Social Class and Income Inequality is Associated with Morality: Empirical Evidence from 67 Countries

A fundamental characteristic of modern societies is economic inequality, where deprived individuals experience chronic economic scarcity. While such experiences have been shown to produce detrimental outcomes in regards to human judgment and decision-making, the consequences of such scarcity for our morality remain debated. We conduct one of the most comprehensive tests of the relationship between experiences of relative chronic economic scarcity and various measures linked to morality. In a preregistered study, we analyse data from a large, cross-national survey (N = 46,450 across 67 countries) allowing us to address important limitations related to measurement validity and external validity in past research. Our ndings demonstrate that experiences of relative chronic economic scarcity, as indexed by (1) low subjective socioeconomic status at the individual level, and (2) income inequality at the national macro level, predict higher levels of moral identity, higher morality-as-cooperation, a larger moral circle, and importantly; more prosocial behaviour. The results appear robust to several advanced control analyses. Finally, exploratory analyses indicate that observed income inequality at the national level does not signicantly moderate the predicted effect of subjective socioeconomic status. Our ndings have vital implications for understanding human morality under chronic resource scarcity.

regional economic inequality might be an important moderator 36,37 . Vitally, these studies have focused on U.S. data, thereby limiting the generalizability of the obtained ndings.
As a direct consequence of the mixed ndings and the contextual variance, in this pre-registered analysis, we conduct one of the most comprehensive tests of the relationship between experiences of relative economic scarcity and measures linked to moral psychology and behaviour. We rely on a large, largely representative, cross-national survey (N = 46,450 across 67 countries; see Fig. 1 for a country and region overview of the sample), thus providing a unique opportunity to achieve three important main objectives. Speci cally, this research design (1) maximizes external validity, providing higher generalizability than prior studies which typically rely on data from a single country; (2) enables a statistically well-powered test of the relationship between experiences of economic scarcity, as indexed by socio-economic status at the individual level and GINI coe cients at the macro level, and morality across cultural and national contexts; and (3) allows us to examine whether the level of economic inequality at the national level moderates the relationship between SES and morality. This is an important extension of past research, as prior studies have been characterized by low external validity from underpowered laboratory experiments [38][39][40][41][42][43] , replicability concerns in the form of null-ndings, results that are in the opposite direction as those reported in the original research 34 , and potential contextual sensitivity by variations in time, culture, or location [44][45][46][47] , which could explain part of the inconsistencies in the literature 19,23 .
To probe the internal validity and contextual sensitivity of our results, we rely on advanced statistical methods in the form of multilevel modelling and cross-validations using machine learning algorithms, in order to identify robust individual and country-level associations between our measures of relative economic scarcity and morality. In doing so, we contribute with a robust crosscultural and externally valid extension of previous research 28, 29,34,35,44,48,49 on how experiences of chronic resource scarcity might affect moral psychology and behaviour. A contribution which, we argue, has important implications for our understanding of how scarcity affects human behaviour in general.

Results
We begin by analysing the relationship between relative scarcity and morality using multi-level modelling, both on the individual level (SES) and on the group level (GINI), and whether there might be any country-level differences in such a relationship. Table 1 reports full models for our four dependent variables, with standardized b-coe cients allowing for a direct comparison of the effect sizes and model t statistics. Sample sizes for each model are reported, given that these varied slightly due to some participants being able to refrain from replying to certain measures in the survey. As a robustness check, we also ran our models with imputed data where the missing values were estimated using non-parametric random-forest estimations. The results remained robust to this imputation of data and are depicted in the Supplementary Table S3.  Figure 3. However, it is important to note that economic inequality does not meaningfully moderate the effect of low SES on any of our indicators of morality. Thus, we cannot claim that individuals living with low SES in more economically unequal countries exhibit more moral behaviour and a higher moral character than their more a uent counterparts.
Instead, our results indicate that exposure to relative resource scarcity on the individual level (low SES) might partially elicit the same effects as resource scarcity on the national level (GINI).
Variance partition coe cients 50 for the models indicated that the majority of the variance could be attributed to the individual level (Moral Identity = 87.4%, Morality-as-Cooperation = 90.6%, Moral Circle = 95.4%, Prosocial Behaviour = 92.3%) with variance at the country-level ranging from 4.6% at the lowest (Moral Circle) to 12.6% at the highest (Moral Identity). We assume that some of this lack of country-level variance is due to common-methods variance in ating the estimates at the individual level 51,52 . Additionally, this lack of country-level variance indicates that the relationship between relative economic scarcity and moral character and behaviour might be fairly robust across different cultural contexts. The robustness of these cross-cultural relationships also aligns with recent ndings using the same data to investigate differences in donation behaviour, in-group favouritism, and age across the 67 countries 53 .
When exploring the relationship between the moral measures and the individual-level measure of relative resource scarcity (subjective SES) on the country and region-level, the directions of the effects generally remained stable. Nevertheless, notable country-level differences in effect sizes were observed (Supplementary Tables S1, S2 Tables S4, S5, S6, S7).
Overall, and despite these cross-national differences, our results indicate robust associations between individual-level (SES) and country-level (GINI) chronic economic scarcity and validated indicators of moral character and behaviour. Hence, contrary to current theoretical paradigms concerning resource scarcity and moral behaviour 6,15,17,19,20 and our pre-registered prediction (https://aspredicted.org/727eq.pdf), we nd evidence for the notion that relative chronic economic deprivation is not associated with a "depletion" of moral character or that it makes individuals less prosocial, but rather results in more moral outcomes.

Discussion
To date, research investigating how chronic economic scarcity might affect human morality has been mixed and at times contradictory. Based on research showing that scarcity impedes cognitive functioning 10,12 , directs attention towards short-term grati cation 18,54-56 , and increases economic risk-taking 57-62 together with a selection of recent ndings highlighting a possible causal link between individual experiences of resource scarcity and unethical behaviour 19,20,35,63,64 , our pre-registration postulated that individuals with higher SES, living in more economically equal societies, would elicit greater moral character and behaviour than their lower-class counterparts. However, we found the exact opposite. Aggregating the data across 67 countries, socioeconomic status was negatively associated with moral identity (Fig. 1a), morality-as-cooperation (Fig. 1b) as well as prosocial behaviour in the form of donation intentions towards national and international charities (Fig. 1d), while being positively associated with the size of one's moral circle (Fig. 1c). As such, individuals with less nancial resources not only seem more inclined to perceive themselves as moral individuals (i.e., moral identity), but also seek to project such behaviour towards their peers and in-group (i.e., morality-as-cooperation, moral circle, and prosocial behaviour). Importantly, we show that this relationship, at least to some extent, holds on the national level such that individuals living in countries with high economic inequality (GINI), and thus a greater degree of relative economic scarcity, report a stronger moral identity (Fig. 2a) but also a greater moral circle (Fig. 2c). These associations are highly robust in cross-validations with supervised machine learning algorithms (10-folds, 200 repetitions; Supplementary Table S8).
Yet, the contrast between our pre-registered predictions and the empirical results raises the question of how our results be interpreted? Concerning prosocial behaviour, a selection of ndings shows that chronic scarcity (i.e., low social class) can increase prosocial behaviour 8,28,31,44 . As previous research (ref. 27,28,48,65 ) has argued that individuals living with lower social class have an increased contextual social orientation (vs. individualistic), the link between prosocial behaviour and relative resource scarcity appears meaningful in that resource deprived individuals, while having less available resources themselves, may still exhibit more prosocial behaviours towards others, as such prosocial actions could aid in generating better future life-outcomes (i.e., by reciprocation).
The relationship we nd between chronic economic scarcity (SES and GINI, respectively) and moral identity suggests that perceiving oneself as a moral individual (internalized moral identity) and exhibiting moral behaviour towards others (symbolized moral identity) is more important for individuals living with less available resources. Lower class individuals could act more moral, because they are more sensitive to their social environment as their life tends to be in uenced by forces which they cannot necessarily control (e.g., relying on government policies, help from charity organizations, and decisions of job managers) 27,66 ; see also 67 . Acting as a moral individual might not be as important if you perceive the world from a more individualistic perspective, which people of higher social class tend to do 27 .
The link we nd between relative resource scarcity and morality-as-cooperation indicates that the moral valence of cooperation principles receives higher importance in populations where resources are scarce. At a general level, managing external constraints and depending on other individuals requires some, albeit differing, degrees of cooperation in order to gain fruitful outcomes 68 .
Consequently, morality is considered a central foundation of cooperative behaviour 69,70 and one of the main functions of morality is to promote fruitful cooperation [68][69][70][71][72][73] . The concept of morality-as-cooperation used in the analysis conceptualizes that certain forms of cooperative behaviour, such as helping a family or group member, reciprocating, and sharing resources are considered morally good in all cultures 69,74 . Our results indicate that individuals living with chronic economic scarcity are more inclined to value whether someone helped a member of their family or worked to unite a community when they decide on whether something is right or wrong 69,74 . Thus, our ndings on morality-as-cooperation suggest that resource deprived individuals are particularly prone to consider their external environment when making moral decisions, likely because they know that they depend on such cooperative connections to obtain more favourable life outcomes.
Regarding the link between chronic economic scarcity and the size of one's moral circle, we nd that individuals with lower social class tend to have a smaller moral circle than individuals with a greater access to economic resources (Fig. 2c), while individuals living in more economically unequal societies tend to have a larger moral circle (Fig. 3c). From an evolutionary perspective, the observed positive relationship between SES and the size of one's moral circle might be explained by models of parochial altruism 75, 76 . In other words, resource deprived individuals may act increasingly altruistic towards their in-group in order to facilitate successful cooperation, but display less moral behaviour towards out-groups in order to successfully protect in-group members.
While the results of the present study originate from a large, representative, cross-cultural sample and appear highly robust, the magnitude of the associations reported herein is relatively small and the general explanatory power of our models is modest by conventional standards. Nevertheless, psychological and cognitive phenomena related to human morality are expected to be in uenced by a plethora of different factors [77][78][79][80] , which means that small effect sizes are to be expected as long as these phenomena are not examined in controlled lab conditions, but rather in real-world settings [81][82][83][84][85] . Recently, research has started to adopt the approach of examining human psychology using more externally valid data sources as the increased computational power of personal computers has allowed scholars to assess very large datasets, with advanced statistical tools, such as machine-learning and neural network models 86,87 . For example, recent scholarly work relying on such methods has documented small but highly robust associations between physical topography and human personality traits 88 . Therefore, while the effect sizes from our analysis are small, this does not imply that they lack practical relevance [88][89][90] . Effect sizes that are considered small by arbitrary standards can have a large impact when evaluated over time 77,91 or at scale [92][93][94] . This is particularly true for human psychology, where effects can accumulate over time, thus underscoring the fact that while an effect might be small when measured at a single point in time, it can have large ultimate consequences 77 . Also, psychological processes, especially regarding morality, are characterized by "di cult-to-in uence" dependent variables, which emphasize that robust small effects are theoretically important 95 . For instance, our ndings demonstrate that an increase of one standard deviation in SES is associated with a decrease of 2.84% in donation value towards national and international charities, which might seem trivial when considered at the individual level, but can have large consequences for societal outcomes at the population level 77,80,96,97 .
A nal note should be that the dataset used in the present investigation was collected during the COVID-19 pandemic. While the general idea regarding the pandemic seems to be that "COVID-19 does not discriminate," recent studies have shown that vulnerable individuals, such as those with discriminated ethnicities or those who are economically disadvantaged, have higher mortality rates than their less vulnerable counterparts 98,99 . The results of the current study not only show a general link between relative resource scarcity and human morality, but also suggest that this association is present when people with the least resources experience an extraordinary increase in the level of risk and exposure to threat. As research has argued that hostile environments motivate people with less available resources to engage in prosocial behaviour 28,100 , our ndings may therefore be stronger than similar investigations conducted during pre-or post-pandemic times or data collected in the absence of other public crises (e.g., nancial recessions, droughts, terrorism attacks, and wars).
In conclusion, the present research demonstrates that social class and income inequality predicts multiple dimensions of human morality. These ndings underline the complex impact that social class and inequality have on the way individuals morally respond and act. We consequently urge future research to further investigate how moral character and behaviour might be affected by experiences of chronic resource scarcity and how such knowledge might be utilized to help the ones with the least, the most.

Methods
The present study was pre-registered on AsPredicted before the data was accessed (https://aspredicted.org/727eq.pdf, August 3 rd 2020). While we generally adhered to the pre-registered analysis plan, there are some deviations, which should be noted.
Speci cally, for our main analysis we employed multilevel correlation analysis, nested OLS regressions, multi-level modelling (Linear Mixed Effects models) and cross-validations instead of standard Pearson's correlations and simple OLS regression, thus addressing the same questions as pre-registered but with more sophisticated and robust methods.
The data was obtained from the International Collaboration on Social & Moral Psychology of COVID-19 (ICSMP) 101 . This project was a large-scale collaboration between more than 200 researchers from 67 different countries with a goal to create an online survey to measure psychological factors underlying the attitudes and behavioural intentions related to COVID-19. The project received ethical approval from the institutional review board at the University of Kent (ID 202015872211976468) and informed consent was obtained from all participants prior to their voluntary participation in the study. The dataset contains self-reported demographics and social and moral psychology data from 46.450 individuals from 67 countries and 5 different regions of the world. Each national team responsible of collecting data in their country translated the English survey into their nations' language using the standard forward-backward translation method. Every participating country was asked to collect data from at least 500 participants, nationally representative with respect to gender and age. The data was collected between April-May 2020 and was administered using an online survey. Every participating individual answered questions regarding demographics and self-reported public health behaviours as well as a series of psychological measures. Scale order was randomized for every participant. The dataset was cleaned by the lead methodologists from the ICSMP project for the initial publication using the dataset 101  In addition to data from the ICSMP, we obtained the most recent GINI Indexes from the World Bank 102 , for every country included in the study, with few exceptions. Taiwan GINI data was obtained from Statista 103 , Cuban GINI data was obtained from Reuters 104 and New Zealand and Singapore GINI data was obtained from Knoema 105,106 . Region names was obtained from the World Bank Development Indicators 107 .

Variables
Moral Identity was measured using a scale of 10-items 96 such as "It would make feel good to be a person who has these characteristics", which would be answered based on a description of a person who has the characteristics: "caring, compassionate, fair, friendly, generous, helpful, hardworking, honest, kind". Each item was measured using a 10-point slider with three labels: 0 = "Strongly disagree", 5 = "Neither agree nor disagree", 10 = "Strongly agree". Items 3 and 4 were reverse scored. Results were aggregated into a single-scale (Cronbach's a = .729), instead of two subscales (internalization and symbolization), as in the original publication developing the scale 96 . Hence, our aggregated measure of moral identity indicates 1) how important moral identity is to one's self-de nition (internalization) and to what degree individual's express this moral identity (symbolization), but does not distinguish between the two respectively.
Morality-as-Cooperation was measured using a 7-item scale adapted from ref. 108 . Each item represented one question out of three from each of the seven "relevance items" from the Morality-as-Cooperation questionnaire 108 . The questions chosen from the original scale were the ones with the highest predictive validity 101 . Individuals were initially asked the following: "When you decide whether something is right or wrong, to what extent are following considerations relevant to your thinking?". Here, the "family" item was labelled "Whether or not someone helped a member of their family". The "group" item was labelled "Whether or not someone worked to unite a community". The "reciprocity" item was labelled "Whether or not someone showed courage in the face of adversity". The "deference" item was labelled "Whether or not someone deferred to those in authority". The "fairness" item was labelled "Whether or not someone kept the best part for themselves". The "property" item was labelled "Whether or not someone kept something that didn't belong to them." Each item was measured using a 10-point slider with three labels: 0 = "Strongly disagree", 5 = "Neither agree nor disagree", 10 = "Strongly agree". All 7-items were aggregated into our single measure of Moralityas-Cooperation (Cronbach's a = .732). In the original publication developing the scale, test-retest correlations for the full scale was shown to range from .79 to .89 108 .
Moral Circle was measured using a single-item scale with 16 levels 109 , asking participants to indicate the extent of their moral circle, where moral circle means "the circle of people or other entities for which you are concerned about right and wrong done toward them." The scale ranges from 1 = "all of your immediate family" to 16 = "all things in existence". Test-retest reliability of a similar scale has previously been shown to be . 61. 110 Prosocial behaviour was measured using a hypothetical choice task with three items. This task asked individuals how much (in percent), if given a median income, they would be willing to 1) keep to themselves, 2) donate to a national charity and 3) donate to an international charity. We formed our measure of prosocial behaviour, by aggregating the second and third item, across individuals.
Subjective socioeconomic status (SES) was measured using the single-item MacArthur ladder scale 110 . This scale uses a picture of an 11-step ladder and asks participants where, in their country, they would stand if the top indicated the people who are the best off -those who have the most money, the most education and the most respected jobs, while the bottom are the people who are the worst off -those who have the least money, least education, and the least respected jobs or no jobs. Participants indicated their standing in their respective society from 0 (absolute bottom) to 10 (absolute top).
GINI Index, as measured by the World Bank, is based on primary household survey data obtained from statistical agencies and World Bank country departments 102 . It measures the amount of income inequality, where 0 = total equality and 100 = total inequality. For more info on speci c measurement and methodology, see PovcalNet from the World Bank (iresearch.worldbank.org/PovcalNet/index.htm).

Correlations
Due to the nested structure of our data, the correlations between the variables; subjective socioeconomic status (SES) 110 , moral identity 96 , morality-as-cooperation 108 , moral circle 109 prosocial behaviour were calculated using multilevel Pearson's correlations with country as the random intercept. We also calculated grouped correlation coe cients for every country and region in the dataset, in order to identify country-level differences of interest (Supplementary Tables S1 and S2). Correlations between our dependent variables and the independent variable "GINI Index" was calculated as single-level Pearson correlations without the multilevel nesting, as the GINI Index would not be different per individual measure, as it constitutes a country-level measure.

Multilevel Models
We performed two speci c forms of multi-level modelling.
Firstly, we performed linear mixed effects modelling 111 , where we regressed our dependent variables with our two main independent variables, covariates and country as the random intercept. Two-tailed signi cance testing (a = .05) was applied for all analyses. For ease of reporting and interpretation, we standardized parameters to report b-coe cients and 95% con denceintervals of our analysis.  Supplementary Figures S1, S2, S3, S4.

Cross Validations
As a robustness check to access the predictive power of our models, we applied 10-fold cross validation, with 200 repetitions on all multi-level models. Cross-validation is a form of supervised machine-learning which splits the dataset into K number of independent datasets (in our case 10) and then uses every dataset in turn as the validation set, where the other K-1 datasets then act as the calibration sets. Each fold of the data leads to a different estimate of the Root Mean Square Error (RMSE) of the model and therefore the process is repeated multiple times (in our case, 200) to get reliable estimates. While cross-validation can be used as a procedure in model-selection, in this article we used the procedure to validate the robustness of our models 112 . That is, our cross validations provide con dence in the reported ndings, by illustrating that the included model results are in fact the ones with the lowest RMSE. The full results of all cross validations can be found in the Supplementary Table S8.

Data availability
The data that support the ndings of this study are available on the OSF page of the original article (ref. 101     World map of ICSMP survey. The map highlights the countries and regions who collected data for the ICSMP Survey. Sample sizes are scaled to colour. Grey areas identify areas where it was not possible to obtain samples.

Figure 2
Country and region-level relationship between moral identity, morality-as-cooperation, size of moral circle, prosocial behaviour and socioeconomic status. a, Association between socioeconomic status (SES) and degree of individual-level moral identity. b, Association between SES degree of morality-as-cooperation. c, Association between SES and the size of one's moral circle, where size indicates the circle of people or other entities for which one is concerned whether right or wrong is done towards them. d, Association between SES and prosocial behaviour, where such behaviour is measured as the amount of money (out of a median income) one would be willing to donate to a national or international charity.

Figure 3
Country and region-level relationships between moral identity, morality-as-cooperation, size of moral circle, prosocial behaviour and level of income inequality. a, Association between level of income inequality (GINI) and how much moral identity individuals exhibit and report to have. b, Association between level of income inequality (GINI) and morality-as-cooperation. c, Association between level of income inequality (GINI) and the size of one's moral circle, where size indicates the circle of people or other entities for which one is concerned whether right or wrong is done towards them. d, Association between level of income inequality (GINI) and prosocial behaviour, where such behaviour is measured as the amount of money (out of a median income) one would be willing to donate to a national or international charity.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.