Where the Head Meets the Heart: ‘Enlightened’ Compassion Lies Between Big Five Openness/Intellect and Agreeableness

Though the term “enlightened compassion” is new, at least two traits from the literature capture central features of the construct as we conceive of it. Highlighted in the opening quote by Johnson (1993), the first such trait is moral imagination and its relatives moral creativity (Haste, 1993), wise reasoning (Grossmann et al., 2010), and integrative complexity (Tetlock, 1986). Moral imagination is the envisioning of novel and varied solutions to moral problems. Rather than blindly accepting the moral status quo, morally imaginative people actively explore intractable dilemmas. They identify potential moral concerns, critically reflect on their personal standing, consider the diverse perspectives of those affected, and generate possible solutions (Kingsley, 2011). The moral imagination tradition thus captures enlightened compassion in terms of thinking flexibly about moral problems to advance more communally acceptable solutions.

Captured in the opening quote by Singer (1981), a second relevant trait is moral expansiveness and its relatives universalism (Schwartz et al., 2012) and (low) generalised prejudice (Crawford & Brandt, 2019). Moral expansiveness refers to the scope of one’s moral boundaries, in terms of the number and range of entities to which one ascribes moral worth. Rather than fixating on the welfare of kith and kin, morally expansive people also feel a responsibility for socially distant others—those with whom they are not acquainted or to whom they are not similar. This includes members of marginalised groups, citizens of foreign communities, nonhuman animals, and nature (Crimston et al., 2016). The moral expansiveness tradition thus captures enlightened compassion in terms of extending one’s duty of care inclusively to a variety of human and nonhuman targets.

We regard moral imagination and moral expansiveness as “sibling constructs” (Lawson & Robins, 2021), each highlighting a different way that compassion can take on a more open-minded or enlightened form. Yet because they have emerged from separate research traditions, potential connections between these traits have been overlooked. Such siloing of research literatures is common in many areas of psychology, especially when trait constructs are derived through a “top-down” mode of inquiry—as was the case for moral imagination and moral expansiveness. Within this mode, researchers zoom-in on their trait of interest (Silvia & Christensen, 2020), anchoring their operational definition and measure to the most pertinent a priori considerations. The goal of the top-down approach is to carve out the most salient and distinctive features of a construct, emphasising the ways in which it is conceptually unique.

An alternative and complementary way to study a new trait construct is to unearth it from the “bottom-up” by operating within the boundaries of comprehensive personality trait taxonomies (Silvia & Christensen, 2020) such as the B5 (John et al., 2008). Such taxonomies organise and condense the otherwise unwieldy trait universe via the specification of basic traits (see Figure 1)—the empirical dimensions that reliably emerge from patterns of covariation within representative samples of all trait constructs (Costa & McCrae, 1992). Within the bottom-up mode of inquiry, researchers zoom-out on their trait of interest (Silvia & Christensen, 2020), anchoring their operational definition and measure to the content of pertinent basic traits. The goal of the bottom-up approach is to assimilate all relevant features of the construct that cohere empirically (Hofstee et al., 1992; Stanek & Ones, 2018). In addition to uncovering new traits from the content of more basic traits, a bottom-up approach can be used to pinpoint the location of existing traits derived in a top-down fashion within taxonomic frameworks and in relation to existing constructs (e.g., Bainbridge et al., 2022; Credé et al., 2016; Silvia & Christensen, 2020).

Although the top-down approach is useful for exploring novel constructs, an overemphasis on novelty can lead a field toward fragmentation (Balietti et al., 2015; Burghardt & Bodansky, 2021) and construct proliferation (Le et al., 2010), as ever more traits are introduced into the literature. Construct proliferation hinders scientific progress because it tends to result in “jangle fallacies”, or the application of distinct labels to traits or measures that are conceptually and/or empirically indistinguishable (Kelley, 1927). The bottom-up mode of inquiry mitigates these problems by capitalising on accumulated knowledge regarding the basic trait(s) most pertinent to a new trait of interest. This enhances the efficiency of studying new traits in at least three ways. First, it allows researchers to use prior studies of the content, correlates, and consequences of the relevant basic trait(s) to inform how they operationally define and measure their focal trait. Second, it ensures subsequent findings are integrated within a shared framework and vernacular for organising and describing trait constructs. Third, it reduces the probability of jangle fallacies because existing basic traits and their correlates are taken into explicit consideration, making it easier to document their empirical overlap (Bainbridge et al., 2022; Goldberg & Mervielde, 1999; John et al., 2008; Stanek & Ones, 2018). Amid these benefits, there are also limitations to the bottom-up mode of inquiry. By focussing on the similarities between traits, the bottom-up approach risks oversimplifying the trait universe by abstracting over meaningful distinctions that would otherwise be captured using a top-down lens. For this reason, the bottom-up and top-down approaches are inherently complementary—a healthy literature will contain a balance of both.

Existing conceptualisations of what we call enlightened compassion, such as moral imagination and moral expansiveness, were rigorously derived from the top-down, but the literature currently lacks a complementary effort to unearth enlightened compassion from the bottom-up. To address this gap, we attempt to derive a measure of enlightened compassion from the content of the B5 (John et al., 2008)—the most widely adopted taxonomy of personality traits (see Figure 1).¹

Among the B5 “domains”, Agreeableness is the most explicitly other-regarding, capturing the tendency to be benevolent and amiable (Graziano & Tobin, 2009). Agreeableness is thus the most obvious dimension within the trait universe to begin our search for enlightened compassion. Like all B5 domains, Agreeableness can be subdivided into a further set of basic traits, which offer a more nuanced partitioning of domain-level variance. As Figure 1 highlights, Agreeableness divides into Politeness and Compassion (Crowe et al., 2018; DeYoung et al., 2007). Whereas Politeness captures compliance with social norms, Compassion describes concern for others’ welfare (DeYoung, 2015b; Zhao et al., 2017). Therefore, as implied by its label, the latter appears more relevant to enlightened compassion, subsuming tendencies such as empathic sensitivity and helpfulness (DeYoung et al., 2007).

Crucially, high scorers on Compassion may not always express their concern for others in enlightened ways. In principle, a compassionate person could be moved to address moral problems, yet default to simplistic heuristics when proposing solutions (e.g., maximising either freedom or equality at all costs). Similarly, they might behave prosocially toward acquaintances yet neglect the needs of targets beyond their social ties (e.g., donating exclusively to local charities). Though compassionate, such sentiments and behaviours would be rigid and parochial and thus unenlightened (see also Bloom, 2016; Bruneau et al., 2017). Therefore, although Compassion is likely a necessary ingredient of enlightened compassion, it may be insufficient to capture the construct with adequate precision.

To count as enlightened, we argue that compassion must be imbued with open-mindedness, or a willingness to explore novelty and diversity. This is one of the core features of Openness/Intellect (Christensen et al., 2019), a domain within the B5 taxonomy describing the tendency to be curious and imaginative (DeYoung, 2015a). At the aspect level, Openness/Intellect divides into Intellect and Openness (DeYoung et al., 2007; Woo et al., 2014), with Intellect describing logical and rational engagement (e.g., “formulate ideas clearly”) and Openness describing aesthetic engagement, imagination, and absorption (e.g., “love to reflect on things”) (Christensen et al., 2019; DeYoung, 2015b). Research has shown that open-minded tendencies—including participation in deep reflection and introspection, and tolerance of alternative perspectives and lifestyles—cluster under Openness rather than Intellect (Woo et al., 2014). Therefore, as implied by its label, Openness appears more relevant than Intellect to enlightened compassion.

Integrating these considerations, it seems the most plausible location of enlightened compassion within the B5 taxonomy lies between the Compassion aspect of Agreeableness and the Openness aspect of Openness/Intellect. This is further bolstered by the fact that, among the B5 domains, Agreeableness and Openness/Intellect both emerge as the strongest meta-analytic correlates of traits that are conceptually related to enlightened compassion—including universalism (Parks-Leduc et al., 2014), low prejudice (Connelly et al., 2014; Crawford & Brandt, 2019; Sibley & Duckitt, 2008), and environmentalism (Soutter et al., 2020). In technical terms, we are proposing that enlightened compassion is an interstitial trait of Compassion and Openness. This means it ostensibly shares substantial overlap with both of these basic traits, and should therefore be located between them within the B5 taxonomy (Saucier, 1992).

Because the B5 model has a hierarchical structure (see Figure 1), there are three locations between Compassion and Openness where enlightened compassion might lie: directly above these traits (as a higher-order domain), adjacent to them (as an intermediate aspect), or directly below them (as a lower-order facet). The B5 maps traits hierarchically based on their content breadth, with traits at the higher levels describing broader (vs. narrower) classes of thoughts, feelings, desires, and actions compared to lower-level traits (Goldberg, 1993). The content breadth of enlightened compassion appears narrower than both Compassion and Openness, capturing a subset of compassionate tendencies that are open-minded. It therefore seems most appropriate to posit enlightened compassion as a lower-order facet nested below these aspects within the B5 taxonomy.

Importantly, there may be no definitive means to determine whether a construct resides at the domain-, aspect-, or facet-level of the B5. Indeed, the number of levels within the personality trait hierarchy is itself indeterminate. Nevertheless, we can specify patterns of findings that would be consistent with our suggestion that enlightened compassion—as a subset of compassionate tendencies that are open-minded—is best regarded as a lower-order facet of the B5. Specifically, an implication of the bandwidth-fidelity trade-off (Gleser et al., 1965) is that measures of relatively narrow traits residing lower in the hierarchy will show incremental validity over-and-above measures of relatively broad traits located above them. In other words, a facet scale should demonstrate higher fidelity in tests of validity, relative to broader bandwidth scales at the aspect level, and likewise for aspect scales relative to domains (e.g., see Danner et al., 2021). Thus, if enlightened compassion is indeed a facet of the B5, we would expect it to predict theoretically relevant traits and behaviours over-and-above the B5 aspects.

We define enlightened compassion as the tendency to show regard for others in a manner that is open-minded rather than rigid or parochial. Across three studies drawing on seven samples, we aim to unearth enlightened compassion from the bottom-up by evaluating whether it can be represented within the B5 taxonomy as an interstitial facet below the Compassion aspect of Agreeableness and the Openness aspect of Openness/Intellect. In Study 1, we develop and assess the structural validity of a targeted enlightened compassion scale, and examine its content validity by assessing convergent and discriminant associations with basic personality traits from the B5 taxonomy. In Study 2, we examine the convergent validity of the enlightened compassion scale in terms of its associations with conceptually similar traits developed in a top-down fashion, including moral imagination (Kingsley, 2011) and moral expansiveness (Crimston et al., 2016). In Study 3, we examine the criterion validity of the enlightened compassion scale in terms of its associations with a relevant and consequential behaviour: willingness to donate money to a wide variety of needy human and nonhuman targets. In Studies 2 and 3, we also examine whether enlightened compassion demonstrates incremental validity over Compassion and Openness. Such incremental validity would suggest that the former trait is indeed a higher-fidelity facet of the latter two, based on principles of the bandwidth-fidelity trade-off in hierarchical models of personality traits (Cronbach & Gleser, 1957).

Table 1 provides an overview of the samples used in our three studies. There are seven distinct samples overall (total N = 2,522), each comprising either students enrolled in an Australian undergraduate psychology course (Samples B, D, E, and F), or Amazon Mechanical Turk (MTurk) workers residing in the US (Samples A, C, and G). All surveys were administered via Qualtrics^TM and included multiple attention checks. Ethical approval for each sample was granted by the Human Research Ethics Advisory Group of the School of Psychological Sciences at The University of Melbourne).

Data, analysis code, pre-registrations, materials (including attention checks and a complete list of administered measures), and a Supplementary File (containing additional analyses and descriptive statistics) are available via the Open Science Framework (OSF) at https://osf.io/qze53/?view_only=192697d7bf22470ab7eb9fc18ec65f93. Some administered measures were not relevant to the aims and hypotheses of this project and are hence not analysed. Only some samples had an accompanying pre-registration (see Table 1), and only some hypotheses within those samples were pre-registered. We note any pre-registered hypotheses in the introductory section of each study, and summarise deviations from our pre-registrations in Appendix A. Statistical power considerations are described in the method section of each study. Analyses were conducted in R (R Core Team, 2020).

In Study 1, we aimed to develop a novel enlightened compassion scale assessing the tendency to show consideration and concern for others in an open-minded way (i.e., B5 Compassion imbued with B5 Openness). Crucially, we aimed to unearth this scale from the bottom-up, by deriving items from the content of Compassion and Openness scales, rather than from the theoretical conceptualisation we laid out in our Introduction. We pre-registered this exploratory exercise (https://aspredicted.org/blind.php?x=cd8xg9), noting we would employ a range of techniques across multiple samples to arrive at a final scale. After finalising the scale and assessing its structural validity (i.e., model fit and reliability), we explored its content validity by evaluating its i) convergence with Compassion and Openness and ii) divergence from the aspects of Extraversion, Neuroticism, and Conscientiousness. These content validity analyses allowed us to evaluate whether our scale is interstitial to Compassion and Openness within the B5 taxonomy.

After constructing an enlightened compassion item pool (see Measures section), we reduced it to a final scale across two phases. In Phase One, we used an initial sample to a) eliminate local dependencies within the item pool via Unique Variable Analysis (UVA), b) assess the dimensionality of the reduced item pool using Parallel Analysis with Principal Axis Factoring (PA), and c) further refine the scale by selecting the items that converged most strongly with both Compassion and Openness. In Phase Two, we confirmed the structural validity of our final scale by a) evaluating its model fit and reliability in a new sample via Confirmatory Factor Analysis (CFA) using the weighted least squares means and variance (WLSMV) estimator, and b) assessing its measurement invariance across additional independent samples.

For Phase One we used our largest sample, Sample A (see Table 1), which provided a strong subject-to-item ratio (>40:1) for the PA model (see Supplementary File; Costello & Osborne, 2005). In Phase Two, we used our next largest sample, Sample B (see Table 1), which provided a strong N:q ratio (>30:1) for the CFA model (see Supplementary File; Kline, 2015). Finally, we examined our scale’s convergence with and divergence from the B5 aspects using Samples A through G, all of which were well powered to detect an average-sized individual difference correlation of r = .20 (Gignac & Szodorai, 2016), with power for our smallest (Sample D; N = 190) and largest (Sample A; N = 774) samples at 79% and 99% respectively (see Supplementary File).

Basic traits. Basic traits from the B5 taxonomy were assessed using the Big Five Aspects Scales (BFAS; DeYoung et al., 2007) and Big Five Inventory-2 (BFI-2; Soto & John, 2017). Basic traits from the HEXACO, a prominent alternative to the B5 (Ashton et al., 2014), were also assessed. For brevity we focus on the BFAS; results for the HEXACO and BFI-2 are in the Supplementary File.

BFAS (Samples A to G). The 100-item BFAS assesses the B5 aspects with 10 items each. This includes the Compassion (e.g., “feel others’ emotions”) and Politeness (e.g., “respect authority”) aspects of Agreeableness, and the Openness (e.g., “enjoy the beauty of nature”) and Intellect (e.g., “think quickly”) aspects of Openness/Intellect. Participants responded to each item from 1 (strongly disagree) to 5 (strongly agree), with each participant’s scores for each B5 aspect computed as their mean response across constituent items.

Enlightened compassion (Samples A to G). We developed an initial pool of 24 items to capture enlightened compassion (Table 2). Each item was designed to blend B5 Compassion with B5 Openness, and was informed by the item content of the measures of these aspects contained within the BFAS. During this process, we perceived three themes within the item content, and thus generated an equal number of items within each theme to ensure a broad sampling of enlightened compassionate content. Thus, eight items covered the tendency to reflect on and envision ways to help others and improve society, a further eight covered the tendency to extend compassion to marginalised or foreign people, and the remaining eight covered the tendency to extend compassion to nature and cultural artefacts.² These themes were not informed by any specific a priori conceptualisation of enlightened compassion, nor were they intended to form specific subscales. Participants responded to each item from 1 (strongly disagree) to 5 (strongly agree).

Local dependence. We first reduced the item pool by identifying instances of local dependence between two or more items, which would indicate that they contain redundant information (Edwards et al., 2018). Such redundancy reduces a scale’s bandwidth, narrowing its conceptual coverage. Local dependence was assessed in Sample A via UVA, a network psychometrics approach for detecting local dependence when the underlying internal structure of the scale is unknown (Christensen et al., 2022). We used the recommended threshold of 0.20 and iteratively identified local dependencies until none remained. Among the 24 items in the enlightened compassion item pool, UVA identified five pairs of locally dependent items (see Table 2), prompting us to eliminate one item per pair from the pool. Within each pair, we eliminated the item with the more narrow/specific (vs. broad/general) focus.

Dimensionality. Using Sample A, we performed PA on the remaining 19 items in the enlightened compassion item pool to explore its dimensionality. Although our item pool spanned three conceptual themes, parallel analysis detected up to five factors (Figure 2). However, only the first factor produced an observed eigenvalue that was >1 and clearly above its simulated eigenvalue. The break in the scree plot of observed eigenvalues also suggested a single factor (Figure 2), prompting us to settle on a one-factor solution. This single enlightened compassion factor (extracted using maximum likelihood estimation) explained 31% of variance in the items.

Scale formation. We sought to further reduce the item pool into a final scale that balanced content breadth with brevity. To determine which of the 19 items to retain, we regressed each on Compassion and Openness in Sample A (i.e., 19 separate regression models) and used the resulting squared multiple correlations (R²) to identify items that overlapped most strongly with Compassion and Openness. To preserve adequate content breadth, we retained the four items within each of our three initial conceptual themes with the largest R² values (see Table 2). These 12 items were selected for the final Enlightened Compassion Scale (EC Scale; Appendix B).

Model fit. We used an independent sample, Sample B, to evaluate the model fit of the EC Scale via CFA (see Figure 3). The scaled goodness-of-fit indices were adequate but did not consistently meet heuristic benchmarks (see Brown, 2015), which is a common occurrence in personality measurement (Hopwood & Donnellan, 2010). The SRMR was .08, which met the ≤ .08 benchmark, but the RMSEA (.11) and CFI (.87) fell just short of their respective benchmarks (.10 and .90). Importantly, these indices were broadly similar to those observed for major B5 scales (e.g., see Table 8 in Soto & John, 2017), suggesting the fit of the EC Scale is less than ideal, but comparable to prominent scales in the field.

Measurement invariance. To assess the consistency of the EC Scale structure (i.e., its measurement invariance) across samples, we compared the CFA model from Sample B to an equivalently specified model within each of our remaining samples (Samples C to G). For each sample, we examined four types of measurement invariance: configural (equivalent factor structure), metric (equivalent factor structure and factor loadings), scalar (equivalent factor structure, factor loadings, and intercepts), and strict (equivalent factor structure, factor loadings, intercepts, and means).

With Sample B, the EC Scale demonstrated metric invariance in Samples C, D, and E. Further, scalar and strict invariance was found for Samples D and E. Samples F and G did not meet the threshold for metric invariance. Based on RMSEA, SRMR, and CFI, configural invariance was adequately supported for all samples, though not strongly, on all three criteria. Specifically, all SRMRs were ≤ .08, and all RMSEAs were ≤ .10, but the CFIs (all ≥ .80) fell short of the .90 benchmark.

Samples that demonstrated at least metric invariance with B differed from those that did not in a few notable ways. In terms of gender identity, Sample B was majority woman identifying (70%) with Samples C, D, and E being similar in majority woman identifying (⁠$≥$ 60%). In contrast, Samples F and G were either more closely split between man and woman identifying (Sample E: 45% and 54%, respectively) or were majority man identifying (Sample G: 60% and 40%, respectively).

In terms of ethnicity, Sample F and G had higher representation of Asian (51%) and Caucasian (73%) ethnicities relative to Sample B (41% and 51%, respectively). It’s notable, however, that Sample E had similar ethnicity characteristics to Sample F and achieved strict invariance suggesting that ethnicity differences may not be driving the differences in invariance. Similarly, Sample G had a much higher average age (35.80) than Sample B (19.82) but Sample C was similar in age characteristics to Sample G and still achieved metric invariance suggesting that age differences may not be driving the differences in invariance. In sum, the gender identity composition of the Samples seems to be a potential factor for the invariance of some Samples relative to others with Sample B.

Reliability. We estimated the internal consistency of the EC Scale in Samples A to G via McDonald’s omega and Cronbach’s alpha. Table 4 shows that the reliability of the EC Scale was solid across samples, with all values ≥ .80.

To perform our content validity analyses in the remainder of Study 1, we computed an overall enlightened compassion score within Samples A through G for each participant by taking their mean response across the EC Scale items.

Convergence with Compassion and Openness. Enlightened compassion was strongly and significantly correlated with Compassion (.46 ≤ r ≤ .65) and Openness (.51 ≤ r ≤ .66) within each sample (Table 5), producing a sample-weighted average correlation of r = .57 for both Compassion and Openness. These correlations suggest that the EC Scale successfully captured a blend of Compassion and Openness content.

Divergence from nonfocal B5 aspects (see Supplementary File). Sample-weighted average correlations across Samples A through G revealed that, as expected, enlightened compassion was negligibly associated with the aspects of Neuroticism and Conscientiousness (sample-weighted average rs ≤ .12). Semipartial correlations revealed that, once their shared variance with Compassion or Openness was removed, neither Politeness (-.13 ≤ srs ≤ .09), Intellect (.00 < srs < .18), nor the Enthusiasm aspect of Extraversion (-.05 ≤ srs ≤ .11) were associated with enlightened compassion.

In summary, the EC Scale covers a blend of Compassion and Openness as intended, while excluding content from conceptually distinct basic traits. This is in line with our suggestion that enlightened compassion is interstitial to Compassion and Openness within the B5 taxonomy.

Findings from Study 1 support the structural and content validity of our measure of enlightened compassion. In Study 2, we aimed to assess the convergent validity of the EC Scale through examining its bivariate correlation with candidate trait correlates from the literature (e.g., moral imagination; moral expansiveness). Most candidate correlates were measured in at least two of our seven samples to gauge the replicability of our findings. We hypothesised that enlightened compassion would correlate positively with all candidate trait correlates, and pre-registered these hypotheses for moral expansiveness and moral imagination (the most theoretically pertinent of these correlates) in Sample C (https://aspredicted.org/blind.php?x=y9b7wm). In addition to these hypotheses, we also evaluated the incremental validity of our EC Scale over-and-above the Compassion and Openness aspects of B5 Agreeableness and Openness/Intellect. According to the bandwidth-fidelity trade-off (Cronbach & Gleser, 1957), narrower traits should predict relevant criteria more strongly than broader traits. As a putative facet measure, our EC Scale should therefore share associations with each candidate trait correlate over-and-above the Compassion and Openness aspects below which it is ostensibly nested.

Beyond these focal analyses, we pre-registered an additional hypothesis that Compassion and Openness would themselves be positively correlated, despite being nested under different B5 domains (see Sample A pre-registration; https://aspredicted.org/blind.php?x=cd8xg9). This hypothesis sought to confirm previous evidence (see DeYoung et al., 2007) that Compassion and Openness share a cross-domain correlation similar in magnitude to the within-domain correlations between Compassion and Politeness, and between Openness and Intellect (rs ~ .40). Although this correlation does not itself provide evidence for an intersitial facet below these aspects (rather, it implies a potential interstial domain above them, between Agreeablneess and Openness/Intellect, as we discuss later), it partly inspired our thinking about enlightened compassion.

Study 2 drew on Samples A through G (see Table 1), which were each sufficiently powered to detect an average-sized individual difference correlation (as described in Study 1).

Focal traits (Samples A to G). To improve the robustness of our incremental validity analyses, we modelled Compassion, Openness, and enlightened compassion as latent variables (Westfall & Yarkoni, 2016) using factor scores derived from EFA. Remaining BFAS aspects were measured as sum-scores as described in Study 1.

BFAS (Samples A to G). Within each sample, the set of Compassion items and the set of Openness items (described in Study 1) were each subjected to EFA. A single Compassion factor (for the Compassion EFAs) and Openness factor (for the Openness EFAs) was extracted within each sample using maximum likelihood estimation, and scores for each participant were saved using the regression method.

Enlightened compassion (Samples A to G). Within each sample, the EC Scale items (described in Study 1) were subjected to EFA. A single enlightened compassion factor was extracted within each sample using maximum likelihood estimation, and scores for each participant were saved using the regression method.

Candidate trait correlates. We assessed several conceptually relevant traits from the literature that were derived via a top-down approach. This included traits that tap central features of enlightened compassion (i.e., moral imagination, moral expansiveness, and universalism), as well as traits that seemed particularly compatible with, but not inherent to, enlightened compassion (i.e., low social dominance orientation, low speciesism, ecological worldview, identification with humanity, solidarity with animals, and connectedness to nature).³

Moral imagination (Samples A, B, C). The 32-item Moral Imagination Inventory (MII; Kingsley, 2011) assesses the tendency to think critically and creatively about moral problems to advance communally acceptable outcomes (e.g., “solving an ethical problem requires me to integrate many different points of view, not just my own”). Participants responded to each item from 1 (strongly disagree) to 6 (strongly agree), with each participant’s moral imagination score computed as the sum of their responses.

Moral expansiveness (Samples A, C, E, F, G). The 30-item Moral Expansiveness Scale (MES; Crimston et al., 2016) measures the degree to which one cares about and feels responsible for the welfare of a wide variety of targets. The 30 targets are grouped within ten 3-item subscales: family and friends (e.g., “partner/spouse”), ingroup (e.g., “somebody from your neighbourhood”), revered (e.g., “charity worker”), outgroup (e.g., “member of opposing political party”), stigmatised (e.g., “refugee”), villains (e.g., “terrorist”), high sentience animals (e.g., “cow”), low sentience animals (e.g., “fish”), plants (e.g., “rose bush”), and the environment (e.g., “coral reef”). In Samples F to G, we added an artefacts subscale comprising three targets: “historical building”, “work of art”, and “work of literature”. Participants rated each target as either 0 (outside the moral boundary), 1 (outer circle of moral concern), 2 (fringes of moral concern), or 3 (inner circle of moral concern), with each participant’s moral expansiveness score computed as the sum of their responses.

Universalism (Samples D and G). Similar to moral expansiveness, the 8-item universalism subscale from the Portrait Values Questionnaire 5X (PVQ-5X; Schwartz et al., 2012) assesses the degree to which one endorses universal concern for all humanity and nature as a guiding principle. This includes items tapping social concern (e.g., “he/she wants everyone to be treated justly, even people he/she doesn’t know”) and ecological concern (e.g., “he/she strongly believes that he/she should care for nature”). Participants responded to each item from 1 (not at all like me) to 6 (very much like me), with each participant’s universalism score computed as their mean response.

Social egalitarianism (Samples D and F). The 16-item Social Dominance Orientation Scale (SDO7; Ho et al., 2015) assesses the tendency to endorse intergroup inequality and dominance (e.g., “an ideal society requires some groups to be on top and others to be on the bottom”). Participants responded to each item from 1 (strongly oppose) to 7 (strongly favour). For interpretational ease, we reverse scored each participant’s responses then computed a social egalitarianism score from the mean of their recoded responses.

Species egalitarianism (Samples D and F). The 6-item Speciesism Scale (Caviola et al., 2019) assesses the tendency to elevate the interests of humans over the equivalent interests of animals (e.g., “morally, animals always count for less than humans”). Participants responded to each item from 1 (strongly disagree) to 7 (strongly agree). As with the SDO7 scale, we reverse scored each participant’s responses then computed a species egalitarianism score from the mean of their recoded responses.

Ecological egalitarianism (Samples D and F). The 15-item New Ecological Paradigm Scale (NEP; Dunlap et al., 2000) assesses the tendency to reject an anthropocentric view of nature (e.g., “plants and animals have as much right as humans to exist”). Participants responded to each item from 1 (strongly disagree) to 5 (strongly agree), with each participant’s ecological egalitarianism score computed as their mean response.

Identification with humanity (Sample G). The 9-item Identification with All Humanity Scale (IWAH; McFarland et al., 2012) assesses the tendency to feel a shared identity with all humans (e.g., “how often do you use the word ‘we’ to refer to people all over the world?”). Participants responded to each item from 1 (never/not at all/not at all close/nothing in common) to 5 (always/very much/extremely close/lots in common), with scale anchors varying based on item content. Each participant’s identification with humanity score was computed as the sum of their responses.

Identification with animals (Sample G). The 5-item Solidarity with Animals Scale (Amiot & Bastian, 2017) assesses the tendency to feel a shared identity with animals (e.g., “I feel close to other animals”). Participants responded to each item from 1 (strongly disagree) to 7 (strongly agree), with each participant’s identification with animals score computed as their mean response.

Identification with nature (Sample G). The 14-item Connectedness to Nature Scale (CNS; Mayer & Frantz, 2004) assesses the tendency to feel a shared identity with nature (e.g., “I think of the natural world as a community to which I belong”). Participants responded to each item from 1 (strongly disagree) to 5 (strongly agree), with each participant’s identification with nature score computed as their mean response.

Additional measures. We also measured general self-reported prosocial behaviours in Samples A to G, and traits capturing empathy and awe in Samples E and F (see Supplementary File for results).

Figure 4 depicts bivariate correlations between the EC Scale and the candidate trait correlates. These ranged from r = .33 (ecological egalitarianism) to r = .76 (universalism), indicating strong convergent validity. For comparison, bivariate correlations between each B5 aspect and the candidate trait correlates are also shown. In almost all cases, Compassion and Openness were the largest aspect-level correlates, but EC Scale correlations were consistently the largest.

Semi-partial correlations (see Supplementary File) demonstrated that both Compassion and Openness had significant unique correlations with most candidate trait correlates when controlling for each other, and when controlling for either Politeness or Intellect. This suggests that these candidate traits are located interstitially between Compassion and Openness in the B5 taxonomy, in line with our claim that they each capture central features of enlightened compassion.

Table 6 reports semipartial correlations between enlightened compassion (controlling for Compassion and Openness), Compassion (controlling for enlightened compassion and Openness), and Openness (controlling for enlightened compassion and Compassion) and the candidate trait correlates. For enlightened compassion, virtually all semipartial correlations were significant, and consistently moderate to large in magnitude. Conversely, for Compassion and Openness, semipartial correlations were mostly non-significant and near-zero, with few exceptions that were significant and moderate. This pattern of associations indicates that enlightened compassion has incremental validity over-and-above both Compassion and Openness, suggesting that it is best construed as a (higher fidelity) facet nested below Compassion and Openness within the B5 taxonomy.

The trait correlates assessed in this study were selected because they fall within the conceptual scope of enlightened compassion. Accordingly, we might expect these correlates to themselves cohere empirically as part of a broader family of “sibling constructs” (Lawson & Robins, 2021). To explore this, Table 7 reports intercorrelations among the trait correlates within each sample. These correlations ranged from r = .24 to r = .69, with an average of r = .44. This average correlation is larger than 90% of effects reported in personality psychology (Gignac & Szodorai, 2016), and similar to the correlation between any pair of within-domain B5 aspects (DeYoung et al., 2007). Thus, it seems reasonable to regard these traits as a family of constructs capturing different features of enlightened compassion.

Though they are nested under uncorrelated domains within the B5 taxonomy, we also examined whether Compassion and Openness are themselves correlated. Confirming our pre-registered hypothesis, and replicating a finding originally reported by DeYoung et al. (2007), Table 8 shows that Compassion and Openness were significantly positively correlated in every sample, yielding a sample-weighted average correlation of r = .40. We discuss the implications of this correlation further in the General Discussion.

Study 2 showed that our enlightened compassion measure had good convergent validity based on its moderate to large correlations with conceptually related traits, and incremental validity over-and-above B5 Compassion and B5 Openness. In Study 3, we aimed to assess the criterion validity of our EC Scale through examining its bivariate correlations with a theoretically relevant behaviour—expansive charitable donations. Specifically, we designed a novel task in which participants could make donations to a wide range of charitable causes (e.g., poverty, animal welfare, cultural conservation). We examined the strength with which the EC Scale correlated with this criterion, and, as in Study 2, assessed the incremental validity of the association over-and-above Compassion and Openness. We pre-registered these hypotheses (https://aspredicted.org/blind.php?x=m8u77e) in addition to more specific hypotheses summarised in Table 9.

Study 3 drew on Sample G (see Table 1), the only sample in which our behavioural criterion was measured. This sample was adequately powered to detect an average-sized individual difference correlation (as described in Study 1).

Focal traits (Sample G). Enlightened compassion, Compassion, and Openness were measured as latent variables, as described in Study 2.

Expansive charitable donations (Sample G). We designed a Non-Parochial Charitable Donation Task (NPCD Task) to assess one form of enlightened compassionate behaviour—willingness to donate money to a wide variety of needy human and nonhuman targets. The task comprised 29 charitable causes spanning eight conceptual categories, summarised in Table 9 (NB: the conceptual categories were not shown to participants). We selected these charitable causes on a priori grounds to represent a range of important issues that addressed both traditional and progressive concerns. We piloted a preliminary version of the task in Sample E (see Supplementary File for these results). Verbatim instructions for the final task are provided on the OSF and summarised below.

NPCD Task instructions. Participants were first allocated $1.00. They were then presented with a list of 29 charitable causes and told that one of these causes would be randomly selected by the researchers and revealed at the end of the study. No deception was involved in this process, and this was made clear to participants. Participants were instructed to indicate, for each charitable cause, how much of their $1.00 (either $0.00, $0.25, $0.50, or $1.00) they would be willing to donate to that cause [versus keep for themselves] if it were to be randomly selected at the end of the study. Each charitable cause was accompanied by a brief description (e.g., for Cancer Patients: “Donations to this cause would go to a charity that supports people with cancer”) and presented in a randomised order. We did not name any specific charities in these descriptions as we were interested in participants’ responses to the broader causes. At the end of the study, the randomly selected charitable cause (in our case, Cancer Patients) was revealed. Based on the participant’s response to this charitable cause during the task, the researchers then donated the specified portion of the participant’s $1.00 to a relevant charity (in our case, The American Cancer Society) and paid the participant the remainder. Participants’ donation decisions were anonymous.

NPCD Task scoring. For each participant, we computed an expansive charitable donation score by summing their donation decisions across all charitable causes. Scores could therefore range from 0 to 29 in increments of 0.25. Given the charitable causes spanned eight conceptual categories, we also computed subscores for each participant by summing their donation decisions across the four causes within each category. Subscores could therefore range from 0 to 4 in increments of 0.25. Associations among charitable causes and among charitable cause categories are reported in the Supplementary File.

NPCD Task checks. We embedded a mock cause—Wealthy People (“Donations to this cause would indicate that you’re not paying attention, so please click $0.00 to pass this attention check”)—among the charitable causes as an attention check to catch participants who indiscriminately donated amounts >$0.00 without reading the charitable cause descriptions. Eleven participants were excluded based on this check, leaving N = 217 in the final sample. Before making their donation decisions, participants also completed a comprehension check based on a set of sample responses. They could not proceed to the task until they correctly identified how much of the $1.00 would be donated based on the sample responses, assuming the 2^nd listed charitable cause was later randomly selected. Despite the inclusion of this comprehension check, we anticipated that one source of miscomprehension would be the failure to realise that each of the 29 donation decisions are independent (i.e., that the $1.00 does not have to be split across the charitable causes, but rather resets for each charitable cause). Therefore, after completing the task, participants whose donation decisions summed to exactly $1.00 received a message notifying them of their potential miscomprehension and giving them the option to revise their responses. Only one participant met this criterion, suggesting miscomprehension was not an issue in our sample.

Additional measures (Sample G). As pre-registered (https://aspredicted.org/blind.php?x=m8u77e), we also measured moral expansiveness, universalism, identification with humanity, identification with animals, identification with nature, and the B5 traits (as described in Study 2).

Figure 5 shows the total amount nominally donated to each charitable cause across the sample, with Captive Animals yielding the largest total nominal donation ($56.75) and Sacred Artefacts yielding the smallest ($29.50). Charitable causes from the afflicted category were among the highest earning, while those from the non-natural and marginalised categories were among the lowest earning. Figure 6 displays the distribution of expansive charitable donation scores. The modal score was 0, meaning many participants (36%) nominally donated $0.00 to all charitable causes. Among remaining participants, scores were spread evenly across the scale. Because expansive charitable donation scores were distributed nonnormally, we report Spearman correlations (⁠$ρ$⁠) for all bivariate analyses involving the NPCD Task.

To check the convergent validity of the novel NPCD Task, we examined its associations with several conceptually relevant traits from the literature (i.e., moral expansiveness, universalism, identification with humanity, identification with animals, and identification with nature). Expansive charitable donation behaviour was positively and significantly associated with all these traits, with correlations being approximately average in size (.17 ≤ $ρ$ ≤ .27), and comparable to the average correlation between measures in psychology that rely on different assessment methods (e.g., behavioural vs. self-report; Hemphill, 2003).

As shown in Table 9, the bivariate correlation between the EC Scale and expansive charitable donation was significant, and small-to-medium (⁠$ρ$ = .17, p = .013), as were its correlations with donations for each individual charitable cause category (.14 ≤ $ρ$s ≤ .23). This confirmed our pre-registered hypotheses and supported the criterion validity of the EC Scale. Surprisingly, none of our pre-registered predictions were confirmed for Compassion, which had nonsignificant bivariate correlations with overall expansive charitable donation behaviour as well as each individual charitable cause category. Some pre-registered predictions were confirmed for Openness, including a significant bivariate correlation with overall expansive charitable donation behaviour (⁠$ρ$ = .15, p = .023).

Table 9 also reports semipartial correlations between expansive charitable donation behaviour and enlightened compassion (when controlling for Compassion and Openness), Compassion (when controlling for Openness and enlightened compassion), and Openness (when controlling for Compassion and enlightened compassion). As in Study 2, enlightened compassion displayed incremental validity over-and-above these B5 aspects. The semipartial correlation between enlightened compassion and expansive charitable donation behaviour was small-to-medium and significant for both the overall NPCD task score (sr = .21, p = .002) and for each charitable cause category (srs ≥ .14) except the “non-human” category. In contrast, the semipartial correlations that Compassion and Openness shared with expansive charitable donation behaviour were all non-significant and approaching zero.

In summary, enlightened compassion showed evidence of criterion validity based on its association with expansive charitable donation behaviour. Enlightened compassion also showed incremental validity over-and-above Compassion and Openness, which suggests (based on the bandwidth-fidelity trade-off) that it is an interstitial facet nested below these B5 aspects.

Enlightened compassion describes the tendency to show regard for others in a manner that is open-minded rather than rigid or parochial. We construe this flexible and inclusive form of compassion as a kind of moral exceptionality (Lawn et al., 2022) that may allow people to integrate competing interests and extend goodwill beyond conventional boundaries when confronted with moral problems. Our goals in this paper were to articulate a thorough conceptualisation of enlightened compassion—drawing together the siloed literatures focussed on moral imagination, moral expansiveness, and other constructs—and to develop a valid and reliable measure of this trait using a bottom-up mode of enquiry. To this end, we produced an Enlightened Compassion Scale (EC Scale), grounded within the most conceptually relevant traits from the B5 taxonomy, and provided an initial evaluation of its structural, content, convergent, and criterion validity.

We found that enlightened compassion converges strongly and consistently with the Compassion aspect of B5 Agreeableness and the Openness aspect of B5 Openness/Intellect, but not with other B5 aspects (Study 1). We also found that enlightened compassion correlated with conceptually relevant traits (e.g., moral imagination and moral expansiveness; Study 2) and with a theoretically pertinent behavioural criterion (i.e., expansive charitable donation; Study 3). Moreover, enlightened compassion explained incremental variance over-and-above Compassion and Openness when predicting these traits and behaviours (Studies 2-3). In line with the bandwidth-fidelity trade-off (Cronbach & Gleser, 1957), this suggests that enlightened compassion can be located interstitially below the Compassion aspect of Agreeableness and the Openness aspect of Openness/Intellect within the B5 taxonomy.

Our bottom-up approach to developing the EC Scale complements previous top-down derivations of enlightened compassionate traits (e.g., moral imagination; moral expansiveness), and locates them within the organising framework of the B5. Because these existing derivations had been developed in a top-down fashion within somewhat separate literatures, their conceptual and empirical links with each other—and with more basic personality traits—had not been documented. The shared vernacular and organising framework for trait constructs provided by the B5 taxonomy (Bainbridge et al., 2022; Stanek & Ones, 2018) allowed us to consolidate and extend (rather than duplicate) existing work on enlightened compassionate traits.

Across multiple samples in Study 2, we found that the EC Scale derived from Compassion and Openness converged with two distinct conceptualisations of enlightened compassion—moral imagination (Kingsley, 2011) and moral expansiveness (Crimston et al., 2016)—along with a broader suite of conceptually related traits. All of these traits, including the EC Scale, were moderately to highly intercorrelated, suggesting they can be regarded as a common family of constructs. Within the B5 taxonomy, these traits converged most strongly with Compassion and Openness, and more weakly with remaining B5 aspects. Importantly, these correlations with Compassion and Openness approached zero after controlling for the EC Scale, suggesting that enlightened compassion may capture the common core of these constructs within the space described by the B5. This knowledge will enable future research focussed on any of these traits to draw on a much larger body of theoretical and empirical work, including the vast B5 literature pertaining to Agreeableness (especially its Compassion aspect) and Openness/Intellect (especially its Openness aspect).

Our research also complements other recent efforts to explore the interstitial space between Agreeableness and Openness/Intellect. Specifically, Christensen et al. (2019) speculated that open-mindedness—the willingness to embrace novel and diverse ideas, lifestyles, and experiences—may be an interstitial facet of these domains. This proposal was partly inspired by the robust finding that both Agreeableness and Openness/Intellect are associated with social open-mindedness, or inclusiveness toward marginalised groups (Connelly et al., 2014; Crawford & Brandt, 2019; Sibley & Duckitt, 2008). In line with these ideas, our findings show that traits related to social open-mindedness (e.g., social universalism and social egalitarianism)—including those pertaining to non-human targets (e.g., ecological universalism and species/ecological egalitarianism)—indeed converge with aspects of both Agreeableness and Openness/Intellect, and with our EC Scale. Tendencies related to social open-mindedness may thus form part of a single interstitial facet lying below the Compassion and Openness aspects of Agreeableness and Openness/Intellect—the facet we have labelled enlightened compassion.

Relatedly, in Study 2 we tested and confirmed the hypothesis that Compassion and Openness are themselves moderately positively correlated. Despite falling under different B5 domains, the correlation between Compassion and Openness is similar in magnitude to the within-domain correlations between Compassion and Politeness, and between Openness and Intellect (rs ~ .40; DeYoung et al., 2007). Therefore, just as the latter two correlations imply Agreeableness and Openness/Intellect, the apparent shared variance of Compassion and Openness implies the existence of a potential domain-level trait lying above these aspects within the B5 taxonomy (i.e., in the interstitial space between Agreeableness and Openness/Intellect). This rotational variant of Agreeableness and Openness/Intellect would likely include enlightened compassionate content—but being a domain-level trait, we expect that it would span a broader range of sensitivity and reflectiveness related tendencies. We encourage future work that explores these potential implications of the robust correlation between Compassion and Openness.

One important caveat to acknowledge when drawing these conclusions is that we can identify no definitive test of our claim that enlightened compassion is a facet-level construct within the B5. For instance, an intuitively appealing approach would be to compare three structural equation models—one comprising Openness, Compassion, and EC as three correlated factors (i.e., representing EC as an aspect); another regressing Openness and Compassion on EC (i.e., representing EC as a domain); and a third regressing EC on both Openness and Compassion (i.e., representing EC as a facet). Unfortunately, these three models are statistically interchangeable, and offer no degrees of freedom to change their (equivalent) model fit. We therefore argue that enlightened compassion is a facet-level construct on conceptual grounds, based on our definition of this trait as a subset of compassionate tendencies that are open-minded. We furthermore defend our claim that the EC Scale assesses an empirical B5 facet primarily based on its incremental validity above and beyond the B5 aspects.

Although we obtained promising evidence for the structural, content, convergent, and criterion validity of our enlightened compassion measure, we caution that our conclusions may not generalise beyond the samples, methods, and procedures reported here (see Simons et al., 2017). For example, we derived the EC Scale using items from a single B5 measure—the Big Five Aspects Scales. Attempts to reproduce our scale using alternative indices of Compassion and Openness may not necessarily yield similar support for the validity of our scale. Furthermore, the support we obtained for metric invariance of the EC Scale across samples must be weighed against the fact that four of our samples comprised students attending the same Australian university, and three comprised MTurk workers residing in the US. We therefore cannot be certain that the internal structure of the EC Scale will generalise beyond such samples. Additionally, despite the comprehensiveness of our assessments of content and convergent validity of the EC Scale, we examined criterion validity in relation to just one behaviour—expansive charitable donation. It remains to be seen whether the EC Scale will predict a wider range of theoretically relevant behaviours, including behaviours outside the lab.

Finally, our participant samples were all drawn from Western, Educated, Industrialized, Rich, and Democratic (WEIRD; Henrich et al., 2010) populations. Among other distinctions, WEIRD cultures are typically more individualistic (vs. collectivistic) than non-WEIRD cultures. Pertinently, Openness/Intellect has been found to be the least culturally generalisable B5 domain (Rolland, 2002). This is particularly evident in collectivist cultures, where the defining features of Openness/Intellect are likely less salient (i.e., less adaptively important) and less coherent (i.e., less functionally interdependent; Schwaba, 2021). Given that enlightened compassion is a facet of Openness/Intellect, it would be unwise to assume that individual differences can be coherently measured along this dimension, or that such individual differences function in the same way, within non-WEIRD cultures.

Beyond assessing the replicability and generalisability of our findings, one pertinent avenue for future research concerns the potential for enlightened compassion to be cultivated. Though volitional trait change is relatively understudied, evidence to date suggests that people can indeed volitionally shift their Agreeableness and Openness/Intellect levels to a small degree over time (Hudson et al., 2020). Future research might therefore investigate whether people can increase their levels of enlightened compassion, which interventions might best support this process, and what individual and social benefits might flow from such efforts. One potential intervention we find intriguing is loving-kindness meditation, a contemplative practice that involves progressively directing empathic feelings toward a widening circle of beneficiaries (Hofmann et al., 2011). Such practices seem like promising candidate tools for developing enlightened compassion, and could be examined as such in future extensions of our research.

Progress on many pressing moral problems may require a meeting of the “head” and “heart”—the exercise of open-mindedness to expand our goodwill beyond familiar frontiers. We call this concept enlightened compassion, which the present studies suggest is a measurable facet of personality located below the Compassion aspect of B5 Agreeableness and the Openness aspect of B5 Openness/Intellect. We argue that enlightened compassion may capture the common core of a cluster of socially open-minded traits—such as moral imagination and moral expansiveness—in terms of their shared location within the B5 taxonomy. We encourage future work to further evaluate the structure, content, and behavioural correlates of enlightened compassion, explore the cross-cultural generalisability of the present findings, and assess the viability of cultivating enlightened compassionate tendencies.

Contributed to conception and design: E.C.R.L, S.M.L, K.Z., L.D.S.

Contributed to acquisition of data: E.C.R.L, L.D.S.

Contributed to analysis and interpretation of data: E.C.R.L, A.P.C, L.D.S.

Drafted and/or revised the article: E.C.R.L, S.M.L, K.Z., A.P.C, L.D.S.

Approved the submitted version for publication: E.C.R.L, S.M.L, K.Z., A.P.C, L.D.S.

The project was supported by a grant from the John Templeton Foundation through The Beacon Project, Wake Forest University. E.C.R.L. was supported by an Australian Government Research Training Program Scholarship.

None.

All measures and materials, preregistrations, participant data, analysis scripts, and summaries of supplementary analyses can be found on this paper’s project page on the OSF: https://osf.​io/​qze53/?view_​only=​192697​d7bf2​2470ab​7eb9fc1​8ec6​5f93

We thank Tif Duong, Victoria Leach, and Vivian Hao-Wen Yang for diligently coding relevant manuscripts; and Timothy Bainbridge for several helpful insights.

Our conceptualisation of enlightened compassion evolved as we sharpened our knowledge of trait hierarchies. Some of our pre-registered analyses were written in an open-ended manner to allow for this. Specifically, we began the project with an open mind about whether enlightened compassion should be construed as a) a higher-order domain derived from the shared variance of Compassion and Openness, or b) a lower-order facet that blends parts of Compassion with parts of Openness. Our pre-registrations thus refer to both options interchangeably, stating that enlightened compassion would be measurable either as a) a sum score or latent factor extracted from BFAS Compassion and Openness items [which maps onto construal a)] or b) using our EC Scale [which maps onto construal b)]. We later reasoned that it is more sensible to conceptualise a) and b) as measures of two distinct but nested constructs. Option a) is a measure of a broader construct than enlightened compassion, and was not the focus of this paper. Option b) is a measure of enlightened compassion, a narrower subset of the broader trait, and was the focus of this paper. We arrived at this conceptual distinction after writing the pre-registrations.

Additional sample specific deviations from our pre-registrations are summarised below:

We pre-registered two hypotheses relevant to Study 1 not mentioned in the manuscript. First, we predicted that Compassion and Openness would each correlate with the Enthusiasm aspect of Extraversion. Second, we predicted that all three aspects would be reducible to one dimension (noting that if this were not the case, we would explore the viability of reducing Compassion and Openness to one dimension, and the viability of reducing Compassion and Enthusiasm to one dimension). After collecting Sample A, we neglected our focus on Enthusiasm for two reasons. First, the correlation between Compassion and Enthusiasm has already been studied by DeYoung et al. (2013), who conceptualise the intersection of these aspects as “affiliation”. Second, as noted in the pre-registration, the correlation between Openness and Enthusiasm is typically small (we indeed find this in our samples; see Supplementary File), suggesting that Compassion, Openness, and Enthusiasm are not reducible to one dimension, but rather two: Compassion/Openness, and Compassion/Enthusiasm.

We pre-registered three hypotheses relevant to Study 2 not mentioned in the manuscript. Namely, we predicted that intellectual humility, social mindfulness, and the dark triad (reversed) would correlate positively with enlightened compassion. Because these constructs are less conceptually central to enlightened compassion, we relegate these analyses to the Supplementary File. The pre-registration also mentioned that a novel “Affirmative Action at Work” task was administered for exploratory purposes. The development of this task is ongoing as part of another project, hence it is not analysed herein.

We pre-registered two sets of hypotheses relevant to Study 3 not mentioned in the manuscript. Namely, we predicted that a) Compassion would have a significant semipartial correlation with expansive charitable donation (and select charitable cause categories) when controlling for Openness, and b) Openness would have a significant semipartial correlation with expansive charitable donation (and select charitable cause categories) when controlling for Compassion. Because Compassion and Openness had negligible and nonsignificant bivariate correlations with expansive charitable donation, we report these semipartial analyses in the Supplementary File. In addition to these hypotheses, we pre-registered exploratory analyses involving a selection of items from the Eugene Springfield Community Sample (Goldberg, 1999) that DeYoung et al. (2007) previously identified as correlates of both Compassion and Openness. These analyses are part of another ongoing project, hence the results are not reported herein.

Here are a number of characteristics that may or may not describe you. For example, do you agree that you are someone who would ‘often think about how my compassion could be put to good use’? Please choose a number for each of the below statements to indicate the extent to which you agree or disagree that it describes you.

1 = Strongly disagree;

2 = Slightly disagree;

3 = Neither agree nor disagree;

4 = Slightly agree;

5 = Strongly agree.

1. Often think about how my compassion could be put to good use.

2. Empathise with people who are very different to me.

3. Feel no emotional connection to artistic works [R].

4. Imagine ways I could make a difference in the world.

5. Care about members of society who are often forgotten.

6. Am saddened by the way humans are treating the environment.

7. Channel my creativity towards good causes.

8. Often think about people who are suffering in other countries.

9. Feel upset when historical or cultural artefacts are damaged.

10. Can’t be bothered coming up with creative ways to help others [R].

11. Consider the perspectives of minority groups when thinking about social issues.

12. Feel emotionally connected to animals.

Items with an “R” require reverse scoring.