paying substantial sums to, for instance, prize winners if that seems warranted. Trying out unusually large prizes or other forms of compensation may have a substantial value of information.**

* To be clear, I’m not arguing that this should lead to some jobs having lower compensation. My argument is that some jobs should have higher compensation.

** Thanks to Daniel Eth for this point.

Thanks to Ryan Carey, Daniel Eth, and George Rosenfeld for helpful comments.