have you tried cluster analysis on the question “Which factors were important in ‘getting you into’ effective altruism, or altering your actions in its direction?”? Since it is multi-select, if most people actually selected more options, information about whether there are some patterns/clusters of factors that are important for people to feel they “got into” EA seems to me like it might be more informative than simple ranking of each factor’s frequencies independently.
This might be valuable to try for both, the whole sample as well as specifically for more engaged subgroups as Ben suggested.
We briefly looked into this and there was basically no interesting relation between “getting into” EA via the different factors. This isn’t particularly surprising to me, since a priori, for most of these factors, I wouldn’t expect getting more involved in EA via one factor to make you more likely to get involved in EA via a different factor relative to the other factors. For example, if you get more involved in EA via 80,000 Hours, should this make you more likely to get involved with local groups or online EA than if you got involved via GWWC? Or, if you get more involved in EA via 80,000 Hours, should this make you more likely to get more involved in EA via local groups than get more involved via online EA? For the most part, I’d expect not, and that was pretty much what the data shows. Most forms of getting more involved are just weakly positively correlated with most other forms of getting more involved, which is what I’d expect (if you get more involved in EA in any one way you are just slightly more likely to report getting involved in any other way).
The only variables between which there seemed to be any substantial connection (and even these were weak) was that a local group being important for one getting more involved was associated with a personal contact being important for one getting more involved. This makes sense, and is arguably almost trivially true, since it seems like being involved in a local group necessarily entails having personal contacts. In the cluster analysis you could arguably_vaguely_ make out a ‘nothing much selected’ cluster, a ‘local group and personal contact (but also a modest amount of online, 80K, GWWC, and book, blog etc.)’ cluster and an ‘everything else inc. especially online EA, 80K, GWWC, and book, blog etc.)’ cluster, but as noted, there wasn’t much definition within the clusters.
Conversely, a case where I think that cluster analysis does make sense and where the results were more informative was looking at categories of group membership as we do here.
I think another instance where looking at the relationships between reporting getting more involved in EA in different ways and other outcomes is the series of Multiple Correspondence Analyses that we included in the Cause Selection post. If you examine these, then you can see, for example, that there is a correspondence between “getting more involved” via online EA, a personal contact or local group and lower prioritisation of climate change (and the converse pattern for AI/x-risk).
With regards to the question of looking at more engaged subgroups, we looked at the relationship between first hearing about EA in different ways and levels of involvement previously. You’ll note that we found little evidence that different routes into EA lead to substantially higher or lower proportions of people becoming highly engaged EAs. Rather, the number of highly engaged EAs seemed pretty closely proportional to the total number of EAs recruited from each source.
When looking at the multi-select question, we do find some significant differences, though as in the case of ‘first heard’ data, we would expect this to be confounded, most importantly by how long respondents have been in EA. Still, it seems like significantly more people who report getting more involved via a book or blog, etc., EA Global, local EA group, online EA or personal contact are members of the EA Forum than not (across all time periods). You can also compared the raw percentages of people involved in different ways across Forum members and non-Forum members. I don’t think this suggests anything in particular about any causal relations between being involved in one way and being involved in another though.
Thanks, David! My intuition a bit different—most people I talked to about how got they got into EA mentioned multiple factors, so I was curious whether there would be general patterns/strong clusters of e.g. “offline EA (e.g. local groups + personal contact)” vs “Online EA” or clusters based on type of activity/contribution, e.g. “donation cluster (Givewell + GWWC + TLYCS + ACE)” vs “direct work cluster (80K + EA global)” vs “producing/reading research (books/articles + SSC + LW)”. I see it doesn’t seem to be the case for the whole sample base don your analysis.
Regarding the more engaged subgroup—while I perceive analysing of “where people first heard about EA” as interesting, the question of “which factors actually got people into EA” seems to me as more important. The more that most people I talked to mentioned that first point of contact with EA didn’t cause them to “get into it” right away, but only after encountering some other factors they finally made a step to “dive into it” or “get involved”. Therefore, looking whether there are some patterns of these factors for more vs less involved people seems to me potentially promising, I agree that “how long people have been in EA” will probably add a lot of noise to this as these patterns could have changed over time. Not sure whether the sample size would be big enough to analyse each year separately or whether there are some other ways to control for it.
Using cluster analysis for the group membership seems to me like a good example of such approach.
Thanks David. I agree the online vs offline idea is pretty intuitive. Arguably you can vaguely see some signs of this in the data, but not to a great extent. As the membership post shows, there’s quite a lot of overlap between online (EA Facebook) and offline (Local Group):
looking whether there are some patterns of these factors [how people report getting more involved] for more vs less involved people seems to me potentially promising
Yes, this is what the analysis in the previous comment aims to do. And yes, you are correct that controlling for time in EA is very difficult because of the very small numbers in the earlier years of EA, hence in the analysis above we just display results splitting by those joining before/after 2016.
Hi David,
have you tried cluster analysis on the question “Which factors were important in ‘getting you into’ effective altruism, or altering your actions in its direction?”? Since it is multi-select, if most people actually selected more options, information about whether there are some patterns/clusters of factors that are important for people to feel they “got into” EA seems to me like it might be more informative than simple ranking of each factor’s frequencies independently.
This might be valuable to try for both, the whole sample as well as specifically for more engaged subgroups as Ben suggested.
Hi David,
We briefly looked into this and there was basically no interesting relation between “getting into” EA via the different factors. This isn’t particularly surprising to me, since a priori, for most of these factors, I wouldn’t expect getting more involved in EA via one factor to make you more likely to get involved in EA via a different factor relative to the other factors. For example, if you get more involved in EA via 80,000 Hours, should this make you more likely to get involved with local groups or online EA than if you got involved via GWWC? Or, if you get more involved in EA via 80,000 Hours, should this make you more likely to get more involved in EA via local groups than get more involved via online EA? For the most part, I’d expect not, and that was pretty much what the data shows. Most forms of getting more involved are just weakly positively correlated with most other forms of getting more involved, which is what I’d expect (if you get more involved in EA in any one way you are just slightly more likely to report getting involved in any other way).
The only variables between which there seemed to be any substantial connection (and even these were weak) was that a local group being important for one getting more involved was associated with a personal contact being important for one getting more involved. This makes sense, and is arguably almost trivially true, since it seems like being involved in a local group necessarily entails having personal contacts. In the cluster analysis you could arguably_vaguely_ make out a ‘nothing much selected’ cluster, a ‘local group and personal contact (but also a modest amount of online, 80K, GWWC, and book, blog etc.)’ cluster and an ‘everything else inc. especially online EA, 80K, GWWC, and book, blog etc.)’ cluster, but as noted, there wasn’t much definition within the clusters.
Conversely, a case where I think that cluster analysis does make sense and where the results were more informative was looking at categories of group membership as we do here.
I think another instance where looking at the relationships between reporting getting more involved in EA in different ways and other outcomes is the series of Multiple Correspondence Analyses that we included in the Cause Selection post. If you examine these, then you can see, for example, that there is a correspondence between “getting more involved” via online EA, a personal contact or local group and lower prioritisation of climate change (and the converse pattern for AI/x-risk).
With regards to the question of looking at more engaged subgroups, we looked at the relationship between first hearing about EA in different ways and levels of involvement previously. You’ll note that we found little evidence that different routes into EA lead to substantially higher or lower proportions of people becoming highly engaged EAs. Rather, the number of highly engaged EAs seemed pretty closely proportional to the total number of EAs recruited from each source.
When looking at the multi-select question, we do find some significant differences, though as in the case of ‘first heard’ data, we would expect this to be confounded, most importantly by how long respondents have been in EA. Still, it seems like significantly more people who report getting more involved via a book or blog, etc., EA Global, local EA group, online EA or personal contact are members of the EA Forum than not (across all time periods). You can also compared the raw percentages of people involved in different ways across Forum members and non-Forum members. I don’t think this suggests anything in particular about any causal relations between being involved in one way and being involved in another though.
Thanks, David! My intuition a bit different—most people I talked to about how got they got into EA mentioned multiple factors, so I was curious whether there would be general patterns/strong clusters of e.g. “offline EA (e.g. local groups + personal contact)” vs “Online EA” or clusters based on type of activity/contribution, e.g. “donation cluster (Givewell + GWWC + TLYCS + ACE)” vs “direct work cluster (80K + EA global)” vs “producing/reading research (books/articles + SSC + LW)”. I see it doesn’t seem to be the case for the whole sample base don your analysis.
Regarding the more engaged subgroup—while I perceive analysing of “where people first heard about EA” as interesting, the question of “which factors actually got people into EA” seems to me as more important. The more that most people I talked to mentioned that first point of contact with EA didn’t cause them to “get into it” right away, but only after encountering some other factors they finally made a step to “dive into it” or “get involved”. Therefore, looking whether there are some patterns of these factors for more vs less involved people seems to me potentially promising, I agree that “how long people have been in EA” will probably add a lot of noise to this as these patterns could have changed over time. Not sure whether the sample size would be big enough to analyse each year separately or whether there are some other ways to control for it.
Using cluster analysis for the group membership seems to me like a good example of such approach.
Thanks David. I agree the online vs offline idea is pretty intuitive. Arguably you can vaguely see some signs of this in the data, but not to a great extent. As the membership post shows, there’s quite a lot of overlap between online (EA Facebook) and offline (Local Group):
Yes, this is what the analysis in the previous comment aims to do. And yes, you are correct that controlling for time in EA is very difficult because of the very small numbers in the earlier years of EA, hence in the analysis above we just display results splitting by those joining before/after 2016.