People react differently when encountering effective altruism for the first time. Some immediately find the ideas appealing and want to learn more, whereas others are much less enthusiastic about them. We suspect that people who immediately find the ideas appealing are more likely to become highly engaged EAs. Let’s call these people proto-EAs. What makes someone a proto-EA?
In a series of surveys, we studied the moral psychological factors that predict immediate interest in effective altruism. We found that the E (“effectiveness-focus”) and the A (“expansive altruism”) are psychologically distinct factors. Both are required to make someone a proto-EA. But only a few people score highly on both. We hope that a deeper understanding of the psychology of proto-EAs can prove practically useful for the community.
In previous research, we and others have found that most people’s intuitions don’t align with effective altruism. For example, most people donate to emotionally appealing charities even if they know that other charities do more good. In part, this is explained by motivations that aren’t aligned with effective altruism. And while it may be almost impossible to change people’s moral motivations (at least not with brief and cheap interventions), it turns out that there are wide individual differences in people’s proto-EA inclinations. So instead of trying to change the preferences of people in general, it may be more promising to target the subset of the population that is especially positively disposed toward effective altruism.
A potentially relevant pre-existing psychological measure is the Oxford Utilitarianism Scale (OUS, Kahane et al., 2018). This scale has two factors: impartial beneficence (complete impartiality and extreme self-sacrifice for the greater good) and instrumental harm (permissive attitude towards active harm for the greater good). However, we believe that these measures aren’t ideal to measure proto-EA inclinations. First, effective altruism doesn’t require an extreme level of self-sacrifice (e.g., giving away almost everything including your kidney) or complete impartiality (e.g., valuing distant people equally as your loved ones). A moderate form of altruism (e.g., giving away 10% of your income or choosing an impactful career you enjoy) combined with an expanded moral circle (e.g., prioritizing your loved ones but also taking distant others into consideration) are sufficient and likely more psychologically sustainable. Second, effective altruism doesn’t necessarily require instrumental harm (e.g., actively killing one person to save five people). Instead it simply requires a focus on effectiveness and the willingness to make the trade-offs required to prioritize the more effective helpful option over the less effective helpful option.
Based on these considerations, our goal was to develop a new scale that captures the moral psychological tendencies that predict relevant EA outcome measures (see below). Note that we did not attempt to capture all psychological factors that predict proto-EA inclinations but only the most relevant moral attitudes. Further, please note that our scales are not meant to define effective altruism conceptually or philosophically, but rather to capture the psychological tendencies that explain the variance in finding effective altruism appealing and interesting.
We ran a series of ten studies. Here we report the most relevant findings. More details about our studies including the study materials can be found in our Supplementary Materials.
We started by devising a long list of “items” (i.e., statements that survey takers could indicate their agreement with) that capture different moral aspects of effective altruism. In an intermediate step, we removed items that were not predictive of relevant EA outcome measures (these outcome measures include overall agreement with the key principles of EA, interest in learning more about EA, willingness to change one's career in order to have more impact). We then collected more data through Amazon Mechanical Turk (US citizens) and conducted exploratory factor analysis (a statistical method to identify which items relate to the same underlying psychological factors).
First, we noticed that items capturing altruism and an expanded moral circle load onto the same factor (i.e., seem to be part of the same underlying psychological trait). For instance, people who have a wide moral circle (e.g., who care about animals or future generations) are also more altruistic in general (e.g., they are willing to give away personal resources). We refer to this factor as “expansive altruism”, defined as people’s willingness to use some of their own resources (time or money) to help others, even if they are far away and not emotionally salient.
After a series of analyses (exploratory factor analyses, confirmatory factor analyses, and internal reliability analyses), we decided on the following items for the expansive altruism sub-scale. In our surveys, we measured how much participants agreed with each item on a scale from 1 (strongly disagree) to 4 (neither agree nor disagree), to 7 (strongly agree).
- As long as my and my family’s basic material needs are covered, I want to use a significant amount of my resources (e.g., money or time) to improve the world.
- I am willing to make significant sacrifices for people in need that I don’t know and will never meet.
- People in wealthy countries should donate a substantial proportion of their income to make the world a better place.
- I would make a career change if it meant that I could improve the lives of people in need.
- We should put a lot of emphasis on the well-being of people who will live thousands of years from now, even relative to the well-being of people who live today.
- From a moral perspective, the suffering of all beings matters roughly the same, no matter what species they belong to.
Responses (N = 259) were normally distributed. The mean score of the expansive altruism scale was 4.4 (SD = 1.1, α = 0.80, CFI = .97), which means that the average response score was slightly above the midpoint (4). Overall, these results show that the willingness to give away personal resources to help others, including distant others, is a coherent psychological construct with large individual differences. This is in line with previous research.
We believe that expansive altruism is a particular type of altruism. In contrast to other forms of altruism, expansive altruism is driven by a desire to not just help proximate and emotionally salient individuals, but also distant individuals that may be less emotionally salient. A similar type of distinction is made by psychologist Paul Bloom in his book Against Empathy (2017), in which he contrasts empathy-driven altruism with, what he refers to as, “rational compassion”. Our studies support the notion that such a form of rational compassion exists. However, our studies don’t suggest that this sort of compassion entails a lack of empathy. To the contrary, we find that expansive altruism strongly correlates with empathy (i.e., the trait “Empathic Concern”; see below).
Someone who scores low on expansive altruism is unlikely to be a proto-EA. However, not everyone who scores high on expansive altruism is a proto-EA. There are many people who agree that it’s important to help distant individuals who are not sympathetic towards effective altruism. This suggests there must be other relevant factors that distinguish between different types of expansive altruists.
Consistent with this possibility, we discovered another psychological trait that captures how much people prioritize effectiveness and their willingness to make the uncomfortable trade-offs that this can require when allocating altruistic resources. We refer to this factor as “effectiveness-focus”. Below is our six-item solution for the effectiveness-focus sub-scale. Before responding to the items (on a scale ranging from 1 to 7), participants read the following introductory paragraph.
Imagine a situation where you intend to do good (e.g., to improve others’ lives or the world) with a certain limited amount of resources available (e.g., your time or money). You can decide how to allocate your resources by choosing from different options that all do good. The stakes are high. In such a situation, when you can choose between different options of doing good …
- helping one person is less valuable than helping two people to the same extent.
- the most important consideration is effectiveness — choosing the option that does the most good per resource invested.
- you should follow evidence and reason to do what is most effective, even if you emotionally prefer another option
- it would be the right choice to refrain from helping one person if that makes it possible to help a larger number of people.
- you should usually help a large group of people over a smaller group, even if it seems unfair.
- it would be wrong to do something that only does some amount of good if there is an alternative course of action that would do much more good
Responses (N = 275) were normally distributed. The mean score of the effectiveness-focus scale was 4.4 (SD = 1.1, α = 0.83, CFI = .96), which means that the average response score was slightly above the midpoint (4). Overall, these results show that the tendency to focus on effectiveness in the pro-social context is a coherent psychological construct with large individual differences. Expansive altruism and effectiveness-focus were weakly correlated (r = .23). This shows that even though the two factors are positively associated, there are many people who score high on one but low on the other. Proto-EAs score high on both (see below).
A few comments about the effectiveness-focus scale: We deliberately chose relatively extreme items that focus on particularly uncomfortable aspects of effectiveness since such items were necessary to get enough variance and predictive power. This is because a large majority agrees with the importance of effectiveness in the abstract if the uncomfortable trade-offs are not made explicit.
Notably, we found that some items with an obligation framing (e.g., “it would be wrong ...”) yielded high variance and were particularly predictive. Remember that our goal was not to create items that together define effective altruism conceptually or philosophically, but rather to identify items that are predictive of our outcome measures. Thus, our inclusion of these obligation items does not mean that we think that effective altruism should be defined in terms of an obligation to help effectively.
Also, note again that effectiveness-focus is not the same as instrumental harm (e.g., actively killing one to save many). Effectiveness-focus doesn’t require active harm but simply the willingness to prioritize the more effective option (and in turn deprioritize the less effective option). And notably effectiveness-focus and instrumental harm are also psychologically different: they correlated only weakly (r = .34; see below).
Some effective altruist readers may wonder why many people disagree (i.e., score low) with the effectiveness-focus items. One possibility is that some people have relatively stronger counteracting preferences. Another possibility is that they are more likely to rely on their emotions than on reflection (cf. the dual process model in moral judgments due to Greene, 2014). It’s also possible that some people don’t want to disparage any type of altruistic action since they don’t consider it required to help in the first place (at least in some contexts). Distinguishing between these explanations would be an interesting topic for further research.
To validate the scales we tested them with three different populations: US general population (MTurk participants), undergraduate students (NYU), and effective altruists. We will now report each of these studies in turn.
Sample #1: US General population (MTurk)
Correlations and predictiveness
Across several studies, we tested how expansive altruism and effectiveness-focus were associated with other factors. The materials (e.g., the giving tasks) can be found in our Supplementary Materials.
In short, here is what we found:
- Expansive altruism and effectiveness-focus were positively associated with each other, r = .23 (p < .00001).
- Political liberalism (i.e., being left-wing) and young age predicted proto-EA tendencies.
- Men were more likely to score high on effectiveness-focus, whereas women were more likely to score high on expansive altruism. Women were slightly less likely to agree with EA overall (r = -.18, p = .004).
- Both factors predicted choices in various charitable giving tasks. For example, both expansive altruism and effectiveness-focus predicted the total amount participants were willing to donate to a highly effective charity instead of to their favorite charity. These measures show that our scales can predict behavioral choices.
- Both factors predicted greater endorsement of EA and interest in learning more about it. Expansive altruism also predicted increased intentions of taking EA-inspired action.
A general note about our correlational findings: Keep in mind that these are initial results and that more studies with larger sample sizes are needed to confirm the robustness of these associations. We report the unadjusted raw p-values without correcting for multiple tests, which means that some correlations could appear statistically significant just by chance. We therefore wish to emphasize that these results should be interpreted with caution. In particular, we believe that one shouldn’t read too much into any of the relatively small correlations, since it is possible that they won’t hold up in future studies.
To look at correlations with demographics, we pooled our data across two studies with US participants (N = 534; 46% female / 54% male; Mage = 41.5)
|Women vs. men||.12*||-.21****|
Note. Gender was coded as 0 for man and 1 for woman. Political liberalism ranged from 1 (“very conservative”) to 5 (“very liberal”). Note that the significant correlations with education level and income held even after controlling for age.
* p < .01, ** p < .001, *** p < .0001, **** p <. 00001
Charitable giving tasks
We presented US participants (N = 259; 49% female / 51% male; Mage = 41.8) with a range of charitable giving tasks that capture typical obstacles to effective giving. Positive correlations indicate that the two factors predict an increased tendency to choose the more effective option. (The task instructions can be found in our Supplementary Materials.)
|Total amount donated instead of keeping it||.34****||.09|
|Fraction donated to effective (vs. favorite) charity||.29****||.26***|
|Total amount donated to effective charity||.34****||.19*|
|Helping statistical victims vs. single identifiable victim||-.05||.29****|
|Deprioritizing one group in order to help larger number of victims||-.03||.22**|
|Wrongness of donating ineffectively||.08||.40****|
|Effectiveness vs. equity||-.01||.21**|
|Higher expected value vs. safer outcome with lower expected value||-.05||.16*|
|Concentrating donation to more effective charity over splitting||.05||.25***|
|Higher impact at the margin||-.08||.20*|
Note. The expected value task responses ranged from 0 to 3.
* p < .01, ** p < .001, *** p < .0001, **** p <. 00001
EA interest and intentions
We presented the same US participants (N = 259) with a short text summarizing the key ideas of effective altruism. And we then asked them some questions, such as whether they like these ideas and agree with them, and whether they would be interested in learning more or in taking action.
|M (SD)||Expansive altruism||Effectiveness-focus|
|Overall agreement with EA||4.8 (1.4)||.32****||.49****|
|Interest in learning more||4.1 (1.8)||.47****||.21**|
|Interest in signing up for an EA newsletter||24% yes (0.43)||.30****||.11|
|Interest in reading a book about EA||56% yes (0.5)||.17*||.11|
|Would consider donating 10%||45% yes (0.5)||.42****||.07|
|Openness to career change||48% yes (0.5)||.49****||.19*|
Note. The response scale for the first two items ranged from 1 (Strongly disagree) to 7 (Strongly agree).
* p < .01, ** p < .001, *** p < .0001, **** p <. 00001
We also conducted regression analyses to test if the two factors together were predictive of the outcome measures. Indeed, we did find that both factors were predictive of EA interest and intentions. In other words, if you know both scores, you can predict better whether someone is a proto-EA better than if you know only one score, but knowing only one score is better than knowing nothing.
Other psychological constructs
Finally, the same participants (N = 259) completed some other psychological scales that we considered potentially relevant. This included the Oxford Utilitarianism Scale (OUS; Kahane et al, 2019), the Footbridge Trolley Problem (Greene, 2014), and the Empathic Concern Scale (Interpersonal Reactivity Index; Davis, 1983).
|Footbridge Trolley Problem||.08||.31****|
* p < .01, ** p < .001, *** p < .0001, **** p <. 00001
We asked participants with a college degree (N = 134) which subject they had studied. Expansive altruism and effectiveness-focus did not correlate with college subject. Participants with a STEM degree were more interested in reading an EA book (r = .21, p = .02) and more likely to consider donating 10% (r = .22, p = .01). Participants with a humanities degree were less likely to consider donating 10% (r = -.18, p = .04). There were no other significant correlations with measures of EA interest/intentions.
Identifying the most likely proto-EAs
To identify the participants who were most positively disposed toward effective altruism, we selected participants who had a score above a certain bar on both expansive altruism and effectiveness-focus. We did this with a sample of 534 participants for the demographic measures and a subset of 259 participants for the outcome measures. We used three different bars: higher or equal to 4, higher or equal to 5, and higher or equal to 6. Remember that 4 was the midpoint (neither agree nor disagree), which means that any score above 4 indicates a positive tendency towards EA values.
|Mean scores on both scales||All||4 and above||5 and above||6 and above|
|Fraction of total sample||100%||49%||14.0%||3%†|
|Gender||46% female / 54% male||43% female / 57% male||27% female / 73% male||40% female / 60% male|
|Overall EA agreement||4.8||5.4||5.8||5.9|
|Learning more about EA||4.1||4.7||5.0||5.4|
|Open to career change||48%||63%||78%||100%|
† Please note that the sample size in the last column is very low and that therefore these statistics are not robust. We report them nevertheless for completeness sake. The sample sizes were 534, 261, 75, 15 for the demographic variables and 259, 120, 32 and 7, for the EA outcome measures respectively. Political liberalism ranged from 1 (“very conservative”) to 5 (“very liberal”). Overall EA agreement ranged from 1 (strongly disagree) to 7 (strongly agree). Interest in learning more about EA ranged from 1 (not at all interested) to 7 (very interested).
Sample #2: NYU undergraduate students
We recruited 96 NYU undergraduate students through the NYU business school (55.1% female / 44.9% male, Mage = 19.9). 40% studied business, 21% economics, and 39% something else.
At the beginning of the study we asked them whether they had heard of the social movement called “effective altruism” before. 69% responded with “no”, 23% responded with “kind of”, and 8% responded with “yes”. Participants who responded with “yes” or “kind of” were asked to explain what effective altruism is. Apart from one response that mentioned earning-to-give, the text responses suggested that participants did not know what effective altruism was.
NYU students scored significantly lower on expansive altruism (M = 4.1, SD = 1.1) than MTurkers (M = 4.4, SD = 1.1, p < .01). NYU students scored similarly high on effectiveness-focus (M = 4.3, SD = 1.1) as MTurkers. 39% of NYU students had a mean score of 4 or higher on both scales, 6% had a mean score of 5 or higher on both scales, and 2% had a mean score of 6 or higher on both scales.
We also measured Actively Open-Minded thinking (AOT) (M = 5.2, SD = 0.7). From an ongoing study by Matti Wilks et al., we know that Giving What We Can members score higher on AOT than MTurk participants.
Expansive altruism and effectiveness-focus did not correlate significantly (r = .11, p = .29), unlike in the MTurk sample. AOT correlated positively with effectiveness-focus (r =.24, p = .02) but not with expansive altruism (r = .05, p = .63).
Male students scored higher on effectiveness-focus (M = 4.7, SD = 1.0) than female students (M = 4.1, SD = 1.0, p < .01). By contrast there were no significant gender differences on expansive altruism. There were also no significant correlations with age.
EA interest and intentions
Students’ responses on the EA interest and intentions measures were similar to those of the MTurkers. Expansive altruism and effectiveness-focus were again predictive of many of the key variables. However, Actively Open-Minded thinking was not predictive for most measures. One possible explanation is that epistemic attitudes are less important in an initial confrontation with effective altruism but that they become relatively more important with a deeper engagement with effective altruist ideas, e.g. regarding cause prioritization.
|Mean (SD)||Expansive altruism||Effectiveness-focus||AOT|
|Overall agreement with EA||4.8 (1.3)||.46***||.35**||.10|
|Interest in learning more||3.9 (1.6)||.41***||.20||.12|
|Interest in signing up for an EA newsletter||11% yes||.26*||.16||.10|
|Interest in reading a book about EA||29% yes||.19||.26*||.12|
|Would consider donating 10%||63% yes||.47***||.06||.03|
|Openness to career change||43% yes||.33*||.08||.30*|
* p < .01, ** p < .001, *** p < .0001
Sample #3: Effective altruists
We recruited 226 effective altruists through social media (34% female / 63% male / 2% other; Mage = 28.8). As expected, effective altruists scored much higher than MTurkers on both expansive altruism (M = 5.6, SD = 0.9) and effectiveness-focus (M = 6.0, SD = 0.8). 95% of effective altruist participants had a mean score of 4 or higher on both scales, 81% had a mean score of 5 or higher on both scales, and 33% had a mean score of 6 or higher on both scales.
Participants indicated to what extent they identify as effective altruists on a scale from 1 (not at all) to 5 (a great deal) (M = 4.0, SD = 1.0). We found that identification with effective altruism correlated positively with scores on both expansive altruism (r = .37, p < .0001) and effectiveness-focus (r = .35, p < .0001). It is noteworthy that the correlation coefficients were roughly equally strong, suggesting that both factors are equally important in predicting identification with effective altruism. This provides further validation of the scales, in addition to the fact that both scales predict relevant EA outcome measures in non-EA samples.
40% of participants identified a “great deal” (maximum level) with effective altruism. Accordingly, those participants scored particularly high on both expansive altruism (M = 6.0, SD = 0.8) and effectiveness-focus (M = 6.3, SD = 0.7).
Younger effective altruists scored higher on expansive altruism (r = -.18, p < .01), but age didn’t correlate with effectiveness-focus. There were no significant correlations with education. There were also no robust correlations between gender and expansive altruism and effectiveness-focus, but men tended to identify slightly more with effective altruism (r = -.14, p = .04).
We asked participants which cause area they consider most important. 48% identified as longtermists (selecting “existential risk mitigation / longtermism”), 34% selected “global poverty”, 12% selected “animal welfare”, and 7% selected “other”. Longtermists identified more strongly with effective altruism (M = 4.2, SD = 1.0) than poverty-focused participants (M = 3.8, SD = 1.0, p = .001). Accordingly, longtermists scored higher on expansive altruism (M = 5.9, SD = 0.8) and effectiveness-focus (M = 6.2, SD = 0.6) than global poverty-focused participants (expansive altruism: M = 5.3, SD = 0.9; effectiveness-focus: M = 5.8, SD = 0.9; both p < .01). (The sample of animal welfare-focused participants was too small to conduct robust analyses.)
A deeper understanding of the psychology of proto-EAs is both practically and theoretically relevant.
First, the study of proto-EAs could prove useful for recruitment and marketing. It could help us better understand who proto-EAs are, what their traits and demographic features are, where we could find them, how to assess them, and how to best target them. More concretely, it could help us estimate the potential market size of effective altruism. How many proto-EAs are there? Less than 0.1% of the population or more than 20%? Where—in which countries, universities, and regions—do we find them? And how does this differ between different populations, e.g., between students and non-students or across different countries? For example, perhaps we learn that there are many more proto-EAs (per capita) in certain countries, in certain professions, at certain types of universities, or in particular study subjects than in others. Then we could start investing more resources into outreach efforts in these areas. And a fine-grained understanding of the different factors of effective altruism could help to optimize the efforts to spread our core values. The studies we presented here can be seen as a first step into this direction. However, our studies can only provide preliminary answers to a subset of the above questions. Much more research is needed to answer all of them (see the Future research section below).
Second, getting a better understanding of the different drivers of interest in effective altruism is theoretically useful. For example, we have found that there are at least two psychologically distinct factors that predict an interest in effective altruism, and it’s likely there are more (similar to the “big 5” personality factors). Such insights can guide future research by generating theory-derived hypotheses. More generally, a more fine grained theoretical understanding of the psychology of proto-EAs can help us research the origins, the underlying drivers, and the obstacles of each aspect of effective altruism in more detail. It can help to explain why certain people, but not others, will find some aspects of effective altruism more appealing than others. Ultimately, these theoretical insights could be useful for practical purposes.
As mentioned above, we see these studies as an initial step towards a better understanding of the psychology of proto-EAs. They have several limitations, as do the scales we have presented.
First, expansive altruism and effectiveness-focus are likely not the only important proto-EA factors. We have focused exclusively on moral aspects but other aspects may be at least as important (see Future research section).
Second, our scales rely on self-reports and thus aren’t particularly useful for assessment in competitive contexts (e.g., applications to EA fellowships) because they are easily gameable. For example, the expansive altruism scale plausibly rewards self-deception and punishes humility/scrupulosity. In competitive assessment contexts, other measures, such as performance-based tests, would likely be more useful. Our scales, or adjusted versions of them, might nonetheless serve as short screening tools in some contexts.
Third, our scales might be too coarse-grained and broad for more specific forms of targeting. For example, the psychology of proto-longtermists could be different from that of global poverty focused proto-EAs. That being said, we found that items related to global poverty, longtermism and anti-speciesism all correlated positively.
There are numerous possibilities for future research. Let us just briefly discuss two of them.
First, future research could conduct large scale representative surveys with different populations to estimate the potential market size of effective altruism. We plan to conduct such a representative survey with a large sample of students (e.g., at a US university using students from all different types of study areas). (Please get in touch if you have ideas or want to help out.) But it would also be valuable to conduct representative surveys with different populations. This could include surveying people across different cultures, or across different types of professions, such as philosophers, economists, medical experts, health care practitioners, etc.
Second, future research could improve the proto-EA measures and extend them. This includes the measurement of non-moral factors such as epistemic attitudes (e.g., truth-seeking/“Scout Mindset”), cognitive ability, analytical reasoning skills relevant for effective altruism, and potentially also certain personality traits, such as honesty/humility, conscientiousness, or “determination”. It also includes the development of assessment tools for competitive contexts.
We found two psychologically distinct moral factors that predict whether someone is a proto-EA. First, such a person is particularly willing to give away personal resources (time or money) to help others in need, even if they are far away. Second, such a person is focused on effectiveness when allocating their altruistic resources. We found that both expansive altruism and effectiveness-focus significantly predicted positive attitudes and interest towards effective altruism in non-EA samples. And we found that both scales predict stronger identification with effective altruism in pre-existing effective altruists. In short, both the E and the A are required to make someone a proto-EA, but only few people score highly on both.
We thank Matt Coleman, David Moss, and Matti Wilks for their helpful comments.
An alternative version of this item could read: “From a moral perspective, a given amount of suffering matters roughly the same, no matter what species the being experiencing it belongs to.” This item clarifies that the suffering amount is held constant. Note that we didn’t test this version.
Note that we initially also considered other possible moral factors. But we found that this two-factor solution was the most robust solution.
Note also that all our items were positively worded which may have inflated the mean scores of our scales due to acquiescence bias, the tendency of respondents to agree with positively worded statements. Social desirability might have also artificially inflated the scores, especially of the expansive altruism scale.