The Impact you can (and can’t) Make in an Hour
This is my contribution to this month’s blogging carnival, with the theme of “blind spots”.
Thanks to Ben Kuhn for substantial help organizing and editing this post.
There’s a trope in the EA world of comparing every activity to spending the same amount of time working and donating the money earned to charity. The implication is that one shouldn’t do any potentially charitable activities unless they meet this extremely high standard. Some examples:
Eliezer uses a canonical example from economics of a lawyer who volunteers in a soup kitchen.
This naive hour-for-hour comparison adequately captures the actual consequences of decisions that trade off time, and as I’ll discuss in this post, I think they unfairly discourage us from using our time other ways. We need ways to make decisions about how we spend our time that take into account how we’re spending our time overall, not just any particular hour.
Why the naive hour-for-hour tradeoff model doesn’t work
Non-linear return on work
These comparisons assume that return on work is linear, that is, the marginal hour at work is worth the same amount of money as the average hour at work. This isn’t usually true. People who earn salaries usually earn their salary as long as they fulfil the minimum requirements for their job. Even if working more than the minimum helps you earn more, there’s no reason to think that it does so at the same rate.
For a clear example, consider a teacher. They earn $40,000 a year and have 12 weeks off during the summer, so their wage for the 40 weeks that school is on is $1,000/week. During the summer, they can try to get a second job, but they probably can’t do better than $10/hour, bringing them to $400/week during the summer. Since teacher salaries are usually based on seniority and not skill, putting extra time into school-related things probably doesn’t pay off monetarily. Even though the average wage is about $860/week, their marginal income is more like $400/week. The comparisons above, then, give twice the lost income a teacher could reasonably expect from a lost hour of work.
What if I’m a software engineer? Since working more in tech probably helps you primarily by increasing the probability that you are promoted, it’s harder to estimate your expected reward. Some considerations:
Raises in the tech industry are usually small; you usually increase your pay only by switching jobs.
If you work hard and climb the tech career ladder faster, you enjoy higher salaries for longer, multiplying the effect of your work.
Tech salaries have a long tail. If you end up the CTO of a successful company, you make a lot of money, but very few people achieve that. One thing this means is that moving from the 60th percentile of tech workers to the 65th is probably much less valuable than moving from the 94th to the 99th.
It’s also probably much easier to move from the 60th percentile of tech workers to the 65th than from the 94th to the 99th, because the other people in the 94th percentile of tech workers are also pretty awesome.
It’s not obvious that working more will help you get ahead at all in tech (although I think it’s likely), and if so, what kind of work. Learning new technical skills, expanding your open source resume, working for high-prestige companies, expanding your network, and investing in leadership and interpersonal skills all seem plausible routes forward, and I think any of them could be the best, and any of them could have almost no effect.
Given this, my guess for the value of a marginal hour for a tech worker is something between a quarter and a half of their average salary. Again, this is enough of a difference that we could be significantly misled by valuing their time at their average hourly wage, as the examples above do.
Counterfactual spending
When I see these comparisons, they always compare the good done by volunteering to the amount of good that could be done with the full amount of money that the potential volunteer would earn. This comparison makes some sense in the abstract—if I’m willing to take a week off work to do good, I should be willing to work an additional week and donate the full amount, not just part of it. However, if I usually donate 10% of my earnings to charity, I will probably only donate 10% of my weeks earnings if I work that week.
Note that this doesn’t apply to everyone—if you donate all of your earnings above an amount that you are certain to make, you can claim that every marginal dollar would be given to charity. However, if you don’t use that system, and I believe most of us don’t, you should either expect to donate your usual percentage to charity, or to do some complicated accounting.
Counterfactual Time Use
Another assumption these comparisons make is that time time spent on one of these activities comes only out of work time. I think this is often false, and we should generally guess that volunteer time comes partly out of work and partly out of recreation time.
Since people usually make commitments at work to get particular things done, they pretty much have to work long enough to get those things done, so even if someone take Tuesday morning off to walk to the voting booth, they are increasing their probability of working late a future week.
In addition, since many volunteer activities more closely resemble recreation than work, they may not only funge mostly against recreation time, but have the same positive effects as recreation. That is, if I take two weeks off of work to donate a kidney, I may return to work more rested and productive than if I had worked those two weeks.
Adding it Up
Say that I earn on average $100 per hour worked, and I’m considering taking on some volunteer activity. If I assume that I can earn in a marginal hour half of what I do in an average hour, and if I work an extra hour that I’ll donate half of the proceeds, and that the volunteer activity I’m considering funges half against work and half against other non-work activities, then the amount I expect my donation to decrease by is not $100/hour, it’s $12.50/hour. This order of magnitude difference could easily make the difference between “worth it” activities and “not worth it” ones.
More importantly, though, how a particular time commitment affects your earnings depends a lot on the situation. Every time we make this type of comparison, we’re making assumptions. Mine are more explicit, and I think they’re more realistic, but even so they only apply to some situations. Someone with a flexible job taking on a small time commitment is likely to have their earnings change by essentially zero. Someone who is paid by the hour who has to trade their shift away to meet a new commitment probably loses essentially their full wages for that hour. No single number comparison will capture all or even most situations.
All this somewhat begs the question, though—how should we manage the trade-off between work and other potentially impactful activities? It seems like for most people, working and donating a portion of their salaries is the most impactful thing they can do, and it seems likely that working harder is a path to being able to donate more. It also seems likely that many of us are at risk of overcommitting to volunteer activities to the point where they do interfere substantially with our jobs. I think we can do better than naive hour-by-hour comparisons for understanding time tradeoffs.
Better Models for Time Tradeoffs
Instead of comparing the uses of every hour individually, people trying to maximize their earnings should have policies for how they spend their time. A simple and good example is a budget. I might commit to spending 50 hours/week on work (or on investing in work skills), 10 hours/week on other EA related activities, and use the rest of my time how I want. Using this system would allow me to still do activities that I want to do (voting, volunteering) in my free time, and feel confident that I’m still doing a good job at work and investing in future earnings. If I overspend in one category, I can consider it a loan from the future, and pay it back later.
Understanding counterfactual use of time
I’ve said that when deciding to spend time on something, it doesn’t make sense to assume that in the counterfactual, that same time would be used for work. However, I don’t have a high-certainty idea of what I would do in the counterfactual. Budgeting time gives us the opportunity to be able to answer that question better.
When you’re considering modifying a budget to add an ongoing time commitment, you can compare two plans directly, and be able to make claims with reasonable certainty about what you would do in the counterfactual if you don’t take on a commitment.
Jeff Kaufman wrote about keeping monetary choices donation neutral (they also mention how they would approach some time tradeoffs).
For many of us, donating money is going to be the most effective thing we do by a noticeable margin. If this is true of you, this probably means putting work in its own category, since regularly choosing to do something besides work with your work time is probably not donation neutral. If you don’t think voting is an effective use of your time, but you still want to do it, go to the polls, but make sure you also meet your other time obligations for the week.
This all leaves a question, though. My initial criticism was of a way we communicate about time tradeoffs (making naive one-off-comparisons). I’ve talked about a better way for people to frame their decisions about how they spend their time, but I’ve still deprived them of a catchy way to demonstrate the relative impact of different activities.
Better Ways to Talk about Time Use
Overall, I’d rather see the impact of different activities compared to money than to time. While the good different people can do with an hour may vary from person to person, the good they can do with a dollar can’t, so dollars makes a better currency for impact than time. I also think comparing the impact of an activity to the impact of a donation is similarly vivid—Scott Alexander uses in very well in this recent post to make the same point.
I also think it’s also reasonable to categorize an activity as “not very effective.” I think the examples of soup kitchen work and voting both count as good things that are far enough from being effective we should probably never call them EA activities. This doesn’t mean you should never do them, only that you shouldn’t put them in your “altruism” budget.
Conclusion
I think there’s a larger pattern that’s worth pointing out here. Ben wrote about how the idea of comparative advantage causes no one in the movement to do things that need to be done. Even though every individual career decision may make sense, in the context of the EA movement they fail to meet some needs. Similarly, making decisions about how to spend our time without context causes us to allocate our time in ways that aren’t as valuable as we’re portraying them.
I think there’s a tendency to discover a useful tool for thinking about how we should behave, and then apply it with insufficient scrutiny to too many situations. We should spend more time stepping back and investigating the actual results of our decision-making algorithms, and looking for algorithms that result in states that we are happier with.
You can earn $40/hour as a personal tutor without any qualifications beyond an ivy-league degree—I’m sure qualified teachers could command as high an hourly rate.
That’s reasonable—I think some teachers probably can do this as a part-time thing on top of teaching. Keep in mind though, that tutoring work is highly variable by time of year, not very available in some places (I knew zero students with private tutors of any kind in my hometown in Kansas), and can have high transportation overheads. Another way to look at is that if an option that good were available to all teachers, you’d never see teachers getting food service jobs.
In either case, my main point is that the way someone earns marginal income can be completely different than the way they earn their average dollar.
This does not weaken your main thesis, but it might add some nuance—there are definitely times when you’re marginal wage greatly exceeds your mean wage. For example, one time I spent two days earning $295/hr. …I also think there was one time I changed something on my tax forms that saved me $30/mo and only took a minute, so—six months later—that was effectively $10800/hr.
Peter, I don’t think it makes a ton of sense to count those as “wages” per se, since they’re not scalable, repeatable, knowable a priori, or even (often) reproducible between people.
In fact, I’m not even sure they’re marginal. For instance, nobody thinks “I think I shouldn’t donate a kidney because I could otherwise spend that sick leave changing things on my tax forms and earn millions of dollars.”
Hey Ruthie,
Thanks for the post, I agree with what you’ve said, and think this is a great post. I think that I can add some extra considerations relating to average and marginal wages though. Yes, sometimes your marginal time is worth less than your average wage. But in other situations, it’s worth more.
As a junior doctor, hospitals place a premium value on your marginal time: you’re paid $25 per hour during the week and 1.5 rate for overtime, and then about 3x for covering shifts. Partially this is because doctors’ skillset is uncommon and partially it’s because they want to ensure continuity of care for each patient—the doctor on after-hours has treated the patient before and knows some of their story. When continuity matters, there’s a reason to pay extra for your marginal time.
In investment banking, in addition to the scarce skillset, there’s a need for staff to be up-to-date with a wide range of aspects of a deal. Similar in consulting. In tech, if you’re working with a complex and abstract problem, sometimes you might want one person to understand the whole codebase, and I’d speculate that’s one among many reasons that at high-levels, some coders are paid a lot more than others. Whether or not your current employer would reward you for working an extra 20 hours per week, it could make you able to perform a larger range of tasks.
Apart from work, you could use your time to gain skills like sales or programming, or to grow your network. Presumably, when we say it’s important for young people to grow their career capital, this is what we’re talking about. Or your could learn about effective charity.
Finally, if you want to start a new company, that’s just one of those sorts of jobs that will require extra work and that might offer returns above your average over 40 hours. But even before you quit your job to be a full-time founder, you’re going to need to use a lot of that marginal time to develop the company as a side-project.
There’s also arguments about getting a second job to earn more money, although that can be messy.
On this view, work would be something that people do not because they think it’s offering them good value for their time but because of necessity, or because of security.
Sure, there are a bunch of arguments that push the other way—if you work 40+ hours, you get tired, and such. But I think it might be rash to guess that marginal time is only worth 25-50% in tech. For many people, it’d be the reverse—their marginal time is worth more than their wage. If so, then using average wage as a first approximation might not be so misleading after all.
Hi Ryan!
My estimate of the value of a marginal hour in tech was based on my experience and pretty general reasoning, and I have pretty low confidence in it, so I think there’s plenty of room for disagreement. However, it’s still my best guess, and I have considerable certainty that a marginal hour in tech is worth less than an average hour. One place my intuition comes from is this: if I work 80 hours a week instead of 40 (or invest in career capital for that long), do I expect my lifetime expected earnings to double? This seems pretty unlikely to me. A typical senior software engineer salary is around $150K, so if I can get there working 40 hours a week, I would need to be earning at least $300K at the end of my 80 hours a week career (and probably more to make up for not immediately earning twice a typical junior salary now). Very few tech workers (or people who started their careers as tech workers) earn that much, and although working 80 hours a week may increase your probability of getting there, it certainly doesn’t increase that probability all the way to 1. That doesn’t prove that your 41st hour isn’t worth more than your 40th, but you have to hit diminishing returns at some point.
In general though, my main point is that marginal pay works differently from average pay, so you need to actually evaluate what you think a marginal hour of work is worth. Your right that for some people a marginal hour is worth more than the average hour, and for some people it happens to be basically the same, but that’s not true for most people. If you think that your marginal hour is worth the same as your average hour, you should by all means use that number when deciding how to use your time.
Maybe my impressions are atypical, but I think this is on the pretty low end at least for solid tech companies in SF? Like there are places (e.g. Dropbox, Google?) that sometimes pay 150K straight out of college, and most of the intermediate/senior people I know (admittedly not many and not representative) make way more than that.
It sounds about right to me, the 90th percentile for software developers in the US is $143k-150k excluding overtime.
Interesting. In terms of predicting Ruthie’s salary as a senior engineer, I don’t know whether to trust my anecdata (which presumably is sampled from closer to the same distribution as Ruthie), or the BLS, which is clearly a lower-variance estimate but possibly of the wrong quantity (systems programmers across the country).
The BLS also has regional data if you check the second link. Seems the BLS thinks programmers in the Bay Area make perhaps 1.3x the national average. But Bay Area programmers don’t just work for highly profitable tech companies… there are non-software-company salaries and salaries at less profitable tech companies pulling the average down. It’s also not clear whether the BLS is taking in to account equity compensation, which is obviously very significant at many Bay Area tech companies.
Not to nitpick, but this:
“Given this, my guess for the value of a marginal hour for a tech worker is something between a quarter and a half of their average salary. Again, this is enough of a difference that we could be significantly misled by valuing their time at their average hourly wage, as the examples above do.”
read like a bit of a strawman to me, because I’d read a couple of the examples and hadn’t noticed it. When I tried to go to the linked examples to discover whether they did in fact use average, rather than marginal, wage, the first link is broken and the fourth seems to link to the third (the third, as far as I can tell, doesn’t discuss voting).
With respect to kidney donation, all I see is talk about ‘lost wages’ in a somewhat opaque way that could mean marginal or average, but the implication is marginal (e.g. it talks about paid leave greatly lessening this cost) and so I think the charitable interpretation would be to assume marginal.
Finally, with respect to vegetarianism, I don’t see any explicit discussion of time/wage cost at all? In fact Katja’s explicit calculation all seems to be in dollars, as you (correctly, in my view) recommend.
I do of course agree that we should be thinking on the margin, and apologise if I’ve missed something.
Thanks for pointing out the broken links! They should be fixed now.
The relevant sentence from the first example:
There’s some discussion between me and some others on the kidney donation post that I think is also pretty clear. http://effective-altruism.com/ea/ay/kidney_donation_is_a_reasonable_choice_for/1ex
I read both of these as discussing average salaries, although you’re right no one says so explicitly. If everyone is already thinking on the margin and occasionally writing in unclear ways, then great.
I see, I didn’t look at the comments section. I agree that there people are talking about average wage.