# Replaceability arguments are overblown

January 2014

A commonly-listed “core idea” of effective altruism is replaceability: it’s okay to take a job in an unethical industry because if you don’t, someone else will, so the counterfactual effect of the unethical industry on the world is the same whether or not you take the job. In practice, the impact of this argument tends to be extremely exaggerated, for at least three reasons.

## Threshold hiring

Many firms, including e.g. earning-to-give darling Jane Street, do not hire to fill quotas. They hire as many qualified people as come their way. If you take a job at Jane Street, it won’t cause them to deny someone else a job–they’re expanding as fast as they find people who can do the work. Threshold hiring may not be the most market-rational thing to do (they should be spending more resources on hiring), but:

1. There’s an extreme shortage of STEM talent, so plausibly Jane Street couldn’t get additional talent even if it raised its pay substantially.
2. Resource allocation is a hard enough problem that we should expect a fair amount of irrationality to persist, especially in industries with high margins (hence less competitive pressure towards purely rational allocation).
3. Even if it is irrational, at the end of the day that’s what they do, so that’s what we should base our decisions on.

If you work at one of these firms, then, in the counterfactual, they actually have fewer resources and can do fewer unethical things. (Note: I separately also strongly believe that Jane Street does not do unethical things, and is hence still a good choice for earning to give.)

Points 1-2 also suggest that threshold hiring may be especially prevalent among firms that are good for earning to give. My experience in software and finance is consistent with this, but also highly anecdotal.

## Supply and demand

Even if you work for a firm that doesn’t practice threshold hiring (i.e. has a demand curve, like a rational microeconomic actor), replaceability still doesn’t hold.

Let’s consider the simplest model that actual economists use: there’s a demand curve for labor made of a bunch of firms, and a supply curve made of a bunch of people (i.e. applicants). By applying, you shift the supply curve down. Then the proportion of the gains internalized by producers (applicants) is $$\frac{e_d}{e_s+e_d}$$ and the proportion internalized by consumers (firms) is $$\frac{e_s}{e_s + e_d}$$, where $$e_s$$ is the elasticity of supply and $$e_d$$ is the elasticity of demand. If demand is highly elastic (e.g. in “threshold hiring,” talent elasticity of demand is infinite) then producers capture a significant portion or all of the gains in resources that you create by joining the market. Even in many markets that don’t practice threshold hiring, talent elasticity is rather high.

## General illegibility

The above model is highly simplified. But even in the more general case, firms’ hiring decisions are highly opaque. The labor market is inefficient due to massive information asymmetries (employees don’t have salary data; firms can’t evaluate employees). Evaluating the counterfactual cleanly and generally, as the linked post attempts to do, is nearly impossible. Replaceability may apply in specific cases, but it depends far, far more on the details than on general considerations of labor movement.

## Conclusion

In most practical circumstances, the replaceability consideration fails to hold or is impossible to assess. People choosing their career altruistically should be cautious when applying it, and should under no circumstances use it as a fully (or even reasonably) general counterargument to “ends-justify-the-means” objections to earning to give. There are likely other counterarguments sufficient to show that earning to give is not a net negative, and the general principle of counterfactual reasoning is sound, but the replaceability consideration itself is highly suspect.

Enjoyed this post? Get notified of new ones via email or RSS. Or comment:

Public (as ) submit ⤇

Vollmer

Have you read Paul’s take on this? http://rationalaltruist.com/2013/01/22/replaceability/

Benjamin Todd

I think replaceability often gets misused in this way (‘your net direct impact in a harmful career is zero, because you’re replaceable’). However, we’ve never said this at 80,000 Hours (for instance, see Will’s paper on Earning to Give, my thesis, the post you reference at the top, our analysis of replaceability for doctors). We’ve only made the claim that your true (counterfactually adjusted) impact in a job is typically less than what you appear to do directly (which is what I mean by ‘replaceability’).

I stand by this, because most fields have diminishing returns (i.e. the 1000th doctor achieves less than the 10th doctor), and by joining a field, you’re basically adding to the margin (after making some efficiency assumptions). In addition, in neoclassical job markets, adding to the demand for a job by 1 will only increase the number of jobs supplied by 0.2-0.8 (for the reasons you state, the precise number depending on the characteristics of the job market), reducing your impact by roughly another factor of 0.2-0.8 (also depending on how well ‘value produced’ and ‘economic value produced’ are correlated in the sector). However, this reduction in impact doesn’t seem strong enough to mean that you can treat your direct contribution as ‘negligible’.

Overall, I agree it’s complicated. Paul’s analysis (http://rationalaltruist.com/2013/01/22/replaceability/) is the best attempt I know to model it. Stepping back, there seems to be more important topics to put your effort into, like: how high priority is the cause I’m supporting? how good can I be at this career? how much career capital can I gain?

Aaron Tucker

Thanks for writing this! This is totally a fact that takes experience to find out, and I don’t remember seeing anyone make it explicit in the EA blogosphere.

I think that there’s also a complementary point, that if replacability is false because firms are actively trying to hire people and failing to find them, that the same is likely true in charities, and that top charities are likely to be talent limited as well as cash limited.

Paul

As Ben remarks, I think you are definitely right that even on naive a priori arguments, there is a non-negligible effect from working in finance. I would probably also agree that many and perhaps most invocations of replaceability by EA’s make the analysis worse rather than better.

I’m skeptical about this use of the illegibility argument (which comes up in many contexts). See discussion here for my take (it’s not my replaceability post). I agree that illegibility pushes you away from the superficially obvious model. But it’s not obvious why it pushes towards “If I do X, there is one more person doing X.” I think the default expectation is that it pushes towards the simple economic answer (since as far as I can tell that’s what you should expect a priori), and we should only deviate from this if we have a clear enough picture to justify deviation.

If people appear to be threshold hiring, I would be quite hesitant to suggest that elasticities are infinite, for reasons closely related to illegibility. After all, how did this threshold get set? Even if there was only a single employer and they used threshold hiring religiously, I would still expect that after taking into account your effect on where they set the threshold in the future, the net effect is much closer to the supply and demand picture. In this case, discussions with people in finance suggests a consensus that there are diminishing returns for the kind of work Jane Street is doing, they just operate at the level of the sector rather than the firm.

As it turns out, threshold hiring is actually what microeconomic models are expected to look like from the inside for many firms, potentially including Jane Street. You want to hire someone iff they make them money. In the regime where diminishing returns are operating for the entire sector rather than for a single firm, this means that they should have a threshold for the quality of hires which depends on the general crowdedness of their sector, rather than on how many people they’ve hired. So you should expect the threshold to slowly raise as more people enter finance and to fall as people leave, offsetting your contribution. But because the situation is so complicated, I wouldn’t want to try and reason about the effects in detail, really I’d just want to think in terms of elasticities.