November 16, 2017
In this 2015 article, Owen Cotton-Barratt argued that a very important question for social movements is how to grow. He argued that to avoid hostility, it may be wise to prioritize improving people's inclination towards the movement ("advocacy") over making the movement more widely known ("publicity").
I wrote this article two-and-a-half years ago. Although I still agree with its points, I would write a somewhat different article today, in framing and substance.
On the framing, this article uses effective interventions to relieve global poverty as the main example of an idea one might to bring attention to. At the time this seemed the canonical central example for the effective altruism community, who were a good fraction of my audience. Even then I guessed it was not the most important idea to draw attention to, and I have become more confident since. Of course all of the arguments still apply to effective poverty interventions; I just wouldn't lend them so much implicit endorsement.
On the substance, I omitted an important consideration: that larger movements come with increased overhead costs for communication and coordination within the movements. Combined with diminishing returns from effort within an area, this reduces the value of making a movement very large: doubling the size might be substantially less than doubling the value. Moreover if one realizes that different people have different amounts to contribute to work on a given problem, the effect is that the value of bringing different types of people in varies significantly. The ideal for a given community might be to include as many as possible of the people who are very well placed to contribute, and not so many people beyond that so as not to impede coordination between that core. For some ideas or problems that might in practice mean quite a small community; for others a large one. The central example of global poverty made this omission less salient since it benefits from more supporters even when large.
Nonetheless, I think the original article contains useful ideas which I continue to make use of in my own thinking. I hope it remains useful to others.
Owen Cotton-Barratt, November 2017
Movement growth may be very important for young social movements. It’s obvious that movement building cannot always be better than direct work, but knowing how to compare them presents a challenge.
In this article I introduce and explore a model of movement growth which tracks individuals’ awareness of and inclination towards the movement. I aim to understand when movement building activities are better or worse than direct work, and apply the model to give my views on movement growth in the effective altruist and related communities.
In the first half of this paper I introduce a model for thinking about movement growth, and terminology to refer to critical concepts. We model individuals as having varying levels of awareness about the movement, and varying inclinations towards it. We assume that these two characteristics can represent the major drivers of interaction with the movement. We explore the consequences of this Awareness/Inclination Model (AIM), particularly looking at the long-term counterfactual effects of direct work compared to ‘publicity’, which aims at increasing awareness of the movement, and ‘advocacy’, which aims at improving inclinations towards the movement. This involves analyzing different possible long-term trajectories the movement may be on.
If we accept the model, this has some general implications:
In the second half of the paper, I apply the conceptual tools developed in Part 1 to answer questions about how to find the best work for the young effective altruism movement and related areas. The conclusions here are not certain, but represent my informed best judgement. Some of them are driven purely by qualitative considerations, and some are based in part on numerical estimates.
My conclusions are:
Acknowledgements: Thanks to Tom Ash, Daniel Dewey, Seb Farquhar, Michelle Hutchinson, Josh Jacobson, Toby Ord, and Kerry Vaughan for helpful conversations and comments.
Suppose you have discovered a great way to help the world. Now you face a trade-off: how should you balance pursuing this directly against spreading the word and convincing others to help pursue it?
The basic problem is similar for many possible ways of helping the world. Perhaps you’ve worked out that washing your hands before surgery helps your patients’ survival rate. Perhaps you’ve realised that treating men and women equally will help welfare levels and productivity. Perhaps you think that giving money to charities that have been shown to do a great deal with it is an effective way to help people. In each case you might have more impact by growing the movement of people who care about the issue than just doing your part of the direct work,1 but it may be challenging to see how to compare these against each other.
In order to make the best decisions, it’s useful to think about how valuable growing the movement is compared to the direct work. This is very hard to know for sure, since we never get to see how all the counterfactuals play out. But I think we can improve our intuitions by understanding some simple models and getting a clearer picture of the kind of dynamics that may be at play.
Section 1, the majority of the paper, aims to give the reader a set of conceptual tools for thinking about movement growth. We introduce the Awareness/Inclination model, and use it to explore the possible life-cycles of movements without intervention, and the counterfactual effects of interventions.
In Section 2 we combine these insights with personal impressions of the effective altruism movement to provide recommendations.
In order to understand the counterfactual impact of a marginal intervention to foster movement growth, it’s helpful to first have a picture of what we expect to happen without that intervention. Generally this will be of the form: (i) the movement grows as more people become aware of it; (ii) as it grows more direct work is done; (iii) at some point the movement size plateaus.
I’m going to introduce a simple model of the space in which this is happening. This will allow us to analyze the effects of different kinds of intervention.
People will know a variety of different things about the movement, and have different attitudes towards it. As a modeling assumption we will compress these down to two dimensions: awareness (how much they know about the movement) and inclination (how favorably disposed they are towards the movement, or how favorably disposed they would be if they knew more).
Then a person (at a fixed point in time) will be represented by a point in this space. Some examples are shown in Figure 1:
Of course over time people may move around in this space. This will usually be driven by finding out more about the movement. So when movement occurs, we should expect it to be primarily in the left-to-right direction of the diagram. As they learn more, people may also become a bit more favorable or hostile to the movement. It seems to me that significant changes in inclination without significant increases in awareness are likely to be relatively rare during the growth phase. This is a substantive assumption which could be worth exploring, but the conclusions of the model do not require it.
Examples of some trajectories individuals could take through this space are shown in Figure 2:
Early on in the life of an idea a large majority of the population will be ignorant of it. We cannot in general say how favorable they will be. Movements where the people on the left are clustered towards the top and away from the bottom have more potential for growth, and less potential to lead to a vocal opposition. It is the people in the top-right of our diagram who are both knowledgeable and favorably inclined who will actively pursue and promote the idea, and may be regarded as the movement built around that idea. This is illustrated in Figure 3:
As the movement starts to grow, some people will become more aware of it, moving along trajectories to the right and perhaps changing inclination slightly (Figure 4).
In our example, the growing awareness resulted in a substantially larger movement, as well as someone moving into the bottom-right corner of the diagram, becoming knowledgeable about the movement and also hostile to it. Why should this section matter? Consider what drives the increases in awareness and changes in inclination. At the level of the model it seems that more people being knowledgeable and favorable to the movement will help to increase the spread of awareness, and that more of the knowledgeable people being favorable rather than hostile will help to nudge people’s inclination in a positive direction. So having a larger, more active opposition to the movement may decrease the frequency of positive inclination-shifts as people learn more (and increase frequency of negative inclination-shifts). It may also affect how the movement is able to influence public policy: a small highly favorable group with no opposition may be better able to shift policy than a larger group with a vocal opposition.
What is the long-term picture? There are four stable situations that can come about in the mid- to long-term as the number of people informed about the movement grows:
Scenario A – Saturation: There is a certain amount of work that should usefully be done on the issue at hand (perhaps each year, or perhaps in total). When the movement gets large enough, this work gets done, and there is no need to make further people aware; the movement is the correct size and more growth might even be bad. Existing people drift away from the movement as they see less need to be involved. E.g. fundraising for specific projects.
Scenario B – Collapse: Something happens to cause negative inclination towards the movement. This could be an inevitable consequence of increased awareness, or could be caused by some particular event. This affects both existing supporters and people who do not know about it. A subcategory of collapse is stagnation, where there is no real opposition but people lose their strong positive inclination, and nobody works to bring new people into the movement. In either case the negative inclinations stops the movement from growing, and it eventually collapses. E.g. Marxism
Scenario C – Controversy: In this scenario almost everyone knows a reasonable amount about the movement. A fraction of society is favorable enough to be actively involved, but many people are not in support. Most large movements can expect to be here. E.g. abortion
Scenario D – Acceptance: It becomes socially normal to support the movement. Almost everyone knows something about it and agrees.2 E.g. ending the use of lead in consumer products
The marginal effect of an intervention may be quite different depending on which long-term scenario we are in. For this reason it’s important to have a good understanding of which are the most likely. To help think about the effects of interventions, we’ll consider the two intervention archetypes illustrated in Figure 7.
Intervention X – Publicity: This intervention focuses on raising awareness of the movement. It is consistent with the view that all publicity is good publicity. E.g. performing stunts that are widely reported.
Intervention Y – Advocacy: This reaches fewer people than Publicity, but is concerned chiefly with improving their inclination towards the movement. It is closer to brand management. E.g. writing a persuasive opinion piece for a newspaper.
Both of these interventions affect the population roughly uniformly; in practice many interventions may target subgroups.
The Saturation and Acceptance scenarios are perhaps the simplest to analyze. In both cases there is a natural eventual size of the movement, so marginal work on movement growth will not change that size, but change when it is realized. This suggests that the focus should be on increasing awareness rather than improving inclination: Publicity has a bigger effect than Advocacy on reaching the end state quickly. There is still a question of how good increasing awareness is compared to direct work.
The curve of movement size against time before intervention (black curve in Figure 8) thus has a horizontal asymptote. Suppose that we intervene to make the movement grow by k people. We will assume for now that our intervention just moves the entire curve forward in time (red intervention and blue curve in Figure 8).
What is the total effect of this? The growth of the movement is advanced in time by a certain amount a (see Figure 9). This will produce extra direct work corresponding to the area between the blue and black curves.3
This area can be shown to be equal to a times the difference between current movement size and eventual movement size (see Figure 9 – so long as k is small the area of the extra grey triangle is negligible). If we value direct work equally regardless of when it happens, this gives a simple way of understanding the value of acceleration: it is only necessary to estimate the current rate of growth and the eventual movement size. Then:
Benefit = (eventual movement size – current movement size) * advancement period
For example, suppose the movement currently has 10,000 people in it, is growing at 10 people per day, and is projected to reach an eventual size of 100,000 people. Then an intervention which gains 10 extra new members beyond what would have happened anyway for the movement accelerates the process by a day, so long as it also advances the background awareness of the movement at the same rate.4 This means the value of this advancement is the same as the direct work produced by 90,000 people in the movement for a day – which is the same as one person for 250 years. So the value of just one person joining can be estimated as their future direct work for around 25 years.
Some caveats. A person being worth about 25 years seems reasonable, but this is a bit of a coincidence of the numbers. The value is quite sensitive to eventual movement size. If the movement will only reach 20,000 people, each person added only equates to 2.5 years of direct work. If it will reach 10 million, each person added (while it’s currently only 10,000 people) produces a whopping 2,500 years of direct work, because they pull forward the future hump of the distribution by a macroscopic amount. Uncertainty about eventual movement size will complicate this, but much of the expected value may be driven by scenarios where the movement gets very large, if these are realistic. In practice interventions won’t advance all parts of the awareness process equally, so just counting new members is likely to overestimate the value (though probably by less than an order of magnitude). And if we value direct work differently at different times (or differently according to how much has been done), that would change the analysis, most likely in the direction of making direct work better relative to movement building.
For these scenarios it seems that movement growth can compare very favorably to direct work, and that Publicity is often better than Advocacy.5 The major exceptions are when the movement has already done most of its growth, or if the movement is simply never expected to get large.
The Controversy scenario in some ways admits a similar analysis. Treating the eventual size of the movement as known and fixed, we can estimate how good accelerating movement growth is in terms of direct work. But an important difference is that we may be able to affect the eventual size of the movement. In Saturation and Acceptance, movement growth can be very important, and it is generally worthwhile to speed it up as much as possible by whatever means are most effective. In Controversy, long-term stability is achieved when everyone knows about the movement but people have different opinions about it. It may be possible to change this eventual distribution of opinions, and this could have a large effect.
For example, suppose we have a movement of 10,000 people growing at 10 people per day, and projected to reach a size of one million people. Suppose there is an opportunity which would bring in 300 new people – accelerating growth by a month – but by being controversial, reduce the pool of people interested by 1%. At the current movement size, putting off 1% is just 100 people, so attracting an additional 300 seems like a great deal. But if the change persists and affects the stable level, this would be reduced by 10,000. Projecting twenty years past the plateau of movement size, the intervention will destroy more than twice as much direct work as it creates in this model.
Since we hypothesize that a major driver of improving inclination towards the movement is having lots of people with high inclination and few people who are opposed, in Controversy it may be more important to focus on improving inclination towards the movement as a first priority, and raising awareness only as a second priority. Advocacy is probably a preferable intervention to Publicity.
In Figure 10 we suppose we have a metric for inclination towards the movement such that the population is normally distributed.6 The horizontal bands show standard deviations (z-scores), and the dots and numbers show how many people lie in each band, from a population of around a billion people.
Changing the threshold required for eventual participation in the movement can have a large effect size. For example, suppose in one scenario the movement only catches on in those who are 3.8 standard deviations towards favourable or higher, whereas in another the movement catches on in those who are 3.7 standard deviations towards favourable or higher. Shifting from the first to the second scenario seems like a small change in required attitude, which might be achievable by shaping the right positive view of the movement. But its effect is to add 35,000 people to the eventual size of the movement, on a base of 72,000 – almost a 50% increase! So if how positive people will eventually be towards the movement can be adjusted, that could be crucial (see Figure 11).
Interventions which change the likelihoods of different long-term scenarios may also be crucial. Changing scenarios is most plausible between Collapse and the other scenarios. This could have a very large total effect. Avoiding movement collapse may be difficult to do with positive interventions, but it becomes extra important to avoid events which could catalyze the creation of opponents of the movement and drive movement collapse.
In the above, we’ve imagined that changing the framing around the movement can affect the size it grows to, but that the underlying message and the direct work associated are essentially the same. It’s also possible that by choosing a slightly different movement to promote, for example by excising less popular ideas, we could change the growth.
This may be a powerful tool. Because the eventual movement size may be so sensitive to – among other factors – how appealing the ideas are (as in Figure 11), it could be valuable to look for opportunities to adjust and improve this. It’s also a tool that should be used very carefully, however – the value of spreading a different movement may be substantially different, and it could be too easy to throw the baby out with the bathwater by removing unpopular components.
The model we have used necessarily makes a number of simplifications. I think these have generally been worthwhile, in allowing us to go further and faster through the analysis without distorting the picture too much. Nonetheless, it’s worth being aware of possible limitations or shortcomings, and it may be that future work could explore relaxing these assumptions.
Restricting to the 2-dimensional AIM space meant that we could not represent more complex attitudes, such as distinguishing between someone who has strong positive affect towards the movement but is not currently taking action on this, and someone who has much more mixed feelings but is currently pursuing some direct work. Much of the time I think these two dimensions suffice to capture the important distinctions, but there may be cases where they fail.
We assumed that people tend to move in AIM space only when presented with something external, and cannot move to the left. In fact people may drift around without interacting with others, and in particular may forget and so slowly lose awareness. It seems likely to me that the pictures presented are still roughly correct on average for growing movements, but further consideration of this point might show extra benefits to some interventions.
Perhaps it is easier to affect inclination by large amounts in some groups such as students. The current model doesn’t make any assumptions about this, but doesn’t make it easy to model. An extension of the model could perhaps provide easier ways to think about heterogeneity in the population beyond the position in AIM space.
When we looked at the marginal effect of a larger movement today, we assumed that this would accelerate the movement growth process fairly uniformly. This is reasonable if growth of the movement is driven primarily by people in the movement. If growth is driven significantly by some exogenous factor (for example uptake of a new technology, reaction to political events, or flow from a related movement) then the long-term effect of growing the movement today could be much smaller. Changing the dynamics so that the long-term state alters could still be very important.
Our example interventions Publicity and Advocacy were quite a long way removed from concrete actions. What might they look like in reality? Since the effects are all about changing minds (via awareness and inclination), interventions will tend to take the form of communication – although they could also include taking actions that can usefully be reported.
Example intervention might include:
The last example may be surprising: we started out setting movement building against direct work, and are now using direct work as an example of movement building! What’s going on here?
The answer is that while direct work has some movement growth effects, it is usually not a very effective way of movement building, so there is a meaningful choice between direct work and (effective) movement growth. However if the ratio of direct work to movement growth activities is too low, this could hinder the growth of the movement, either because it gives a natural line of criticism, or because it is harder to tell a clear story of what the movement is about. In such cases direct work may become the most effective form of movement growth at the margin – albeit with a slant towards direct work that can be communicated clearly or sets a good example.
As a final note, when choosing what to do, it’s important to keep the concept of comparative advantage in mind. Even if you think a bit more movement growth is the most effective thing to be doing at the margin, if you have much better opportunities to pursue direct work, that could be better. Think about the portfolio of work done by those involved in the movement. Direct work should generally be done by those best at it, and movement building by those best at that. The importance of movement building relative to direct work sets the location of the split between these activities and shows what people with intermediate skill sets should be working at.
The effective altruism movement is concerned with using evidence and reason to do the most good it can in the world. It is currently quite small (in the thousands of people, probably the low thousands) and growing rapidly (for example membership of Giving What We Can has roughly doubled each year).
A key decision-relevant parameter is how big the movement will get. It seems that it would be hard to run out of effective things to do, so Saturation is impossible. Collapse, Controversy, and Acceptance are all possibilities. Movement collapse is probably unlikely, but it is very important to avoid.
Knowing how big the movement might get would tell us whether we’re likely to end up in scenario Controversy or Acceptance and, if Controversy, what the scale of the eventual plateau might be. I think there is a lot of uncertainty around the eventual size. It could be valuable to do market research to estimate what the uptake rates could get to in different slices of society, although this is not straightforward.
If, as seems to me most likely, we are heading towards Controversy, it is also crucial how much we could hope to affect the eventual size of the movement. Again, this is hard to estimate, but might be done with market research, seeing how much effect different framings have on people’s willingness to engage.
The higher we think the plateau is, the more we should care about movement growth relative to direct work. The more we think we can affect the level of the plateau, the more we should prefer quality to speed of growth, focusing more on inclination than awareness. And the higher the discount rate we place on direct work, the more we should care about direct work compared to movement growth, and the more about spreading awareness than improving inclinations.
My impressions are that it is possible the movement could get very large – millions of people within twenty to fifty years. Until we have ruled this out, it is important to keep movement growth a priority, and to weight improving inclinations more than increasing awareness in many trade-offs. Focusing on improving inclinations means that we should be careful with our framing, trying to explain how we help people to do things they already wanted to do.
Effective altruism has several facets, as people pursue direct work in multiple areas. Dealing with developing-world poverty has been perhaps the largest focus area of the movement. I think that at present movement growth is not significantly limited by a lack of direct work in poverty, and this work is not dramatically more useful now than in a few years. The implication of this is that it may be better at the margin if we put a few more resources into movement growth rather than direct work in poverty. Of course this statement has several caveats:
Does the same argument apply to say that effective altruism should reduce its focus on direct work altogether to prefer movement building? Not entirely. I think that the movement’s growth may be limited by credibility, by ability to assess cause areas or interventions, or by direct work in other areas such as engagement with policy. In general the direct work that is done will have an effect on the way that people perceive the movement, so may affect awareness and will certainly affect inclination.
Suppose the effective altruism movement currently has 3,000 people doing an average of $5,000 direct work per person per year (present movement size equivalent $15M/yr), and is growing by 10 people per day.
There is great uncertainty over eventual movement size. Possibilities, excluding collapse:
The mean is approx $10B/yr.
Adding a person to the movement ~ bringing this forward by 0.1 days. Adjust down to 0.025 days to account for the fact that there are lots of steps in getting people towards movement.
0.025 days * $10B/yr = $700,000 worth of direct work (excluding movement building effects from direct work).
Perhaps adjust this down a bit for future work holding less value than current work, and down a bit more for people doing otherwise useful work. Say $300k for getting a person into the movement.
That was for spreading awareness. What about improving inclination? (Numbers for this are if anything even more crude.)
Suppose increasing inclination of everyone who currently knows something substantial about the movement (say 10,000 people, which includes many people not in the movement) by 1σ leads to opinions being 0.1σ higher when the movement reaches plateau, and that that change lasts 20 years.
Then causing a noticeable (0.2σ, corresponds to ~1.5cm height) shift in one person’s opinion leads in expectation to 2×10^-6 σ shift in population opinion. That leads to an expected ~200 people at 90th percentile, and around 25 people in expectation total, causing about $1.2M extra direct work. This should be adjusted down for later work being less valuable than early work – perhaps to $600k.
This is a large effect for improving opinions. It’s probably most important for improving the opinions of people near the top and bottom of the inclination spectrum, as those are the ones who are likely to communicate most about the movement; therefore I suggest increasing this estimate for those people as well as for influential people, and decreasing it for those in the middle.
It might be better to model shifts in opinion as decaying with time and when spreading to new people. This will probably lead to a lower overall estimate of the value of changing opinions, since the decay will be biggest in scenarios where the movement gets very big and inclination is particularly important (a mock-up in Excel suggests it reduces impact by a factor of 4). Say $150k for a noticeable shift in an opinion.
All of these numbers are basically pulled out of my head and are neither carefully modeled nor terribly robust. Nevertheless, they are suggestive, and I’d be mildly surprised if they were wrong by much over an order of magnitude. It’s important to remember that we excluded the movement growth effects of direct work in order to make the comparison, so direct work will tend to compare more favorably in practice than these suggest.
Key empirical questions for better numerical estimates:
Here I’ll try a little more interpretation, to discuss what this means when deciding between actions. These views are informed by the theory but not entailed by it:
How does this analysis translate over to other young movements, or other focus areas of effective altruism? For such linked movements, it generally seems that collapse and controversy are the most likely outcomes, that many people are relatively ignorant of it, and that the community of people who might care about it is much larger than the community that currently does. So for the same reasons as before it seems that movement growth is important, and that inclination may be more important than awareness. In cases where many of the eventual routes to impact are likely to involve policy, focusing on inclination and avoiding making opponents could be particularly important.
However, I think there is a significant disanalogy between effective work in global poverty and some more speculative areas such as existential risk reduction. Effective interventions in global poverty are well-demonstrated, so the growth of attention is not significantly limited by direct work. In contrast, the growth of attention for existential risk reduction is substantially limited by a lack of credible direct work in the area. Therefore, in order to foster movement growth, one of the highest value activities is to do direct work, and to set an agenda for direct work, that clearly helps with the problem. Choosing work that demonstrates credible ways to help, and explaining the value of this work, is probably more important than choosing the direct work with the biggest actual effect in the short term.^[See Ord: http://www.fhi.ox.ac.uk/the-timing-of-labour-aimed-at-reducing-existential-risk/] There may also be risks that need to be addressed in the short term, which could increase the discount rate on direct work, and so make direct work more valuable relative to movement building.^[See http://www.fhi.ox.ac.uk/Allocating-risk-mitigation.pdf]
This is the last in our series of articles setting out the key ideas in effective altruism. If you'd like to find out more, read some of the latest research from the community. Or see how you can contribute and get involved.
You might also want to sign up for the monthly Effective Altruism Newsletter, which shares the latest news in EA (global progress, new charities, high-impact job opportunities, and more).