Has 80,000 Hours justified its costs?

NOTE: This piece is now out of date. More current information on our plans and impact can be found on our Evaluations page.

---

We set up 80,000 Hours because we thought it could become one of the most effective charities in the world. The idea was to achieve a multiplier – with a small amount of our time, we could enable thousands of others to spend their careers on whatever is most effective, and achieve thousands of times as much as we could individually.

In this post, we examine whether 80,000 Hours has generated enough impact to justify its costs over our history, and make some rough estimates of our multiplier.

Because it’s hard to estimate what would have happened if 80,000 Hours had never existed, all of these estimates are very uncertain, and can be debated. However, there are multiple ways we’ve plausibly justified our costs to date. In this document, we sketch out some of these pathways. We’re not aiming to be fully rigorous. Rather, consider the examples as a group. If only a few turn out to be genuine cases of impact, we’ll have justified our costs many times over.

Is this the wrong question?

80,000 Hours is a startup. Asking whether we’ve justified our costs to date is like asking whether Google was profitable in 2000. The aim of 80,000 Hours is to grow, and have a far larger impact years in the future.

Most of the value of donations to 80,000 Hours comes from the chance that these donations enable us to grow 10-times or 100-times.

That said, we think 80,000 Hours is already “profitable”, in the sense that each year we generate enough impact to justify our costs in that year. This means that even 80,000 Hours fails to grow, it’s still a good use of funding. That’s what we’ll discuss in the rest of this post.

(Read more about the marginal vs. growth approach to evaluating non-profits.)

How many plan changes have there been?

In total, we’ve tracked just over 1,500 impact-adjusted significant plan changes (IASPC). In our plan change analysis you can read more about how we track them, and what they consisted of on average.

Our costs to date

Direct costs

We’ve spent about £700,000 as of Nov 2016.

In addition, there have been about 28 person-years spent on the project (that we’ve tracked). This includes freelancers, interns, and volunteers as well as full-time staff.

This means the ratio of costs to plan changes is £470 and about one person-week.

Opportunity costs

The person-years have an opportunity cost, because if we hadn’t worked on 80,000 Hours, we could have worked on a different project instead. As a lower bound, we could have earned to give. In a previous review, the team estimated they would have donated about £18,000 per year if they hadn’t worked at 80,000 Hours. That would suggest the opportunity cost of time is about £504,000. So our true costs are probably at least double the £700,000 figure.

As a ballpark estimate, let’s suppose for now that our total costs have been £1.5m, or £1000 per plan change. If we hadn’t existed, most of this would have gone to GiveWell-recommended, meta and existential risk charities otherwise.

Donations in kind

We’ve also benefited from contributions by local student groups, especially those who have helped run our workshops. In 2016, we estimate that they spent about 500 hours helping 80,000 Hours, not including time they would have spent otherwise and time we paid for. This, however, was only 5% of the labor that went into 80,000 Hours over 2016, and probably has a lower opportunity cost than staff time.

In addition, we receive some input from other groups, such as GWWC, though again, it’s relatively small.

We also receive discounted services from companies (listed here), though, for the most part, the value of these only accounts for a few percent of the budget.

One exception is that Google gives us $40,000 of free AdWords per month as part of their Grantspro program that can be spent on keywords with a value up to $2. If valued at market rate, this would be over half of our budget, though we’d never pay this much if they weren’t free.

Another exception is that this year we received pro bono legal advice on visas, which would have been worth about $10,000 (3%).

We ignore donations in kind in the rest of the estimates.

Complications with the estimates

£1000 is an overestimate of the cost per plan change, because even if we stopped working on 80,000 Hours tomorrow, the website and community would continue to cause about 50 plan changes per month (at least for some time).

Moreover, we don’t track every plan change – some people never fill out our survey, while some might be influenced indirectly and not know it was due to us.

This metric also doesn’t account for all our impact. Small shifts aren’t counted – to be a significant plan change someone must clearly switch from one option to another. Second, we help our users in other ways, such as making them more ambitious, altruistic and successful.

What’s your prior?

Before we go into the estimates, two notes to bear in mind:

1. A significant plan change is a big deal. Each change means someone at least spends years of time differently, and that has a knock on effect on the rest of their career. £1000 is small compared to the value of this time. To quantify very roughly, most of the plan changers are college graduates, so will have lifetime incomes over £2 million, so £1000 is only 0.05% of that.¹

2. We think lots of the advice in our guide can increase someone’s lifetime impact several fold, and is not provided by other careers advisors. Some examples include: (i) our framework for problem selection (ii) earning to give as an option (iii) the concept of flexible career capital (iv) suggestions for specific paths few people consider, such as becoming an academic research manager.

If you agree that some of the plan changers are really changing behaviour and that our advice is better than what people normally receive, then it’s likely that we’re worth funding.

Now we’ll make some quantified estimates.

Additional donations

Giving What We Can pledges

Our most quantifiable impact comes from raising money for high-impact charities. One way we do this is by encouraging people to take the Giving What We Can pledge to give 10% of lifetime income to effective charities.

We do this by giving the arguments for the pledge in the career guide and workshop, and there are reminders in our newsletter and tools.

In 2016 we tracked 115 people who took the GWWC pledge due to 80,000 Hours. This is 8% of our plan changes for the year, and 13% of our impact-adjusted plan changes.

Doing this incurs some costs for Giving What We Can (GWWC), because they also sometimes speak to the pledgers, though it’s unlikely to be more than 100 hours per year (~1% of the labor going into 80,000 Hours), so we’ll ignore it in the calculation.

We track this figure by emailing people who take the pledge and say that they first heard about GWWC through 80,000 Hours, and ask them how likely it is that they would have taken the pledge if 80,000 Hours didn’t exist. If they say they would have taken it soon anyway, then they’re not counted. In addition, we track people who say in our impact surveys that they now intend to take the pledge due to 80,000 Hours, and who then become members.

(We also know that 5-10 people take the pledge each month after having clicked through to GWWC’s website from 80000hours.org. We don’t count these directly but they provide independent evidence that we’re persuading 60-120 per year.)

We think 115 is a reasonable estimate of how many extra people took the pledge due to 80,000 Hours. You could argue that people underestimate how likely they would have been to join GWWC anyway a couple of years later. On the other hand, there are also people we helped to persuade that GWWC later pushed over the edge, and so are attributed to GWWC instead of us. The two effects should roughly balance. (We discuss how to handle cases where two people cause the same outcome in more depth here.)

As of Dec 2016, GWWC estimates that each pledge is worth £58,000 of additional donations to top recommended charities. This accounts for (i) money that would have been given otherwise (ii) people who drop out from the pledge (iii) time-discounting.²

In 2016, the funding went 54% to GiveWell-recommended charities, 39% to meta, 5% to existential risk research, and 2% to factory farming.

If 13% of the IASPC take the GWWC pledge, then that’s £7540 per IASPC (13%*£58,000), compared to our historical costs of £1000. (GWWC also incurs some costs gaining these pledgers, but unlikely more than about £50.)

Taking these numbers at face value would mean it has been significantly more effective to fund 80,000 Hours than to donate to GiveWell-recommended charities.

However, there are many complications with this estimate. GWWC’s counterfactual adjustment is based on asking the pledgers how much money they would have given to effective charities if GWWC had never existed (and is about 50%). It’s not clear we should use the same percentage.

On the other hand, the pledgers who come through 80,000 Hours seem to be higher quality than average. For instance, they’re more likely to log their donations on the site. Second, we expect they’re higher income, because we do more to promote earning to give.

We also expect that more donate to meta-charities, factory farming and global catastrophic risks because we explicitly advocate for these causes.

The earning to give community

80,000 Hours encourages some users to earn more so they can give more – “earning to give”.

We identified 20 of the largest donors who have made a significant plan change, and asked them (i) how much they’ve given after making the plan change (ii) how much they intend to donate over the next three years (including the chance that they stop earning to give) (iii) whether they’ve made a lifetime pledge to give, and (iv) how much they attribute to us (i.e. how much extra they gave to equally effective charities due to 80,000 Hours).

15 responded, and here’s what we found:

What’s the multiplier?

If you only count historical donations, on this basis, our multiplier is about 0.5 i.e. we’ve spent two dollars for each extra dollar donated. This is because we incurred costs of £1.5m to gain donations to £675,000.

However, for this to be a reasonable estimate of our impact, you’d need to assume that all the donors stop giving immediately, which is unlikely because most of them intend to continue, and ⅔ have taken a public lifetime pledge to give.

If we add in donations 2017-2019, then the total attributed to us is £6.7m, making the multiplier 4.5. Expected donations 2017-2019 are much larger because our users are young, so their incomes are still growing rapidly.

Note that we’ve included opportunity costs in the £1.5m cost figure. If we were only counting financial costs, then the multiplier would be about 9. Most estimates you see in the community do not include opportunity costs, though these are often larger than the financial costs, because people work at far below their market rate. This means most ratios are overstated by about two-fold (the lower the salaries, the worse the overstatement).

A multiplier of 4.5 is probably still an underestimate, because it ignores donations beyond 2019. For all the donations who made a lifetime pledge to give, we made a very rough estimate of their expected lifetime income, and multiplied it by the percentage they’ve publicly pledged. This produced £51m of lifetime donations. If we use the same percentage attribution as we used for the 2017-2019 period, then the total attributed to us is £19m, which would make the multiplier 12.7.

Where will the donations go?

We made rough estimates of the breakdown between area for each donor, and found the following:

• 79% meta-charity and existential risk organisations.
• 20% GiveWell-recommended charities.
• 1% other.

It’s likely some will be redirected into things no-one has thought of yet, but we didn’t try to estimate this.

Problems with the estimate

The percentage that the donors attribute to us is the most difficult part to estimate. Most of the donors were influenced by several people and organisations, so many would have earned to give even if 80,000 Hours never existed. Our hope is that if we ask lots of people, the errors will average out. The overall average percentage is in the 30-50% range, which seems reasonable.

There are probably systematic biases that we can’t average away. It’s unclear, however, which way these biases go. Some people will inflate the figures because they’re telling us what we want to hear, but others might deflate the figures because they don’t want to admit being influenced, or are not aware how they were influenced (since 80,000 Hours is responsible for the idea of earning to give becoming widespread).

More importantly, there are probably other earning to give donors who were influenced by us who aren’t counted because they’re not tracked as a significant plan change. The donors tracked here only account for about 20% of the capacity of the meta-charity-focused earning to give community. But as the organisation that has done the most to promote the idea, it’s likely that a significant fraction of donations by the earning to give community wouldn’t have happened otherwise.

Another major difficulty with the estimates is taking into account the opportunity cost of earning to give. If these people hadn’t earned to give, they might have pursued direct work otherwise, so the value of that work needs to be subtracted. Our guess is that this doesn’t reduce the value of the donations by more than 50%, this is because:

• Everyone in the survey thinks they’re having more impact earning to give than they would have otherwise (and this seems plausible to us too).
• Many would have pursued a similar career, but donated a smaller percentage, or given to less effective charities.
• One person switched from software engineer to entrepreneurship, so is likely having a larger direct impact.

However, some of the people earning to give could have pursued academic careers otherwise, and it’s possible they might have had more impact like that.

We also haven’t factored in time-discounting, though 5% discounting over 15 years only reduces the total value by about 50%, so it’s not a major factor except in the lifetime donations estimate.

Overall, there’s a lot of uncertainty in these estimates, but we think it’s plausible that 80,000 Hours has justified its costs purely on the basis of these earning to give donors.

Overlap with Giving What We Can pledges

One third of the large donors have not taken the Giving What We Can pledge, and many of the others have pledged less income than they actually intend to donate, so these weren’t counted in the GWWC pledge estimates. Combining these two effects suggests that 40-50% of the earning to give donations are additional to the GWWC estimates.

Looking at it the other way around, the top 10% of donors account for about 75% of the money given by the GWWC pledgers, so if we only look at top earning to give donors, then we’re missing 25% of the total. Either way, 25-50% of the donations don’t overlap.

Very large future donors

3 of the 1,500 significant plan changes expect to donate over $100m to effective charities over their lifetime, and say they have been influenced by us (especially in their choice of cause), though have not yet made significant donations. They are not included in the estimates above.

If our advice and support helps any of these three spend this money a little more effectively, then the multiplier will be much higher again. This is all very uncertain, but it shows that there’s potentially a lot of extra upside in the estimates.

We’ve also spoken to plenty of other very large donors through 80,000 Hours, who haven’t yet made a significant plan change, but we expect many of them will end up giving differently. These are mainly people Will MacAskill met through the promotion of Doing Good Better, and people we’ve met through the Y Combinator community.

Our influence here arises more through personal connections than through programs run by 80,000 Hours, however, the more 80,000 Hours succeeds, the more impressive people we meet, so many of these connections wouldn’t have happened if 80,000 Hours did not continue to grow. We also met two of the three large donors at events we ran.

Founder’s Pledge

Founder’s Pledge encourages entrepreneurs to donate at least 2% of their profits on exit to charity and has raised over $170 million of legally-binding pledges (based on current market valuations) in just a couple of years. It also seeks to involve the entrepreneurs in direct work.

David Goldberg, the founder, came across effective altruism through Peter Singer’s TED talk. He then invited Will MacAskill to speak at a conference on behalf of 80,000 Hours. Will introduced him to the research on effective giving, which led to a partnership between CEA and Founder’s Pledge in which CEA advises the entrepreneurs on where to donate. CEA also introduced them to several donors, who later made large grants to Founder’s Pledge.

David has estimated that 33% of the funds will now go to non-profits in the effective altruism community, and several large donations have already been made. David says “overall, 80,000 Hours has been instrumental. It will increase my lifetime impact many times over.” (Read more.)

In addition, two of the organisation’s early hires are plan changes. Ben Clifford is one of the key people involved in signing up new entrepreneurs. Ben says reading about Founder’s Pledge on our blog made him more likely to take the job. The other is currently doing a trial.

Let’s guess that the existence of 80,000 Hours will mean that an extra 1% of the money goes to GiveWell-recommended charities or better. 1% of $170m is $1.7m, which is £1.4m. This roughly justifies all our costs to date. Moreover, the total size of the pledged donations are growing rapidly e.g. Founder’s Pledge just partnered with Y Combinator.

This is not a robust estimate, but doesn’t seem unreasonable. The true figure could easily be many times higher.

Raising for Effective Giving

Raising for Effective Giving (REG) is a meta-charity that raises money for effective charities from poker players. Poker players pledge at least 2% of their winnings to charities suggested by REG. In late 2014 and 2015 it raised $1,146,683 while incurring direct costs of around $127,000. Its staff could have donated approximately $60,000 had they earned to give instead. This suggests a net gain of $1m so far.

Around half of this money has gone to poverty reduction and animal charities recommended by GiveWell or Animal Charity Evaluators, with the rest going to existential risk reduction and meta-charities. It is likely to continue moving at least $500,000-$1m to charity each year.

80,000 Hours increased the probability of it existing significantly and/or caused it to be founded sooner:

• Jonas Vollmer, who started REG along with Adriano Mannino, Tobias Pulver and Ruairi Donnelly, was convinced to switch from earning to give in medicine to direct work, in significant part due to our directly relevant online content such as How to create the world’s most effective charity.
• Tobias Pulver who helped get REG off the ground, also reported that choosing to work in an effective altruist organisation rather than on global poverty in government or an NGO was attributable to the ideas and community produced by 80,000 Hours.
• Ruairi Donnelly received coaching from 80,000 Hours and others who advised him to do direct work rather than earning to give. However, there’s some chance this would have happened anyway.
• Jonas, Ruairi and Tobias all went on to have major management positions in the Effective Altruism Foundation and play a large role in the Germany/Switzerland EA community.

Would REG have happened if 80,000 Hours had never existed?

• It is likely that REG or another fundraising project would have been founded by the Swiss effective altruism community, albeit later or with fewer staff.
• Adriano Mannino also would have worked on REG or something similar in our absence, and was the person who brought the connections to the poker world.

Overall, we think it’s plausible that 80,000 Hours sped-up REG’s progress by 6 months, which, based on REG’s current scale, has a value of about $270,000 to top charities ($350,000 minus costs of 23%), or £216,000.

Plan change impact

So far, we’ve ignored the impact of over 80% of the plan changes, because we’ve only looked at extra donations. The aim of 80,000 Hours, however, is to solve talent gaps rather than funding gaps.

Although it’s much harder to measure whether you’re solving talent gaps, we think they’re often more pressing, and they’re also more neglected by the effective altruism community. We’ll now look at some of the impact we’ve had through career changes.

We expect that the impact of career changes is fat tailed, so the most high impact 5% of examples account for perhaps 50% of the total impact. So, in this section, we look at case studies of especially high-impact changes.

There’s a huge amount of uncertainty in all of these case studies. There were multiple influences on each person, so it’s hard to know the counterfactual. However, many of the cases alone could justify our entire costs to date. So, even if you just think one or two of these were real examples of our impact, then we’ve justified our costs.

Moreover, this is an underestimate of the total impact, because it ignores everything that the plan changers achieve in the future.

Global Priorities Project

The Global Priorities Project (GPP) carries out research into how to compare global priorities, and advocates for the adoption of top priorities by governments and other large funders.

In Dec 2015, £2.5bn of UK aid spending was reallocated in the direction it was advocating. GPP estimates this was equivalent in terms of QALYs to £1.4 – £24bn donated to Against Malaria Foundation. It’s hard to estimate GPP’s effect on the spending and how effective it will really be, but it’s not out of the question that GPP had more impact than the entire effective altruism movement so far.

In addition, GPP made some of the first efforts to develop policy around existential risks, for instance, creating a policy paper that was read at the highest levels of the UK government, and holding the first conference for European policy makers on existential risk. These efforts could easily lead to far more impact again. Finally, GPP did novel global priorities research.

80,000 Hours has contributed to GPP in the following ways:

• The first staff member hired by the project, Owen CB, decided to switch into this research field from pure mathematics in part due to 80,000 Hours. Owen says it’s unlikely he’d have the job if 80,000 Hours had never existed (though it was more driven by the community and thinking about the ideas than using the guide or coaching).
• GPP was initially managed by Niel Bowerman, another significant plan change.
• In 2016, GPP hired two researchers who have made significant plan changes, and were influenced towards global priorities research by 80,000 Hours. Max Dalton would be studying philosophy otherwise, and helped to prepare the cost-effectiveness estimates used in the DFID proposal. Ben Garfinkle would have done theoretical physics otherwise.
• We helped them raise £50,000 of seed funding.
• We produced research arguing that a project like GPP is high-priority, building early support for the organisation.
• We helped prepare the initial business plans for the organisation.
• GPP was incubated in CEA, an incubator we co-founded.

Would GPP’s impact have happened if 80,000 Hours had never existed?

• GPP itself probably wouldn’t exist, but CEA’s policy efforts were started by Toby Ord, so some measure of policy work would have happened otherwise.
• The Future of Humanity Institute was also involved in the existential risk policy efforts, so some of that would have likely happened otherwise.
• However, it’s difficult to get policy change implemented without research and organisational credibility, which was aided by GPP’s existence and work done by their staff. In particular, the proposals concerning research spending in international aid were prepared by GPP staff, and the existential risk policy conference was run by GPP staff.
• Even if GPP never existed, 80,000 Hours expanded the total pool of talent working on global priorities research, so these people might have contributed in another organisation.

Putting all this together, we think 80,000 Hours made it significantly more likely that this impact happened, and this could easily justify our costs many times over.

Note that GPP has now been merged with the rest of CEA, as the “policy team” and “fundamentals team”.

Animal Charity Evaluators

Animal Charity Evaluators (ACE) is the leading meta-charity focused on animal welfare. It does research into the most effective ways to reduce animal suffering and recommends charities to donate to. So far, it has raised over $3m for its recommended charities (up from $1m last year), and has a team that’s larger than 80,000 Hours itself. ACE is being considered for a grant from Open Philanthropy, so seems set to grow further.

ACE likely wouldn’t exist without 80,000 Hours. Here’s how we helped:

• Animal Charity Evaluators grew out of Effective Animal Activism, which was founded by Eitan Fischer during an internship at 80,000 Hours.
• We contributed to the initial concept for the charity and introduced it to its initial donors.
• The project was also initially contained within 80,000 Hours and received substantial aid from the team, for instance, we built the first version of the website.
• Two of the current staff are plan changes due to 80,000 Hours, and were influenced towards working at effective altruist organisations by us. One of them learned about the job through us.
• One of the 80,000 Hours team members, Rob Wiblin, was a founding board member of the organisation.

Might ACE have happened otherwise?

• Yes, the idea to set up an animal focused meta-charity was an obvious one, since a significant fraction of the community were focused on this cause, because there was no group making charity recommendations for them. Other groups would have likely done something similar otherwise.
• Brian Tomasik, one of the first donors and board members, has given seed funding to several other animal-focused meta-charities.
• Jon Bockman, the CEO and first full-time employee, came from the animal community, rather than 80,000 Hours, so may have done something similar eventually otherwise.
• However, based on the above, it’s likely that 80,000 Hours still accelerated the process, perhaps by several years. Given that ACE is now directing $2m of donations per year, which is £1.6m, even a one year speed-up would be enough to cover our total costs to date.

Technical AI safety research pipeline of 50 people

We think one of the most pressing problems is AI safety research, and one of the biggest constraints facing the area this is technical researchers who are concerned by existential risks. So one of our priorities from 2015 was to start building a pipeline of researchers.

We used our career review (viewed over 8,000 times) and workshop (attended by over 2,000 people, mainly at global top 20 universities) to gather 60 leads – people who have relevant expertise and seem serious about working on safety issues (e.g. have read Superintelligence, care about existential risk). In total, the pipeline consists of:

• 21 working as software engineers
• 6 working as data scientists
• 3 completing a relevant PhD (maths, physics, CS or computational neuroscience)
• 14 currently applying to relevant grad school
• 13 completing a relevant masters
• 12 completing a relevant undergraduate

(Sums to 69 because options not mutually exclusive.)

The majority of these people say they were influenced by us towards this path, and 46 have reported significant plan changes.

We took the most promising 50 or so people, and gave them one-on-one advice, either by Peter McIntyre, our new head of coaching, or David Krueger, a machine learning researcher who has worked at the FHI (and as of Dec, we pay to work as a freelance mentor).

One of these people has already taken a job at a top AI organisation. This person says they were influenced towards AI by us, and wouldn’t have applied without our encouragement.

We’d be surprised if several didn’t end up as successful safety researchers. People in the AI risk community have said they’d trade a new safety researcher for $1m per year in donations, so if we get one, then we’ll have justified our costs.

Many of the others will work on capacity building, earning to give or policy instead. In the next year, we’d like to start a mentor network in AI policy too.

AI risk community capacity building

Additional plan changes include:

• Niel Bowerman is the Assistant Director of the Future of Humanity Institute, where he carries out day-to-day management of the organisation, including much of its fundraising and outreach. FHI is one of the most important organisations working on existential risks, and has been bottlenecked by management capacity for much of its history. Niel became involved in effective altruism through 80,000 Hours and Will MacAskill back in 2011-2012 (read more).
• One researcher at DeepMind who wouldn’t be focused on AI risk if it weren’t for us.
• One researcher at CSER with policy experience wouldn’t be focused on AI risk or have a policy background if it weren’t for 80,000 Hours.
• Nick Robinson works at ASI data science bootcamp, which has been funded by Jaan Tallinn, and aims to train up safety-concerned researchers. He’s highly influenced by us, and wouldn’t have this job otherwise.

We’ve also been able to use 80,000 Hours to meet lots of influential people in the tech community. In particular, we’re part of the Y Combinator network, which includes Open AI, and we have met many of the most influential people involved.

Building the Effective Altruism community

One of our main aims with 80,000 Hours is to build the effective altruism community. We think building the community is likely more effective than working directly on global problems because, among other reasons, it’s more flexible – the community will change where it focuses as we learn more about which problems are most pressing.

We think the effective altruism community as it exists today is going to go on to achieve a huge impact. 80,000 Hours is one of the main groups that has built the community, and was first to use the term in outreach in 2012, so some reasonable fraction of the community’s total impact wouldn’t have happened if 80,000 Hours had never existed.

We help the community both by getting more people involved, and by helping people who are already involved to be more effective.

How we grow the community

We get more people involved because our career guide and workshop introduces all the key concepts in effective altruism, encourages people to get involved in the community, and helps them take jobs at effective altruist organisations.

Some results:

• In the last 12 months, 80000hours.org was the largest effective altruist website by traffic,³ and the newsletter is bigger than all the other effective altruist organisations put together at over 88,000.
• In 2016, 43% of plan changers say they were influenced to “become more involved in the EA community” (based on a sample). If this applies to the entire 1,400 plan changes over 2016, then it’s 600 people.
• 80,000 Hours was the largest and highest quality source of applicants for Effective Altruism Global this year (measured in terms of acceptance rate).
• Many key student group leaders are plan changes and would have been less likely to do effective altruism outreach otherwise.
• Many key staff at effective altruist organisations are plan changers and many wouldn’t have the job if it weren’t for 80,000 Hours. Just in 2016, we tracked 9 people who took jobs at effective altruist organisations due to us. The two most senior staff at CEA besides the trustees, Tara Hedley and Michael Page, probably wouldn’t have the job otherwise.

How we help the community be more effective

We also help to make the existing community more effective. One way we do this is by helping people who are already involved in the community choose more effective careers, with our general career advice. About 9% of the plan changes are dedicated members of the community, over 10 per month.

In addition, we do novel research that moves the community’s understanding forward, and are the only group in the community focused on identifying the most pressing talent gaps as opposed to funding gaps. Some of our research consists of our career reviews and problem profiles, but we’ve also helped to contribute many key strategic considerations, including:

• The importance of career choices in effective altruism. Much of the community was previously highly focused on donations. 80,000 Hours is the main group to make career choice a key part of effective altruism.
• The concept of talent gaps
• Rules of thumb for coordination
• Replaceability and it’s importance.
• Earning to give, its pros and cons, and who should do it.
• The quantitative problem framework
• The growth approach.

All of this means that 80,000 Hours is probably contributing a significant fraction of the impact of the community.

Cost-effectiveness

At the same time, we’ve done this with a comparatively small budget. In 2016, 80,000 Hours only had 22% of the budget of CEA and 6% of the budget of GiveWell.⁴ When compared in terms of “effective altruism community growth per dollar”, 80,000 Hours has a good claim to have been one of the most effective organisations.

Would all of our impact been caused by the community otherwise?

This point also answers the most common criticism of our plan changes, which is that they would have been caused by the rest of the community anyway. But since 80,000 Hours also grows the community, impact that’s due to the community is also partly due to us.

A more reasonable approach is to assign a fraction of the community’s impact to 80,000 Hours, where you estimate the fraction based on the size of our contribution compared to other groups (read more).

In addition, only 9% of the plan changers say they’re already heavily involved in the community, so it’s implausible all of the impact would have happened otherwise.

What’s the marginal cost per plan change?

The ratio of costs per IASPC has fallen four-fold in the last four years. This is because we’re learning how to create plan changes more efficiently, and benefiting from economies of scale (see our post on increasing returns).

In this document, we’ve used a figure of £1000 per plan change, including opportunity costs.

In contrast, in 2016, the ratio of financial costs per plan changes was £250, or if you also include opportunity costs, perhaps £500 per plan change.

Projecting out the future cost per plan change based on these figures is difficult. The 2016 ratio is too high because it ignores future plan changes that result from work done in 2016 – if we stopped working on 80,000 Hours tomorrow, we’d continue to produce about 50 IASPC per month (at least for 6+ months).

The ratio is also too low because it ignores that we would have had some of the plan changes in 2016 anyway due to work done in previous years. However, our guess is that this is the smaller effect, because if we hadn’t worked on 80,000 Hours over 2016, our estimate is we would have only caused about 30 IASPC per month.

So, we think £250 is in the right ballpark, and is probably too high.

As further evidence, in 2016, we hired Peter McIntyre and he gave one workshop a week, creating plan changes for about £210 each over the year (and less if you consider the long-run impact of the workshops). This is an indication of the kind of opportunities we have at the margin.

Finally, note that the largest determinant of our cost per plan change over 2017 will simply be how aggressively we invest for growth. If we make lots of investments that will pay off in 2018 but not 2017, then it will drive up the ratio of cost to plan changes over 2017. This wouldn’t, however, mean that we were becoming less cost-effective.

We also think it’s likely we’re underspending right now. £250 per plan change is very cheap, and it’s plausibly worth spending 5-times as much. For this reason, we want to significantly ramp up our rate of spending until the price at the margin starts to rise. This will mean we’re less cost-effective at the margin, but have a much larger total impact.

Conclusion

The nature of our work means that each example of impact is debatable, however, the number of ways that 80,000 Hours has made an impact are sufficient that – even at a conservative estimate – we have likely justified our costs many times over.

Turning to funding, 80,000 Hours has probably had a positive multiplier when just assessed in terms of how many people take the GWWC pledge. In addition, according to the estimates above, it will have raised £6m from the earning to give community by 2019 (of which 50-70% overlaps with the GWWC pledgers), £1.4m from aiding Founder’s Pledge, £240,000 from aiding REG, and £1.6m from aiding ACE (which also does valuable research). After this, there’s still potential upside from very large donors.

This funding, however, is probably a minority of the impact, and arises from a minority of the plan changes. 80,000 Hours has also contributed to policy through GPP, has helped to build the AI safety community, and found many of the key people who work at effective altruist non-profits. In addition, 80,000 Hours has been one of the main groups that has built the effective altruism community.

When compared to donations to GiveWell-recommended charities, we’re 90% confident it has been more effective to donate to 80,000 Hours so far. 80,000 Hours has also plausibly been a large multiplier on existential risk, animal and other meta-charities.

We’ve tried to share the reasoning behind this here, though it’s hard to communicate all the information we have about the plan changes. If you’re interested in donating, you can do so here.

Finally, we address the question of whether 80,000 Hours is likely to remain cost-effective here.