National Societal, Economic, and Religious Factors Underlying Perceived Happiness

Abstract

Various explanatory factors for perceived happiness have been identified in cross-country comparisons, but their relative importance and causal relationships remain unknown. Associations and causality were studied between 17 societal, economic, and religious variables contributing to happiness scores on the World Happiness Report (WHR) in 147 countries. In a correlation analysis, the strongest association with national happiness scores was observed for the median gross domestic product (GDP)/capita (r squared 0.69), followed by the mean GDP/capita (0.65), tertiary education (0.42), corruption (−0.41), democracy index (0.35), religious non-affiliation (0.31), and consanguinity (−0.25). Bayesian statistical analysis indicated strong causal cascades originating from democracy (a positive effect) and, to a lesser degree, corruption (a negative effect) during the years 1945–1985 to subsequent national happiness levels, with a delay of several decades. The results were confirmed to have good reproducibility with Bootstrap resampling, indicating robust findings. Although the median GDP/capita had the strongest association with happiness in the cross-sectional analysis, it was not identified as a primary causal factor in the Bayesian analysis, in which high levels of democracy and low levels of corruption were entangled with median GDP/capita as apparent prerequisites for perceived happiness. Our results suggest that the median GDP should be used as an indicator of the economic prosperity of citizens instead of the currently established mean GDP when studying the links between economy, democracy, and well-being across countries.

1. Introduction

Since time immemorial, humankind has striven for happiness and well-being. Definitions of happiness have differed from the time of the ancient Greeks to the medieval Christian world, the modern era, and the present. Individuals’ struggle for happiness in particular has increased in postmodern society, where they need to form a basis for and determine the meaning of their own lives themselves.

Various studies have measured happiness in different ways. Happiness is a crucial human disposition that denotes a construct characterizing positive feelings and overall well-being in individuals. Happiness can be perceived as a relatively stable state of overall well-being linked to diverse facets of an individual’s life, distinguished from more transient emotions like joy or sadness [1]. There are several potential pathways leading to increasing well-being and happiness in human societies. The widely acknowledged Welzel and Inglehart human development sequence framework suggests that in poor and underdeveloped societies, survival is a priority, ensured by strict control and order [2]. When living conditions improve, individuals can start to pursue non-materialistic goals, such as independence, self-expression, and freedom of choice. Higher income, and better education and healthcare enable empowerment of people, and when emancipated individuals value greater autonomy, societies develop towards democracy. However, since economic development and democratic rights are entangled and interact with each other, it is difficult to differentiate if economic growth predominantly leads to democratization or vice versa.

The World Happiness Report (WHR) calculates the perceived happiness of citizens of each country and has become the established indicator of comparative well-being in over 140 countries. Because it is the most widely referenced dataset for country-specific perceived happiness, it was used in this study. The WHR uses the Cantril Ladder as its measurement method [3]: individuals evaluate both their life and life satisfaction on a scale from 0 to 10, with 0 referring to the worst possible life and 10 to the best possible life. Preceding research has proposed factors underlying happiness scores in different countries. Happiness has mostly been explained by economic, social, and health-related factors, personal rights, and freedom, as well as low levels of corruption. These studies [3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22] are summarized in Table 1. Only five studies included more than 100 countries, which may have been the reason for many contradictory findings between studies, such as a positive [3,9,14,17,18,22], negative [5,8,19], or neutral [4] associations between mean gross domestic product (GDP) per capita. These results possibly depend on which particular countries were included, since the mean GDP per capita does not reflect the income level of ordinary people if the wealth is concentrated to few individuals of the elite. The most consistent finding has been the association between low corruption level and happiness [3,4,7,9,16,18,21,22]. In the WHR, variance in national happiness is primarily explained by social support from friends and family, followed by the log mean gross domestic product (GDP) per capita, health and life expectancy, the freedom to make life choices, generosity, and perceptions of corruption. One of the proposed explanations for Nordic countries scoring highest in the World Happiness Report has been mutual trust in other people and authorities [23,24,25,26]. Concerning these variables, the first comprehensive data were published in 2006 with the Gallup World Poll on social support conducted by the WHO in 2000, regarding healthy life expectancy, and in 2010 with the Gallup World Poll on generosity. Since the aim of our study was to investigate long-term causal cascades originating more than 30 years prior to the outcome, these variables were not included in our analyses. Religions have been found to have positive [4,12,13,18], negative [10], or mixed [6,11] associations with perceived happiness. Therefore, we also included religions as variables in our study. No previous studies have investigated the comparative importance of the contributing factors, including the median income instead of mean income per capita, or used time-series data to investigate the causal temporal evolution in perceived happiness.

Table 1. Review of the literature.

Study	Variables	Findings
Mookerjee & Beron, 2005 [4] 60 countries	Gini coefficient, corruption, economic freedom, religion, and civil liberty.	Religion and gender have an important link to the happiness levels of countries. Income levels are not related to happiness levels in rich or poor countries.
Di Tella & MacCulloch, 2006 [5] 37 countries	GDP per capita, real annual income per capita, and real household income.	Happiness economics and method of measurement are not sufficient to improve welfare or happiness.
Bjørnskov et al., 2008 [6] 70 countries	GDP per capita, religion, education, overall governance, climate, and Gini coefficient.	Overall governance and policies affect life satisfaction negatively. Of all religions, Buddhism is the most detrimental to life satisfaction. A bicameral political system, communist past, share of Protestants and Roman Catholics in population, number of years of country’s independence, and infant mortality are the most important factors associated with life satisfaction.
Helliwell & Huang, 2008 [7] 75 countries	GDP per capita, government quality, religion, corruption, civil and political rights and liberties, age of democracy, and general trust.	The quality of a country’s government has a greater effect on life satisfaction than mere economic factors. General trust in society and other sociocultural, sociopolitical, and socio-economic factors contribute to life satisfaction with governance and personal income.
Easterlin et al., 2010 [8] 37 countries	GDP and financial satisfaction.	A higher rate of economic growth does not lead to an increase in happiness in rich or poor countries in the long run.
Kahneman & Deaton, 2010 [9] 150 countries	Marital status, religiosity, student status, health conditions, intoxicants, gender, and income.	Wealth does not increase well-being and happiness after a certain point has been reached.
Heukamp & Ariño, 2011 [10] 64 countries	Education, age, marital status, GDP per capita, Gini coefficient, population, suicide, abortion, and religion.	Lack of corruption increases subjective well-being. Climate also has an impact on people’s well-being. Muslim countries are the unhappiest countries in the world.
Brulé & Veenhoven, 2014 [11] 49 countries	Social, economic, actual and potential freedom, and civil liberties.	Roman Catholicism is associated with a hierarchical society that discourages personal and other types of freedom, which does not increase perceived happiness in the same way as Protestantism.
Florea & Caudill, 2014 [12] 5 countries	Religion, income, employment status, education, age, gender, and marital status.	Religion increased people’s well-being after the transition from a communist system to capitalist economical system.
Ngamaba & Soni, 2017 [13] 100 countries	Protestantism, Roman Catholicism, Eastern Orthodoxy, Buddhism, Judaism, Islam, small religions and irreligion, GDP, and Gini coefficient.	Perceived happiness and well-being were most influenced by the country’s economic and cultural landscape. Protestants, Roman Catholics, and Buddhists were the happiest of all religious groups. Eastern Orthodox Christians were the most unhappy denomination.
Jebb et al., 2018 [14] 164 countries	Questionnaires for individuals, GDP, income, life evaluation, and subjective well-being.	When income reaches a certain level, individuals’ well-being and happiness do not continue to increase; 60.000–90.000 USD is the limit for maximum happiness.
Dominguez & Lopez-Noval, 2020 [15] 77–101 countries	Religiosity/religion, GDP per capita, health, collective social capital, freedom, and social support.	Only the socio-institutional aspect of religion affects well-being considerably. An average level of religiosity has a positive link to mediocre life satisfaction.
Li & An, 2020 [16] 126 countries	GDP per capita, unemployment rate, religious and ethnic fractionalization, and corruption.	Corruption decreases both subjective well-being and the country’s well-being. It also influences well-being via freedom of religion, freedom of expression, freedom of assembly and association, and equal opportunities. In high-income and democratic countries, corruption has a significant (negative) impact on happiness, while in low-income countries, this effect does not exist or is not significant.
Muresan et al., 2020 [17] 26 countries	Income, unemployment, education, health, social status, and culture.	Happiness increases with individual income until a threshold of 27,913 Euros per year (rounded to 35,000 USD) in European countries. Smaller power gaps, high individualism, a low level of uncertainty avoidance, and high indulgence statistically increase the level of happiness.
Hamzaoğlu, 2022 [18] 60 countries	Religiosity, importance of god, trust, financial satisfaction, life satisfaction, education, corruption, and freedom of choice.	Health, security, commitment to religion, financial satisfaction, freedom of choice, income, and living in a secure and democratic environment have positive and significant impacts on happiness. Perceived corruption is negatively correlated with happiness.
Kudrna & Kushlev, 2022 [19] 2 countries	Income, employment status, marital status, and health.	Continuous treatment of income showed a null relationship with happiness, whereas the categorization of income showed that some of those with higher incomes reported feeling less happy than some of those with lower incomes. Lowess and spline regressions suggested null results overall, and there was no evidence of a relationship between income and happiness.
Sánchez-Rodríguez, Ángel et al., 2023 [20] 71 countries	Gini index, national income, gender, age, perceived meritocracy and economic inequality, and current well-being.	Perceived economic inequality predicts poor personal well-being. National income moderates the effects of happiness, meaning, harmony, and religion.
Araki, 2024 [21] 152 countries	Freedom of choice, perception of corruption, GDP per capita, and social support.	Social support is substantially linked to both happiness and unhappiness worldwide. Freedom of choice predicts a lower risk of suffering and higher chances of thriving and positive affect, especially in advanced economies. Corruption negatively predicts happiness, particularly in the richest countries.
Behera et al., 2024 [22] 166 countries	GDP per capita, perception of corruption, freedom of choice, and social support.	Per capita income, social support, and freedom to make life choices positively impact happiness, while air pollution exposure has a negative impact. Happiness enhances during the initial phase of economic growth in a country but stays stable or starts to decrease after reaching a certain threshold.
World Happiness Report (WHR) 2025 [23] 147 countries	Social support, GDP/capita, political and cultural values, trust, healthy life expectancy, corruption, and generosity.	The happiest people are altruistic and benevolent. Social support and mean GDP/capita had the strongest association with perceived happiness. Low level of corruption and high level of democracy were associated with perceived happiness.

Several cross-sectional happiness studies have examined the correlation of explanatory variables and happiness, but causality cannot be analyzed by investigating associations only. To the best of our knowledge, no previous study has attempted to examine the causal cascade underlying differences in comparative happiness among countries. Therefore, in this study, we used Bayesian causal inference to discover which variables contribute most significantly to national happiness levels. Ultimately, the variables identified in the aforementioned 20 studies were included, plus one additional variable, the median GDP per capita. This variable was included because the mean GDP does not reflect the income level of common people if the income per capita has a skewed distribution. In contrast, the Gini coefficient shows the relative distribution of income but not the absolute level of median income. Therefore, median GDP per capita can be considered a good indicator of the materialistic prerequisite for empowerment in the aforementioned Welzel–Inglehart human development sequence framework. No a priori hypothesis was set concerning the comparative importance of the variables for perceived happiness, i.e., we had a hypothesis that any variable could be the most important factor associated with the WHR happiness outcome, and the comparative contributions and interactions of all specific variables were studied in the same analysis. Therefore, the issue of multiple comparisons was addressed with appropriate statistical methods.

2. Methods

2.1. Data Sources

To facilitate time-series analysis with a reasonable amount of data, information on the variables was gathered at 5-year intervals. Data were retrieved from the following public databases: country happiness rankings from the World Happiness Reports 2012 [27], 2018 [28], and 2025 [3]; religion percentages from the Pew Research Center [29]; annual Gini index data from GapMinder [30]; median GDP and mean GDP per capita from Our World in Data [31]; democracy indexes from Our World in Data [32]; timelines of world religions from Schultz et al. [33]; corruption indexes from Our World in Data [34]; education statistics from BarroLeeData (the latest data was available from the year 2010 [35]); the number of Christian denominations (countries classified as Protestant, Orthodox, or Roman Catholic) from the U.S. Department of State Office of International Religious Freedom [36]; and consanguinity statistics from World Population Review [37]. Each annual World Happiness Report includes data from three previous years; therefore, the 2025 report covers the years 2022, 2023, and 2024 retrospectively but not the year 2025 itself.

The associations between the newest variables available for the Gini index; median GDP/capita; mean GDP/capita; corruption; democracy index; primary education; tertiary education; consanguinity; percentages of Christians, Buddhists, Hindus, Muslims, and non-affiliated individuals; and the religion timeline (duration of practiced religion in each country) and WHR 2025 results were investigated using Spearman correlation analysis with SPSS version 29.0.2.0. The level of significance was set at p = 0.05:14 = 0.00357 with Bonferroni correction. The causality between all time-series variables was then studied using Bayesian statistics.

2.2. Bayesian Analysis

2.2.1. Purpose and Principles

A Bayesian analytical framework [38,39] was employed to examine how countries’ religious background and contemporary socio-economic status relate to national happiness levels, measured with the WHR Happiness Index. Bayesian analysis (BA)—implemented through Bayesian networks (BNs)—provides a powerful and flexible approach to investigating complex systems, especially when data are limited, uncertain, or contain missing values. Unlike traditional frequentist methods, BA can effectively handle non-linear relationships and small sample sizes by incorporating prior knowledge and uncertainty directly into the modeling process [40,41,42,43,44].

A key strength of BA lies in its ability to incorporate causal assumptions from observational data, offering a structured framework for reasoning about potential cause–effect relationships [45,46]. This makes BA particularly suitable for social and political research, where data are often observational and confounding factors are difficult to control. Consequently, Bayesian approaches have become increasingly popular in empirical political science and the social sciences as they allow researchers to address data challenges in a more transparent and principled manner than competing methods [47,48]. Although temporal ordering rules out reverse causation, residual bias due to unmeasured factors cannot be eliminated even in the Bayesian approach.

BNs are represented as Directed Acyclic Graphs (DAGs) composed of nodes that represent random variables and arcs that denote conditional dependencies between them. Each node is associated with a conditional probability distribution, quantifying the strength and direction of dependencies between connected variables. The directionality of arcs—from “parent” to “child” nodes—enables the modeling of both causal and non-causal (predictive or explanatory) relationships, depending on the research context.

Moreover, BNs allow researchers to control or manipulate variables within the model to test the impact of specific factors while holding others constant. By setting particular variable states or combinations, it is possible to explore “what-if” scenarios—a valuable feature for assessing the potential influence of religion-related factors on socio-economic outcomes [40,41,42,46,47].

2.2.2. Implementation Steps

The first step in the analysis involved constructing a fully connected network of manifest (observed) variables using the Maximum Spanning Tree algorithm with the Minimum Description Length (MDL) score. The Maximum Spanning Tree (MST) algorithm was used only as an initial exploratory step to identify groups of strongly associated variables for subsequent cluster construction. Since the objective at this stage was not to infer the final BN structure but to reveal the strongest dependency patterns among variables, the default BayesiaLab parameter settings were considered appropriate. The resulting MST served solely as the basis for defining clusters, after which the latent variable model was estimated independently. Variables Hap12, Hap18, and Hap25 (“Hap” refers to happiness, and the numbers refer to the years 2012, 2018, and 2025) were excluded at this stage to ensure they remained outside the clusters. After validating the clusters through multiple clustering procedures in Probabilistic Structural Equation Modeling (PSEM), a set of latent variables representing each cluster was obtained, with a purity of approximately 85% [38,49].

PSEM in BayesiaLab extends traditional Structural Equation Modeling (SEM) and factor analysis into a fully probabilistic framework. In this approach, relationships among variables—both manifest (observed) and latent (unobserved)—are expressed as conditional probabilities rather than fixed linear equations. Leveraging BayesiaLab’s clustering and structure learning algorithms, latent variables are automatically inferred as hidden nodes that group together religious and socio-economic characteristics of countries exhibiting similar patterns. These latent nodes function analogously to factors in conventional factor analysis, representing unobserved dimensions that account for shared variance among observed indicators [38,50].

In the next phase, Hap12, Hap18, and Hap25 were reintroduced into the model, and the latent variables were temporally ordered using Temporal Indexes (TIs). The latent variables were then classified into seven time-based groups, beginning with religious background factors, to reflect temporal causality. The outcome variable, Hap2025, represented the final observed state. Each latent variable was defined according to the manifest variables exerting the strongest influence upon it [38].

Finally, an unsupervised learning algorithm, Taboo Order, was applied to construct a causal BN that identified dependencies among latent variables and their connections to the outcome. A Taboo Order search was performed with a list size of 10 and a post-processing list size of 71, parameters that control the number of candidate solutions retained during the main search and the subsequent refinement stage, respectively. Larger values enable a broader exploration of the search space at the expense of increased computation time. No forbidden or forced arcs were in use.

In the graphical representation, all manifest (observed) variables were integrated with the latent variables. The resulting model is visualized as a Directed Acyclic Graph (DAG), where both manifest and latent variables are interconnected and converge on the outcome variable—the Happiness Index (2025). We analyzed the amount of mutual information shared between the explanatory factors and the target node, as well as total effects on the target (Happiness 2025).

A robustness assessment of the arcs connecting the remaining variables through the implementation of the Bootstrap resampling method was carried out within BayesiaLab version 11.7.4 (Bayesia, Changé, France, 2023). Bootstrapping was performed using the Taboo Order algorithm, using a list size of 10 and 100 tests. This assessment exclusively incorporated arcs that were consistently present across all networks derived from the Bootstrap resampling, thereby ensuring that a 70% presence criterion was met.

3. Results

3.1. Correlation Analysis

Table 2 shows the Spearman correlations between the latest national happiness scores and 14 explanatory continuous variables. The three remaining categorical variables related to subtypes of Christianity (Catholic, Protestant, and Orthodox) were included in the BA only. The strongest positive correlation with the Happiness 2025 score was observed for the median GDP/capita (r squared 0.69), followed by the mean GDP/capita, (0.65), tertiary education 2010 (0.42), democracy (0.35), belonging to an Unaffiliated religious group (0.31), and Christianity (0.08). Alternatively, a negative association was observed for corruption 2023 (−0.41), consanguinity (−0.25), Islam (−0.15), and the Gini index 2025 score (−0.08). Religion timeline, Buddhism, Hinduism, and primary education each had weaker associations (r-squared values of less than 0.05), and their associations with happiness were not statistically significant. Strong collinearity (r > 0.8) was observed between the mean and median GDP/capita (r = 0.85), and the median GDP/capita had the strongest association with most positive outcomes. Concerning democracy, the r-squared was 0.58 (r = 0.76, 95% CI 0.66–0.83), while it was only 0.23 for the mean GDP/capita (r = 0.48, 95% CI 0.35–0.59). The correlation between median GDP and happiness is shown in Figure 1. Among countries with a median GDP/capita above 30.000 USD, 27 of 27 (100%) had a national happiness score above the median happiness score (5.868), but among those countries with a median GDP/capita below 10.000 USD, only 4 of 31 (13%) countries had a happiness score above the median happiness score. In the partial correlation analysis including the variables with the strongest associations with Happiness 2025, median GDP/capita retained a moderate correlation with the Happiness 2025 score (r = 0.36, p < 0.001, df = 83, r² = 0.13), with an r-squared value of 0.13 when controlling for democracy, corruption, and tertiary education. In contrast, democracy (r = 0.16, p = 0.13, df = 83, r² = 0.03), tertiary education (r = 0.16, p = 0.14, df = 83, r² = 0.03), and corruption (r = −0.02, p = 0.87, df = 83, r² = 0.00) had negligible correlations with happiness when the other three variables were controlled for (controlling for the median GDP/capita had the strongest diluting effect).

Table 2. Associations between explanatory variables and the WHR 2025 happiness score. Spearman correlations between the WHR happiness score and 14 explanatory variables. The level of statistical significance with Bonferroni correction is set at p < 0.00357. Gin 25 = Gini index 2025; Median 24 = median GDP/capita 2024; GDP/cap24 = mean GDP/capita 2024; Cor 23 = corruption 2023; Demo 24 = democracy 2024; Edu Prim 10 = primary education 2010; Edu Tert 10 = tertiary education 2010; Hap 25 = Happiness 2025; Consang = consanguinity; Christ perc = percentage of Christians; Buddh perc = percentage of Buddhists; Hindu perc = percentage of Hindus; Unaffiliat perc = percentage of Unaffiliated individuals; Islam perc = percentage of Muslims; and Relig tim = religion timeline.

Spearman’s Rho		Correlations
		Gini 25	Median 24	Gdp Cap 24	Cor 23	Demo 24	Edu Prim 10	Edu Tert 10	Hap 25	Consang	Christ Perc	Buddh Perc	Hindu Perc	Unaffiliat Perc	Islam Perc	Relig Tim
Gini 25	Correlation Coefficient	1.000	−0.472 **	−0.369 **	0.301 **	−0.228 **	0.363 **	−0.467 **	−0.281 **	0.078	0.201 *	−0.037	−0.008	−0.094	−0.164 *	−0.450 **
	Sig. (2-tailed)		<0.001	<0.001	<0.001	0.004	<0.001	<0.001	<0.001	0.517	0.011	0.640	0.918	0.240	0.039	<0.001
	N	159	99	154	157	157	136	136	144	72	159	159	159	159	159	159
Median 24	Correlation Coefficient	−0.472 **	1.000	0.853 **	−0.770 **	0.759 **	−0.338 **	0.731 **	0.833 **	−0.470 **	0.137	0.068	−0.044	0.625 **	−0.315 **	0.397 **
	Sig. (2-tailed)	<0.001		<0.001	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001	0.173	0.504	0.667	<0.001	0.001	<0.001
	N	99	100	98	100	100	92	92	94	44	100	100	100	100	100	100
Gdp cap 24	Correlation Coefficient	−0.369 **	0.853 **	1.000	−0.668 **	0.480 **	−0.231 **	0.698 **	0.808 **	−0.311 **	0.164 *	0.192 *	0.193 *	0.436 **	−0.243 **	0.288 **
	Sig. (2-tailed)	<0.001	<0.001		<0.001	<0.001	0.008	<0.001	<0.001	0.010	0.040	0.016	0.016	<0.001	0.002	<0.001
	N	154	98	157	154	154	133	133	144	68	157	157	157	157	157	157
Cor 23	Correlation Coefficient	0.301 **	−0.770 **	−0.668 **	1.000	−0.708 **	0.263 **	−0.543 **	−0.644 **	0.439 **	−0.180 *	−0.125	−0.128	−0.522 **	0.333 **	−0.174 *
	Sig. (2-tailed)	<0.001	<0.001	<0.001		<0.001	0.002	<0.001	<0.001	<0.001	0.023	0.114	0.107	<0.001	<0.001	0.028
	N	157	100	154	160	160	137	137	145	70	160	160	160	160	160	160
Demo 24	Correlation Coefficient	−0.228 **	0.759 **	0.480 **	−0.708 **	1.000	−0.144	0.444 **	0.590 **	−0.611 **	0.375 **	−0.108	−0.060	0.520 **	−0.434 **	0.003
	Sig. (2-tailed)	0.004	<0.001	<0.001	<0.001		0.094	<0.001	<0.001	<0.001	<0.001	0.173	0.455	<0.001	<0.001	0.968
	N	157	100	154	160	160	137	137	145	70	160	160	160	160	160	160
Edu Prim 10	Correlation Coefficient	0.363 **	−0.338 **	−0.231 **	0.263 **	−0.144	1.000	−0.334 **	−0.162	0.332 **	0.028	−0.030	−0.001	−0.260 **	0.013	−0.084
	Sig. (2-tailed)	<0.001	<0.001	0.008	0.002	0.094		<0.001	0.067	0.007	0.742	0.731	0.994	0.002	0.875	0.328
	N	136	92	133	137	137	138	138	129	65	138	138	138	138	138	138
Edu Tert 10	Correlation Coefficient	−0.467 **	0.731 **	0.698 **	−0.543 **	0.444 **	−0.334 **	1.000	0.650 **	−0.442 **	0.080	0.172 *	0.071	0.465 **	−0.166	0.348 **
	Sig. (2-tailed)	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001		<0.001	<0.001	0.350	0.043	0.407	<0.001	0.052	<0.001
	N	136	92	133	137	137	138	138	129	65	138	138	138	138	138	138
Hap 25	Correlation Coefficient	−0.281 **	0.833 **	0.808 **	−0.644 **	0.590 **	−0.162	0.650 **	1.000	−0.502 **	0.288 **	0.101	0.072	0.556 **	−0.393 **	0.191 **
	Sig. (2-tailed)	<0.001	<0.001	<0.001	<0.001	<0.001	0.067	<0.001		<0.001	<0.001	0.225	0.383	<0.001	<0.001	0.020
	N	144	94	144	145	145	129	129	147	66	147	147	147	147	147	147
Consang	Correlation Coefficient	0.078	−0.470 **	−0.311 **	0.439 **	−0.611 **	0.332 **	−0.442 **	−0.502 **	1.000	−0.645 **	0.181	0.346 **	−0.733 **	0.726 **	0.245 **
	Sig. (2-tailed)	0.517	<0.001	0.010	<0.001	<0.001	0.007	<0.001	<0.001		<0.001	0.126	0.003	<0.001	<0.001	0.037
	N	72	44	68	70	70	65	65	66	73	73	73	73	73	73	73
Christ perc	Correlation Coefficient	0.201 *	0.137	0.164 *	−0.180 *	0.375 **	0.028	0.080	0.288 **	−0.645 **	1.000	−0.318 **	−0.294 **	0.349 **	−0.661 **	−0.307 **
	Sig. (2-tailed)	0.011	0.173	0.040	0.023	<0.001	0.742	0.350	<0.001	<0.001		<0.001	<0.001	<0.001	<0.001	<0.001
	N	159	100	157	160	160	138	138	147	73	163	163	163	163	163	163
Buddh perc	Correlation Coefficient	−0.037	0.068	0.192 *	−0.125	−0.108	−0.030	0.172 *	0.101	0.181	−0.318 **	1.000	0.407 **	0.073	−0.092	0.133
	Sig. (2-tailed)	0.640	0.504	0.016	0.114	0.173	0.731	0.043	0.225	0.126	<0.001		<0.001	0.355	0.241	0.089
	N	159	100	157	160	160	138	138	147	73	163	163	163	163	163	163
Hindu perc	Correlation Coefficient	−0.008	−0.044	0.193	−0.128	−0.060	−0.001	0.071	0.072	0.346 **	−0.294 **	0.407 **	1.000	−0.172 *	0.184 *	0.138
	Sig. (2-tailed)	0.918	0.667	0.016	0.107	0.455	0.994	0.407	0.383	0.003	<0.001	<0.001		0.028	0.018	0.079
	N	159	100	157	160	160	138	138	147	73	163	163	163	163	163	163
Unaffilat perc	Correlation Coefficient	−0.094	0.625 **	0.436 **	−0.522 **	0.520 **	−0.260 **	0.465 **	0.556 **	−0.733 **	0.349 **	0.073	−0.172 *	1.000	−0.624 **	−0.126
	Sig. (2-tailed)	0.240	<0.001	<0.001	<0.001	<0.001	0.002	<0.001	<0.001	<0.001	<0.001	0.355	0.028		<0.001	0.109
	N	159	100	157	160	160	138	138	147	73	163	163	163	163	163	163
Islam perc	Correlation Coefficient	−0.164 *	−0.315 **	−0.243 **	0.333 **	−0.434 **	0.013	−0.166	−0.393 **	0.726 **	−0.661 **	−0.092	0.184 *	−0.624 **	1.000	0.226 **
	Sig. (2-tailed)	0.039	0.001	0.002	<0.001	<0.001	0.875	0.052	<0.001	<0.001	<0.001	0.241	0.018	<0.001		0.004
	N	159	100	157	160	160	138	138	147	73	163	163	163	163	163	163
Relig tim	Correlation Coefficient	−0.450 **	0.397 **	0.288 **	−0.174 *	0.003	−0.084	0.348 **	0.191 *	0.245 *	−0.307 **	0.133	0.138	−0.126	0.226 **	1.000
	Sig. (2-tailed)	<0.001	<0.001	<0.001	0.028	0.968	0.328	<0.001	0.020	0.037	<0.001	0.089	0.079	0.109	0.004
	N	159	100	157	160	160	138	138	147	73	163	163	163	163	163	163

** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

Figure 1. The association between the national median GDP/capita 2024 (in US dollars) and the WHR 2025 happiness score (scale 0–10). Correlation between median GDP/capita 2024 and the WHR happiness score 2025 (Spearman r = 0.833, p < 0.001, N = 94). Among countries with a median GDP/capita above 30.000 USD, 27 of 27 (100%) had a national happiness score above the median happiness score (5.868), but among those countries with a median GDP/capita below 10.000 USD, only 4 of 31 (13%) countries had a happiness score above the median happiness score.

3.2. Bayesian Analysis

Descriptions of the discretization of variables and the content of clusters in the BA are provided in Supplementary Tables S1 and S2. Because a limited number of countries were included, rough discretization of continuous variables (low vs. moderate vs. high) was employed instead of a more granular analysis. Numerical variables were discretized into three categories using the R²-GenOpt algorithm in the BayesiaLab package. Continuous variables were discretized using the BayesiaLab R²-GenOpt algorithm, which automatically determines cut-points that maximize the amount of information retained from the original continuous variable with respect to the target variable, thereby minimizing the information loss caused by discretization. The Bayesian results for the comparative contributions (standardized total effect on the target) of explanatory factors to the 2025 national happiness scores, indicated by the size of each factor, are shown in Figure 2. This figure is based on a Bootstrap analysis, in which only the causal links replicated in more than 70% of iterations are shown. The sole causal link to happiness was observed from the democracy cluster for 1995–2024 to the happiness score of 2012, which was subsequently linked to the happiness score of 2025. While the causality between the 1995–2024 democracy cluster and happiness in 2012–2025 is slightly spurious due to partial temporal overlapping, the causal cascades leading to the democracy 1995–2024 cluster are more explicit. This cluster receives input from the democracy 1945–1985 cluster through the corruption 1945–1985 and 1995–2023 clusters (democracy and corruption antagonizing each other). The democracy 1945–1985 cluster is also entangled with cluster F10 (related to religions) and the primary education 1955–1995 cluster, which subsequently contributes directly and via the tertiary education loop to the median GDP/capita 1985–2024 cluster and subsequently to the democracy 1995–2024 cluster, which impacts happiness outcomes. The most robust corresponding numerical results of the analysis are shown in Table 3. Among clusters, the greatest standardized effects on Happiness 2025 were observed for cluster F9 (democracy 1995–2024), cluster F2 (median GDP/capita 1985–2024), and cluster F3 (democracy 1945–1985). Concerning single factors, the strongest standardized effects on Happiness 2025 were observed for democracy 2024, 2015, 1995, and 2005 (positive effects), followed by corruption 2005, 2015, and 1995 (negative effects) and median GDP/capita 2005 (a positive effect). Among the single factors within each cluster preceding the 2025 happiness outcome at least by 5 years, the most important ones were democracy 2015 in cluster F9 (standardized total effect on Happiness 2025 0.26), corruption 2005 in cluster F6 (−0.19), and median GDP/capita 2005 in cluster F2 (0.17). Within the democracy 1995–2024 cluster, even the earliest factors in the years 1995 (0.24) and 2005 (0.24) made substantially greater contributions to Happiness 2025 than any factors in other clusters. The results for mutual shared information with the target node (Happiness 2025) are reported in Supplementary Table S3, showing the highest values of normalized mutual information. The highest relative significance was observed for cluster 9; democracy 2024, 2015, 2005, and 1995; cluster 6; and corruption 2005. These findings were consistent with those on the total standardized effect, as shown in Figure 2 and Table 3. The variable abbreviations are defined in Supplementary Table S4.

Figure 2. The causal networks contributing to the WHR 2025 happiness score. The relative importance of factors contributing to the WHR happiness scores 2012–2025 are indicated by the size and color of each factor. Size represents the importance of the factor in the entire model and color in the pairwise effect on the target (Hap25). Node colors indicate the variables’ total effect on the outcome variable, including both direct and indirect effects. Green signifies the highest impact, yellow indicates intermediate effects, and red signifies the lowest impact. The blue lines (arcs) between nodes depict the relationships between variables, with line thickness corresponding to Kullback–Leibler’s divergence. Only the Bayesian causal links surviving 70% replication in Bootstrap resampling analysis are shown. The democracy cluster 1945–1985 (a positive effect), corruption clusters 1945–1985 and 1995–2015 (negative effects), cluster F10 (a negative effect), primary education 1955–1985 and tertiary education 1975–2010 clusters (positive effects), and median GDP/capita cluster 1985–2024 (positive effect) contribute to happiness outcomes via the democracy 1995–2005 cluster. The greatest standardized total effects on Happiness 2025 were observed for the democracy 1995–2024 (F9), median GDP/capita 1995–2024 (F2), and democracy 1945–1985 (F3) clusters. The detailed numerical results are shown in Table 2. Node abbreviations are as follows: Hap for happiness, Demo for democracy, Cor for corruption, Unaffiliated perc for the percentage of Unaffiliated individuals, Median for the median GDP, EduTert for tertiary education, EduPrim for primary education, Gini for the Gini index, Hindu perc for the percentage of Hindus, Orth for Eastern Orthodox, Religtim for religion timeline, Consang for consanguinity, Cath for Roman Catholic, Prot for Protestantism, Othr for religions other than Christian denominations, Christ perc for the percentage of Christians, Relig for religion, Buddh perc for the percentage of Buddhists, and Islam perc for the percentage of Muslims.

Table 3. Numerical results for standardized total effects on the target (WHR Happiness 2025 score). The unit in the analysis of the total effect on the target indicates the amount of change as standard deviations (SDs) in the target, corresponding to change of one SD in an explanatory factor. Abbreviations of nodes are Demo for democracy, Cor for corruption, Median for median GDP/capita, and EduPrim for primary education; F9 is the cluster consisting of Demo1995–2024, F2 consists of Median1985–2024, F8 consists of Cor1945–1985, F3 consists of Demo1945–1985, and F4 consists of EduPrim1955–1995.

Node	Standardized Total Effect	G-Test (Data)	df (Data)	p-Value (Data)
Hap 2018	0.6317	91.4401	4	0.0000
Hap 2012	0.6287	90.8285	4	0.0000
F9	0.2891	58.6134	6	0.0000
Demo 2024	0.2871	52.9800	4	0.0000
Demo 2015	0.2590	35.2340	4	0.0000
Demo 1995	0.2397	35.1269	4	0.0000
Demo 2005	0.2394	30.1580	4	0.0000
Cor 2005	−0.1881	32.6645	4	0.0000
Cor 2015	−0.1828	33.4501	4	0.0000
F2	0.1813	47.4530	4	0.0000
Cor 1995	−0.1781	32.2926	4	0.0000
Median 2005	0.1712	35.0983	4	0.0000
Cor 2023	−0.1674	26.1613	4	0.0000
Median 2015	0.1638	48.6543	4	0.0000
Median 2024	0.1631	43.3545	4	0.0000
Cor 1985	−0.1422	16.1968	4	0.0028
F3	0.1416	32.5929	6	0.0000
Cor 1975	−0.1388	7.0972	4	0.1308
Demo 1975	0.1366	33.2514	4	0.0000
Median 1995	0.1339	31.6493	4	0.0000
Edu Prim 1965	0.1338	51.1415	4	0.0000
F4	0.1333	52.6966	10	0.0000
Demo 1965	0.1326	39.7038	4	0.0000
F8	−0.1302	32.5244	10	0.0003
Demo 1955	0.1287	39.2683	4	0.0000

4. Discussion

In a cross-sectional analysis, the strongest positive correlations with happiness scores were observed for the median GDP, mean GDP, tertiary education, and democracy. The Unaffiliated and Christianity variables were also associated with happiness in a positive way. On the contrary, high corruption, consanguinity, and Muslim religion had a negative correlation with happiness levels. Our results on the mean GDP per capita, democracy, Christianity, and Unaffiliated religion/world view, as well as the corruption and Muslim variables, align with those of previous studies [7,10,13,51]. However, previous studies did not report that the median GDP per capita has the strongest association with happiness. As shown in Figure 1, the association was somewhat non-linear: a steep correlation showing that the national happiness is low if the median GDP/capita is below 10.000 USD. Since the Gini coefficient had only a weak correlation with happiness, the results suggest that the absolute income level, rather than equally distributed poverty, is important. In addition to having a greater overall impact on happiness than the mean GDP per capita, the median GDP also showed substantially stronger associations with a high democracy index and low corruption level than any other variable, including the mean GDP. The causality between prosperity and democracy has been studied by several researchers, including the recent Nobel laureate Daron Acemoğlu, who showed with his collaborators that extractive institutions prevent economic development [52]. Countries with this type of institution do not respect political and civil rights such as property rights, which is typical for undemocratic countries with excessive corruption. In these studies, the mean GDP per capita has been used as the indicator of national income levels instead of the median GDP per capita. Our results show that the mean GDP explained only 23% of the variance in democracy, while the median GDP explained 58% of the variance. This may be referred to as the “Qatar effect”, revealing that a high mean GDP per capita does not necessarily mean a high median GDP per capita as well. This indicates that both average economic well-being and within-population distribution of income is important. Our results suggest that increasing the median GDP per capita in countries where it is below 10.000 USD would have the greatest beneficial effect on happiness, as can be seen from the non-linear relationship shown in Figure 1. Our results suggest that the median GDP per capita should be used as the indicator of the economic prosperity of ordinary citizens instead of the currently established mean GDP per capita when studying the links between economy, democracy, and well-being across countries. Our results suggest that median GDP per capita is a suitable indicator of the materialistic prerequisite for empowerment in the Welzel–Inglehart human development sequence framework.

Since correlation does not imply causation, our aim was also to observe causal links using BN analysis. To the best of our knowledge, this is the first cross-country causal model study of the factors affecting perceived happiness. We observed causal links from the democracy index cluster 1945–1985 to subsequent national democracy, corruption, and happiness levels 50 to 80 years later. The robustness of the results was confirmed with Bootstrap resampling analysis at a 70% reproducibility level. This suggests that the causal effect of democracy on national happiness levels is observed with a delay of up to two generations. Our cross-sectional analysis was well in line with the WHR reports showing that national log mean GDP/capita plays a crucial role in perceived happiness. Although the median GDP/capita had the strongest association with happiness in the cross-sectional analysis, it did not appear as a primary causal factor in the BA. As seen in Figure 2, the democracy cluster 1945–1985 (F3) contributed via education clusters to median GDP cluster F2, which subsequently contributed to democracy cluster F9, which further contributed to happiness. This implies that economy and democracy are entangled, with continuous interaction with each other. Therefore, our results only partly support the Welzel–Inglehart human development sequence framework, which suggests a predominantly one-way causal pathway from economic development to democracy via emerged emancipatory values. Our results suggest that both cultural (mass values following increasing safety and economic growth) and institutional (democratic institutions enhancing economic growth) are equally plausible explanations for welfare and happiness.

There are multiple possible explanations for the scores achieved by the happiest nations in the world. The happiest countries according to the WHR are mostly parliamentary democracies where women’s and minorities’ rights and equality are upheld and citizens enjoy a high degree of other liberties. Many of the happiest listed countries were also Protestant countries, which were leaders in the industrial revolution [53]. It has also been speculated that growing respect for women and children during the early modern period contributed to the rise of natural science in Europe [54] and, subsequently, prosperity and happiness in the Western world. The most democratic nations can be found in Europe, where democracy first flourished in ancient Greece and Rome, Venice, and the Netherlands in the early modern era; a long history of democracy seems to have paved the way for functional democracies which, in turn, have promoted happiness and prosperity.

The world’s happiest countries share two worldview or religious variables: Unaffiliated and Christianity. This is supported by World Happiness Report results showing that the top-scoring nations are secular societies in which the state is separated from both institutional religion and religion altogether. In these countries, citizens may or may not hold an institutional or individualistic worldview of their own, which is in line with the idea of empowerment and emancipation in the Welzel–Inglehart framework [2]. Protestant Christianity was associated with the happiest and wealthiest countries. This was in line with the results of previous studies, which have indicated low corruption rates as a possible way to explain the happiness rankings [55,56]. The individualistic and non-hierarchical nature of Protestantism has facilitated individuals’ own activity and economic freedom, with less involvement on the behalf of government in comparison to nations with more hierarchical religions [33]. It has even been suggested that Western Christianity has bolstered the progress of its surrounding society by prohibiting cousin marriages and inhibiting the formation of tribal and clan societies, thereby increasing mutual trust within large and diverse populations [33]. However, it must be acknowledged that although most of the happiest nations were Northern European Protestant Christian countries, many African countries with a Christian majority were among the least happy. This underlines the role of other factors, such as colonial legacy and ethnic fragmentation; these factors may also be important for countries with Muslim, Hindu, and Buddhist majorities.

Our results suggest that there is a minor causal cascade from a high national proportion of Muslim citizens to low democracy indexes and low national happiness scores; the unhappiest countries in the world, such as Afghanistan, Syria, and Somalia, are Muslim-majority nations. This may be explained by corruption [55,56], religious hierarchy [33], a close bond between the state and religion [56], authoritarian forms of government, Islam’s problematic relationship to the concepts of democracy and rights [51,57,58,59,60], and stagnation, especially in regard to economics, science, and freedom of thought [61]. Since the proportion of Muslims in the world populations has changed very little during preceding decades, reverse causality between Islam and national happiness is unlikely [62]. However, this may be the case concerning the relationship between religiously Unaffiliated individuals and happiness because the proportion of people belonging to Unaffiliated groups has increased over recent decades, especially in high-income countries with high happiness scores [62]. Political turmoil, conflicts, and their effects over the past few decades may have also affected the happiness scores of Muslim countries. However, wars and civil wars are largely attributable to a lack of democracy, as democratic countries rarely start wars [63]. In addition to Islam, many Muslim-majority countries and their surrounding regions have a large variety of both religious minorities and ethnic groups that have tightly knit, clan-based communities and links to consanguinity. However, we acknowledge that our results showed that neither Muslim, Christian, or any other religious affiliation (non-affiliated) made any major contribution to happiness in the Bayesian causal analysis, and in the cross-sectional analysis, only religious non-affiliation had a substantial association with happiness.

The low levels of both happiness and democracy and the high corruption scores of the unhappiest countries may also be explained by different cultural factors (such as tolerance to corruption and a long history of authoritarian governance) and their status as former colonies. The exploitation of former colonies by colonial powers has also hampered the development of the former. However, it can be argued that previous colonial status does not explain all poor scores for the unhappiest countries, given that nations such as Liberia, Ethiopia, Afghanistan, and Iran were able to retain their independence during the colonial period.

5. Strengths and Limitations

The primary aim of our analysis, the standardized total effect on the target, is considered a good measure of causality, but it may overlook subtle effects. Therefore, we also conducted shared mutual information analysis, which is more sensitive but implies associations more so than causality. We observed fairly consistent results between the analyses, indicating that we had not missed important connections between explanatory factors and happiness. There were only 147 countries in the analysis, which forced us to use a rough three-grade discretization (low vs. moderate vs. high) of the continuous variables, since a more granular discretization of countries would have made subgroups too small for the BA. Discretization inevitably involves some loss of information, and alternative approaches are available. However, discretization is a well-established practice in BN modeling, particularly when the objective is to identify probabilistic dependency structures rather than to estimate continuous functional relationships. In this study, the variables were discretized into three states using the R²-GenOpt algorithm implemented in BayesiaLab, which optimizes the predictive information retained with respect to the target variable. The objective of the present study was to describe the structure of dependency among the variables rather than to model precise continuous response functions. Future work will extend this framework using causal BNs and intervention-based analyses.

The strength of our study was that it was free of a specific hypothesis, making it possible to study all potential variables in one analysis. The downside of this approach was that we had to use strict statistical procedures to control for type 1 errors by using Bonferroni correction and Bootstrap resampling. This may have led to a type 2 error, i.e., only the most robust results remained statistically significant, and weaker signals may have remained undetected. Due to the number of countries and limited amount of data, only the strongest phenomena were identified. For instance, nothing may be stated about the reliability of any potential effects of Hinduism because the study included only one country with a Hindu majority. Differentiating the subgroups of Christianity, Islam, and Buddhism was not possible because it would have led to overwhelming complexity in the analyses. Therefore, the aggregation of these subgroups may have diluted the findings if the different subgroups had opposite types of associations with happiness. These limitations of our study may have led to an inability to detect subtle causal cascades between variables. Our study utilized 17 different variables, and some of the data were insufficient and did not encompass every single country. Therefore, the results may be partly explained by hidden unknown factors.

6. Conclusions

Our results suggest that median GDP should be utilized as the indicator of the economic prosperity of citizens instead of the currently established mean GDP when studying the links between economy, democracy, and well-being across countries. In light of this study’s results, it would be beneficial to invest in fostering democracy and combatting corruption. These two factors do not completely explain happiness levels on their own, but they form a basis to increase perceived happiness. Increasing democracy and reducing corruption may be difficult in the least happy countries, as many have harsh authoritarian governments, and dictatorship is typically based on a lack of democratic rights and corruption, with advantages for civil servants and the army. However, governments in semi-authoritarian countries might be willing to promote anti-corruption efforts to increase the well-being of their citizens.