How Valuable Are GiveWell Research Analysts?
Update 2016-05-18: I no longer entirely agree with this post. In particular, I believe GiveWell employees are more replaceable than this post suggests. I may write about my updated beliefs in the future.
Edited 2016-03-11 because I’ve adjusted my estimate of the value of global poverty charities downward, which makes working at GiveWell look worse.
Edited 2016-03-11 to add a new section.
Edited 2016-02-16 to update the model based on feedback I’ve received. Temporal replaceability doesn’t apply so I was underestimating the value of research analysts.
Summary: The value of working as a research analyst1 at GiveWell is determined by:
Temporal replaceability of employees
- How good you are relative to the counterfactual employee
- How much good GiveWell money moved does relative to where you could donate earnings
- A lot if you care most about global poverty, not as much if you care about other cause areas
- How directly more employees translate into better recommendations and more money moved
- This relationship looks strong for Open Phil and weak for GiveWell Classic
If you believe GiveWell top charities are the best place to donate, working at GiveWell is probably a really strong career option; if you believe other charities are substantially better (as I do) and you have good earning potential, earning to give is probably better.
While GiveWell employees probably aren’t directly replaceable2, they’re still temporally replaceable. Here’s what I mean. At some point either GiveWell will run out of money, or it will have to slow/stop expansion and level off expenditures. That means if you start working at GiveWell, you’re accelerating the rate at which GiveWell maxes out.
Right now GiveWell is hiring approximately one employee per month, so it’s probably reasonable to say that a marginal hire causes GiveWell to max out one month sooner.
If you assume you’re as good as the counterfactual future employee, the value of working at GiveWell is the value of the work you do during that one extra month. That means your value is small unless you’re better than the counterfactual future employee (my formal model below justifies this intuition). For each month you work, that’s work that would have gotten done one month later without you. But the employee who would have done that month next work can now do work that would have been done the month after, and so on. This is almost3 the same as if you added a new block of work that would not have been done otherwise.
It’s possible that this same reasoning applies to donations in some cases. I don’t believe it does in general, since GiveWell’s hiring and funding situation is fairly unusual, but are there other specific situations where it applies? I wrote a bunch of stuff about this and then deleted it because it was too convoluted, and then wrote a bunch of stuff again and deleted it because it was still too convoluted. So I’ll just leave this question as an exercise for the reader.
Value of Marginal Hires
It’s probably reasonable to assume that, given sufficient resources, GiveWell will eventually find all the best giving opportunities. (I would have been less confident about this a few years ago since old GiveWell ignored non-human animals, but modern GiveWell is better about this.) That means that by working at GiveWell, you cannot expect to cause GiveWell to move money to causes that otherwise wouldn’t get any; you can only cause it to move that money sooner. So the value of working at GiveWell is dependent on how much sooner you expect the money to get moved.
That means your primary added value over the counterfactual employee is that you move money to the best causes more quickly. So we get the value of your contributions by looking at how long everything you do would have taken without you, and subtracting how long it takes with you. Then we time-discount your contributions over this period.
It’s not clear how additional employees translate into more money moved. Marginal employees probably have a pretty trivial effect on increasing money moved to top charities, and a small effect on changing the top charities themselves4. The work on Open Phil has a more direct linear relationship between research time and money moved. (Note that even if you don’t work on Open Phil, you still allow GiveWell to do more Open Phil work by letting other staff members allocate more time to it.)
Right now about 95% of GiveWell’s money goes to its top charities, and only about 5% goes to Open Phil. I expect this to change based on what Holden has written on the Open Phil blog, although I don’t know how rapidly or by how much. If Open Phil starts making larger grants, we should probably attribute most of the value of the new grants to program officers. That’s bad news for research analysts (because they don’t have as much an effect on grants) and good for program officers (because they have a lot of authority over large grants). So even if we expect Open Phil grants to increase by a lot, the value of research analysts won’t increase by as much since the change will be primarily driven by program officers.
Value of Money Moved
How much good GiveWell does heavily depends on your estimate of the value of its money moved. If you believe that GiveWell top charities are the best place to donate, then working at GiveWell is relatively valuable. If you believe that animal charities or AI safety research matter more, then working at GiveWell looks less compelling.
This formula gives the value of working as a GiveWell research analyst in terms of dollars donated to your favorite charity.
- money_moved_classic: Money moved per employee by GW Classic
- money_moved_op: Money moved per employee by Open Phil
- adjust_classic: Dollars to your favorite charity equivalent to $1 to GW Classic recommendations
- adjust_op: Dollars to your favorite charity equivalent to $1 to Open Phil grants
- movability: Effect of marginal employees on increasing GW Classic money moved
- discount_rate: Discount rate on donated money
- delay: Time (in years) between employee hires
- years_working: Number of years you will work at GiveWell
- counterfactual: How much of your value would be lost by switching to the counterfactual future employee (0 means you are the same, 1 means the counterfactual employee is worthless)
adj_money_moved = (money_moved_op * adjust_op)
+ (money_moved_classic * adjust_classic * movability)
replacement_time_saved = delay
ctrfac_time_saved = (years_working - delay)
* (1 / (1 - counterfactual) - 1)
replacement_discount = 1 - (1/(1+discount_rate))^replacement_time_saved
ctrfac_discount = 1 - (1/(1+discount_rate))^ctrfac_time_saved
value = adj_money_moved * replacement_discount
+ adj_money_moved * ctrfac_discount / (years_working - delay)
adj_money_moved = (money_moved_op * adjust_op) + (money_moved_classic * adjust_classic * movability) value = adj_money_moved
(This model is simpler than my previous model because temporal replaceability was adding a lot of complexity.)
I created a Guesstimate sheet implementing this model. The sheet includes some estimates on each of these inputs.
When I make my best estimates for all the inputs, I get that me working at GiveWell is equivalent to
$12,000 $110,000 $68,000 donated to my favorite charity, assuming I’m as good as the counterfactual future employee5.
There are a few reasons why you might disagree with this estimate:
- You believe GiveWell top charities are better relative to the best charities than I do
- You apply different time discounting to donations
- You believe you’re much better than most GiveWell employees
The result of this model is highly dependent on how good you believe GiveWell top charities are relative to the best charities. If you believe GiveWell top charities are the best charities, then working at GiveWell is probably better than earning to give unless you plan on donating millions of dollars. If you go way to the other extreme and believe that one particular cause is thousands of times better than GiveWell top charities, then working at GiveWell looks less appealing. Of course, you could work at GiveWell and try to move more money toward your favorite cause, which is a strategy my model doesn’t account for. I don’t believe this is a huge factor because you probably couldn’t have much influence unless you were an unusually good researcher, but I can’t make that claim with much confidence.
This model no longer cares about how good you are relative to the future counterfactual employee. That does matter, but it doesn’t matter until GiveWell runs out of room for capacity, so this model roughly accurately tells how much good you do until that point comes.
If marginal employees aren’t worth much, why has GiveWell been so valuable?
Organizations seem to have rapidly declining marginal utility of additional employees. We can observe this more easily in for-profit companies: founders capture a huge portion of the wealth generated, and early employees get a lot more than late employees. Even for giant companies like Microsoft or Google, founders usually get around 20% of the wealth and early employees get around 1%. I won’t write the percent that late employees usually get because it will contain too many zeroes after the decimal point.
We should not be surprised to see something similar for nonprofits. Maybe 50% of the value of GiveWell came from the fact that Elie and Holden created it, and marginal employees today add much less value.
Plus GiveWell has the fairly unusual situation where it barely does any advertising or outreach. Presumably early on it could get more donations by producing more impressive or convincing research, but I don’t believe that’s true anymore. Quality of research has not prima facie improved in several years. Marginal improvements in quality of research probably have tiny effects on increasing money moved at this point.
Other grantmaking organizations
There’s one major disadvantage to working at GiveWell: it’s going to make grants to effective organizations whether you work there or not. Perhaps there are other organizations making large grants that would be amenable to an effectiveness-driven mindset, but currently allocate their grants suboptimally. You might be able to do more good by working at an organization like that. I haven’t looked into this so I don’t know how easy it would be to get a position or influence grants, but this is a potentially promising option. If you can work at a grantmaking organization where the counterfactual grants are a lot worse than the ones you would make, you might be able to do a lot more good than you would at GiveWell.
Significance of this model
According to this model, working at GiveWell is equivalent to making a pretty sizable donation. But I don’t think this automatically means you should work at GiveWell rather than earn to give. Here are a few reasons why:
- This estimate is pretty non-robust. In the course of developing it, I changed the estimate by a factor of 10 in both directions. I wouldn’t be surprised if it’s still off by an order of magnitude up or down. Conversely, the case for earning to give is pretty straightforward.
- It’s generally easier to move from the for-profit sector to the non-profit sector than the reverse, although it’s plausible to me that for-profit employers would look more highly on GiveWell than on most non-profits.
- If you identify substantially better giving opportunities in the future than what you’re currently aware of, it’s easy to redirect your own money that you’re donating, but it would be relatively harder to move Good Ventures money.
- You make less money working at GiveWell so there’s greater personal risk.
There are some other reasons in favor of working at GiveWell:
- You’d learn more about how to investigate charities.
- If you work on the Open Philanthropy Project, you might be able to direct grants toward more effective areas. (I don’t know how much influence research analysts have on this.)
- Money moved per employee could increase a lot in the future.
Overall my best guess is that I’m indifferent between working at GiveWell and donating $50,000 a year. But I’m pretty uncertain about this.
Most of the reasoning in this post applies to non-research positions as well, but I focus on research analysts. ↩
GiveWell currently has enough funding to hire every strong candidate it interviews. Hiring is largely constrained by its ability to find strong candidates, so if you apply and get hired you’re probably not replaceable. ↩
It’s not quite the same because you aren’t simply adding a new unit of work, you’re shifting over all the work. So the proper way to compute the value is to take the sum of the time-discounted savings from pushing up this month’s work, plus the time-discounted savings from pushing up next month’s work, and so on to infinity. The result is slightly less than one (it depends on parameters but probably something like 0.95). So the value of adding a new month of work is about 0.95 times the value of that month’s work. ↩
This is based on the fact that GiveWell top charities have barely changed over four years in spite of spending maybe 50,000 hours of staff time researching them. ↩
I used a slightly more complex model to calculate these numbers. I broke down GiveWell’s historical money moved and grants by category and made an adjustment for each category, then used these to predict future adjusted money moved under the assumption that Open Phil will significantly expand. ↩