EA Survey 2019 Series: How many people are there in the EA community?post by David_Moss · 2020-06-26T09:34:11.051Z · score: 71 (30 votes) · EA · GW · 9 comments
- It is uncertain how many people there in the EA community and what proportion of these the EA Survey manages to sample
- We compare EA Survey numbers to other data sources and estimate that we sampled around 40% of highly engaged EAs, and fewer less engaged EAs
- Based on this, we estimate there are around 2315 highly engaged EAs and 6500 (90% CI: 4700-10,000) active EAs in the community overall.
How many people are there in the EA Community?
How many people there are in the EA community remains largely unknown. The two most recent EA Surveys received about 2500-2600 EA-identifying respondents, but we don’t know what proportion of the total EA community this represents. There are also multiple different criteria we could use to define the EA community.
David Nash [EA(p) · GW(p)] has estimated that there are around 500-2000 people who “either attend multiple EA events each year or contribute to online discussion, and some proportion of people who work at an EA-related organisation.”
David Denkenberger [EA(p) · GW(p)], conversely, noted that local groups have around 2124 regular attendees (defined as 25% of events or an event every two months) based on estimates by local group organisers in the 2019 Local Groups Survey [EA · GW], which is higher than the upper bound of David Nash’s estimate.
Denkenberger also noted that there are 4400 Giving What We Can members in total and 843 self-identified GWWC members in the 2018 EA Survey (19% of the total membership) and so, if one were to assume this represented the response rate for the EA Survey as a whole, this would suggest that there were around 13,000 EAs.
That said, we suspect that Giving What We Can membership is not a good basis for making this kind of inference about what portion of the EA Survey we sampled. It seems likely that many people who have taken the Giving What We Can pledge are not engaged in, and perhaps do not identify with, the EA community. Luckily, we have access to a number of different proxies that we think serve better.
These results converge fairly closely on around 40%, provisionally suggesting we sampled around 40% of the real populations in these cases.
However, we need to take into account the fact that all of these groups would be fairly high engagement on average, probably around levels 4-5 on the self-reported engagement scale [EA · GW] below.
- 1) [Audience] No engagement: I’ve heard of effective altruism, but do not engage with effective altruism content or ideas at all
- 2) [Followers] Mild engagement: I’ve engaged with a few articles, videos, podcasts, discussions, events on effective altruism (e.g. reading Doing Good Better or spending ~5 hours on the website of 80,000 Hours)
- 3) [Participants] Moderate engagement: I’ve engaged with multiple articles, videos, podcasts, discussions, or events on effective altruism (e.g. subscribing to the 80,000 Hours podcast or attending regular events at a local group). I sometimes consider the principles of effective altruism when I make decisions about my career or charitable donations.
- 4) [Contributors] Considerable engagement: I’ve engaged extensively with effective altruism content (e.g. attending an EA Global conference, applying for career coaching, or organizing an EA meetup). I often consider the principles of effective altruism when I make decisions about my career or charitable donations.
- 5) [Core] High engagement: I am heavily involved in the effective altruism community, perhaps helping to lead an EA group or working at an EA-aligned organization. I make heavy use of the principles of effective altruism when I make decisions about my career or charitable donations.
On the whole, we would expect to sample higher proportions of more engaged EAs, at around the 4-5 level, due to higher engagement causing higher motivation and them also being more likely to encounter the EA Survey more often. Conversely, we would expect people matching the description of levels 1-2 to be less likely to encounter the survey and to have less reason to take it.
Looking at the EA Survey results, this is what we see, with slightly more at level 3 than 4 or 5, but fewer at level 1 or 2.
So this suggests, as we would expect, that the EA Survey samples relatively more people from the moderately engaged ‘participants’ group and upwards, than from less engaged groups.
Nevertheless, for estimating the size of the EA community we are probably mostly interested in respondents who identify at least with level 3. To reiterate: this corresponds to someone who has “engaged with multiple articles, videos, podcasts, discussions, or events on effective altruism” and who “sometimes consider the principles of effective altruism” when they “make decisions about... career or charitable donations”, which seems a plausible minimal characterisation of involvement in the EA community. Someone selecting level 2, conversely, has engaged with a few articles/podcasts etc., but may not even necessarily “sometimes consider the principles of effective altruism” when making decisions. Of course, being part of the “EA community” in this sense is not a criterion for being effective or acting in an EA manner- for example, one could donate to effective charity, without being involved in the EA community at all- but it still seems valuable to estimate how many people are involved in the EA community specifically in this narrower sense.
Since the proxies we used to estimate our sampling rate predominantly apply to level 4-5 EAs, we can estimate the size of this higher engagement population. There were 926 level 4-5 EAs in our sample, so if we sampled around 40% of this population, we would expect there to be 2315 in this population in total. Of these, 1140 would be estimated to be in the highest level of engagement (level 5).
We don’t know exactly what proportion of level 3 EAs the EA Survey sampled, though as noted, we would expect it to be somewhat lower than the portion of level 4-5 EAs. If we sampled 30% of level 3 EAs (a slightly lower rate than for more engaged EAs), then we would estimate that there are 2443 level 3 EAs of this kind and 4758 EAs (level 3-5) overall. Conversely, if we sampled many fewer level 3 EAs than 4-5s (let’s say only 10% or 4 times fewer than 4-5 EAs) then we would estimate there to be 7330 such EAs, and almost 10,000 (9,645) EAs overall.
Looking at our slightly more detailed guesstimate model, we see that the estimated 90% CI for the total EA community (in the level 3-5 sense defined above), is between 4,500 and 10,000, with a mean around 6,500.
This suggests that there are roughly 2000-3000 highly engaged EAs in total- quite a lot, in some ways, but still very small in others. For example, this would only be a fraction of the student population of most universities.
Likewise the estimated size of the EA community more broadly, worldwide, is only 5000-10,000 or about the size of a small town or university. Indeed, the global EA community, even according to the larger estimates, only has about as many members as the weekly attendance of a mid-ranking mega-church. These kinds of numbers seem worth bearing in mind when thinking about the reach and scope of the EA community.
The annual EA Survey is a project of Rethink Charity with analysis and commentary from researchers at http://rethinkpriorities.org/. This essay was written by David Moss. Thanks to Aaron Gertler, Peter Hurford, Jason Schukraft and Neil Dullaghan for comments.
Other articles in the EA Survey 2019 Series can be found here
We are informed that people in the joint CEA/80K/GPI slack were polled as to whether they took the EA Survey. 11/28 (39%) of respondents answered affirmatively. ↩︎
613 EA Survey respondents identified as members of the EA Forum. The EA Forum population was estimated to be 1522 based on the total active accounts (viewed at least 1 post) in the 6 months prior to May 2020. This is 8 months after the 2019 EA Survey was run, so the EA Forum population might have been lower at the time of the EA Survey, so this might be an overestimate of the Forum population. On the other hand, respondents might have identified as members of the EA Forum even if they hadn’t viewed even a single post in the last 6 months, which would lead to this figure being an under-estimation. ↩︎
The EA Survey had 907 self-identified local group members. Organisers responding to the Local Groups Survey estimated that, in sum, there were 2124 regular attendees at their events, giving a figure of 42.7%. We suspect this is likely to be an overestimate due to a tendency for organisers to ‘round up’ their estimates, when they didn’t have precise figures. ↩︎
EA Survey respondents were asked to specify which local group they were a member of, if they were a member of a local group. Since many local groups are quite small (the median number of members, captured by regular attendees is 10), many of these figures would be quite noisy, so we looked for self-reported members of 5 of the largest groups. Of course, this is not an independent population from the local group members estimated in footnote 3. However, these are independent measures, traceable to specific groups, so this seems to offer some additional evidence. ↩︎
68.2% of local group members in the EA Survey selected 4-5 on the engagement scale and 26.4% selected level 3. ↩︎
It is not clear to me that we would expect to see a higher response rate from EAs at level 5 (e.g. EA org employees) than level 4 (the next highest level of engagement). While their higher engagement level might be expected to predict higher motivation to take the survey all else being equal, it is possible that they are busier, and anecdotally we know that some senior figures in EA are worried that they would be easily identifiable from their responses. (Note that while we understand the general concern, we do not think there is any risk of individuals being identifiable in the redacted public dataset, and if individuals are worried about being potentially identifiable to the survey creators, we would recommend that they take the survey, but skip any specific questions (such the city and org they are employed in) that they worry would be easily identifiable). ↩︎
These have at most read a few articles or spent a few hours in total on EA websites (level 2) and may merely “have heard” of EA, but have no engagement with it at all (level 1), meaning that they many of whom may not even identify with EA even broadly (and as such, would not even be included in our analysis of the EA community). Naturally it is particularly difficult to estimate the number of people who had heard of EA or visited EA websites but have no other engagement with EA, but even based on the totals of those visiting the 80,000 Hours or GiveWell websites see EA Growth Metrics 2018 [EA · GW], it is clear that we would have sampled only a tiny fraction of one percent of this population. ↩︎
The majority of EA Survey respondents who identify with level 3 have made a donation influenced by EA principles (~80%), read an EA book (~55%) and just under 40% have changed their career plans based on EA principles, but have not undertaken any other of the EA activities listed. As such, self-reported engagement at level 3 on the scale is still a fairly permissive criterion. By contrast, majorities of level 4 respondents have completed those activities and large numbers (~50%) have also attended EAG or organized an EA event, and increasing numbers have undertaken a number of other activities. See more details in our post on Engagement Levels [EA · GW]. ↩︎
We grant that some people who fit this description wouldn’t count as part of the EA community more narrowly construed, since it seems possible to have engaged with multiple EA texts, events etc. and to “sometimes” use EA reasoning, while not endorsing EA ideas or really identifying with EA according to a stricter standard. Exactly where to draw the line is open to question and depends on your practical interests in referring to ‘the EA community’, though we think that the majority of level 3s would count as part of the broader EA community in most views. Conversely levels 4-5 offer a neat and natural grouping corresponding to more actively engaged EAs. ↩︎
As we reported in our 2018 post on EA Growth Metrics [EA · GW], >24,000 individuals appeared to donate to GiveWell alone in 2017 and, doubtless, many more individuals have donated to other EA-aligned charities. Thus, this figure is already higher than the upper limit of our estimate of the number of people involved in the EA community, but we would expect many donors to GiveWell, Giving What We Can etc., to not be at all involved in EA. ↩︎
Of course, only a subset of those who would self-identify with the highest level of engagement would be engaged in EA in specific ways associated with high engagement. For example, only 202 of the 456 respondents who identified with the highest level of engagement reported being current employees of EA-aligned orgs, so we would estimate that there are around 505 such people in the population as a whole. Similarly, we would estimate that there have been around 730 organisers of local groups. People could construct their own estimates using the number of EAs who report having completed different activities [EA(p) · GW(p)]. Interestingly, the number we sampled and proportion we estimate that we sampled was roughly equal across both level 4 and 5 of self-reported engagement, suggesting similar the end of the funnel containing ‘contributors’ and ‘core’ is flat rather than narrowing, which may be counterintuitive. However, this is not the case when looking at sampled numbers of EAs who have completed specific different activities, which offers a more fine-grained sense of engagement. There, as the image in the linked ‘Activities’ section shows, we see a clear tapering, with the fewest current number of EAs being current employees of an EA-aligned org, slightly more having received 80,000 Hours careers coaching, volunteered at an EA org or attended EA Global, and several (3-5x more) have read an EA book or made an EA-influenced donation. ↩︎
We might expect the rate of sampling to vary across levels of engagement within the level 3 category (i.e. higher rates among those who are almost level 4, who are highly engaged and considering attending an EA Global conference, applying for career coaching, or organizing an EA meetup and lower rates among those closer to level 2, who have read a few things but are otherwise not particularly engaged). But, similarly, the proportion of EAs who meaningfully count as part of the EA community follow a similar pattern. So overall, the proportion by which we are under-sampling those we want to count as part of the EA community in this analysis might be expected to be roughly in the middle. ↩︎
Guesstimate will generate slightly different values each time. ↩︎
Comments sorted by top scores.