The pursuit of well-beingFootnote 1 has long fascinated individuals. Empirical evidence suggests that happiness yields numerous benefits across multiple life domains (for a review, see Lyubomirsky et al., 2005). For example, happier individuals have better cardiovascular health, more robust immune responses, and longer life expectancies (Boehm et al., 2012; Marsland et al., 2001; Stone et al., 1987). Furthermore, happy individuals are more likely to possess resources highly valued by society, such as positive social connections, fulfilling romantic relationships, higher incomes, and better work performance (Biswas-Diener & Diener, 2001; Diener & Seligman, 2002; Karney & Bradbury, 1995; Reis et al., 2000; Stutzer & Frey, 2006; Wright & Staw, 1999). They also exhibit greater resilience in adversity (Tugade & Fredrickson, 2004) and a higher likelihood of achieving positive life outcomes (Lyubomirsky et al., 2005).

Given these benefits, it becomes imperative to identify and understand the pivotal determinants that underpin well-being. Indeed, research has explored correlates across diverse domains such as personality traits (e.g., extraversion, neuroticism; Anglim et al., 2020; DeNeve & Cooper, 1998; Ozer & Benet-Martinez, 2006), attitudes and beliefs (e.g., optimism, self-esteem, self-affirmation; Alarcon et al., 2013; Diener & Diener, 1995; Emanuel et al., 2018), lifestyle (e.g., religiosity, leisure engagement; Hackney & Sanders, 2003; Kuykendall et al., 2015), and demographic characteristics (e.g., age, wealth; Caporale et al., 2009; La Barbera & Gürhan, 1997). Moreover, Choi et al. (2021) have recently advanced the notion that the essentialist beliefs about happiness (EBH), which view happiness as genetically predetermined and immutable, exert (detrimental) effects on individual well-being. In addition to individual study findings, meta-analyses further contribute to the overall trends; among these factors, positive perceptions of self and others, sociability, likability, prosocial behavior, adaptive coping, and physical health are potent predictors of well-being (Lyubomirsky et al., 2005). However, these meta-analytic results, by their very nature, have been critiqued for metaphorically ‘mixing apples and oranges,’ due to the challenge of comparing disparate studies and samples (Carpenter, 2020; Wortman, 1983). This underscores the need to investigate the determinants of well-being within a unified sample.

To overcome this constraint, machine learning offers a highly effective statistical method for feature selection and regularization compared to traditional regression models, particularly when identifying the most significant factors of complex phenotypes (e.g., well-being) across multiple domains. This data-driven method employs an advanced feature selection algorithm optimized for dissecting intricate variable relationships among highly correlated predictors and reducing the dimensionality of the model. Given our research goal to identify and prioritize predictors of psychological and subjective well-being measures across diverse domains, we adopted a machine learning (ML) approach to address the limitations of traditional regression methods, such as overfitting, by adding a penalty to the regression coefficients while maintaining predictive accuracy (Dwyer et al., 2018; Elhai & Montag, 2020). Despite ML approaches across various domains, such as emotion (Prout et al., 2020), personality (Gladstone et al., 2019), health (Kim et al., 2015; Lee et al., 2020a), and workplace dynamics (Sajjadiani et al., 2019), its use for systematically comparing potential predictors of happiness remains surprisingly sparse. One recent study attempted to utilize ML with a panel dataset to elucidate the relative importance of various predictors of subjective well-being within the United States (Margolis et al., 2021). The analysis identified individualistic or personally expressive aspects, such as self-esteem, control, and autonomy, among the key positive predictors of subjective well-being in the U.S. population. Conversely, subjective stress reactivity emerged as the most significant negative predictor. Furthermore, although demographic factors and subjective socioeconomic status remain important, they play a comparatively minor role in predicting subjective well-being (Margolis et al., 2021). Nevertheless, considering the cultural and national variations in well-being and its predictors (Diener et al., 1999; Hofstede, 1984; Schimmack et al., 2002; Suh & Oishi, 2004), questions about the generalizability and the broader applicability of these findings, especially in other cultural contexts such as South Korea (hereinafter referred to as Korea) persist. To fill the gap, we employed the ML approach specifically to ascertain the most predictable determinants of happiness within the Korean population. Furthermore, we aim to assess the extent to which findings from studies conducted in Western cultures are applicable to, or can be generalized across, distinct cultural contexts.

In Korea, a country with a collectivistic orientation and prevalent relational-interdependent self-construal, internal self-attributes (e.g., self-esteem) or a sense of personal control might not carry the same predictive power for well-being as observed in the U.S. (Diener & Diener, 1995; Kim et al., 2003; Suh, 2000). Alternatively, social status, easily discerned or judged by others, could play a pivotal role in determining well-being (Curhan et al., 2014b; Kim et al., 2010; Leung & Cohen, 2011; Wirtz & Scollon, 2012). Despite potential cultural differences, the possibility of similar rank ordering of well-being predictors in the Korean context should not be overlooked. For instance, contrary to expectations, collectivistic Koreans view the contribution to their community and society as a relatively low priority (Park, 2009), and similarly, the U.S. and Korea share comparable patterns in beliefs about well-being and actual well-being experiences (McMahan et al., 2014). Yet, due to the limited sample representation in previous studies (e.g., older women living alone; Kim et al., 2019) and the paucity of explicit comparisons of predictor strength, the question remains open. Thus, the quest to identify the most potent determinants of well-being continues; the answers, particularly for Korea, still seem elusive.

We aimed to fill this gap in the literature by (1) using a sample of Korean adults, (2) examining potential variations in a comprehensive set of more than 30 predictors of well-being, and (3) integrating advanced ML approaches. In pursuing these objectives, it is imperative to consider one important issue: the optimal assessment of well-being.

1 Conceptualization and Measurement of Well-Being

Well-being is a multifaceted construct without a single, universally accepted definition. Historically, from Aristotle’s era to the present day, scholars have employed diverse approaches to conceptualize and operationalize this construct. In Nicomachean Ethics, Aristotle made a distinction between the good life (eudaimonia) and pleasure (hedonia) (Aristotle, ca. 350 B.C.E./1999). Building on Aristotle’s foundation, modern psychologists posit two traditions in well-being research: psychological well-being (PWB) and subjective well-being (SWB).

Although both these traditions fall under the well-being umbrella (Disabato et al., 2016), they highlight different aspects of wellness (Fowers et al., 2010; Keyes et al., 2002; Ring et al., 2007; Thorsteinsen & Vittersø, 2020). PWB focuses on the pursuit and realization of one’s potential, characterized by positive psychological functioning across six dimensions: autonomy, environmental mastery, personal growth, positive relationships, purpose in life, and self-acceptance (Ryff, 1995). Conversely, SWB underscores an individual’s evaluative perspective of life, capturing global life satisfaction, the experience of positive affect, and the absence of negative affect (Deci & Ryan, 2008). The dimensions of PWB show weak to moderate correlations with SWB (Keyes et al., 2002; Ryff & Keyes, 1995), solidifying the idea that while they are conceptually related, they remain empirically distinct.

Previous research, however, has largely overlooked the identification of potential predictors and their relative importance in PWB compared to those of SWB. For example, Margolis et al. (2021) treated the subscales of the PWB as predictors for SWB. This might be influenced by the perspective that eudaimonic constructs, such as those measured using the PWB scale, should be viewed more as causes than as constituents of well-being. Meanwhile, Keyes et al. (2002) sought to discern the distinctive contours between PWB and SWB; however, their exploration was confined to sociodemographic variables, such as age and education, and the five-factor personality traits (i.e., the Big Five). More recently, Joshanloo et al. (2021) identified distinct associations between hedonic and eudaimonic well-being in a large Korean sample, analyzing five predictors including long-term planning, self-control, sensation-seeking, grit, and intellectualism. Nevertheless, this list of predictors is by no means exhaustive.

It is noteworthy that the conceptions of well-being discussed here, largely rooted in Western philosophies, have been shown to be culturally acceptable and widely utilized across diverse socio-cultural and linguistic contexts (Diener, 1996; Ryff, 2013). For instance, measures of PWB have been validated in countries such as the U.K. (Abbott et al., 2006), China (Cheng & Chan, 2005), Germany (van Dierendonck, 2004), France (Salama-Younes et al., 2011), Italy and Belarus (Sirigatti et al., 2013), Portugal (Fernandes et al., 2010), Spain (van Dierendonck et al., 2008), Taiwan (Lin, 2015), and Türkiye (Akin, 2008); similarly, SWB measures in Brazil (Gouveia et al., 2009), China (Bai et al., 2011), France (Blais et al., 1989), Germany (Arrindell et al., 1991), Portugal (Sancho et al., 2014), Mexico (López-Orteg et al., 2016), and Türkiye (Durak et al., 2010). Importantly, PWB and SWB have also been examined in Korea among diverse groups, including the general adult population (Kim et al., 2001), married women (Kim & Kim, 2000), and patients with schizophrenia (Yoo, 2014); these measures have been utilized in numerous studies (e.g., Jyung et al., 2021; Lee et al., 2020b, 2021). Despite broad consensus on the applicability of these well-being concepts, formidable challenges remain in interpreting cultural nuances—a point we will revisit in the General Discussion section.

1.1 Present Study

Despite debates on the distinctions between these two traditions (e.g., Kashdan et al., 2008; Waterman, 2008), we argue for a comprehensive examination of both. Understanding the essence and determinants of well-being necessitates an examination of both traditions and their relationships with predictors. This study seeks to offer insights into whether PWB and SWB can be distinguished based on their determinant factors. To establish the most important predictors of the two pivotal aspects of happiness, we assessed PWB and SWB, taking into account more than 30 potential predictors. These spanned categories such as demographics, health, personal attributes, and social connectedness. To ensure analytical precision, we employed the Least Absolute Shrinkage and Selection Operator (LASSO), an ML algorithm specifically designed to discern multivariate profiles of psychological and lifestyle factors (Tibshirani, 1996). The LASSO has a unique ability to apply a penalty term that sets certain coefficients to zero, facilitating automatic variable selection among multiple predictors. The LASSO was chosen over traditional multiple regression analysis because it provides superior variable selection and generalization through coefficient shrinkage, and handles multicollinearity more effectively. Our study aimed to identify the most influential predictors of PWB and SWB by employing the LASSO to enhance model generalization, given the correlations among our predictors. Leveraging this methodology, we adeptly pinpointed the primary determinants for PWB and SWB using fold cross-validation methods that enhance the model’s performance, thereby increasing its predictive power. To the best of our knowledge, no previous empirical research has applied ML techniques to explore potential variations in such a comprehensive set of predictors for both PWB and SWB, specifically within the Korean context.

We report how we determined our sample size, all data exclusions, and all measures relevant to the analysis. The data and analytic code necessary to reproduce our results are available at https://osf.io/bqa98/?view_only=8e24cab5423346bc94a972560509332d. All participants provided written informed consent before participation, and the Institutional Review Board of Seoul National University approved all procedures.

2 Method

2.1 Participants and Procedures

The data for this study were derived from the Korean Adult Longitudinal Study (KALS), which conducted its first wave of data collection from 2017 to 2018. Our sampling strategy focused on Seoul residents, ensuring a balanced distribution across various age groups (ranging from 20s to 60s), genders, and different geographical regions within Seoul. We partnered with a research firm based in Seoul, Korea, for participant recruitment. Participants were selected through random-digit dialing (RDD) of mobile phone numbers, adhering to specified criteria for age, gender, and residential location. Time and resource constraints limited the number of participants that could be recruited. Given our focus on panel data and the nature of machine learning models, which operate without prior hypotheses or assumptions, there were no a priori considerations about sample size or power. A total of 561 Seoul residents aged 20–69 years were recruited and two participants were excluded due to incomplete surveys. This resulted in a final sample size of 559 participants (280 males, 279 females; Mage = 44.87, SDage = 13.82). All participants fully responded to every question, ensuring no missing data for any variables.

3 Measures

3.1 Psychological Well-Being and Subjective Well-Being

Two measures of well-being were administered: Ryff’s PWB scale (Ryff & Keyes, 1995) and Diener’s SWB scale (Diener et al., 1999). The PWB scale is an 18-item measure assessing six aspects of eudaimonic well-being: autonomy, environmental mastery, positive relations with others, purpose in life, self-acceptance, and personal growth. In the present study, we combined the scores of these aspects into an overall PWB score, with the scale demonstrating adequate reliability (α = 0.79). The SWB scale is specifically designed to capture the cognitive and affective aspects of hedonic well-being. It was measured using the Satisfaction with Life Scale (SWLS) (Diener et al., 1985) and Positive and Negative Affect Schedule (PANAS) (Watson et al., 1988). The SWLS is a five-item measure of perceived satisfaction with one’s life (α = 0.87), whereas the PANAS is a 20-item measure of affect balance, comprising 10 items for positive affect (e.g., interested, excited; α = 0.84) and 10 for negative affect (e.g., afraid, nervous; α = 0.86). These well-being measures have been repeatedly validated and used in our recent studies among Korean populations (e.g., Jyung et al., 2021; Lee et al., 2020b, 2021), demonstrating acceptable internal consistency (all αs ≥ 0.77).

3.2 Predictors of Well-Being

Our goal was to identify the most significant predictors of well-being and analyze their relative impacts. Therefore, we selected 32 relevant predictors from the available panel survey questionnaire, all of which have established correlations with well-being in the field and span various key domains. First, we focused on personality traits, specifically the five broad dimensions of personality (Anglim et al., 2020; Steel et al., 2008) and the concepts of individualism and collectivism (Ahuvia, 2002; Suh & Oishi, 2002; Tov & Diener, 2007; Veenhoven, 1999). Second, the psychological assets and liabilities domain comprises eight predictors: gratitude (Wood et al., 2010), grit (Vainio & Daukantaitė, 2016), meaning in life (Grouden & Jose, 2015), optimism (Augusto-Landa et al., 2011), self-esteem (Diener & Diener, 1995), self-control (Margolis et al., 2021), materialism (La Barbera & Gürhan, 1997), and social comparison (Wheeler & Miyake, 1992). Third, we included essentialist beliefs about happiness, recently noted for their influence on happiness levels (Choi et al., 2021). The fourth and fifth domains address major well-being predictors such as social connectedness (Kim & Sul, 2023; Jose et al., 2012)—assessed via perceived social support and number of friends—and mental and physical health (Martín-María et al., 2017; Steptoe et al., 2015; Yazdani et al., 2018)—measured through depression, loneliness, body image, and physical symptoms. The sixth domain explores lifestyle factors, including volunteering (Van Willigen, 2000), travel frequency (Friman et al., 2017), religious affiliation (La Barbera & Gürhan, 1997), and leisure activities (Brajša-Žganec et al., 2011). Lastly, we incorporated subjective socioeconomic status (Howell & Howell, 2008; Tan et al., 2020) for current and childhood, as well as demographics (Caporale et al., 2009; Ferrer-i-Carbonell, 2005; Lee et al., 2022) such as age, gender, educational attainment, and income. A complete list of the predictors, along with their corresponding scale references, is summarized in Table 1.

Table 1 Descriptions and descriptive statistics of predictors included in analyses
Full size table

3.3 Statistical Analysis

Pairwise Pearson’s correlation coefficients and corresponding p-values between the continuous variables were computed, and a correlation plot was generated using the ‘corrplot’ package in R (Wei, 2021). For the ML analysis, we utilized the LASSO model (Tibshirani, 1996) to identify multivariate patterns of candidate factors predicting PWB and SWB. Using the LASSO model, which applies a penalty term that shrinks the coefficients of less relevant predictors towards zero, we prioritized the factors predicting PWB and SWB, considering the relationships among them from the most important to the least important. The LASSO-based ML protocol reported in previous studies (Ahn et al., 2014; Ahn & Vassileva, 2016) uses the ‘easyml’ R package (Hendricks & Ahn, 2017), which is a user-friendly wrapper of various ML packages, including the glmnet R package (Friedman et al., 2010). In short, this protocol divided the dataset into a training set (67%) and a test set (33%), and was repeated 1,000 times to ensure robust correlation scores across different divisions of the training or test sets (Ahn et al., 2014; Ahn & Vassileva, 2016). As an index of prediction accuracy, we determined the correlation score for the mean training and test sets for PWB and SWB.

4 Results

4.1 Correlation Between the Variables

We began our analysis by conducting Pearson’s pairwise correlations between continuous predictors arranged in order using hierarchical clustering. Several noteworthy patterns emerged from the correlation diagram (see Fig. 1). First, essentialist beliefs about happiness (EBH) were positively correlated with age but negatively correlated with gratitude, optimism, and self-esteem. Furthermore, self-esteem is strongly associated with several predictors, including emotional stability, grit, conscientiousness, self-control, and social support; however, it showed a negative correlation with depression and loneliness. Monthly income was correlated with subjective socioeconomic status (SES), openness, education, and childhood SES. Finally, monthly physical symptoms were highly associated with depression and loneliness. We excluded binary variables such as sex, religious affiliation, and leisure pursuits from the correlation analysis. The exact coefficients and p values are provided in Supplemental Table 1.

Fig. 1
figure 1

Correlation between the predictors. SWB = subjective well-being; PWB = psychological well-being; SES = socioeconomic status; EBH = essentialist beliefs about happiness

Full size image

4.2 LASSO-Based Machine Learning for Psychological Well-Being and Subjective Well-Being

The ML analysis, utilizing the LASSO approach, revealed multivariate profiles of the predictors for PWB and SWB, respectively. Due to the high correlation among the variables, we used the LASSO analysis to select a sparse model that prioritizes the most predictive well-being variables. When we evaluated the correlation scores for PWB using randomly selected training and test sets (with 1,000 iterations) comprising 67% and 33% of the data, respectively, the average correlation score was 0.82 for the training dataset and 0.78 for the test dataset (see Supplemental Fig. 1). The model’s performance indicated an average correlation score of 0.77 for the SWB training dataset and 0.73 for the test dataset (see Supplemental Fig. 2).

4.2.1 The Most Influential Predictors of PWB vs. SWB

First, we prioritized the factors that predict PWB or SWB individually, identifying the most influential factors for each aspect of well-being. Among the predictors in the current model, EBH, which views happiness as a fixed state, exerts the most significant negative influence on PWB. Conversely, the level of meaning in life was strongly and positively related to PWB, followed by self-esteem (see Fig. 2A). Regarding SWB, subjective SES and personality traits of emotional stability (vs. neuroticism) were the strongest positive predictors of SWB (see Fig. 2B). Conversely, depressive symptoms negatively predicted SWB, as indicated by the largest absolute value of the coefficient estimate.

Fig. 2
figure 2

Multivariate patterns of demographic and psychological measurements that predict PWB (A) and SWB (B). Error bars indicate 95% confidence intervals

Full size image

4.2.2 The Shared Predictors of PWB and SWB

From the 32 measurements assessed, we identified 12 shared factors that concurrently and significantly predicted both PWB and SWB. Individuals demonstrating elevated levels of collectivism, conscientiousness, emotional stability, extraversion, openness to new experiences, gratitude, grit, meaning in life, self-esteem, as well as engaging in frequent travel and volunteering, reported higher levels of PWB and SWB. Conversely, higher levels of social comparison and depressive symptoms were associated with lower levels of PWB and SWB.

4.2.3 Exclusive Predictors for PWB and SWB

Acknowledging the shared but distinct natures of the two well-being dimensions, we identified variables that exclusively predicted either PWB or SWB. Subjective SES, the most powerful predictor of SWB (see Fig. 2B), was found to be insignificant regarding PWB (see Fig. 2A). Moreover, age, gender, materialism, and religion are also exclusively significant factors for SWB. In contrast, variables including educational attainment, self-control, individualism, optimism were only successful predictors of PWB, but were not significantly associated with SWB (see Fig. 3 for a summary of the results). Given the possible disparities between the estimates from the LASSO approach and pairwise correlation analyses (refer to Supplemental Table 1), certain findings (such as the lack of significant relationship between income and PWB) may not be replicable compared to previous findings. It is important to acknowledge that the current findings hinge on the use of the LASSO algorithm, an approach that automatically selects variables significantly influencing the response variable—in this case, PWB and SWB—while concurrently disregarding variables deemed less impactful. This methodology inherently prioritizes more significant variables, which may potentially overshadow lesser, yet still meaningful, variables in the model. For example, while subjective SES surfaced as a significant SWB predictor amidst other variables in the model, it was not influential enough to predict PWB. This does not mean it is unimportant, rather there are more influential predictors. We will discuss the strengths and limitations of this model in depth in the discussion section.

Fig. 3
figure 3

Overview of variables predicting either PWB or SWB. AE = appearance evaluation; chSES = childhood SES; Comparison = social comparison; Control = self-control; EBH = essentialist beliefs about happiness; ES = emotional stability; Friends = number of friends; PS = physical symptoms; SE = self-esteem; SS = social support; Travel = travel frequency; sSES = subjective SES

Full size image

4.2.4 The Factors Least Influential (or Insignificant) for PWB and SWB

Among the considered variables, social support, monthly income (an object measure of SES), and appearance evaluation were not predictive for either PWB or SWB. Although some of them exhibited significant pairwise correlations with either PWB or SWB (as shown in Supplemental Table 1), their weights were ultimately penalized to zero in the LASSO model due to the penalty term that eliminates less significant variables.

5 General Discussions

Our study leveraged a comprehensive survey dataset, drawn from a Korean sample, to examine the significant distributions of diverse determinants influencing well-being. Given the multifaceted nature of happiness, this study delineated two related yet distinct aspects of well-being: eudaimonia and hedonia. With high predictive accuracies of 0.78 and 0.73, respectively, the LASSO-based ML model prioritized the relevant predictors from a pool of 32 potential candidates. Echoing earlier correlational and ML-driven research in the U.S., predictors such as emotional stability, extraversion, gratitude, and depression were significantly associated with both well-being aspects. Despite a significant but moderate correlation between PWB and SWB (r = .57, p < .001), our results highlighted the distinct primary determinants for each. For PWB, the most salient predictors were EBH, meaning in life, and positive self-evaluation of worth (i.e., self-esteem). In contrast, SWB was predominantly influenced by depressive symptoms, subjective SES, and emotional stability. While certain primary predictors such as meaning in life, self-esteem, depression, and emotional stability influenced both well-being aspects, there were observable disparities in the magnitude of their influence and their relative importance among other determinants between PWB and SWB. Thus, our findings support the theoretical stance that, although PWB and SWB share conceptual underpinnings, their empirical manifestations differ, particularly in relation to their determinants.

Notably, the LASSO approach identified EBH as a robust negative predictor for PWB, though its influence on SWB was comparatively subdued. This adverse relationship between EBH and well-being was corroborated by our correlational analyses (r = − .46, p < .001 for PWB and r = − .11, p = .007 for SWB), aligning with earlier studies on essentialist views pertaining to well-being and mental health. For instance, individuals with stronger EBH showed reduced motivation to seek happiness and reported lower happiness levels (Choi et al., 2021). Similarly, those predisposed to perceive health and personal traits as fixed showed less engagement in health-promoting and protective behaviours (Schreiber et al., 2020; Zhang & Kou, 2022) and faced heightened risks of mental health issues (Schleider et al., 2015; Schleider & Weisz, 2016).

On a related note, subjective SES, one of the significant contributors to SWB, displayed a distinctive pattern: it had a null effect on PWB. There is solid evidence that individuals’ subjective placement within the socioeconomic hierarchy positively correlates with their SWB, above and beyond objective SES indicators (e.g., Adler et al., 2000; Anderson et al., 2012; Diener et al., 2010; Haught et al., 2015; Kraus et al., 2013; Tan et al., 2020). However, the relationship between subjective SES and PWB has either been largely overlooked in research (e.g., Curhan et al., 2014a) or has shown inconsistent or comparatively weak correlations when explored. In the current study, subjective SES exhibited a small correlation with PWB (r = .33, p < .001; its correlation with SWB was more substantial, r = .58, p < .001), mirroring effect sizes observed in recent studies (e.g., Navarro-Carrillo et al., 2020; Wong & Yang, 2023). In one account, this observed discrepancy may be attributed to the strong association of SWB with the fulfillment of immediate desires or gratification, which often depends on financial resources for acquiring necessities and luxuries; while PWB, on the other hand, integrates experiences spanning the past, present, and future, rendering it less susceptible to the tangible resources derived from social standing (Disabato et al., 2016; Huta & Ryan, 2010; Joshanloo et al., 2021; Kim et al., 2014). For example, present-oriented activities including shopping are perceived as more hedonic, while activities that are future-oriented or involve skill-building are more eudaimonic (Henderson et al., 2013). Likewise, having money to purchase desired items enhances happiness but does not necessarily contribute to meaningfulness—a central component of eudaimonic well-being (Baumeister et al., 2013).

On another note, our study aligns with previous findings suggesting that most demographic factors, including age, gender, education, and income, are relatively weak predictors of well-being (Keyes et al., 2002; Margolis et al., 2021). One possible reason is that subjective perceptions, as opposed to demographic or objective indicators, often capture more relevant and dynamic attributes of well-being. This perspective is supported by research domains such as the study of social class in relation to psychological functioning, which advocates for the use of subjective SES measures over objective ones (e.g., Adler et al., 2000; Curhan et al., 2014b; Singh-Manoux et al., 2005).

The present study further advances our understanding of the determinants of happiness within Korea’s unique cultural framework. While the significance of the predictors seems to remain across different countries, the current study suggests that their relative strength can vary, thus highlighting potential cultural nuances. For example, in the U.S. sample, perceived control over one’s life and autonomy emerged as the most important predictors of SWB (Margolis et al., 2021). In contrast, in the Korean sample from this study, the weight of personal control and autonomy was overshadowed by subjective SES. Additionally, these factors were less predictive of well-being than a collectivistic orientation, which emphasizes communal concern and harmony. This cultural insight suggests that in cultures with a strong emphasis on relational interdependence, like Korea, externally observable factors (e.g., class) and shared societal values could have a more pronounced impact on well-being than individual-focused attributes.

Despite the promising findings, several limitations warrant discussion. First, our dataset, though comprehensive in its coverage of potential well-being predictors from past literature, predominantly consists of subjective characteristics and has a relative dearth of objective ones. Additionally, certain subjective metrics, such as work-related satisfaction and basic psychological needs—both deemed pivotal for well-being—were not included. Future research should seek to incorporate a broader and more balanced mix of both objective characteristics and subjective metrics. Second, as the present study targeted Korean adults, our findings should be interpreted cautiously when generalizing to other countries and cultures. For instance, in countries more individualistic than Korea, the impact of collectivism on PWB might be less pronounced, whereas individualism could have a more significant effect. Third, a more fundamental concern involves potential cultural nuances in the conceptions and measurements of well-being (e.g., Veenhoven, 2012). Although existing evidence suggests that traditional well-being theories based on Western, individualistic, or personally expressive frameworks are not necessarily irrelevant or culturally biased against modern Korean populations (e.g., Joshanloo et al., 2019; McMahan et al., 2014), unresolved questions remain regarding the suitability of these concepts in East Asian contexts, which often emphasize harmony and societal cohesion as essential components to genuine happiness (Markus & Kitayama, 1991; Triandis, 1989). We look forward to future work that replicates and extends this line of inquiry through more nuanced approaches, such as employing recently introduced collective well-being concepts (Hitokoto & Uchida, 2015; Krys et al., 2019) and scales (Suh & Koo, 2011), to further validate the applicability of these measures in diverse cultural settings, including Korea. Fourth, the variable relationships in the analysis can sometimes make the interpretation challenging. For example, even though we found a negative pairwise correlation between loneliness and SWB (r = − .43, p < .001), within the context of our LASSO model analysis, loneliness did not emerge as a prominent predictor, likely owing to its smaller effect size relative to other variables in the model. It is important to note that LASSO offers distinct advantages over traditional regression methods when it comes to prioritizing the relative influence of predictors within the context of other variables, rather than solely focusing on identifying pairwise relationships. Finally, the cross-sectional design limited our capacity to make causal inferences. Given our interest, we sought to examine the role of independent variables such as EBH, as predictors or determinants of well-being. However, it is equally conceivable that happiness could lead to a more significant endorsement of EBH. This methodological challenge underscores the need for future studies to adopt an experimental or multiple-wave longitudinal design to provide a solid foundation for causal inference. The current study advances our knowledge of well-being predictors, building on prior research like the findings of Margolis et al. (2021) within the U.S. populations. To enhance generalizability, future research should involve more extensive and representative samples from Korea, spanning a wider age range from adolescents to the elderly. This approach will help confirm and extend the results observed in this study.

6 Conclusion

This study sought to elucidate the critical determinants of the well-being of Korean adults, a population that has received limited attention in previous research. According to the World Happiness Report (Helliwell et al., 2022), Korea is the third least happy country among 38 OECD nations. Despite having a relatively high gross domestic product and life expectancy, the Korean population seems to experience a conspicuous lack of happiness. Our work offers valuable insights into the most likely causes of relatively low happiness among Koreans and potential interventions aimed at optimizing their overall well-being. Based on an advanced machine learning model, results identified both beneficial factors, such as the influence of perceived social status on hedonic well-being, and detrimental factors, like adopting an entity view of happiness, particularly concerning eudaimonic well-being. Taken together, we anticipate that these findings will advance scholars in designing and analyzing their future well-being research, encouraging a deeper exploration of cultural nuances.