EA Survey 2018 Series: Group Membership


In this post we ex­plore which EA groups EA Sur­vey 2018 re­spon­dents were mem­bers of. We find rea­son­ably large num­bers of EAs are mem­bers of some groups (such as EA Face­book and Lo­cal Groups), but much smaller num­bers are in­volved across many groups.

To­tal mem­ber­ship of differ­ent groups

EA Facebook

The main EA Face­book group had the largest mem­ber­ship among our re­spon­dents, with slightly more than half be­ing a mem­ber. This is un­sur­pris­ing, given that the EA Face­book group has a very large, broad mem­ber­ship (>16,000, though only around 56.9% of these are ac­tive, based on data from the ad­mins).

EA Fo­rum and LessWrong

By con­trast, EA Fo­rum mem­bers were a much smaller group, com­prised of just over 500 re­spon­dents or 20% of the to­tal sam­ple. This find­ing makes sense given that the Fo­rum caters to rel­a­tively more in depth, niche and ad­vanced con­tent and dis­cus­sion. As noted in our pre­vi­ous post, the EA Fo­rum is es­ti­mated to have only around 500 ac­tive mem­bers. A fairly similar num­ber of EAs in our sam­ple were mem­bers of LessWrong.

Lo­cal Groups

One sur­prise from this year’s data was the high num­ber of EAs who re­ported be­ing mem­bers of a Lo­cal EA group. This was 1018 in­di­vi­d­u­als, the sec­ond largest group­ing be­hind the EA Face­book group, and just short of 40% of EAs in our sam­ple. This is a sig­nifi­cant in­crease com­pared to re­ported lo­cal group mem­ber­ship in pre­vi­ous years of the EA Sur­vey (~25% in 2015, ~30% in 2017), al­though com­par­i­sons across differ­ent iter­a­tions of the EA Sur­vey should be treated with ex­treme cau­tion, given pos­si­ble differ­ences in sam­pling (for ex­am­ple, we might have sys­tem­at­i­cally un­der-sam­pled lo­cal group mem­bers in ear­lier years or have dis­pro­por­tionately heav­ily sam­pled them in later years).

That said, it strikes us as plau­si­ble that there would be a real in­crease in the per­centage of EAs be­ing mem­bers of lo­cal groups, given that there has been a large in­crease in the to­tal num­ber of EA lo­cal groups in ex­is­tence ac­cord­ing to both Lo­cal Group Sur­vey 2017 data and the cen­tral database of lo­cal groups (which both find a dou­bling in num­bers of lo­cal groups since 2015).

N.B. no EA Sur­vey was run in 2016.

From Lo­cal Group Sur­vey 2017 data.

This change is strik­ing given that i) the EA com­mu­nity as a whole has of­ten been per­ceived as a largely on­line com­mu­nity, ii) it has of­ten been sus­pected that the EA Sur­vey likely over­sam­ples the on­line com­mu­nity and would un­der­sam­ple those who may be in­volved in EA offline, but are not ac­tive on­line. Each of these may still be true, and yet even if so, it seems there were still many more (around twice as many) EAs in our sam­ple who re­ported be­ing a mem­ber of a lo­cal group than the EA Fo­rum and similar num­bers were mem­bers of lo­cal groups than were mem­bers of the EA Face­book group.

GWWC and TLYCS Pledge

Around 32.4% of EAs in our sam­ple had taken the Giv­ing What We Can Pledge. We will be dis­cussing the Giv­ing What We Can Pledge fur­ther in a fu­ture post. Num­bers of sur­vey re­spon­dents who had taken the The Life You Can Save Pledge, which seems less promi­nent in the EA com­mu­nity, were pre­dictably much lower.

Re­la­tion­ships be­tween mem­ber­ship of differ­ent groups

Most (>83%) re­spon­dents who were mem­bers of the EA Fo­rum, were also mem­bers of the EA Face­book group. This means we can­not, strictly speak­ing, think of group mem­ber­ship within our sam­ple in terms of a larger, broader group of EA Face­book mem­bers, con­tain­ing a more nar­row, pop­u­la­tion of EA Fo­rum mem­bers, though there is broad over­lap. In con­trast, just over 61% of Fo­rum mem­bers are mem­bers of lo­cal groups. Among lo­cal group mem­bers, 31% (320 peo­ple in to­tal) are also mem­bers of the EA Fo­rum (note that this is sub­stan­tially higher than among those who are not mem­bers of lo­cal groups (12.8%)).

EA Face­book and EA Lo­cal Groups were the two groups with the largest mem­ber­ship within our sam­ple and while there was sub­stan­tial over­lap, many EAs who were mem­bers of EA Face­book were not mem­bers of a lo­cal group and vice versa.

De­spite there be­ing similar num­bers of EA Fo­rum and LessWrong mem­bers in our sam­ple, and de­spite the two fo­rums shar­ing some similar ar­eas of in­ter­est, mem­ber­ship of LessWrong and the EA Fo­rum was sur­pris­ingly non-over­lap­ping. Only about 31% of Fo­rum mem­bers were also mem­bers of LessWrong and only 30% of LW mem­bers were also mem­bers of the EA Fo­rum, with only 165 peo­ple in our whole sam­ple be­ing mem­bers of both. The ma­jor­ity of re­spon­dents (1694, 65%) were mem­bers of nei­ther.

Mem­ber­ship clusters

We also performed a cluster anal­y­sis to in­ves­ti­gate clusters within group mem­ber­ship in our data more in­duc­tively, rather than as­sum­ing a pri­ori that cer­tain group­ings were re­lated or in­for­ma­tive com­bi­na­tions.

To de­ter­mine the best num­ber of clusters we used the NB­cust pack­age in R which pro­vides 30 clus­ter­ing in­dices and pro­poses the best scheme by vary­ing all com­bi­na­tions of num­ber of clusters, dis­tance mea­sures, and clus­ter­ing meth­ods. We used eu­clidean dis­tance for the dis­similar­ity ma­trix, and com­pleted the fi­nal clus­ter­ing by cut­ting the clas­sifi­ca­tion tree gen­er­ated by ag­glomer­a­tive hi­er­ar­chi­cal clus­ter­ing with com­plete link­age to cre­ate the de­sired num­ber of groups.

Our anal­y­sis sug­gested that the data could be best un­der­stood in terms of 7 clusters of re­spon­dents, dis­played be­low. The table shows, for ex­am­ple, that among mem­bers of group 1, 0% of peo­ple are mem­bers of the EA Fo­rum, Face­book, LessWrong, The Life You Can Save a lo­cal group, while 19% are mem­bers of Giv­ing What We Can. By con­trast, in group 5, 100% of mem­bers are part of EA Face­book and LessWrong and a lo­cal EA group and 50% are mem­bers of the EA Fo­rum.

How ex­actly to in­ter­pret these clusters in the data is not ex­actly in­tu­itive or straight­for­ward. Nev­er­the­less, we think it high­lights a few group­ings of in­ter­est. For ex­am­ple, it iden­ti­fies a cluster (1) of 709 peo­ple (27.7% of re­spon­dents) who do not re­port be­ing mem­bers of any of the EA groups listed other than a minor­ity hav­ing taken the GWWC pledge. Clusters 2 (16.2%) and 3 (18.54%), con­versely, rep­re­sent clusters that are largely in­volved in EA Face­book, lo­cal groups (and, for group 2, GWWC), but not the EA Fo­rum or LessWrong. Cluster 4 are all lo­cal group mem­bers, but not mem­bers of most other groups. Cluster 5 (6.21%) and cluster 7 (11.98%) con­versely, rep­re­sent two dis­tinct clusters that are highly en­gaged across a se­lec­tion of groups.

Notably these clusters cor­re­spond fairly in­tu­itively with differ­ences that we pre­vi­ously noted in the cause se­lec­tion post. Clusters 1-4 each have a plu­ral­ity of re­spon­dents se­lect­ing Global Poverty as the top cause, whereas 5 and 7 (who have higher mem­ber­ship across a va­ri­ety of highly in­volved groups), along with cluster 6, which is pri­mar­ily LW mem­bers who are oth­er­wise not en­gaged with EA groups, have a plu­ral­ity of re­spon­dents se­lect­ing AI as the top cause. Similarly, cluster 1 the cluster with the low­est level of en­gage­ment, has the high­est sup­port for Cli­mate Change as a cause.

Gen­der Differ­ences in Group Membership

One more sen­si­tive find­ing was the re­la­tion­ship be­tween gen­der and mem­ber­ship of differ­ent EA groups. Across vir­tu­ally ev­ery cat­e­gory of group mem­ber­ship, a lower per­centage of women were mem­bers rel­a­tive to the sam­ple as a whole. The one ex­cep­tion is The Life You Can Save Pledge, which ap­pears to have a slightly higher per­centage of women.

A differ­ent way to look at this is the % of women and men, in our sam­ple, who were mem­bers of the differ­ent groups. Note that the num­ber of re­spon­dents for the cat­e­gories “Other” and “Pre­fer not to an­swer” were not in­cluded in the anal­y­sis be­low as the very small num­ber of re­sponses would limit statis­ti­cal anal­y­sis.

Of course, it is im­pos­si­ble to in­fer much about the cause of these differ­ences from these re­sults, nev­er­the­less, it seems worth not­ing as po­ten­tially rele­vant to on­go­ing dis­cus­sions about de­mo­graphic differ­ences within EA.


Over­all, these analy­ses sug­gest that a fairly large pro­por­tion of EAs, in our sam­ple at least, are not mem­bers of many EA groups. While mod­er­ately large por­tions of our sam­ple were in­volved ei­ther in the EA Face­book group (~50%), a lo­cal group (~40%) or the EA Fo­rum (~20%), rel­a­tively small num­bers of in­di­vi­d­u­als were in­volved in a wide va­ri­ety of differ­ent groups. This seems to ac­cord with a com­mon view that, among EAs, a rel­a­tively small sub­set are highly in­volved, at least in terms of group mem­ber­ship alongside oth­ers with differ­ing lev­els of in­volve­ment.

We will be ex­plor­ing pos­si­ble pre­dic­tors of in­di­vi­d­u­als be­ing more likely to be highly in­volved in EA groups in a fol­low-up post.

[1] Note that the per­centage fe­male for GWWC pledges as a whole (which in­cludes mem­bers who are not nec­es­sar­ily effec­tive al­tru­ists) is es­ti­mated to be roughly 30%- higher than within the effec­tive al­tru­ist sam­ple, though far from gen­der par­ity.


This post was writ­ten and with anal­y­sis by David Moss, Neil Dul­laghan and Kim Cud­ding­ton.

Thanks to Ju­lia Wise for com­ment and to Peter Hur­ford for re­view and edit­ing.

The an­nual EA Sur­vey is a pro­ject of Re­think Char­ity with anal­y­sis and com­men­tary from re­searchers at Re­think Pri­ori­ties.

Sup­port­ing Documents

Ar­ti­cles in the 2018 EA Sur­vey Series

I—Com­mu­nity De­mo­graph­ics & Characteristics

II—Distri­bu­tion & Anal­y­sis Methodology

III—How do peo­ple get in­volved in EA?

IV—Sub­scribers and Identifiers

V—Dona­tion Data

VI—Cause Selection

VIII- Where Peo­ple First Hear About EA and In­fluences on Involvement

Prior EA Sur­veys con­ducted by Re­think Charity

The 2017 Sur­vey of Effec­tive Altruists

The 2015 Sur­vey of Effec­tive Altru­ists: Re­sults and Analysis

The 2014 Sur­vey of Effec­tive Altru­ists: Re­sults and Analysis

No comments.