1 Introduction

Inequalities in income and subjective well-being (SWB)Footnote 1 are pressing issues for scholarship and social policy (Aiyar & Ebeke, 2020; Araki, 2023a; Oishi et al., 2011; Olivos & Jin, 2023; Ono & Lee, 2013). Evidence suggests that individuals with higher income are more likely than their less resourceful counterparts to have better SWB (Clark et al., 2008; Deaton, 2008; Diener et al., 2010a, 2010b; Helliwell, 2003; Kahneman & Deaton, 2010; Kahneman et al., 2006; Musick et al., 2016). Although there is debate about the presence of income satiation (i.e., the income effect disappears at a certain point) (Jebb et al., 2018; Killingsworth, 2021; Stevenson & Wolfers, 2013), it is a near consensus that money can buy SWB to a considerable degree (Delhey & Steckermeier, 2016; McBride, 2001).

Despite such a progressive understanding, one important structure remains understudied: the linkage between low income and low SWB, or what one may call subjective ill-being (SIB). Given the positive income-SWB association as shown in the literature, it seems obvious that the poor face a greater risk of SIB than the affluent. Nevertheless, the fact that high income leads to better SWB does not necessarily mean that economic hardship results in SIB. Indeed, pioneering research in this vein reveals that the construct and mechanisms of well-being and ill-being are not identical (Headey et al., 1984, 1985). Social indicators movements have also detected the significant linkage between economic disadvantage (e.g., unemployment) and negative emotions (e.g., worry, anxiety, and pain) (Anderson, 2015; Glatzer et al., 2015; Land & Michalos, 2018). Notwithstanding, evidence is elusive concerning how low income and SIB are associated.

In exploring this agenda, one essential perspective is gender inequality. A vast literature shows that the structure of SWB, including the strength of income-SWB relationships, differ between women and men (Arrondo et al., 2021; Meisenberg & Woodley, 2015; Stevenson & Wolfers, 2009). This implies that the socio–psychological penalty for the poor may also be formed in a gendered way. In addition, recent research reports people’s SWB is notably influenced by the degree of gender (in)equality (Aassve et al., 2015; Audette et al., 2019; Chen et al., 2023), such that gendered social norms impose greater pressure on individuals and thus undermine their SWB (Preisner et al., 2020). Should this be the case, it is plausible that the association between low income and SIB varies in accordance with macro-level gender parity. However, such potentially gendered structures have rarely been examined in a standardized way. We thus investigate the distribution of low income and SIB with particular attention to its gender difference and heterogeneity across societal-level gender inequality. Methodologically, this analysis also contributes to the literature by examining whether distinguishing between SWB and SIB reveals nuances obscured by the commonly assumed linear income-happiness correlation (Myers & Diener, 2018).

In the next section, we first review prior studies on the link between income and SWB to illustrate the core concept of our inquiry. Drawing on this basic understanding, we further review the literature from a gender perspective to formulate hypotheses. Data/methods are then explained, followed by results and discussion.

2 Income, Subjective Well/Ill-being, and Gender Inequality

In investigating the determinants of SWB, income has long been a central target of analysis. Much evidence shows that individuals with higher income are likely to enjoy better SWB, arguably by realizing their material needs and boosting psychological satisfaction as compared to the poor (Diener et al., 2010a, 2010b; Kahneman & Deaton, 2010; Kahneman et al., 2006). Extending earlier studies on income satiation (Kahneman & Deaton, 2010), Jebb et al. (2018) argued the positive effect of income would not increase beyond $95,000 (on life evaluation) and $60,000–75,000 (on emotional well-being). However, recent evidence suggests SWB keeps rising along with income. Killingsworth (2021), for example, analyzed experienced and evaluative well-being of over 30,000 workers in the United States (i.e., day-to-day feeling and reflection/satisfaction of their life, respectively) and found the substantial association between these two dimensions of SWB and income without explicit satiation.

The favorable income-SWB linkage thus seems evident, especially when comparing high- versus low-income groups at a certain point in time (McBride, 2001). However, longitudinal analyses suggest the SWB level does not necessarily improve even when the income level increases over time, sparking doubts about the linearity assumption (i.e., Easterlin Paradox) (Easterlin, 1974; Easterlin et al., 2010). At the theoretical level, this decreasing marginal utility is often explained by social comparison and the “hedonic treadmill,” whereby people get accustomed to higher levels of economic conditions (Brickman & Campbell, 1971; Clark et al., 2008; Ferrer-i-Carbonell, 2005; Firebaugh & Schroeder, 2009; Stevenson & Wolfers, 2008; Wolbring et al., 2013). Nonetheless, Oishi and Kesebir (2015) found that income inequality operated as the key societal trait that undermined the effect of increased income on SWB. This means that the positive income-SWB association would be supported, at least partially, by a fairer distribution of economic outcomes. Mikucka et al. (2017) also argue that economic development would contribute to SWB so long as social trust is maintained and income inequality declines in the meantime.

It is therefore logical to conclude that income contributes to better SWB, regardless of its dimension (e.g., experienced and evaluative). However, one similar yet substantively different question remains unanswered: are low-income individuals more likely than the affluent to encounter lowered SWB (i.e., SIB)? Although we can say that possessing higher income leads to favorable SWB, it does not guarantee the economically disadvantaged suffer from SIB. This is because, even though the average SWB score is higher among the advantaged (as frequently evidenced by the positive coefficients of high-income groups in regressions), it is still possible for the poor to somewhat avert the SIB risk. Indeed, Headey and colleagues shed light on the distinction between well-being (e.g., life satisfaction and positive feelings) and ill-being (e.g., life dissatisfaction and negative emotions), arguing that their determinants were not parallel (Headey et al., 1984, 1985). Diener (2006) also stressed the importance of paying attention to both well-being and ill-being in research and policy making. Nevertheless, evidence is scarce concerning how low income is associated with the bottom range of SWB.

One oft-cited exception is Clark and Oswald (1994), which analyzed the link between unemployment and mental distress. Although the function of income was inconclusive, they revealed (1) unemployment was generally linked to greater stress; (2) the negative effect of unemployment was relatively smaller among young people and workers in areas with high unemployment rates; and (3) those who were recently unemployed would suffer from greater distress than the long-term unemployed. Subsequent studies also investigated the structure of negative aspects of SWB with attention to such factors as income shocks (Baird et al., 2013) and age (Blanchflower, 2020). Budria (2013) further analyzed the German case and showed that the relatively low-income status and falling below the average income of one’s reference group led to low SWB despite its null effect on high SWB. Meanwhile, Headey and Wooden (2004) found that the effect of wealth on ill-being and well-being was as significant as that of income.

While ill-being has thus been examined by some pioneering work, there are four rooms for further investigations. First, most empirical studies in this vein assess the disadvantage of unfavorable economic status (e.g., low income; unemployment) by focusing on the relatively limited and/or negative coefficient of such attributes for the positive SWB measure (e.g., the 0–10 life satisfaction scale). Although this approach may detect a lower chance of obtaining better SWB among the disadvantaged, it remains unclear whether they are more likely than the advantaged to suffer from SIB. Figure 1 depicts this concept. Panel A contrasts the average SWB level (e.g., measured by the traditional 0–10 Cantril ladder scale) between low-income and high-income groups. This simply indicates a positive income effect on SWB. Meanwhile, Panel B illustrates the predicted probability of a three-tiered measure (i.e., favorable SWB, unfavorable SIB, or neither). One may assume that the chance of favorable SWB (e.g., high LS and positive affect) is higher among the rich than the poor (left figure) regardless of their difference in “neither” (middle figure). However, we know little about the relative risk of suffering from SIB (e.g., low LS and negative affect) among the disadvantaged compared to the affluent (right figure, patterns (a–c)). While it is probable that the pattern (a) or (b) exists (i.e., money buffers SIB), one cannot deny the possibility of (c) or even the reverse trend. Put differently, as with much debate on the presence of income satiation in relation to SWB (Jebb et al., 2018; Kahneman & Deaton, 2010), the downward trend of economic disadvantages and SIB may be nonlinear. This also means methodologically that, if the structure of the SIB risk attributed to low income is simply symmetrical with the link between high income and favorable SWB, linear-model analyses focused on the positive income-SWB trends without attention to the poor’s SIB would suffice in this line of research.

Fig. 1
figure 1

Conceptual model of the association between income and subjective well/ill-being

Full size image

Second, even when the unfavorable SWB indicator (e.g., mental distress, negative affect, and suffering) is used as an outcome variable, income is ignored or merely used as a control due to different research foci. For instance, Adamkovič (2020) examined the effect of income on economic decision-making whilst employing negative affect as an under-theorized control variable. Likewise, although some studies showed a significant association between macro-level economic factors and SIB-related indicators such as suicide rates (Flavin & Radcliff, 2009; Ross et al., 2012), the risk of low income per se is elusive. Third, alongside the average link between low income and low SWB, its gender difference remains understudied. Given the oft-reported heterogeneous structure of SWB between men and women (Audette et al., 2019; Dutta & Foster, 2013; Inglehart, 2002; Verma & Ura, 2022; Zweig, 2015), one may assume that the poor’s likelihood of suffering from SIB differs across gender. In particular, drawing on the evidence from different contexts that shows the larger effect of labor market indicators on SWB among men as compared to women (Arrondo et al., 2021; Bowman, 2023; Van der Meer, 2014), it is plausible that the psychological penalty for the poor is also more significant among men. Thus, we first test two basic hypotheses as follows.

Hypothesis 1

Low-income individuals are more likely than their high-income counterparts to suffer from SIB.

Hypothesis 2

The linkage between low income and SIB is stronger among men than women.

Herein, fourth, it is pivotal to pay attention to societal-level gender inequality given that the degree of gender parity significantly affects SWB (Aassve et al., 2015; Audette et al., 2019; Chen et al., 2023; de Looze et al., 2018). Nonetheless, little is known about how the gendered societal structure moderates the psychological penalty for the poor. In considering how SIB is stratified in connection with the economic disadvantage and gender inequality, one should note that (1) SWB is affected by freedom of choice (Evrensel, 2015; Gehring, 2013; Ngamaba, 2017; Veenhoven, 2000) and (2) macro-level gender inequality may reflect women’s financial autonomy (Mandel, 2009). The ideal of autonomy is the individuals’ disposition of resources to pursue their own ends (Zelleke, 2011). Thus, the lack of agency in more unequal societies could constrain the financial autonomy of women and, in turn, hamper the translation of higher resources into higher levels of SWB. Recently, Hajdu and Hajdu (2018) showed that these gender norms at the individual level moderated the effect of within-couples relative income on SWB. The capability approach (Sen, 1985) also supports this argument, and the body of evidence confirms that higher levels of autonomy may enhance the freedom of choice, which enables individuals to achieve what they value in their lives (Muffels & Headey, 2013). In more gender inegalitarian societies, the disposition of higher income may therefore increase women’s SIB due to frustration of lacking the autonomy to achieve those ends.

Moreover, despite the weakening of male breadwinner and female homeworker models in recent decades incurred by women’s educational and economic empowerment (Cunningham, 2008), evidence shows that (1) the “gender revolution” has been stalled especially in gender inegalitarian societies (England, 2010, 2011; Kan et al., 2022; Scarborough et al., 2019); and (2) labor market disparities persist in terms of motherhood penalties and/or male breadwinner premiums (Nylin et al., 2021; Van Winkle & Fasang, 2020), albeit diminishing in some contexts (Cooke & Fuller, 2018; Fuller & Cooke, 2018; Schechtl & Kapelle, 2023). Importantly, these family models still affect economic outcomes. For instance, Gonalons-Pons and Gangl (2021) showed that the country-level prevalence of male-breadwinner norms would negatively predict men’s unemployment. Likewise, men and women who conform to the chief-earner norms and the homemaker norms, respectively, are likely to be judged as more favorable from a traditional perspective (Gaunt, 2013). Researchers further argue this socially accepted men-as-breadwinner model makes men accountable for their performance in the labor market (Bowman, 2023), thus increasing psychological pressure/stress (Preisner et al., 2020).

As such, considering income as a fundamental element of breadwinning (Tracey Warren, 2007), one may expect that in more gendered societies, the low income level is detrimental to men’s SWB because it violates traditional gender norms of specialization. This perception is consistent with previous studies showing that the mismatch between actual working hours and preferred working hours negatively affects life satisfaction (Mark Wooden et al., 2009). In contrast, in more gender-egalitarian societies, low-income men may not be penalized for deviating from traditional gender norms, while women’s autonomy is also retained. Taken together, these arguments yield to the following moderating hypothesis:

Hypothesis 3

The linkage between low income and SIB is stronger in gender inegalitarian societies for both women and men.

3 Data and Methods

3.1 Data and Variables

We conduct country-specific and cross-national analyses using two large-scale datasets: (1) European Social Ssurvey (ESS) for over 35,000 individuals in 27 countries; and (2) the joint European Values Study-World Values Survey (EVS/WVS) for over 75,000 individuals in 48 countries. Given that the reference year of the ESS data is consistent across countries, the available latest wave of ESS is primarily used. However, considering the conventional 30/30 rule in hierarchical modeling (Bickel, 2006; Hox & McNeish, 2020; Maas & Hox, 2005),Footnote 2 the EVS/WVS dataset is also used for a robustness check.

ESS is an international biennial survey undertaken across European countries, where the nationally representative respondents are selected from the population aged 15 and above. The survey items cover people’s attitudes and behaviors related to politics, religion, ethnicity, trust, and justice among others, alongside their socio-economic attributes. Since its launch in 2002, ESS has been completed 10 times thus far, and multiple waves have been widely used in well-being research (Aassve et al., 2015; Araki, 2022; Glass et al., 2016; Van der Meer & Wielers, 2013).

While 32 countries participated in the latest 10th wave between 2020 and 2022, the dataset is still under adjustment (e.g., some new variables are added; weighting/sampling variables are changed). Meanwhile, the complete data for 29 countries are available from the ninth wave administered between 2018 and 2020, with some of fundamental contextual variables (e.g., macroeconomy, economic inequality, and gender inequality) nested to each respondent (i.e., hierarchical data) (ESS ERIC, 2021).Footnote 3 Among 29 countries, Latvia and Serbia are not included in the following analyses due to the lack of accurate data for marital status and country-level variables, respectively. See Table 1 for the target countries and the sample size. Table 12 also summarizes the list of acronyms used in the current study for reference.

Table 1 Target countries and the sample size (ESS data analysis)
Full size table

Following much research in this vein (Helliwell et al., 2023), the key outcome variable is life satisfaction (LS) measured by the 0–10 Cantril ladder scale. To examine the importance of paying attention to SIB apart from the average LS level in relation to low income, we use both the original LS score (with 10 being the best) and the three-point ordinal scale, where 1 represents “thriving” (7–10 LS points), 2 indicates “struggling” (5–6 LS points), and 3 means “suffering” (0–4 LS points), among which “suffering” represents SIB (Clifton, 2022). For robustness checks, supplementary analyses are performed with different thresholds for SIB. Because the results are largely consistent, one model with the 0–3 LS points being “suffering” is shown in Table 7. Given that the income effect on life evaluation may differ from that on emotions (Kahneman & Deaton, 2010), future research must extend the outcome variable to other dimensions of SIB.

The key predictor is household income, which is the sole income-related variable in the ninth wave ESS dataset. As demonstrated by previous studies (Oishi et al., 2011), household income enables us to analyze those individuals that are not selected into the labor market. This is particularly relevant for women who are more likely than men to be outside of the workforce. To directly capture the relative position of low-income individuals as compared to the rich, three groups are generated based on the income deciles: upper income (8th–10th); middle income (4th–7th); and lower income (1st–3rd). Given the potential bias due to this dichotomization, the deciles as such are also used for a robustness check (see Table 8 in the Appendix).

Other individual-level predictors include age (both the continuous value and its squared term), educational attainment (two dummies for lower secondary or below and upper secondary, with tertiary and post-secondary being the reference), marital status (three dummies for married/cohabited, separated/divorced, and widowed, with never married being the reference), whether having a child, trust (an 11-point scale from 0 being “You can’t be too careful.” to 10 being “Most people can be trusted.”), religiosity (an 11-point scale from 0 “Not at all religious.” to 10 “Very religious.”), declared health (two dummies for bad and good, with fair being the reference),Footnote 4 and employment (five dummies for paid work, education, unemployed looking for a job, retired, and housework, with others being the reference).Footnote 5 All of these controls have proved to be significantly linked to LS in prior research (Araki, 2022; Bartram, 2021a; Başlevent & Kirmanoğlu, 2017; Deeming, 2013; Lim & Putnam, 2010; Van der Meer, 2014). Note that, as Bartram (2021b) argued, controls should generally be limited to confounders (i.e., age, education, and employment in the current case) unless there is a specific purpose to include other variables. Nonetheless, because the results/implications that follow are consistent between models with/without other attributes, the full models are shown in the following sections.

As regards the aggregate gender inequality level, following prior research (e.g., Audette et al., 2019), the Gender Inequality Index (GII) and the Global Gender Gap Index (GGGI) are used for the main analysis and the appendix, respectively. GII is a composite measure consisting of three dimensions: health, empowerment (education and politics), and the labor market. It ranges from 0, which means women and men are equal in these domains, to 1 indicating an extremely gendered condition.Footnote 6 GGGI is also a composite index based on the gender equality level in economic, educational, health, and political domains, stretching from 0 (completely unequal) to 1 (full parity).

To better assess the moderating function of gender inequality, five societal-level traits are incorporated in accordance with recent findings in this vein as well as data availability: (1) macroeconomy (GNI per capita at current prices in 2015 in our analysis), which is considered to be the fundamental factor for both the relative function of income and the average SWB level (Easterlin, 2001; Easterlin & O’Connor, 2022); (2) economic inequality (Gini coefficient of equivalized disposable income); (3) economic freedom (a composite indicator ranging from 0 (worst) to 10 (best) as reported by Vásquez and Porčnik (2019)); (4) aggregate trust (the average trust level in each country derived from the micro data); and (5) the coverage of unemployment benefits, which would support the economically vulnerable group. All of these conditions are widely reported as potential determinants of SWB including their moderating effect on the income-SWB link (Araki, 2023b; Başlevent & Kirmanoğlu, 2017; Easterlin & O’Connor, 2022; Graafland, 2020; Helliwell & Putnam, 2004; ILO, 2021; Mikucka et al., 2017; Ono & Lee, 2013; Veenhoven, 2000).

The reference year is 2017 for GNI, Gini, and economic freedom, whereas the latest value of GII available in the ESS multilevel dataset is in 2015. Trust and the coverage of social protection range from 2017 to 2020 across countries. Given the potential bias incurred by the multicollinearity among level-two predictors and their relatively large number as compared to the country-level sample size, a series of robustness checks are conducted in tandem with the EVS/WVS data analysis (see the Appendix). See Table 2 for descriptive statistics.

Table 2 Descriptive statistics
Full size table

3.2 Analytic Model 1: Country-Specific Analysis

To examine the strength of the linkage between low income and the SIB risk, ordered logistic regression is conducted for women and men in each country with the said three-point scale as the outcome. This operationalization of using the ordinal outcome, rather than employing the “suffering” dummy only, is crucial for the current study as it captures the gradually intensified probability of falling into SIB from thriving to struggling and then suffering. The basic model is describable as follows.

$${Y}_{i}^{*}={b}_{0}+\sum_{1}^{2}{b}_{n}{I}_{i}+\sum_{3}^{19}{b}_{n}{X}_{i}$$
(1)

where i = individual, Yi* = the latent response for individual i, bn = the coefficient of predictor variables, Ii = two income dummies (lower and middle, respectively, with upper being the reference), and Xi = the vectors of other predictors including age, age squared, two education dummies, three marital status dummies, the having a child dummy, trust, religiosity, two health dummies, and five employment dummies. The observed response Yi (three categories consisting of 1 = thriving, 2 = struggling, and 3 = suffering) can be determined by thresholds (\({\tau }_{n}\)) below.

$${Y}_{i}=\left\{\begin{array}{c} 1 \quad if {Y}_{i}^{*}{\le \tau }_{1}\\ 2 \quad if {\tau }_{1}<{Y}_{i}^{*}\le {\tau }_{2}\\ 3 \quad if {\tau }_{2}<{Y}_{i}^{*}\end{array}\right.$$
(2)

The primary focus is on b1 and b2 (i.e., parameters for income dummies), and to make their magnitude comparable across countries, the average marginal effects of three income groups (i.e., the probability of falling into three outcome categories) are calculated (see Fig. 2).

Fig. 2
figure 2figure 2

The probability of suffering by income groups in 27 Countries. The dots and lines indicate the average marginal effects of three income groups and the 95% confidence intervals based on country-specific ordered logistic regressions (Tables 3 and 5). The vertical axis indicates the predicted probability of suffering

Full size image

In addition, linear regression (ordinary least squares: OLS) is performed by replacing the dependent variable with the original LS Cantril ladder scale, using the same predictors as the logistic model. This analysis is essential to confirm (1) whether low income is negatively associated with the continuous LS measure in the present dataset as reported by the literature; and (2) the extent to which the disadvantage of low income is consistent between two models with different foci. As detailed in the results section, from the comparative perspective, the relative effect of the economic disadvantage notably differs depending on the outcome variable (e.g., Country A shows the relatively strong penalty for low income when using the continuous LS score in OLS despite the limited risk of suffering in a logit model; Country B demonstrates a high probability of SIB in a logit model, but the standardized coefficient in OLS is relatively small) (see Table 3 and Fig. 3). It is therefore essential to proceed with multilevel ordered logistic regression as follows.

Table 3 Coefficients of low income in OLS of life satisfaction and ordered logistic regression of subjective ill-being in 27 countries (Extract)
Full size table
Fig. 3
figure 3

Standardized Coefficients and Marginal Effects of Low Income in 27 Countries. Countries are plotted based on two measures (see Table 1 for country abbreviations). The X axis indicates the predicted probability of suffering among the lowest income group based on the ordered logistic regression. The Y axis is the standardized coefficient of low income in OLS of life satisfaction (see Tables 3 and 5 for the original results of regressions)

Full size image

3.3 Analytic Model 2: Multilevel Analysis

Combining individual-level and societal-level data for 27 countries, Model 1 first includes income and other individual-level predictors to see the association between income groups and the probability of suffering as well as struggling and thriving.

$${Y}_{ij}^{*}={b}_{0j}+\sum_{1}^{2}{b}_{n}{I}_{ij}+\sum_{3}^{19}{b}_{n}{X}_{ij}$$
(3)

where j = country, and other components are equivalent to country-specific analysis (Eqs. (1) and (2)). To see whether the result is distorted by any influential country, a supplementary analysis is performed with country dummies before incorporating country-level variables and cross-level interactions (see Table 6).

In Model 2, country-level indicators are added to Model 1 to examine how the link between low income and SIB varies when accounting for societal conditions.

$${b}_{0j}(\text{in equation }(3))={\gamma }_{00}+{\gamma }_{01}{GII}_{j}+\sum_{2}^{6}{\gamma }_{0n}{C}_{j}+{u}_{0j}$$
(4)

where γ00 = the average intercept, γ0n = the coefficient of country-level predictors, GII = the Gender Inequality Index (see the Appendix for the model using GGGI), C = the vectors of country-level controls, and u0j = the country dependent deviation. Denoting bn by γn0 and substituting Eq. (4) into Eq. (3), Model 2 is defined as follows.

$${Y}_{ij}^{*}={\gamma }_{00}+\sum_{1}^{2}{\gamma }_{n0}{I}_{ij}+\sum_{3}^{19}{\gamma }_{n0}{X}_{ij}+{\gamma }_{01}{GII}_{j}+\sum_{2}^{6}{\gamma }_{0n}{C}_{j}+{u}_{0j}$$
(5)

Finally, Model 3 incorporates cross-level interaction terms between two income groups and the degree of gender inequality (as well as other macro-level variables). γ11 in the following equation shows how the magnitude of low income varies depending on aggregate gender parity. As with the country-specific analysis, the shifting marginal effects of income groups (i.e., the predicted probability of suffering) associated with GII are calculated and illustrated (see Fig. 4).

$${Y}_{ij}^{*}={\gamma }_{00}+\sum_{1}^{2}{\gamma }_{n0}{I}_{ij}+\sum_{3}^{19}{\gamma }_{n0}{X}_{ij}+{\gamma }_{01}{GII}_{j}+\sum_{2}^{6}{\gamma }_{0n}{C}_{j}+{\gamma }_{11}L{I}_{ij}{GII}_{j}+{\gamma }_{21}M{I}_{ij}{GII}_{j}+\sum_{2}^{6}{\gamma }_{1n}L{I}_{ij}{C}_{j}+\sum_{2}^{6}{\gamma }_{2n}M{I}_{ij}{C}_{j}+{u}_{0j}$$
(6)

where LI = the low-income dummy, MI = the middle-income dummy, and C = the vectors of country-level predictors. Given the potentially biased estimation by incorporating only random intercepts (Heisig & Schaeffer, 2019), the random slope for low income is further added as a robustness check (see Table 8). As discussed, another set of sensitivity analysis is conducted by using income deciles instead of the dichotomous categories (Table 9) and changing the threshold for suffering (Table 7). All the computations are undertaken with the variance components structure using the robust standard errors and the analysis weight (“anweight”), which incorporates all of the design weight, a post-stratification adjustment, and a population size adjustment.Footnote 7

Fig. 4
figure 4

Predicted probability of suffering by income groups across societal-level gender inequality. Lines indicate the predicted probability of suffering among three income groups (upper, middle, and lower) across levels of societal-level gender inequality (i.e., Gender Inequality Index: X axis) based on Model 3

Full size image

To confirm the robustness of the results, another supplementary analysis for Model 3 is conducted with the joint EVS/WVS data (EVS/WVS, 2021) (see Tables 10 and 11). It is noteworthy that the results are largely consistent regardless of model specification.

4 Results

4.1 Country-Specific Analysis

Table 3 extracts the key results from OLS and ordered logistic regression for women and men in 27 countries, and Fig. 2 illustrates the average marginal effects of income groups on the risk of SIB based on the nonlinear model (see Table 5 in the Appendix for the full results). Among men, 18 countries show a negative association between low income and the upward LS score in OLS at the 5% significance level, whereas the higher probability of suffering (i.e., the positive sign in logistic regression) is confirmed in 17 cases. Meanwhile, low-income women are likely to report lower LS scores in 14 countries, and they face a greater risk of suffering in 19 societies. Accordingly, the disadvantage of low income is observed regardless of gender and the outcome variable (i.e., the continuous LS measure or the three-point SIB scale) in almost half of the target countries. These results generally support Hypothesis 1.

For example, in Croatia (HR), the coefficient of the low-income dummy for LS is negative and statistically significant (B = − 0.596, 95%CI: − 1.139 to − 0.054, P < 0.031 for men; B = − 0.605, 95%CI: − 1.114 to − 0.096, P = 0.020 for women). This means the economically disadvantaged tend to report lower LS scores, corroborating prior research. Likewise, low income shows a substantially positive sign (i.e., greater risk) in the nonlinear model (B = 0.789, 95%CI: 0.197 to 1.380, P = 0.009 for men; B = 0.774, 95%CI: 0.245 to 1.304, P = 0.004 for women). Figure 2 (HR) indeed indicates the higher chance of suffering (i.e., low LS) among low-income people as compared to their affluent counterparts. Other countries with a similar structure (i.e., disadvantage of low-income women and men in both models) include Czechia, Estonia, Finland, France, Germany, Hungary, Slovenia, Sweden, and the United Kingdom.

While these results suggest that the conventional approach focused on the upward LS score suffices in understanding the link between (low) income and SWB/SIB. Nonetheless, some countries indicate a substantial disadvantage of low income only in a linear model without an explicit sign of greater risks in a logit model, and vice versa. In Austria (AT), for instance, the low-income dummy does not demonstrate a significant coefficient despite its negative sign for women in OLS (B = − 0.223, 95%CI: − 0.546 to 0.099, P = 0.175), implying that women with lower/higher income do not explicitly possess lower/higher LS in a continuous term. However, in the logit model, the substantial association between the lowest income group and the risk of SIB is confirmed (B = 0.771, 95%CI: 0.121 to 1.421, P = 0.020). Figure 2 (AT) clearly illustrates the higher probability of suffering among economically disadvantaged women and men. This suggests Austria somehow manages to mitigate income-based inequality among women as far as the average LS score is concerned, but the lowest income tier still faces a greater risk of SIB as compared to higher income groups.

In contrast, OLS indicates that low-income men in Denmark (DK) are likely to possess lower LS (B = − 0.327, 95%CI: − 0.657 to 0.004, P = 0.052), whereas the significant sign of the income-based disadvantage is not confirmed in the logit model despite its positive coefficient (B = 0.612, 95%CI: − 0.605–1.829, P = 0.324). That is, although affluence matters in obtaining higher LS on average, low income is not necessarily penalized in terms of SIB. Indeed, Fig. 2 (DK) shows the risk of suffering among low-income individuals is as low as their wealthier counterparts, with the probability ranging from 1.5% (upper income) to 3% (lower income). It is therefore critical to pay close attention to the extent to which the economically disadvantaged are likely to fall into SIB, alongside the general tendency of the income-LS association. Figure 3 illustrates these cross-national variations in terms of the standardized coefficients and marginal effects of low income in OLS and the logit model, respectively.

Another important perspective here is the gender difference. In such countries as Bulgaria (BG), Norway (NO), and Poland (PL), it is only low-income men who are more likely than their high-income counterparts to face a greater risk of SIB in the logistic regression (e.g., B = 1.085, 95%CI: 0.600–1.570, P < 0.001 for men; B = 0.167, 95%CI: − 0.279 to 0.614, P = 0.463 for women in Bulgaria). Meanwhile, the significant association between low income and SIB is detected solely among women in Island (IS), Lithuania (LT), and Spain (ES) (e.g., B = 0.345, 95%CI: − 0.306 to 0.995, P = 0.299 for men; B = 0.832, 95%CI: 0.094 to 1.570, P = 0.027 for women in Spain). Nonetheless, one should note that this gender gap in the significance of the low-income dummy merely reflects the relative position of low-income individuals as compared to the affluent within each gender group. It is therefore still possible that the predicted probability of suffering is in the same range between women and men even in these countries. Indeed, Fig. 2 displays the identical marginal effect of low income in both gender groups in Bulgaria (approximately 0.45), Norway (0.05), Island (0.04), and Spain (0.08). Meanwhile, the relationship between low income and SIB is stronger among men than women in countries like Lithuania (LT) and Slovenia (SI), whereas the opposite structure (i.e., the larger penalty for low-income women) is observed in France (FR) and Montenegro (ME).Footnote 8 Hypothesis 2 is therefore mostly rejected with the exception of limited cases (e.g., LT and SI).

Given such a stratified structure of economic disadvantage and SIB, an essential question is how this heterogeneity is associated with societal-level gender inequality and other conditions. Multilevel analyses thus become the key as follows.

4.2 Multilevel Analysis

Table 4 summarizes the results of multilevel ordered logistic regressions of SIB. In Model 1 with only individual-level predictors, the low-income dummy demonstrates a positive and statistically significant coefficient for both women and men. This association between low income and SIB holds when (1) including country dummies (Table 6) and (2) controlling for country-level variables (Model 2) (B = 0.911, 95%CI: 0.736 to 1.087, P < 0.001 for men; B = 0.782, 95%CI: 0.551 to 1.013, P < 0.001 for women). These results suggest, as with country-specific analyses, low-income individuals are more likely than their economically advantaged counterparts to suffer as a general tendency.Footnote 9

Table 4 Multilevel ordered logistic regression of subjective ill-being
Full size table

However, this structure is not static; rather, it varies depending on societal contexts as seen in Model 3 with cross-level interactions. For men, the interaction term between low income and the degree of gender inequality (GII) shows the positive and statistically significant sign (B = 4.111, 95%CI: 0.211 to 8.012, P = 0.039).Footnote 10 This means that the risk of SIB among the poor is likely to be higher than the rich in societies with stronger gender inequality. Indeed, Fig. 4 (Men) indicates that the predicted probability of suffering is around 5–7% across three income groups when GII is 0, but it notably varies in accordance with GII (e.g., approximately 12.5%, 24.9%, and 40.0% for upper, middle, and lower income groups in a society where GII is 0.3). This suggests, in tandem with a higher level of gender disparity, the have-nots are penalized more significantly compared to the affluent (although the latter group also faces a higher SIB risk in more gender inegalitarian circumstances).Footnote 11

As regards women, a slightly puzzling result is observed for the interaction between GII and low income: a negative interactional effect with middle income at the conventional level (B = − 5.230, 95%CI: − 10.167 to − 0.293, P = 0.038). This indicates that the likelihood of suffering among the middle-income group is lower as compared to the most affluent group in line with higher levels of gender inequality. However, GII’s main effect is substantially large (B = 9.609, 95%CI: 4.581 to 14.638, P < 0.001), meaning that people are more likely to report SIB in gender inegalitarian societies regardless of income level. Nonetheless, because the slope of such tendencies is steep among the wealthiest (i.e., high-income women are psychologically penalized more significantly in tandem with higher GII), the said negative interactional effect is confirmed among the middle tier (i.e., the probability of suffering does not decrease along with gender inequality even for middle-income people). This structure is clearly illustrated in Fig. 4 (Women).Footnote 12

Thus, along with the significant link between low income and SIB, its elasticity associated with societal-level gender inequality is confirmed, such that Hypothesis 3 is largely supported despite the relatively smaller penalty for low-income women than their affluent counterparts. These results are confirmed when (1) employing a different threshold for SIB (Table 7), (2) including the random slope (Table 8), and (3) using the income deciles as a continuous measure instead of the three categories (Table 9). Furthermore, an analysis of the joint EVS/WVS data largely supports the aforementioned findings (Table 10).

5 Discussion

Scholars have long investigated whether money can buy SWB. Much evidence suggests, despite some heterogeneity across gender and other attributes, the rich are more likely than the poor to enjoy better SWB (Clark et al., 2008; Diener et al., 2010a, 2010b; Kahneman et al., 2006). While such a positive income-SWB link has been widely reported, we know little about (1) the extent to which low-income women and men encounter a higher risk of SIB; and (2) how the magnitude of such penalties for the poor differs depending on societal-level gender inequality. Examining this knowledge gap from a comparative perspective is pivotal to better understand the nuanced structure of economic and socio-psychological disparities.

Using the large-scale international survey data, country-specific analyses first confirm the significant linkage between low income and a higher risk of suffering among women and men in most cases. This suggests that what we understand from an extensive body of linear-model analyses of the positive income-SWB association is generally applicable as the opposite edge of the scale, if not completely symmetrically, to explain the bottom ranges of income and well-being (i.e., the SIB risk among the poor). Exceptions are Denmark and Slovakia, where the lowest income tier does not explicitly face a greater probability of SIB for both gender groups. Importantly, despite the said applicability of the conventional approach focused on the average SWB score, the low income-SIB link is not identical to the favorable income-SWB relationship. For example, low-income women in Austria are not less satisfied with their life on average, but they are more likely than those with higher income to be suffering. In contrast, low-income men in Denmark report averagely lower LS, but they do not see a significant risk of SIB. Shedding light on the association between low income and SIB is therefore critical in this vein.

Moreover, these findings emphasize the conceptual and analytical contribution of distinguishing between SWB and SIB. In response to the pioneering conceptualizations in positive psychology, there has been debate about whether SWB and SIB are two independent factors (Headey et al., 1984) or two sides of a continuum (Wood & Joseph, 2010; Sirgy et al., 2019; Zhao & Tay, 2023). We join the predominant approach of polar opposites but with an important nuance that integrates these two conceptualizations. In its original version, the two-dimensional perspective posited that SWB and SIB have different correlates and causes. For instance, while SWB is strongly predicted by social relationships, SIB is predicted by health conditions (Headey et al., 1985). In our case, we show that the two sides of the continuum are asymmetrically linked to one of their correlates (i.e., income), thus conceptually bridging both perspectives.

Multilevel analyses also show that low income is substantially associated with the higher risk of suffering even when accounting for country-level variables. Nevertheless, alongside the observed direct link between gender parity and SIB, the suffering risk among low-income individuals markedly varies depending on this societal trait (i.e., the moderating effect of gender inequality).

Among men, the disadvantage of low-income individuals is likely to be larger in a society with a higher-level GII. Although further investigations are necessary to establish causality, this exacerbated risk is interpretable as reflecting gender norms. That is, stronger gender inequality (i.e., higher GII) may intensify psychological pressure and stress against men as a breadwinner as examined by a vast literature (King et al., 2020; Preisner et al., 2020). Consequently, those with low income are likely to face stigmas and a greater chance of suffering from SIB. Although their higher-income counterparts also see an intensified SIB risk in a more gender inegalitarian setting, the penalty is particularly large for the low-income tier. For women, the link between the economic disadvantage and SIB is also stronger in countries with greater gender inequality. Although the cross-level interaction between low income and GII shows a negative sign, this is not because the poor’s SIB risk declines but because the upper-income tier concurrently encounters the larger risk of suffering. Put differently, regardless of income strata, women are likely to report unfavorable psychological well-being in gender inegalitarian societies.

More robust analyses are necessary to be conclusive. Nonetheless, based on the said results, one may argue that promoting gender parity would potentially help protect economically disadvantaged people from falling into SIB. One should particularly note that gender inequality could punish low-income men arguably because of persistent gender norms that require a male breadwinner model, which is often obscured by analysis of the linear income-SWB relationship. That is, so long as we focus on the upward trend of income and better SWB, the nuanced socio–psychological penalties for economically disadvantaged men in gendered societies are hardly detectable. It is therefore imperative both substantively and methodologically to pay close attention to how low income and ill-being are associated in certain societal contexts. This also suggests from a policy perspective that fostering gender equality could be beneficial not only for women in general but also for less resourceful men who may suffer from the discourse of masculinity and relevant stigmas incurred by the gender role conflict (O’Neil, 2014).

To advance our understanding of this matter, future research must address several issues. First, longitudinal analyses with panel data are necessary to see how people’s SWB/SIB shifts over their life course and societal transformations. Examining the cases that manage to recover from suffering, alongside its experience per se, should also provide insights into the mechanism of SIB, the function of income, and effective measures to support vulnerable people. Second, the scope of analyses and variables should be extended. Although the primary focus of the current paper is on low income, the association between SIB and other socio-economic disadvantages (e.g., gender/sexual and racial/ethnic minorities, low educational attainment, precarious work or unemployed, and untraditional union formation) is worthy of investigation. Comparing the structure between income-based penalties and other factors, one may better understand how SIB is stratified among individuals in a gendered way. The outcome variable can also be stretched to other dimensions of SWB/SIB. Third, as partly demonstrated in a supplementary analysis (Tables 10 and 11), low- and middle-income countries should be included. It is particularly important to incorporate the Global South, given that prior research (Jebb et al., 2018) shows that cross-regional analysis advances our understanding of SWB/SIB stratification. This may eventually lead to new typologies of societies.

With this potential for further inquiry, the present study provides a foundation to advance scholarship on SIB, as well as SWB, with attention to disadvantaged people. Future research in this vein is encouraged to shed more light on the structure surrounding bottom ranges of economic, social, and psychological statuses. We argue that this line of studies would eventually better inform social policy and contribute to realizing human flourishing and more equitable societies.