Introduction

The roots of flourishing

For decades, indeed centuries, scholars have been intrigued by how a person’s childhood affects their outcomes later in life. Answering this question ideally involves, (a) longitudinal studies that, (b) begin tracking participants during childhood, including (c) with attention to numerous contextual factors (e.g., details of their family), and (d) follow participants into adulthood, with (e) relevant outcomes of interest (e.g., variables relating to wellbeing). For some outcomes, a wealth of studies meet these criteria. On the whole though, these tend to focus on ‘negative’ (i.e., undesirable) outcomes, such as mental health problems. Such is the depth of longitudinal research on environmental factors associated with their onset during childhood/adolescence, for example, that enough meta-analyses now exist to allow an umbrella review of them1. This focus on negative phenomena has characterised academia for most of the past century, but is gradually changing however, with increasing attention to more ‘positive’ (i.e., desirable) outcomes. This interest is not new per se; humanistic scholars like Maslow advocated such an approach over 80 years ago2. Not until the emergence of positive psychology in the late 1990s though did this work begin to receive more widespread hearing, with Martin Seligman using his ascension to the presidency of the American Psychological Association to encourage greater attention to positive phenomena as serious and legitimate topics of scientific enquiry.

As a result, outcomes like happiness and life satisfaction – which we briefly discuss here as an exemplar of the wider literature on flourishing, with the latter being a more comprehensive term that includes states like happiness as facets – have now received extensive attention. This includes the kind of longitudinal work mentioned above that allows scholars to explore the childhood roots of such outcomes. One study3 for example used data from the UK National Child Development Study to assess a cohort of 4,400 children born in 1958 who had been repeatedly surveyed for 50 years, including being asked about life satisfaction (“how satisfied or dissatisfied you are with the way life has turned out so far”) at four points in adulthood (ages 33, 42, 46, and 50). Demographic and socio-economic factors in early childhood notably had a relatively negligible effect, predicting just 1.2% of the variance in life satisfaction. More individual characteristics by contrast – especially childhood behavioural-emotional problems and social maladjustment – were “powerful predictors” of adult satisfaction. Similar findings were obtained in an analysis of 5,124 young participants (aged 11–15) from the British Household Panel Survey4. The researchers began by assessing the determinants of youth happiness – on a scale of 1 (completely happy) to 7 (completely unhappy) – and found contextual factors like family structure had a significant impact. Living with both natural parents, for example, was linked to greater happiness, both relative to a step family (for boys and girls) and living with a single parent (for boys). Most relevantly here, a subset of participants (1,825) went on to participate in the adult panel (i.e., 18 +), which features a question on life satisfaction. While satisfaction was influenced by demographic and socio-economic conditions in childhood, their impact was much smaller than a “youthful personality trait for happiness” (i.e., an “individual effect” in the analysis of youth happiness capturing individual heterogeneity). Youth happiness predicted from socio-economic factors had a coefficient of 0.15 and 0.14 for adult satisfaction for women and men respectively, whereas the coefficient for the youth happiness trait was 0.27 (for both men and women). Such work is not limited to happiness moreover; research has also indicated that perceived childhood experiences of, for example, parental warmth subsequently predict a range of flourishing outcomes, including more positive relationships and other forms of social wellbeing, including social contribution and integration5.

Through such research we are gaining a better understanding of topics like happiness and other domains of flourishing. Other desiderata have received relatively little attention however, in general, and certainly in any comparable longitudinal way to the happiness studies cited above. These neglected topics include “low arousal positive states” (LAPS) such as inner peace (IP). An influential theoretical context for understanding these is Russell’s circumplex model of affective states6, which construes them through the intersection of two parameters: valence (experientially pleasant and approach-inducing, versus unpleasant and withdrawal-inducing), and arousal (high versus low, or active versus passive). Affective states are thus understood as being generated – physiologically, mentally, experientially, etc. – by their interaction7. Juxtaposing the parameters, Russell created a two-dimensional state space with four quadrants: (1) low arousal and negative valence (e.g., depression); (2) high arousal and negative valence (e.g., anxiety); (3) low arousal and positive valence (e.g., calmness); and (4) high arousal and positive valence (e.g., elation). Most relevantly here, when it comes to research on positively-valenced states, the majority of attention has focused on high arousal forms, like enjoyment. Among the most influential constructs in this arena, for example, is subjective wellbeing (SWB)8, comprising a cognitive component (usually understood and measured using constructs of life satisfaction or evaluation) and an affective component (viewed through the prism of positive and negative affect). Most research on the latter though has concentrated on high arousal forms of positive affect. For instance, since 2005 the Gallup World Poll (GWP) has included items pertaining to both components of SWB, yet until very recently, positive affect had just been assessed through high arousal notions like enjoyment. Significantly though, since 2020 these items have been augmented by items pertaining to LAPS, as we explore below. First though, our next section considers why LAPS may have historically been overlooked.

Neglecting LAPS

In attempting to understand the tendency in wellbeing research to focus on high rather than low arousal states, two plausible interlinked explanations are, (a) the Western-centric nature of psychology, and (b) greater importance being placed on HAPS than LAPS in Western cultures, and hence psychology. Let’s briefly consider each in turn. It is increasingly recognized that psychology has historically been Western-centric, with most work published in the top journals conducted by and on people in societies that are relatively “WEIRD” (Western, Educated, Industrialised, Rich, and Democratic)9. While one ought not to simplistically classify societies in a binary way as WEIRD versus non-WEIRD10 – since each element of the acronym is a spectrum upon which countries may be variously situated – most of the world is certainly not as WEIRD as places like the USA, where much of the scholarship in prominent journals takes place. Critics have consequently questioned the extent to which such research is generalizable and universally valid. Although some scholars might retort that people are relatively similar across cultures and share a common human nature, and hence findings from Western contexts can be extrapolated to other locales, many academics would likely agree that the conditions of life can vary dramatically among cultures. It is therefore problematic to draw conclusions about human experience based only on comparatively WEIRD contexts. The issue is not only about participants moreover, but scholars themselves, who are likewise shaped by their context, which will thus influence all aspects of their work, from their choice of topics and methodologies to subsequent analyses and interpretation of data.

Most relevantly here, the Western-centric bias of academia has been implicated by Tsai and colleagues in the relative neglect of LAPS, who suggest the preference for HAPS is a Western-centric concern, whereas by contrast Eastern cultures place greater value on LAPS. Tsai described such preferences as “ideal affect” – “the affective states that people strive for or ideally want to feel”11 – and has observed these across an extensive series of studies, mostly involving college students in America and China12,13,14,15,16,17,18, with similar patterns observed by others19,20,21,22 . That said, new items on LAPS in the GWP – mentioned above and discussed further below – indicate that such emotions may be more universally valued and experienced than these East–West generalisations imply23. Even so, notwithstanding such findings, one can still argue that LAPS have historically received greater valorization and attention in Eastern cultures, for which various explanations have been mooted.

One prominent interpretation invokes another distinction often noted vis-à-vis East–West differences, namely between individualism and collectivism. Generalisations along these lines have been aired for centuries, but especially came to prominence in the 19th Century as Western scholars tended to “other” the East, often in disparaging ways, as notably charted by Edward Said in his critical text Orientalism24. In the modern era though, this binary has been harnessed in a relatively neutral way to differentiate cultures25 and their people’s self-construal26. The distinction has since been explored in hundreds of studies, with numerous meta-analyses, not only of the binary per se, but specific facets, such as its link to subjective wellbeing27. Most relevantly here, it has been suggested that Eastern cultures tend to appraise HAPS as relatively self-aggrandizing and hence disruptive of social harmony, whereas LAPS are more conducive to such harmony19,28. Scholars have also pointed towards other cultural trends, such as the rich history of contemplative practices in Eastern cultures, that may also contribute to their greater valorisation of LAPS29. Whatever the explanation, there has been a relative inattention to LAPS in academia, with research tending to focus on HAPS30. But the situation may be changing amidst a broader concern with redressing the Western-centricity of psychology: a review of positive psychology interventions found that although 78.2% were in Western countries, there was “a strong and steady increase in publications from non-Western countries since 2012,” indicating an encouraging “trend towards globalization” of happiness research31. These dynamics also mean LAPS are receiving more attention, as exemplified by the Global Wellbeing Initiative.

Global wellbeing initiative and the Gallup World Poll

Since 2005 the GWP has collected data annually on wellbeing (and many aspects of life) worldwide. Conceptually though it has mostly still been subject to the Western-centrism that characterises wellbeing research more broadly, with its main metrics being Cantril’s “ladder” item on life evaluation32 and several pertaining to HAPS. To redress these issues, the Global Wellbeing Initiative (GWI), a partnership between Gallup and the Japan-based Wellbeing for Planet Earth foundation, was launched in 2019, focusing on developing items related to Eastern cultures (given the Japanese location of the foundation). The first iteration was included in the 2020 GWP, as analysed in a chapter for the 2022 World Happiness Report33. The module has since been through two substantive iterations34, and by the 2022 GWP was centred exclusively on balance/harmony and LAPS, collectively described as “harmonic principles of wellbeing”35. Most notably, each year it has included an item involving various permutations on the theme of IP: “Did you feel at peace most of the day yesterday, or not?” (2020); “In general, how often do you feel you are at peace with your thoughts and feelings?” (2021); and “In general, how often can you find inner peace during difficult times?” (2022, 2023, 2024). Analysis of these items has revealed numerous notable patterns36.

Of particular relevance here are the contextual factors associated with peace, as revealed by a regression analysis. In that regard, there were intriguing differences between the three peace items, leading to an interpretation that while the latter two are more directly about “inner peace,” the 2020 item is more ambiguous, straddling inner and “outer peace” (i.e., the peacefulness of one’s societal context). Consider for example that poverty only had a significant impact on the 2020 item, both in terms of lacking money for food (B = -0.37) and shelter (B = -0.49). One might suggest that people lacking money for these necessities are indeed likely to experience a lack of outer peace, living in situations that are challenging, which perhaps lies behind these variables having a significant impact on being “at peace most of the day.” By contrast, and perhaps counter to expectations, poverty appeared to have no discernible impact on the two items that tapped more directly into IP. Similarly, the 2020 item was more strongly associated with life factors such as finding it “very difficult to get by” on present income (B = -1.81, versus -0.30 for 2021 and -0.37 for 2022), and having other people to count on (B = 0.58, versus 0.07 for 2021 and 0.15 for 2022). Conversely, the more “inner-oriented” factor of negative emotions had a stronger association with the 2021 (B = -0.44) and 2022/2023/2024 (B = -0.45) items than 2020 (B = -0.18). There were also interesting demographic patterns, but most saliently these had an overall larger impact on the 2020 item.

While these kinds of contextual analyses are interesting and relevant, here we are especially interested in the childhood predictors of peace. The GWP data however is not conducive to that kind of assessment, since it neither has respondents who are children nor asks adults about childhood. Indeed, although academia has paid increasing attention to the childhood predictors of certain flourishing outcomes, such as happiness, this has so far not extended into the domain of LAPS. Metrics pertaining to LAPS have mostly been absent from the kind of rigorous longitudinal work that allows assessment of the childhood predictors of positive outcomes, reflecting their general omission from scholarship on flourishing more generally. That said, there has still been some relevant scholarship. Research indicates that that childhood trauma may lead to a diminished sense that attainment of IP is possible37, while similarly, maladaptive stress coping mechanisms, along with disrupted patterns of homeostasis, contribute to neuropsychiatric disorders that are inimical to IP38. From the other direction, a synthesis of relevant studies led scholars to conclude “Positive early interpersonal experience lays the groundwork for a more peaceful individual life,” and that a “cycle of positive early experience” fosters secure relationships that “promote cognitive and social skills, in turn leading to more peaceful relationships within families and beyond”39. Similarly, IP can be positioned within a set of spiritual factors and experiences that partially mediate the effect of Adverse Childhood Experiences on quality of life in adulthood40, which suggests relationships between IP and a variety of social factors may be reciprocal over the life course. Such research, while relatively sparse, is sufficient to suggest that conditions and experiences in childhood may well have a bearing on IP later in life. More work is needed though – especially rigorous longitudinal designs – to explore this neglected area of inquiry more fully.

To that end, the present paper reports on an assessment of IP included in the Global Flourishing Study (GFS), an intended five-year panel study investigating the predictors of human flourishing across over 200,000 participants from 22 geographically and culturally diverse countries. A distinguishing feature of this study is its longitudinal nature. While there are various laudable endeavours researching flourishing in a global context – such as the GWP – these are mostly cross-sectional, so cannot provide much insight into causal dynamics. Hence the value of the GFS, which involves a longitudinal assessment of a comprehensive battery of items relating to all aspects of flourishing. Furthermore, even the first-wave demographic intake form included retrospective enquires into childhood experiences, allowing for a synthetic longitudinal study of sorts based on the first year of data alone, which is the focus of the present paper. Our dependent variable of interest is IP in adulthood, adapted from the 2021 GWI item: “In general, how often do you feel you are at peace with your thoughts and feelings?" (always, often, rarely, never). The analysis here is guided by three research questions: (1) how do different aspects of a child’s recalled upbringing predict IP in adulthood; (2) do these associations vary by country; and (3) are the observed relationships robust to potential unmeasured confounding, as assessed by E-values? Specifically, we look at 13 different childhood predictors: (1) age (year of birth); (2) gender; (3) marital status / family structure; (4) age 12 religious service attendance (with age 12 chosen so as (i) to allow somewhat better recall, versus say age 6, and (ii) so that there was sufficient time still in the home for the childhood experiences to matter); (5) religious affiliation at age 12; (6) relationship with mother; (7) relationship with father; (8) feeling like an outsider growing up; (9) abuse; (10) self-rated health growing up; (11) immigration status; (12) subjective financial status of family growing up; and (13) race/ethnicity (when available). We have three main general hypotheses (adapted from a generic set of predictions designed to apply across analyses of the GFS as a whole, spanning its myriad outcomes): (1) among the 13 childhood predictors, certain ones will show meaningful associations with IP in adulthood; (2) the strength of associations between the predictors and IP in adulthood will vary by country, reflecting the influence of diverse sociocultural, economic, and health contexts that characterize each nation; and (3) the observed associations between the predictors and IP in adulthood will be robust against potential unmeasured confounding (as assessed through E-values, in some cases suggesting that the observed associations would require strong confounding effects by unmeasured variables to explain away, thus enhancing the credibility of our findings). The study design was pre-registered with the Open Science Framework on November 18th, 2023 (https://osf.io/ztm7r).

Methods

The description of the methods below has been adapted from VanderWeele and colleagues41. Further methodological detail is available elsewhere42,43,44,45,46,47,48,49,50.

Data

The GFS is a study of 202,898 participants (in this first year) from 22 geographically and culturally diverse countries, with nationally representative sampling within each country, concerning the distribution of determinants of wellbeing. Wave 1 of the data included the following countries and territories: Argentina, Australia, Brazil, Hong Kong [S.A.R of China, with mainland China also included from 2024 onwards], Egypt, Germany, India, Indonesia, Israel, Japan, Kenya, Mexico, Nigeria, Philippines, Poland, South Africa, Spain, Sweden, Tanzania, Turkey, United Kingdom, and United States. (Note: Data from Hong Kong (S.A.R. of China) is available in the first wave of data collection. Data from mainland China were not included in the first data release due to fieldwork delays. The first wave of fieldwork in mainland China began in February 2024, and a second wave occured in November–December 2024. All wave 1 and 2 data from mainland China will be part of the second dataset release in March 2025.) The countries were selected to (a) maximize coverage of the world’s population, (b) ensure geographic, cultural, and religious diversity, and (c) prioritize feasibility and existing data collection infrastructure. Data collection was carried out by Gallup Inc. Data for Wave 1 were collected principally during 2023, with some countries beginning data collection in 2022 and exact dates varying by country49. Four additional waves of panel data on the participants will be collected annually from 2024–2027. The precise sampling design to ensure nationally representative samples varied by country and further details are available [49]. Survey items included aspects of wellbeing such as happiness, health, meaning, character, relationships, and financial stability51, along with other demographic, social, economic, political, religious, personality, childhood, community, health, and wellbeing variables. The datasets generated and/or analysed during the current study are available in the Open Science Framework repository upon submission of pre-registration (https://www.cos.io/gfs-access-data). During the translation process, Gallup adhered to TRAPD model (translation, review, adjudication, pretesting, and documentation) for cross-cultural survey research43.

Measures

Childhood antecedents

Relationship with mother during childhood was assessed with the question: “Please think about your relationship with your mother when you were growing up. In general, would you say that relationship was very good, somewhat good, somewhat bad, or very bad?” Responses were dichotomized to very/somewhat good versus very/somewhat bad. An analogous variable was used for relationship with father. “Does not apply” was treated as a dichotomous control variable for respondents who did not have a mother or father due to death or absence. Parental marital status during childhood was assessed with responses of married, divorced, never married, and one or both had died. Financial status was measured with: “Which one of these phrases comes closest to your own feelings about your family’s household income when you were growing up, such as when you were around 12 years old?” Responses were lived comfortably, got by, found it difficult, and found it very difficult. Abuse was assessed with yes/no responses to “Were you ever physically or sexually abused when you were growing up?” Participants were separately asked: “When you were growing up, did you feel like an outsider in your family?” Childhood health was assessed by: “In general, how was your health when you were growing up? Was it excellent, very good, good, fair, or poor?” Immigration status was assessed with: “Were you born in this country, or not?” Religious attendance during childhood was assessed with: “How often did you attend religious services or worship at a temple, mosque, shrine, church, or other religious building when you were around 12 years old?” with responses of at least once/week, one-to-three times/month, less than once/month, or never. Gender was assessed as male, female, or other. Continuous age (year of birth) was classified as 18–24, 25–29, 30–39, 40–49, 50–59, 60–69, 70–79, and 80 or older. Childhood religious tradition/affiliation was had response categories of Christianity, Islam, Hinduism, Buddhism, Judaism, Sikhism, Baha’i, Jainism, Shinto, Taoism, Confucianism, Primal/Animist/Folk religion, Spiritism, African-Derived, some other religion, or no religion/atheist/agnostic; precise response categories varied by country50. Racial/ethnic identity was assessed in some, but not all, countries, with response categories varying by country. For additional details on the assessments see the COS GFS codebook52 or other sources42.

Outcome variable

IP is assessed with one question: "In general, how often do you feel you are at peace with your thoughts and feelings?" The response categories are: always, often, rarely, never. In our analyses, we dichotomized IP as always/often1 vs rarely/never [0].

Analysis

Descriptive statistics for the observed sample, weighted to be nationally representative within each country, were estimated for each childhood demographic category. A weighted modified Poisson regression model with complex survey adjusted standard errors was fit within each country of IP on all of the aforementioned childhood predictor variables simultaneously. In the primary analyses, random effects meta-analyses of the regression coefficients53,54, along with confidence intervals, estimate proportions of effects across countries with effect sizes (risk-ratios) larger than 1.1 and smaller than 0.9, and \({I}^{2}\) for evidence concerning variation within a given demographic category across countries55. Forest plots of estimates are available in the Online Supplement. Religious affiliation/tradition and race/ethnicity were used within country as control variables, when available, but these coefficients themselves were not included in the meta-analyses since categories/responses varied by country. All meta-analyses were conducted in R56 using the metafor package57. Within each country, a global test of association of each childhood predictor variable group with outcome was conducted, and a pooled p-value58 across countries reported concerning evidence for association within any country. Bonferroni corrected p-value thresholds are provided based on the number of childhood demographic variables59,60. For each childhood predictor, we calculated E-values to evaluate the sensitivity of results to unmeasured confounding. An E-value is the minimum strength of the association an unmeasured confounder must have with both the outcome and the predictor, above and beyond all measured covariates, for an unmeasured confounder to explain away an association61. As a supplementary analysis, population weighted meta-analyses of the regression coefficients were estimated. All analyses were pre-registered with COS prior to data access, with only slight subsequent modification in the regression analyses due to multicollinearity (https://doi.org/https://doi.org/10.17605/OSF.IO/ZTM7R); all code to reproduce analyses are openly available in an online repository46.

Missing data

Missing data on all variables was imputed using multivariate imputation by chained equations, and five imputed datasets were used62,63. To account for variation in the assessment of certain variables across countries (e.g., religious affiliation/tradition and race/ethnicity), the imputation process was conducted separately in each country. This within-country imputation approach ensured that the imputation models accurately reflected country-specific contexts and assessment methods. Sampling weights were included in the imputation model to account for missingness being related to probability of inclusion.

Accounting for complex sampling design

The GFS used different sampling schemes across countries based on availability of existing panels and recruitment needs49. All analyses accounted for the complex survey design components by including weights, primary sampling units, and strata. Additional methodological detail, including accounting for the complex sampling design, is provided elsewhere45.

Results

Descriptive statistics

Table 1 provides the distribution of descriptive statistics (weighted counts and proportions). Participant ages ranged the entire adult lifespan (18–80 +). The sex (“gender”) distribution was nearly balanced with 51% female, 48% male, along with a small representation from other gender identities (0.3%). Most participants reported either having a somewhat good or very good relationship with either parent while growing up. Regarding attending religious services growing up, 41% of participants report attending at least once a week while 23% of participants report never attending. Counts and proportions for demographic characteristics weighted to be representative of each country’s population are reported in supplemental Tables S1a-S22a.

Table 1 Nationally representative descriptive statistics of the observed sample.
Full size table

Childhood experiences predicting inner peace

The meta-analytic estimates of how childhood experiences predict IP are reported in Table 2. These results show a significant association between 10 of the 13 childhood candidate predictors and responses about IP (with immigration status being non-significant, and with race and religious affiliation categories varying by country and so no meta-analysis was conducted for these, though results are available by country in the Online Supplement). Childhood experiences associated with IP in adulthood included having a good relationship with parents, having a sense of a comfortable financial status growing up, being in good or excellent health growing up, and more frequent attendance at religious services. These factors were, on average across countries, associated with a higher frequency of IP as an adult. However, these positive associations were not universal across all countries. In India, for example, the effect of having a very/somewhat good relationship with one’s mother was slightly negative to null (RR = 0.89, 95% CI [0.79,1.00]). All country-specific results and variations are given in the Online Supplement, and we comment on the variation across countries further in the Discussion below. Although most of these effects were positive on average, the effect was often not statistically significant within the country-specific analyses. The forest plots provided in our Online Supplement provide additional evidence for the heterogeneity of these effects across countries (see Figures S1-S27).

Table 2 Random effects meta-analysis of regressing inner peace on childhood predictors.
Full size table

Sensitivity of effects to unmeasured confounding

Sensitivity to unmeasured confounding was assessed using E-values, suggesting that some of the observed associations were moderately robust to unmeasured confounding (Table 3). To explain away the estimate for good/somewhat good relationship with mother, for instance, an unmeasured confounder associated with both higher IP and a good/somewhat good relationship with mother with risk ratios of 1.31 each, above and beyond the measured covariates, could suffice, but weaker joint confounder associations could not. To shift the confidence interval to include the null, an unmeasured confounder associated with both higher IP and a good/somewhat good relationship with mother with risk ratios of 1.23 each, above and beyond the measured covariates, could suffice, but weaker joint confounder associations could not. Further, country-specific sensitivity analyses are reported in the Online Supplement (see Tables S1c-S23c).

Table 3 Sensitivity of meta-analyzed childhood predictors to unmeasured confounding.
Full size table

Country comparisons

Finally, an important aspect of the GFS is the way it reveals variation among countries. To illustrate this, consider the most impactful factor on average: self-rated health growing up, as assessed on a five-point scale: poor; fair; good; very good; and excellent. As shown in Table 2 above, relative to the middle category of “good,” RRs range from 0.93 for “poor” and 0.94 for “fair” to “1.04” for very good and “1.07” for excellent. However, within this general overall pattern was considerable country-level variation, as illustrated in Table 4, which shows the respective RRs for the four health categories (relative to the middle category of “good”) for the 22 countries (with details for each country provided in the Supplementary Tables), together with the respective E-values and 95% CIs.

Table 4 Country-level variation in the impact of childhood health on inner peace (relative to “good”).
Full size table

Discussion

This exploratory cross-sectional analysis has shed unique light on the childhood predictors of IP. As indicated above, this outcome has received relatively little attention per se, with low arousal emotions in general being understudied and underappreciated in research on flourishing and its various aspects. It is thus unsurprising that there has been barely any research into its childhood predictors, hence the value of our study. In summary, all three of our main hypotheses were supported, often strikingly so. As a reminder, our first was that among the 13 childhood predictors, certain ones will show meaningful associations with IP in adulthood. Indeed, every predictor – with the sole but notable exception of immigration status – had a significant association with IP when meta-analyzed over the 22 countries. Second, the strength of associations between the predictors and IP in adulthood will vary by country, reflecting the influence of diverse sociocultural, economic, and health contexts that characterize each nation. Third, some of the observed associations between the predictors and an individual’s IP in adulthood will be robust against potential unmeasured confounding (as assessed through E-values). Here we touch in turn on the predictors, beginning with the one with the strongest impact, namely self-rated health growing up. We discuss this factor in some detail as a way of illustrating the nature and nuances of the data. We then consider the other factors more briefly, referencing and extrapolating the points made in relation to health.

First though, we should make an important point about comparing results across countries. Methodologically this can be problematic for various reasons, some of which we note at the end of the Discussion in a paragraph on limitations, but perhaps above all due to the vagaries of language. Translation is a notoriously difficult endeavour, with a wealth of literature showing it can be very hard to find exact equivalents of a given term in another language64. As such, in comparing data from different countries, it is always possible that subtle culturally-influenced linguistic nuances are playing a role, influencing the data. That said, Gallup has considerable experience and expertise in ensuring translations are as accurate as possible, adhering to a TRAPD model (translation, review, adjudication, pretesting and documentation)65. The process of developing translations for the GFS was consequently very detailed and thorough, involving numerous experts in the relevant languages43. Ideally then, the rendering of inner peace – or indeed any concept analysed here – signifies the same phenomenological state in one language as do its equivalents in other languages. It is still possible though that these terms were not precisely equivalent, and perhaps – even if only very subtly – were actually assessing slightly different outcomes. This is not a possibility we can investigate in the present paper, and would require in depth qualitative research to explore, which we hope this paper will inspire. But it is important to bear in mind as we seek to understand apparent differences between nations. For this reason, the main analytic purpose of the GFS was not to make cross-cultural comparisons per se, but to carry out data analysis separately within countries as 22 closely-related cohort studies, and then to conduct meta-analyses across countries. Such meta-analytic approaches do not presume that the items are interpreted identically across countries, but merely are relatively closely related to one another (just as a meta-analysis of closely related interventions that may differ in specific administration, dose, mode, etc.). Thus, while we would argue that the data can still be used for cross-country comparisons, these do need to be interpreted with some caution.

With that caveat in mind, overall, the most impactful factor on average was self-rated health growing up. Relative to the middle category of “good,” the Risk Ratios (RRs) ranged from 0.93 for poor and 0.94 for fair to 1.04 for very good and 1.07 for excellent. An RR can be interpreted as the relative percentage in each category, which in the present paper is calculated in relation to the proportion of people reporting IP. In that respect, although IP was assessed on a four-point scale (never, rarely, often, or always at peace), in our analysis and interpretation we aggregate this – for simplicity and to aid interpretation – into two binary categories, whereby people either have IP (endorsing either “often” or “always” on the peace item) or do not have it (endorsing either “rarely” or “never”). Thus, taking the RR of 1.07 for excellent health as an example, this means that, compared to people who reported that they “only” had good health growing up, the proportion of people with excellent health who have IP is 1.07 times greater than those who do not have IP. Put another way, there is a 7% increase in having IP for those who reported excellent health in childhood relative to those who reported good health (conditional on all other variables in the model). Although this effect size is modest, at the population level it is quite meaningful, and moreover, in some countries the effect size was considerably higher.

One can also examine the robustness of these associations through their E-values61, which pertains to our third main hypothesis. An E-value measures the strength that an “unmeasured confounder” – a variable not included in the analyses – would need to be to “explain away” the observed relationship. In the case of excellent health, the E-value for the estimate was 1.36. This would mean an unmeasured confounder would need to be both, (a) related to peace with a RR of 1.36 (meaning that being one unit higher on the confounder is associated with a 36% increase in having peace), and (b) simultaneously related to childhood health with a RR of 1.36 (where a one unit increase on the confounder is associated with a 36% increase in being in the excellent health category over the good category). It is this simultaneous association of the unmeasured confounder with our outcome (peace) and our predictor (health) that makes the unknown variable a “confounder.” The usefulness of the E-value is that it serves as a benchmark for thinking – in light of existing knowledge and theory – about whether a confounder could exist that does indeed fulfil this criterion of simultaneous association with responses to the health and IP items. Essentially, the closer the E-value to 1, the more likely it is that such a confounder could indeed exist. Conversely, the higher the E-value, the less likely that it does. In the present case, an E-value of 1.36 is judged as high: especially given the observed associations, it is difficult to envisage an unmeasured variable that could plausibly have an RR of 1.36 with both IP and childhood health. That is, this RR would need to both be, (a) much higher than the one observed between excellent childhood health and IP (i.e., 1.07), and (b) apply separately to both childhood health and IP. So, in terms of our third hypothesis, we have some evidence that the observed RR between IP and childhood health is robust to potential unmeasured confounding, so is a “real” effect (i.e., rather than a statistical artefact produced by us not including enough relevant variables in our analysis).

This finding that childhood health is associated with IP in adulthood is unique: we could find no previous study that has explored such a connection, so simply observing it here is a notable addition to the literature. It is worth emphasizing though that we did not actually assess people’s health in childhood itself, but rather their retrospective recollections about their childhood. Crucially, there are indications that people sometimes change their ratings of childhood health over time; an analysis66 found nearly one half of their sample revised this during a 10-year observation period. Older adults who were relatively advantaged (e.g., with socioeconomic resources and better memory) were less likely to revise it, whereas those with multiple childhood health problems were more likely to (either positively or negatively); for example, development of psychological disorders was associated with more negative revised ratings. As such, we must be somewhat cautious in interpreting our data, given we did not measure health in childhood per se, and recall bias might be present. However, for recall bias to completely explain the observed associations of the childhood predictors with adult IP, the effect of adult IP on the retrospective assessments of the childhood predictors would have to be at least as strong as the observed associations themselves67.

Moreover, numerous longitudinal studies have actually measured health in childhood then traced its impact on later outcomes, with a substantial literature showing it does have a substantive effect on myriad aspects of adult life. Given that context, it is reasonable to think – based on our data – that IP is indeed one of the variables affected by it. Much of this existing longitudinal work focuses either on physical health or socio-economic status, with poor childhood health having a long-term detrimental impact on these outcomes68,69,70,71. However, there is some work with more direct relevance to IP, with various studies connecting poor childhood health to mental health issues specifically in later life, particularly depression. One analysis of a nationally representative sample of late midlife adults in the US (N = 3,572), for example, found childhood disability was significantly associated with higher levels of depressive symptoms, suggesting that such people may accumulate more physical impairment over the life course, thus suffering worse mental health in late midlife72. A comparable assessment found people with childhood disability exhibited more depressive symptoms at age 50 compared to those who did not, although there was no difference in the progression of depressive symptoms over time between the two groups, suggesting an initial inequality which was then maintained over the life course73.

Turning now to our second hypothesis, we did indeed observe national differences in the effects of the factors. Many of the studies cited above were in a US context, which is characteristic of the psychological literature as a whole, as elucidated in the Introduction. A particular strength of our research is thus its multinational reach, covering 22 diverse countries. As per our second hypothesis, there was considerable variation in the impact of childhood health, as illustrated in Table 4 in the Results section, which showed the respective RRs for the four health categories (relative to the middle category of “good”) for the 22 countries (with details for each country provided in the Supplementary Tables), together with the respective E-values and 95% CIs. There one can see many notable country-level nuances. For a start, compared to the overall RR range of 0.14 (spanning 0.93 for poor to 1.07 for excellent health), some nations had a much larger range – as much as 0.95 in Turkey – implying that childhood health is a much more significant factor there compared to other countries. More research is needed to explore why this regional variation exists, but it will almost certainly involve considerations such as the provision of healthcare in the various countries. There were also intriguing patterns that are harder to explain and certainly merit further investigation, especially the fact that, in some countries, the RRs seemed “out of order.” One would expect, based on the overall RRs, that relative to people with “good” childhood health, people with worse health (“poor” or “fair”) would have lower IP (RR < 1.00), while people with better health (“very good” or “excellent”) would have higher (RR > 1.00). Seven countries did indeed conform to this linear escalating pattern (Australia, Brazil, Hong Kong, Sweden, Turkey, UK and USA). In the remaining countries though, this pattern was subverted in various ways, where compared to those with good childhood health, some groups with worse health (either poor and/or fair) had higher IP (RR > 1.00), while conversely others with better health (very good and/or excellent) had lower IP (RR < 1.00). Consider Nigeria, for instance, where people with poor childhood health had an RR of 1.19: not only does poor childhood health not detract from IP in adulthood there, but the data imply it actively helps. Moreover, the E-value for this particular observation is 1.67 (while the E-value for the 95% CI is 1.38), suggesting this finding is very robust to potential confounding. We cannot know from our data why this effect is observed, i.e., what is special about Nigeria that childhood poor health seems to actually facilitate IP in adulthood. One could speculate that, at least in some countries, experiencing poor health in childhood either encourages or compels people to develop a certain resilience or other psychological qualities that may give rise to IP. But this still begs the question, namely, what is different about these countries that this effect is observed, and why are similar effects not found elsewhere. Certainly, this is something that demands more in-depth study.

Let us now consider the other variables. While we do not have the space to discuss these in comparable depth to childhood health, we can nevertheless highlight some notable patterns that merit further study. To reiterate, every factor – apart from immigration status – had a significant effect on IP in adulthood. To begin with, family dynamics are important, including having a good relationship with one’s mother (RR = 1.06) and father (1.03), as is having parents who were married compared to either being divorced (0.97), single or never married (0.95), or one or both parents having died during childhood (0.95). The financial situation of the family also matters: relative to people whose families “got by,” those who “lived comfortably” fared better (1.03), while people did worse whose families found it either “difficult” (0.98) or “very difficult” (0.96). These findings accord with extensive scholarship on the importance of these factors for wellbeing, both in childhood itself and moreover in later life. With respect to relationships with parents, for instance, a considerable literature on attachment styles shows the positive impact of “secure” bonds – generally regarded as the optimal type of attachment – on mental health later in life74. So too with marriage: overall, research has consistently shown this to be beneficial for children relative to other possibilities such as divorce/separation, both during childhood itself (Størksen et al., 2005) and over the life course75, though one notes that in some situations – such as conflicted or abusive marriages – divorce may indeed be better option all round76. And again, with the financial aspect, research consistently finds that economic security in childhood is associated with better long term mental health prospects77. Until now, however, these factors had not been linked to IP in adulthood, and thus our work now extends the literature to encompass this outcome.

Perhaps of even greater value in this study is the way it highlights national variation, showing that the impact of these factors differs considerably based on the location. The effect of having a good relationship with one’s mother ranged from (RR =) 0.89 in India to 1.26 in Indonesia, while the impact of having a good relationship with one’s father ranged from 0.93 in Nigeria to 1.16 in Turkey. There was likewise considerable variation pertaining to parental marital status, where compared to having parents who were married, the effect of parents: being divorced ranged from 0.81 in Nigeria to 1.36 in Turkey; being single or never married ranged from 0.79 in Egypt to 1.15 in Australia; and one or both parents having died ranged from 0.74 in South Africa to 1.16 in Mexico and the Philippines. Finally, there was also variation in relation to finances, albeit less so than the other familial dynamics, implying this factor is somewhat less susceptible to cultural influence. Compared to those whose families “got by” financially, the effect of one’s family having “lived comfortably” ranged from 0.95 in Poland to 1.18 in Turkey, while for those who found it “difficult” ranged from 0.91 in Japan to 1.05 in the US, and for those who found it “very difficult” ranged from 0.76 in Turkey to 1.14 in Spain. These regional differences are again fascinating and deserve further study, and will require in-depth enquiry into cultural dynamics to help explain them. Consider for example the impact of having parents who were divorced, with a 0.55 RR differential between Nigeria, where such divorce has a markedly negative impact on the likelihood of experiencing IP in adulthood, and Turkey, where it means one is more likely to have IP compared to people whose parents were married. Accounting for such findings will require detailed exploration into the traditions, values, and practices pertaining to both marriage and divorce in the respective countries. It may be relevant, for instance, that Nigeria has large numbers of both Christians (45.9% of the population) and Muslims (53.5%), whereas Turkey is overwhelmingly Muslim (99%)78. In that respect, it is possible that in some contexts Islam is relatively more accommodating of divorce – albeit still generally not looking favourably upon it79 – than Christianity, and hence overall may be less destabilising to the future equanimity of Muslims than Christians.

Another important variable is religious attendance at age 12. Not only was this associated with adult IP, but moreover an increasing amount depending on the frequency of attendance. Compared to those who never attended, the impact of attending rose from (RR =) 1.03 for those attending less than once a month, to 1.05 for those attending 1–3 times a month, to 1.06 for those attending at least weekly. This aligns with an extensive body of work on the positive impact of childhood religious attendance on subsequent physical and mental health80 and also with scholarship that explores the centrality of peace to many religious traditions22. Again though, our study seems to be the first to link childhood religious service attendance to IP specifically. Also again, however, perhaps even more striking is the regional variation, where the impact of attending less than once a month ranged from 0.84 in Nigeria to 1.17 in Turkey, of attending 1–3 times a month ranged from 0.93 in South Africa to 1.32 in Turkey, and attending weekly ranged from 0.92 in Nigeria to 1.33 in Turkey. Once more we see a striking comparison between – as above – Nigeria and Turkey in particular, where childhood religious attendance in the former seems potentially detrimental to adult IP, while in the latter it strongly supports this later outcome. As with all factors here, the impact of attendance thus may not be uniformly positive, and depends on local socio-cultural factors. In that respect, in-depth work in places like Nigeria will help us better understand why this country in particular seems to buck the overall trend. One wonders for example about the relevance, as noted above, of Nigeria having two main religions – which moreover can often be in tension and even conflict with one another in the country81 – while Turkey is nearly all Muslim, hence lacks comparable internal divisions. It does therefore seem plausible that religious involvement in Nigeria could bring a level of adversity or friction that is relatively absent in Turkey, thus accounting for the significant disparities in the impact of that involvement on adult IP.

The final set of factors that seem impactful for IP are adverse experiences, namely experiencing abuse and feeling like an outsider in one’s family growing up, both with an RR of 0.94. These of course connect with a now vast literature on the long-term detrimental impact of Adverse Childhood Experiences, which are documented to negatively impact a panoply of outcomes later in life, ranging from substance use82 and food insecurity83 to depression84 and even frailty in older adults85. To this literature we can now also add that such adversities also lower the likelihood of experiencing IP as an adult. Again though, the regional variation is striking, where the impact of abuse ranges from 0.80 in Poland to 1.01 in Mexico, while the impact of being an outsider ranges from 0.83 in Brazil to 1.07 in Turkey. Here it seems that, in certain countries (like Turkey), experiencing being an outsider can make it slightly more likely one will experience peace later in life. This seems to echo the finding above regarding poor childhood health, where in select countries, like Nigeria, this raised the chances of people having IP in adulthood. As in that health case, it would appear that, at least in some cultural contexts, adversity can lead people to develop the aptitude or fortitude that leads to a greater propensity to attain peace later in life. Again, we cannot tell from our data what it is about these particular contexts that does perhaps enable that, but this would be a fruitful avenue for future research to investigate.

Finally, there are three factors that are not necessarily about childhood per se, but are nevertheless relevant to childhood, namely, people’s age, sex (referred to in the GFS as “gender”), and immigration status. In one sense of course, these are childhood factors (in that they tell us something about people’s childhood), but from another perspective are factors that pertain to the individual at all life stages. Nevertheless they are worth briefly noting here. Of these, age had the strongest impact. Essentially, the older the participant, the more likely they are to have IP. Compared to people aged 18–24 (i.e., born between 1998 and 2005), those aged 25–29 (1993–1998) had just a marginally higher RR of 1.01, but the RRs rise in a linear way with the age categories, culminating in 1.19 for people aged over 80 (born in 1943 or earlier). These findings could be regarded as reflecting a childhood factor, especially if we interpret the data as being about the time period when people were born, hence being a cohort effect. However, the emergent literature on IP suggests it tends to increase as a function of age33. It is thus perhaps more realistic to interpret the findings here as more a question of the actual current age of the participants. It is nevertheless again still interesting to note regional variation, where the RR of this oldest category ranged from 0.77 in Poland to 1.37 for the UK and US, showing the relationship between age and peace is not universally observed, and like the other factors here is affected by socio-cultural dynamics.

The penultimate variable is sex (“gender”), which ranked second last in terms of impact, where compared to males, females had an RR of 0.98, although the very small percentage of reporting their gender as “other” had considerably lower IP (RR = 0.44). We need to be cautious in interpreting this latter finding, as this group was very small (< 0.1% of the observed sample) within several countries, leading to complete separation and large uncertainty in this estimate (95% CI: 0.13, 1.50). It is nevertheless a strikingly low RR that demands further study. There is by now an extensive literature showing that people who identify as LGBTQ + tend to have lower levels of mental health across the lifespan, from youth86 to older adults87. It is perhaps unsurprising then that this factor then would also affect IP. It is not certain whether the data here constitutes a childhood factor per se, since the item asks people their current gender, not their gender as a child, and it is possible that some percentage who answered “other” now would not have done so in childhood. Yet even if the latter were the case, it is likely that some relevant dynamics may have manifested during childhood (e.g., a sense of gender dysphoria). More research will therefore be needed to look into this finding. As with other factors, it will also be important to investigate the regional variation, where the RR for females ranged from 0.90 in Kenya to 1.13 in Turkey, and for those answering “other” ranging from 0.40 in Indonesia to 1.36 in Mexico. It would be helpful to know, for instance, what it is about Mexico that means people who answer “other” tend to be much more likely to have IP than males or females.

Lastly, one factor had no significant impact on peace, namely immigration status: compared to people born in the country in which they live, those born elsewhere had a RR that was basically equal (1.01). As with age and gender, this is not necessarily a childhood factor, since it reflects a person’s current immigrant status, not that of when they were a child (though it does indicate whether they were born in a different country than the one in which they now live). But it is still intriguing that such status does not seem to have any bearing on IP, which is notable, given that being an immigrant is frequently perceived as presenting challenges that can be detrimental to mental health88. That said, research often finds immigrant mental health is “better than expected”89, and may even be better than native people, a phenomenon remarked on frequently enough to have a label – the “healthy immigrant effect” – which “suggests that immigrants have a health advantage over the domestic-born,” though this usually “vanishes with increased length of residency”90. In our case, while we didn’t observe this kind of effect, neither were immigrants disadvantaged when it comes to IP. Again though, there were also significant regional disparities, with RRs ranging from 0.92 in Egypt and India to 1.19 in Tanzania, so in some countries at least the healthy immigrant effect does seem to play out.

Before closing, there are of course limitations to the study that need acknowledging. First, as flagged above, caution is needed in interpreting cross-national differences as these may be influenced by various socio-cultural factors. We already mentioned some linguistic considerations, but in additional to these are issues such as local, national and international occurrences that might have an effect on the outcome of interest, as well as seasonal effects arising from data being collected in different countries at different times of the year. Second, IP was assessed using a one-item measure, which does not capture its full complexity compared to multi-item scales which would have higher validity and reliability, such as the Peace of Mind Scale20. There is always a trade-off though in survey research between depth and breadth: including multi-item scales would limit the number of constructs assessed, and it was decided that, on balance, any limitations of using single item measures are outweighed – from the perspective of the GFS as a whole – by the value of including a greater number of constructs which together allow a more comprehensive assessment of flourishing. Third, the data in this paper are cross-sectional, which precludes conclusions about the directionality of the associations. The GFS is a longitudinal study, however, and the second wave of data collection is already underway. Moreover, even with this first wave of data we are arguably able to construct a synthetic longitudinal study by retrospectively assessing childhood experience, while also reporting E-values to assess the robustness of our findings to unmeasured confounding. Thus, although retrospective assessments are subject to recall bias, for such bias to completely explain away the observed associations would require that the effect of adult IP on biasing retrospective assessments of the childhood predictors would essentially have to be at least as strong as the observed associations themselves67.

Conclusion

IP has received relatively little attention amidst the proliferation of research into the various aspects of flourishing over recent decades. Hence the value of our study, which is one of the first to explore its childhood predictors, especially from a cross-national perspective. Using a retrospective assessment of childhood experiences in 22 countries, combined with several important current demographics, we found that many aspects of a child’s upbringing do predict peace in adulthood, with the most impactful factor being self-rated health growing up, while the least impactful predictor was immigration status (which indeed was the only factor with a non-significant effect). All the significant relationships documented in this study were robust to potential unmeasured confounding, as assessed by E-values. However, across all factors were important country-level variations that will require further research to understand. We hope future work will help to explain such differences, and more generally that researchers will pay closer attention to IP as an important constituent of human flourishing.