In this academic session, Joan Gass, who holds a Master’s in Public Administration from the Harvard Kennedy School, argues that members of the EA community who want to work in global development should adopt a “venture capital” approach, looking for opportunities to make bets on the cause areas with the highest expected value. In particular, she contends that fostering economic growth in emerging markets and building state capability are high-value areas, while discussing other areas in RCT research and social entrepreneurship that are relatively low in value and should be avoided.
Below is a transcript of Joan’s talk, which we’ve lightly edited for clarity. You may also watch it on YouTube or read it on effectivealtruism.org.
When I started at the Kennedy School, I was in an international development-specific policy program. I approached our program facilitator at the beginning of the program and started talking to him about career decisions. And this is what he said to me:
He had some really strong thoughts about career options and what he thought were the highest-impact things to do.
Some caveats before I begin this:
I wrote this paper for my global development thesis and did it over a three-to-four month period. It is specifically for students who are thinking about careers in global development or public policy. As a result, it doesn't cover the full spectrum of what we typically talk about within effective altruism. In particular, it doesn't take into consideration longtermist thinking or animal welfare. It’s just targeted toward an audience of my peers within the program that I was working in. Also, I don't think it's exhaustive. I chose a few examples of cause areas within global development to illustrate a methodology that I was interested in exploring, but [those don’t reflect] an exhaustive search.
So why does it matter what one chooses to focus on within global development?
I think we've seen some examples of things that people have done within the global development space that have been complete home runs. [Norman Borlaug] invented a key strain of wheat during the Green Revolution Revolution that saved between 10 million and 1 billion lives. Eliminating smallpox averted 1.5 to 2 million deaths per year. And there's some research that shows that four out of five Haitians who have emerged from extreme poverty have done so because they immigrated out of Haiti. Those are some examples of “home runs” from a policy perspective or global development perspective.
At the same time, there have been actions that have been really negative for the world. For example, Brazil had a slowdown in the 1980s that resulted in a current net loss of $7.5 trillion in value.
When thinking about cause prioritization, I took the classic effective altruist [approach of considering] impact, neglectedness, and tractability, and tried to adjust it using metrics applicable to the global development space. I'll run through how I thought about [each area] in the context of my thesis.
Neglectedness is about the people and resources currently [devoted to] an issue. I used proxies related to World Bank funding, with some modifications, to think about how much money was going toward certain spaces. I also did a survey of [Kennedy School] alumni who had been out of my program for a few years to think about how many people were working on a given issue.
When thinking about impact, I looked to see how much improvement in human well-being there might be if a problem were solved.
When thinking about tractability — which is, I think, one of the fuzzier areas to measure — I conducted interviews with experts, looked at historical examples, and, when I was still unsure, did a threshold analysis. I'll give an example of that later. [It involved determining] what percent confidence I had in thinking an area was a good one to bet on. Then, I tried to see if I my analysis was enough to put me above or below that threshold.
What recommendations [did I arrive at]?
Of the universe of options that I looked at, I ended up recommending people focus on:
* New modalities to foster economic productivity
* New modalities or ways to develop state capabilities
* Global catastrophic risks, particularly pandemic preparedness
* Meta EA research on cause prioritization within global development
I'm going to go in-depth on economic productivity, so I'll skip over it for now and talk briefly about the last two to just give you a flavor of what I did for my thesis.
One of the reasons why I emphasize pandemic preparedness is that generally, pandemics disproportionately impact developing countries. Think of how Ebola impacted Liberia. Also, there's a decent amount of evidence and predictions in this area. One that I cited in my thesis is Larry Summers’ talk about how the annual cost of expected pandemics in the next few years, in terms of negative implications on human well-being, is comparable to a range of outcomes you could see from climate change. There's a pretty limited amount of funding currently going into this space, particularly in regards to the “tail risk” scenarios that we often talk about within EA.
Plus, I think there's a lot of potential for [pandemic preparedness] to be tractable. A few days after the Open Philanthropy Project helped fund a Blue Ribbon Study Panel on biodefense, live anthrax was accidentally shipped across the country.
In terms of meta research [on effective altruism], there are some areas that I would have loved to investigate if I’d had more time. These include:
* The “80,000 Hours ”approach to global development
* Promising cause areas like immigration, improving security in low-income areas (theft and insecurity are among the top drivers of suffering in low-income urban areas), and gender-based violence (gender is one of the biggest drivers of violence around the world)
There is definitely a lot more research to be done if folks are interested.
What about areas that I either [felt only moderately excited] about or didn't end up recommending? I ended up in the “medium” range with global health, which might be surprising for an EA audience. I think addressing global health problems is clearly highly impactful. There's a lot of work left to do. But it's not quite as neglected from a funding perspective. So, maybe an additional marginal person who has innovative ideas might make slightly less of a difference than in some other cause areas that are high-impact, tractable, and more neglected.
I think global health is still a pretty solid bet; it depends on your position. It’s also clearly very high in terms of tractability, because we have a lot of historical examples of wins in this space. I recommend you look at global health, particularly if you're more risk-averse in your career.
What about areas that I didn't recommend in my thesis?
Particularly for people in my program [at the Kennedy School], I recommended thinking carefully before going into social entrepreneurship. There are a lot of definitions [of social entrepreneurship]; [for my purposes] it means individuals trying to sell a product that has a double bottom line [i.e., is both profitable and has a positive social impact]. I think that in situations in which you displace a government from providing a public good, that can have longer-term [negative] implications.
One example of this is a social enterprise in India that provided private, clean water for rich households in urban areas. That took away public demand for systems change to bring clean water to the entire neighborhood, including to lower-income households. Therefore, I have concerns with this model when entrepreneurs are targeting monopoly-provided services.
In general, I think there are also concerns within my program that, for folks who are rigorously trained in economics and data analysis, social entrepreneurship [may not give them a] comparative advantage or [reflect] what they're uniquely positioned to do. I think we might have overemphasized it a bit [in the past] due to its prestige and status. I did some analysis of Echoing Green Fellows’ likelihood of success. When I considered their impact 10 years following their fellowships, it was lower than a lot of my peers expected. (Echoing Green is a venture firm that funds social entrepreneurs.) For those reasons, I’m a little bit hesitant, on average, to recommend social entrepreneurship.
Another [potentially contentious question centers on] whether running additional RCTs is beneficial. [For my thesis], I was interested in not just the value of RCTs in general, but the value of an additional marginal person working in this space. Of course, in any given space, there's variance. I think there's probably really high value for someone to do a longitudinal study around the long-term effects of deworming. [EA-aligned funders in global development would likely] pay a lot of money for that, and I would really love for someone to do that. But for an average person entering this space, running an average RCT, what do we think the additional value is?
I came to the conclusion that the value of research might not be super-strong for RCTs because of considerations related to external validity. We might get higher-value research on some macro-level issues and from systems change work or diagnosing administrative and political constraints. I could take 20 minutes to talk about this point, but I'm just going to acknowledge that it's a hot topic within EA.
One way to summarize this is to say that within the global development community, we've thought a lot about service delivery that's not at scale.
For example, what do we think about vocational training, cash transfers, or activities within an NGO context? A lot of what I ended up recommending in my thesis is [related to the idea] that there's value in thinking about how we move from service delivery that’s not at scale to service delivery at scale. How do we iterate to make service delivery at scale happen in the first place? And how do we facilitate broader economic growth?
I'm advocating for something that might look more like a hits-based approach to global development [versus one focused on making marginal improvements, as we’ve done in the past].
In this diagram [see slide above], there are two overlapping graphs. We have a risk-return analysis showing that [higher-risk] activities can have a higher return — but also potentially a higher variance of success. I’ve argued that in the global development space, there’s been a lot of value in thinking about better bets in the micro space. There’s a decent amount of evidence showing that, on average, some of the benefits we thought activities in microfinance [would bring] aren't as strong as those from GiveDirectly [which gives cash directly to people living in extreme poverty to use as they see fit]. We've definitely pushed the efficiency frontier forward.
But I'm also arguing that there might be certain macro things, like economic growth, that have a wide variance of outcomes. Some won’t work at all. And some we might knock out of the park. If we focus on these higher variances, we might see a higher return overall.
I’m going to go a bit more in-depth for one or two minutes.
As I was thinking about economic growth, I was analyzing the impact at stake. If you look at accelerations in economic growth — times when countries’ economies have really taken off compared to what we expected them to do — [those periods have] have created $20 trillion in total value. To put that in context, that's the current annual GDP of Nigeria multiplied by 60 — a really significant amount of value. Imagine if we could figure out how to foster or extend these growth accelerations. This is an area where questions around neglectedness are debated. There's a decent amount of general World Bank funding that's going into [promoting] economic growth, but a pretty limited amount supports more novel methods. Those are the areas that I'm much more interested in.
There are huge debates within economics about whether or not this is tractable. Is it even possible to foster economic growth? In my thesis, I tried to provide some examples showing why I thought it was tractable and why I am really excited about new methods for fostering economic growth.
This fuzzy diagram [see slide above] is an example of a growth diagnostic. It shows ways in which you can try to identify the binding constraints or bottleneck for economic growth to take off in Nepal. Several different folks within the government and the private sector have said that once they had this diagnostic and were aware of the challenges, they were able to focus their political capital, administrative capital, and other resources in [response]. And I think this is an example of a methodology that could be potentially really high-impact.
Another conversation I had was around whether we are confident enough in any given [attempt to spur] economic growth to make it worthwhile. I did some analysis of how much money it would take to try out one of these approaches [relative to the amount of] value we expected to get from it. What would the difference be between trying one of these new methods and just giving money directly to [people in need]?
My example use case was 0.1% growth in GDP in Ethiopia lasting five years. The analysis showed that if there is a 2.5% chance that we can get something right in terms of economic growth, then it's worth a $5 million investment, because that's the equivalent amount of value we would get from donating $193 million to GiveWell. So then my question is: Do we think there's a 2.5% chance that we could actually achieve this outcome? We could do further research around case studies and expert interviews to answer that.
Finally, what do we do with all of this? And at the very end of my thesis, I tried to recommend what people who have gone through this analysis might do next. Maybe you've arrived at similar conclusions — or maybe you've used the same methodology and arrived at different conclusions. [Regardless], I think that there's potentially a four-step process that you could use.
Step 1: Develop your impact hierarchy. Try to be cause-neutral. Think about using something like impact, neglectedness, and tractability to identify what what you think the most important problems are to work on. Rank them.
I think that in the global development space, you should take your nationality into account because that gives you specific leverage on certain issues, especially if you're thinking about government work.
This is just an illustrative example — not my particular cause prioritization — of what an impact hierarchy could look like.
Step 2: Map out your personal fit. For example, this hypothetical person in my example is from India and really excited about microfinance there. It's something that is a core competency and that they've done before. Maybe they think biosecurity is really important, too, but they would need to be in the U.S. to do that. They want to be close to family, so it’s outside of their personal fit. Personal fit, location, skills, and interests could overlap with someone’s prioritization of a few causes — or they might nott.
Step 3: Test the highest-impact option that overlaps with your personal fit. Or maybe if you're in grad school, or have some free time and runway, you might test an option that's just outside your boundary. Think about areas you could and couldn’t stretch to. All of these steps allow you to get more data on where you might be a good fit.
Step 4: The last thing that I think you can do is backwards-engineer how to have a lot of influence in a given area. You can look at people who have been really successful and influential in that field and follow [a similar] career path. Ideally, you have a variety of career paths to choose from, and then you can think about what your next steps might be to get to a place where you will have a maximum amount of influence in that field.
Also, you can download my paper.
A warning: It is 50 pages long, but I also included a four-page policy overview. I would really love comments and feedback. This was my attempt to take some methodologies that we talk about within effective altruism and apply them to a field in which I have background knowledge and the opportunity to reflect on it. But I think there's a lot of work still to be done in this area.
James Snowden [Moderator]: Thanks so much, Joan, for that interesting talk. I'll just give a very brief summary of what I took from the talk and then I’ll ask you some questions.
Joan analyzed a number of different options for smart graduates working in international development using a framework measuring importance, neglectedness, and tractability.
She concluded that four areas look particularly promising:
1. New ways to foster economic productivity
2. New ways to develop state capabilities
3. Global catastrophic risks (in particular, pandemic preparedness)
4. Meta research on talent in global health and development
Joan also outlined a practical, four-step process for choosing a career:
1. Determine which areas you think are most important to work on
2. Identify your own personal fit
3. Challenge yourself to test the highest-impact option, stretching within the boundaries of your personal fit (which I think is very important)
4. Develop a plan to reach maximum influence on the issue
Joan, I think you’ve done a really great job of surveying a number of promising areas and developing practical heuristics for making good decisions about your career. My first question is: If you were someone sitting in this room who wanted to build on your research and make it more robust or expand the scope, what kind of questions would you be asking? And what would be the most practical way to do that?
Joan: I think that there are ways that the methodology could be more robust. I used a few proxies that I mentioned at the beginning of the talk — for example, World Bank funding as a proxy for all global development funding. You could do a more precise estimate of that. And I was looking specifically at graduates of my program. Those graduates aren’t fully representative of the full set of graduates and experienced researchers in the field of development policy. So there's definitely more methodological depth [that someone could add].
I also think there are a lot of additional areas that I didn't have the scope or capacity to explore — not only more approaches to how people might do this meta research, but on topics that I think would be particularly interesting to explore further, like migration and violence.
James: You mentioned a few case studies as examples of some areas you were looking into. I was particularly interested in the case studies on growth and whether there are examples of particular organizations that might've played a causal role in the story there, which might help us think about where we could have an impact.
Joan: In principle, I think there's a lot more high-value, interesting research to be done on growth. What causes growth episodes to take off? How does a fragile state or a conflict zone become more stable?
I don't know if we're ever going to be able to show [with complete confidence] that an individual or organization [caused] growth. But some interesting examples that I was looking at (and would have continued to look at if I’d had time) include the Korean Development Institute. That is a really interesting example. It’s a think tank that helped Korea develop a growth strategy. And in the 1980s, there was a similar organization in Indonesia, where a group of PhD economists helped advise the government. Some were [Indonesians who already worked] in the government. So, I think we have a couple of historical examples that at least point to smart folks who were thinking about economic growth policy and provided input at the right time. That could be promising [to explore].
James: One last question from me before I hand it over to the audience. You mentioned some new approaches [to enabling] states’ capabilities and economic growth. You separated those from older approaches. I'm curious about examples of particularly promising new approaches, maybe from the last 10 years, which we might not know about and which you think could be particularly good areas to work in.
Joan: I think there's value in both [older and newer approaches]. Within the economic growth space, I was particularly interested in this “growth diagnostic” approach — this idea of not just doing whatever promotes growth [and is feasible]. Let’s target the things that we think are likely to promote the _most_ growth. I think maybe that’s why some of this work has had a mixed track record in the past. We could try taking a new approach to state capabilities, governance, and the delivery mechanism. I think there's a lot of really interesting work around facilitated emergence [focusing on problems instead of solutions and, instead of following a set plan, using a process that allows for iteration and learning].
Historically, there have been a lot of people who [travel to a developing area] and conduct a weekend or week-long training [meant to promote growth-friendly policy]. Then they'll leave and everything isn't fixed. I think there are a lot of really interesting questions around how to foster local talent decision-making and build on capacities that already exist in a longer-term, sustainable way. There's also a really interesting question of getting smart, capable people who are nationals of their country further involved in civil service and making a career that's rewarding for them. Those would be areas that I would further explore.
James: Got it. Great. We have a few questions from the audience. One audience member asks: If you're using climate change as a cost comparison across cause areas, is climate change itself also worth considering as a cause area?
Joan: Yeah. Absolutely. I have a pretty small paragraph in my thesis in which I ask: Why isn't climate change in this bracket? I think the impact is comparable to pandemics based on the research that I saw. But climate change is relatively less neglected in terms of funding. I think there are people who would disagree with that. Some people would say that the tail-end risks of climate change are much, much worse than pandemics. I think there's a lot of really interesting research that should be done on climate change. For example, what are the most effective levers? There's a little bit from Project Drawdown on prioritizing interventions, but from a policy perspective, I think that's a really fruitful question.
James: That leads into our next question: What was the response from your colleagues at the [Kennedy School] to this framework? Were they mostly receptive or did you receive a lot of criticism?
Joan: I think people were generally receptive. Some of the conclusions that I make are judgment calls that not everybody agrees with. And that affects tractability. I also think that there's this interesting dynamic that I saw more with the Kennedy School career group: By the time people get to policy school, some have already specialized for several years in a particular cause area within global development. They are open to the intellectual exercise of comparing cause areas, but maybe aren't open to the personal implications of changing areas within the field. Sometimes I'd have conversations [in which I said,] “Pretend that you are just starting over again. How would you feel about some of these questions?”
James: Did you use a similar four-step process to figure out your own personal career path and if so, where has it led you?
Joan: Yeah, that's funny. Honestly, a lot of this was a personal intellectual exercise for me. I came into the Kennedy School believing that the top thing that I could do, especially given my nationality as an American, was to scale up an evidence-based program. I made this pretty large revision; I now believe that some of the other paths I outlined as top recommendations might be higher impact.
And so, the thesis was a journey for me that allowed me to work through that and the implications for my career. Another way to phrase that is my impact hierarchy was in flux. I came in with a certain version of my impact hierarchy and it changed over the course of my time at the Kennedy School.
That has led me to take the meta approach. I’m interested in EA community building. So I think I took a step away from some of these [other options] in part because a lot of the meta questions are super important. I wish more folks were focusing on them.
Hi I'm struggling to understand the threshold analysis. Can anyone help?
It seems like ln(GDP) is the rate of change of GDP. If so, why isn't this compounding? ie why isn't the rate of change of GDP of after 5 years (1+.105*.96)^5 = 1.65 not .48. This would mean it would only need a 2% success chance to be the most effective intervention. I guess I'm missing something, please could someone help.
What is ln(GDP)? What does the ln mean in this context. I don't think it can be natural log. Most naturally it seems to be the rate of change but I'm confused as to why they chose ln().
Why is a 4% discounting rate the right one to choose?
Thank you for your time.
Hi Nathan, I think it is a log function (most likely natural log, but could have been shorthand for some other base).
People often use this when figuring out benefits of gdp, consumption growth etc (eg, I think Givewell assumes that a doubling of wealth is ~ equally good no matter what the baseline wealth is).
The approximate reasoning for this is that we expect there to be diminishing marginal returns to wealth per capita, and there is some weak empirical evidence that a log function specifically fits the data reasonably well.